Bridging the GAP: using social media optimally in public health efforts & the example of Utah during the covid-19 pandemic

In many ways, the COVID-19 pandemic is a unique time in history. Since cases were first reported in China in December 2019 and the early months of 2020, over 217 million cases have been reported globally with over 4.5 million deaths, totals that continue to rise as the pandemic progresses. The United States has contributed to a large portion of these totals, with over 39 million cases and 639,000 deaths. In addition to the widespread effects that permeate every aspect of modern life, COVID-19 is the first large-scale global pandemic that has occurred since the advent of modern technology, including communication technology, health IT on both the hospital and consumer sides, and technology for the analysis of large datasets. The optimal use of this technology is important in a public health setting under normal conditions and vital in a pandemic response. This study will focus on a subset of this modern technology and seek to understand the uses and effectiveness of social media technology in response to the COVID-19 pandemic. It will first establish context using perspectives from the fields of epidemiology and public health and disseminate findings from previous researchers, with specific attention given to studies of the COVID-19 pandemic. It will then draw on psychological research to understand how social media users process information and engage with the platform. By focusing on the pandemic in Utah, it will then survey the ways in which government and public health officials have used Twitter to communicate with the public and promote positive health outcomes. Finally, this study will synthesize these findings and make recommendations for better utilizing social media technology for the purposes of public health promotion in the unique conditions a global pandemic presents.

ABSTRACT In many ways, the COVID-19 pandemic is a unique time in history. Since cases were first reported in China in December 2019 and the early months of 2020, over 217 million cases have been reported globally with over 4.5 million deaths, totals that continue to rise as the pandemic progresses. The United States has contributed to a large portion of these totals, with over 39 million cases and 639,000 deaths. In addition to the widespread effects that permeate every aspect of modern life, COVID-19 is the first large-scale global pandemic that has occurred since the advent of modern technology, including communication technology, health IT on both the hospital and consumer sides, and technology for the analysis of large datasets. The optimal use of this technology is important in a public health setting under normal conditions and vital in a pandemic response. This study will focus on a subset of this modern technology and seek to understand the uses and effectiveness of social media technology in response to the COVID-19 pandemic. It will first establish context using perspectives from the fields of epidemiology and public health and disseminate findings from previous researchers, with specific attention given to studies of the COVID-19 pandemic. It will then draw on psychological research to understand how social media users process information and engage with the platform. By focusing on the pandemic in Utah, it will then survey the ways in which government and public health officials have used Twitter to communicate with the public and promote positive health outcomes. Finally, this study will synthesize these findings and make recommendations for better utilizing social media technology for the purposes of public health promotion in the unique conditions a global pandemic presents. ii TABLE OF CONTENTS ABSTRACT________________________________________________ ii INTRODUCTION___________________________________________ 1 1918 INFLUENZA PANDEMIC 3 CORONAVIRUSES 8 COVID-19 & SOCIAL EPIDEMIOLOGY_______________________ 13 ETIOLOGY: THE ORIGINS OF THE NOVEL CORONAVIRUS 13 RISK FACTORS 20 MEASURING COVID-19 IN POPULATIONS 23 COVID-19 NATURAL HISTORY & PROGNOSIS 33 SYMPTOMS OF COVID-19 44 MILD & SEVERE SYMPTOMATIC CASES OF COVID-19 45 ASYMPTOMATIC CASES: THE TWO-EDGED SWORD 49 THE EMERGENCE OF LONG COVID 53 METHODS OF PREVENTION & CARE DELIVERY 57 RECOMMENDATIONS FOR PUBLIC POLICY 64 WHAT MAKES THE COVID-19 PANDEMIC DIFFERENT _______ 72 COVID-19 ON TWITTER: WHAT RESEARCH HAS FOUND SO FAR 73 PSYCHOLOGICAL CONCEPTS RELEVANT TO THE PROCESSING OF INFORMATION_______________________ 77 ATTENTION 80 ANCHORING 82 CONFIRMATION BIAS 83 PERCEIVED CONFIDENCE 84 MENTAL MODELS 87 PROJECTION & PREDICTION 91 REACTANCE 94 A NOTE ABOUT ADOLESCENT PSYCHOLOGY 98 RISK PERCEPTION 104 AN OVERARCHING MODEL OF PERSUASION 127 MODERN COMMUNICATION TECHNOLOGY IN PUBLIC HEALTH ________________________________________ 133 SOCIAL MEDIA 133 DIVING INTO SOCIAL MEDIA DATA 143 NETWORK EFFECTS 144 CONTEXTUALIZED DATA 149 iii ANALYSIS OF TWEETS: GOVERNMENT AND PUBLIC HEALTH IN UTAH ____________________________ 156 ANALYSIS OF TWEET CONTENT 157 ANALYSIS OF TWEET SENTIMENT: GENERAL COVID-19 PANDEMIC TERMS 172 ANALYSIS OF TWEET SENTIMENT: COVID-19 PROTECTIVE BEHAVIORS 185 ANALYSIS OF ENGAGEMENT 192 UNDERSTANDING COMMENTS 209 USE OF HASHTAGS 217 RETWEETS AND SHARING BEHAVIORS 227 LIKES, FAVORITES, AND REACTIONS 243 USE OF MEDIA 253 CONCLUSIONS AND PROPOSITIONS_________________________261 FINAL THOUGHTS 289 REFERENCES ______________________________________________ 291 APPENDIX A _______________________________________________ 314 APPENDIX B_______________________________________________ 317 APPENDIX C _______________________________________________ 319 iv 1 INTRODUCTION When the novel coronavirus SARS-CoV-2 began to spread around the globe in late 2019 and early 2020, it marked the beginning of the largest pandemic in over 100 years. However, infectious disease has repeatedly left its mark on human history, with the coronavirus pandemic simply being the most recent instance. Pradhan et al. provide a brief history of viral infections: Viral infections [have remained] a critical issue since the year 1000 when the poxvirus spread through a wide region of China. The large death report [of] poxvirus infection is often linked to the black death bubonic plague, which [killed] 75-200 million people in the European continent during the period 13471351. In the year 1901 yellow fever caused by Flavivirus, a filterable agent reported as the first human infected virus responsible for more than 50 million people. [Later], in the year 1918-1920 pandemic, Spanish flu (H1N1)...affect[ed] millions of people...worldwide. (Pradhan et al., 2020, 363). The first reports of what is now called COVID-19 were released in December 2019 with cases initially characterized in both news media and academic sources as pneumonia. As Zhu et al. describe, local health facilities began to report “clusters of patients with pneumonia of unknown cause that were epidemiologically linked to a seafood and wet animal wholesale market in Wuhan, Hubei Province, China” (Zhu et al., 2020, 727). Zhu et al. provide further detail into some of the earliest cases: Three adult patients presented with severe pneumonia and were admitted to a hospital in Wuhan on December 27, 2019. Patient 1 was a 49-year-old woman, Patient 2 was a 61-year-old man, and Patient 3 was a 32-year-old man...Patient 1 reported having no underlying chronic medical conditions but reported fever (temperature, 37° C to 38° C) and cough with chest discomfort on December 23, 2019. Four days after the onset of illness, her cough and chest discomfort had worsened, but the fever was reduced; a diagnosis of pneumonia was based on computed tomographic (CT) scan. Her occupation was retailer in the seafood wholesale market. Patient 2 initially reported fever and cough on December 20, 2019; respiratory distress developed 7 days after the onset of illness and worsened over the next 2 days, at which time mechanical ventilation was started. He had been a frequent visitor to the seafood wholesale market. Patients 1 and 3 recovered and were discharged from the hospital on January 16, 2020. Patient 2 2 died on January 9, 2020. No biopsy specimens were obtained. (Zhu et al., 2020, 729). The following discussion will focus on comparisons between COVID-19 and the 1918 flu pandemic as well as other recent coronavirus outbreaks. 1918 INFLUENZA PANDEMIC The 1918 flu pandemic consisted of three waves: “In spring 1918, the first case of the H1N1 virus appeared in Kansas of the United States after military personnel showed symptoms of fever, cough and headaches” (Agrawal et al., 2020, 129). Some researchers have argued that “the first wave was equivalent to the seasonal outbreaks of influenza. [However,] estimates reveal that the virus strain which mushroomed around the world underwent mutation” (Agrawal et al., 2020, 131). As such, “the second wave was marked by high virulence and fatality. This highly lethal wave emerged in an Army Training Camp near Boston...and fall 1918 led to [the] emergence of many infection cases and deaths, eventually leading to the shortage of nurses in the United States. The third wave began in winter and spring of the year 1919, which steadily receded in the summer of the same year” (Agrawal et al., 2020, 129). Overall, the 1918 flu pandemic “was responsible for 45-50 million deaths across the globe, which accounts for 2-3% [of] the total population, including 675,000 Americans” (Agrawal et al., 2020, 130). One important characteristic of the flu in 1918 is its deviation from typical age-group mortality trends seen in other strains. “Conventionally, the graph of other strains of influenza viruses follows a ‘U-shape’ curve affecting the age groups <20 and >40 years, whereas the H1N1 strain exhibited [a] ‘W-shape’ curve with an extra peak of mortality in 20-40 year age groups” (Agrawal et al., 2020, 130). While multiple hypotheses have sought to explain 3 this phenomenon, “one of the[se] theories supporting such age-specific deaths...proposes that the influenza virus A H1N1 possesses antigenically shifted HA gene, which is novel and the population lacks immunity to it. The low susceptibility of the older age group may be due to the already existing antibod[ies] against the H3 influenza A virus [the virus responsible for the 1889-91 epidemic] protected them to a certain extent” (Agrawal et al., 2020, 130). Javelle & Raoult argue that “if introduced in our contemporary world, the influenza strain that caused the 1918 Spanish flu would probably not repeat the same scenario. Autopsy series have revealed that the majority of deaths at any age resulted from bacterial coinfections with common species of the upper respiratory tract, including Streptococcus pneumoniae. Thus, antibacterial interventions and pneumococcal immunization have substantially reduced morbidity of flu and probably explain the attenuation of influenza pandemic excess mortality during the past century” (Javelle & Raoult, 2021, 78). By contrast, “it appears that..severe and fatal SARS-CoV infections do not result from the combined occurrence of viral and bacterial pneumonia but are [instead] due to a secondary vascular and inflammatory disease in which immune responses dysregulation and host factors have a role” (Javelle & Raoult, 2021, 78). As the flu pandemic of 1918-1920 is the most recent worldwide pandemic prior to COVID-19, many researchers have drawn parallels between these situations, though there are also important differences. Despite the fact that “the H1N1 virus and SARS-CoV-2 accountable for the death-dealing pandemics of 1918 and 2019 respectively differ in their genomic organization and pathogenicity...the viruses share a similar epidemic orbit” (Agrawal et al., 2020, 131). First, though perhaps a bit obvious, the extent of disease spread and devastation caused in both pandemics cannot be ignored. At the time of this 4 writing, COVID-19 has infected over 217 million people and directly caused more than 4.5 million deaths worldwide (World Health Organization, 2021). Countries such as Italy, Spain, the United States, and most recently India have had particular difficulties controlling the spread of the virus, with each country becoming an “epicenter” of the pandemic on at least one occasion. The United States alone has seen 39 million cases of COVID-19 with over 639,000 deaths thus far (World Health Organization, 2021). The degree of devastation in both pandemics is due, in large part, to the fact that “Spanish Influenza virus and SARS-CoV-2 are novel viruses owing to their new strain and lack of immunity in humans against them” (Agrawal et al., 2020, 132). Nichols et al. provide further detail: In the United States, induction camps, cramped quarters, wartime transport [during World War I], and industry generated optimal conditions [for] the flu’s transmission. Around the world, global interconnection had reached an apex in world history such that the flu was able to reach much of the world in a scant four months and to circumnavigate the globe within a year...The flu undermined the war effort and the economy. It strained hospitals to and beyond their breaking points. It disproportionately infected and killed young people between age 18 and 45 years of age” (Nichols et al., 2020, 642). There are many statements in this summary that may easily apply to the COVID-19 pandemic. Though luckily world war has not been a driving force behind the spread of SARS-CoV-2, international connection in other ways, particularly travel, has been, with some of the earliest cases outside of mainland China being documented on cruise ships. Travel-related quarantine measures were implemented in the early months of 2020 in an attempt to stop the spread of the virus, such as one cruise ship that quarantined 3,700 people in Japan for over 14 days or 195 Americans who were required to quarantine on a military base in Riverside, California after leaving Wuhan, China in January (Stieg, 2020; Jordan, 2020). As was the case in 1918, SARS-CoV-2 spread quickly across the globe, 5 and perhaps even faster than the flu pandemic because of the increasing degree of global interconnection, though debates remain exactly when the virus first emerged and infected humans at the end of 2020, an issue that will be discussed later. Many consequences of the 1918 flu pandemic have also occurred in the COVID-19 pandemic, including harming national and international economies as well as straining healthcare systems to and beyond their breaking points. Early reports in the United States of patients experiencing unusual symptoms were also similar in the two pandemics, with the difference here being the origin of those cases. Because public health is the primary focus of this paper, it must be noted that while the channels public health officials use to communicate with the American public have changed greatly, the messages themselves have not seen such change. As one researcher comparing the two pandemics notes, “even the daily reports [are] similar; the number of cases and deaths; dire warnings from public health officials; painful stories of lives lost; heroic actions by healthcare workers; promises of a vaccine; and an eagerness for any sign that the number had plateaued and would start to decrease” (Nichols et al., 2020, 653). The advice of public health officials was also ignored in both pandemics, with some resistance, importantly, coming from the top of the United States government, which “has only exacerbated…[the pandemics]. Woodrow Wilson [in 1918] allowed the Fourth Liberty Loan drive to go forward and did not halt troop movements, both facilitating the spread of the disease. Perhaps most shocking, he never addressed the American people publicly about the pandemic, offering neither guidance nor sympathy” (Nichols et al., 2020, 655). The second wave of the 1918 flu pandemic was particularly devastating, which was fueled, in part, by the behavior of individuals ignoring public health recommendations. Widespread disregard of “doctors and health care officials 6 [who] raised the alarm about...gatherings, asking people to please avoid crowding into trams and joining parades” arose after the end of WWI in November 1918, with “people in many countries turn[ing] out in large numbers to celebrate the armistice” (Nichols et al., 2020, 657). In the case of the COVID-19 pandemic, resistance to the recommendations has taken additional forms, though the failure of effective presidential leadership is one important similarity. Since reports of the virus first emerged, “[Then]President Trump...offered repeated commentary on the pandemic, but much of it has been misguided or divisive. He has confused public understanding...with his frequent misinformation and politicized basic public health protections and practices with his refusal to wear a mask and his tweets calling for the liberation of states under stay-athome orders. One of the president’s many misstatements [was] his repeated claim that the nation was well prepared for, and successfully handling, the pandemic. In 1918, too, Americans were overly optimistic and inadequately prepared, and as a result were shocked by the pandemic and the chaos and suffering it unleashed” (Nichols et al., 2020, 655). In 1918, part of this shock came from the fact that early on, even “public health experts...failed to anticipate the scope of the disease, [and instead] reassured the public with unjustified optimism, and pursued policies based on hope rather than evidence” (Nichols et al., 2020, 654). However, perhaps “the most striking similarity [is] the introduction of measures such as closing schools, prohibiting public assemblies, restrictions on business, and recommendations regarding masks. Whereas early in Covid19, the historical examples from 1918 appeared as a worst-case situation, after more than two months of social distancing, the response to Covid-19 has actually gone much further 7 than was the case in 1918” (Nichols et al., 2020, 653). Indeed, “the advice and policies recommended in 1918 also contained the core of what was recommended in spring 2020: wash your hands, cover your cough, stay away from people, stay home when possible, and seek medical attention at the earliest sign of certain symptoms” though the technology and methods used to implement these measures has changed (Nichols et al., 2020, 654). Importantly, as E. Thomas Ewing notes, the essential messages “[were] present in 1918 and 2020; the challenge is getting people to pay attention, modify their behavior, and be consistent in their actions” (emphasis added), goals which will be the focus of this paper (Nichols et al., 2020, 654). CORONAVIRUSES While referred to colloquially as ‘coronavirus,’ epidemiologically, SARS-CoV-2 is not the only coronavirus, nor is it the only coronavirus to spread and cause concern in the 21st century. At present, there are four general types of coronaviruses: “Alphacoronavirus, Betacoronavirus, Gammacoronavirus, and Deltacoronavirus. These genera are divided by phylogenetic clustering but can easily be identified by their main reservoir host. Alpha- and Betacoronaviruses are known to infect mammalian species, Gammacoronaviruses are known to infect avian species, while Deltacoronaviruses are known to infect both” (Chandra & Chandra, 2020, 466). SARS-CoV-2 “became the seventh member of the coronavirus family,” though only SARS, MERS, and COVID-19 have caused major concern, while “other known human coronaviruses such as HCoV229E (betacoronavirus 1) and HCoV-OC43 (betacoronavirus), are vaguely related to SARS [and] both viruses are primarily associated with the common cold and other minor illnesses” (Afshar et al., 2020, 196; Chandra & Chandra, 2020, 465). Wang et al. provide 8 the following overview of coronavirus outbreaks: Although Coronaviruses (CoVs) have been known for decades, they did not raise great attention in human medicine until the outbreaks of Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS). SARS-CoV first emerged in November 2002 in Guandong province of Southern China and then rapidly spread to 29 countries and regions, infecting over 8000 individuals with a death toll of nearly 800. Ten years after the SARS, MERS emerged in 2012, [which has] caused 2494 human infections with 858 deaths (as of November 2019) and remains a disease of global, and particularly Middle Eastern, public health concern...In December 2019, a new coronavirus (2019nCoV), which is about 70% similar to SARS-CoV, was discovered in the central Chinese city of Wuhan, with 545 cases in 25 provinces being diagnosed (Wang et al., 2020, 354). SARS-CoV-2 bears many similarities to both SARS (also called SARS-CoV or SARSCoV-1) and MERS, particularly in their genetic makeup, common symptoms, and methods of prevention and treatment, regardless of the drastically different course the outbreaks took. Domingo et al. describe SARS-CoV-2 as a “previously-unknown βcoronavirus which shows 88% identity to the sequences of two bat-derived SARS-like coronaviruses, 79.5% identity to SARS-CoV, and about 50% identity to Middle East Respiratory Syndrome (MERS)-CoV” (Domingo et al., 2020, 2). All three coronaviruses are thought to be of zoonotic origin. SARS “can be traced back to a SARS-like coronavirus present in animal species, specifically Himalayan palm civets and racoon dogs, that evolved through the species barrier to humans. This SARS-CoV-like virus was isolated in a live-animal market in Guandong, China in October, 2003” and “spread to other east Asian countries such as Hong Kong, Taiwan, Vietnam, and Singapore, eventually making its way to Toronto, Canada” (Chandra & Chandra, 2020, 465). MERS, by contrast, “predominantly extended to countries in the Middle East, Africa, and East Asia, [affecting] over 27 countries…[with] 2,494 cases and 858 deaths since September 2012” (Chandra & Chandra, 2020, 465). MERS is hypothesized to have derived from 9 Dromedary camels after “the full human genome of MERS-CoV was found in Dromedary camels in Saudi Arabia, along with 15% of a camel derived coronavirus” in 2013 and was subsequently isolated from the same type of camels in Qatar, UAE in 2014 (Chandra & Chandra, 2020, 465). Many scientists also believe “that a Dromedary coronavirus originated from bat coronaviruses through cross-species transmission in the distant past” (Chandra & Chandra, 2020, 465). In the absence of a vaccine, “supportive care, or measures taken to reduce the effects of the disease symptoms rather than the disease itself, is the only currently-implementable practice against any of these coronaviruses” (Chandra & Chandra, 2020, 467). Treatment for COVID-19 presently encompasses “general measures that are recommended for other severe acute pulmonary infections...includ[ing] proper infection control measures (airborne, droplet and contact precautions), supportive therapies (hydration, analgesics, antipyretic, intubation, respiratory support, empirical antibiotics [in the case of bacterial infection)], managing sepsis if present, and also close monitoring” (Afshar et al., 2020, 196). Many treatment methods are similar across all three coronaviruses, as “the clinical symptoms of SARSCoV-2 are much like SARS-CoV and MERS-CoV. The most common include fever, cough, dyspnea, fatigue, myalgia, headaches, and sputum production. Sore throat, chest pain, conjunctival congestion, nausea, vomiting, diarrhea, rhinorrhea, and hemoptysis [are] present in some cases” (Chandra & Chandra, 2020, 467). Multiple vaccines for COVID-19 have been developed and approved for use, though vaccines have not been similarly approved for SARS and MERS, partly due to the limited effect of the outbreaks and the fact that these infections are “quite rare today” (Chandra & Chandra, 2020, 468). Additionally, due to the shared modes of transmission across all three coronaviruses, 10 “preventive strategies for SARS-CoV, MERS-CoV, and SARS-CoV-2 are mainly the same. The gist of these preventive measures is to simply reduce the number of infections in a given population. Most of these interventions involve proper hygiene and reduced social contact. ‘Social distancing,’ such as canceling mass gatherings and closing schools, was the main preventative strategy for SARS-CoV and MERS-CoV and is being implemented today for COVID-19” (Chandra & Chandra, 2020, 468). Given these similarities, it may perhaps be puzzling why SARS and MERS infected only a few thousand individuals and total deaths under 1,000 while SARS-CoV2 has killed more than 3 million people and infected over 160 million more worldwide as of May 2021, just 17 months after the first reports of the virus (World Health Organization, 2021). On July 5, 2003, “the World Health Organization (WHO) announced that the global SARS outbreak was contained” while “MERS has died down drastically since its peak in 2013, [with] sporadic outbreaks….primarily in Saudi Arabia from 2014 to 2019” (Chandra & Chandra, 2020, 465). To understand this difference, it is important to understand the differences between the viruses themselves, particularly with regard to transmission. Afshar et al. argue that “the rapid spread suggests that COVID-19 virus is highly contagious and easily transmissible between humans” (Afshar et al., 2020, 196). Studies have already found that “a key factor in the transmissibility of Covid-19 is the high level of SARS-CoV-2 shedding in the upper respiratory tract, even among presymptomatic patients, which distinguished it from SARS-CoV-1, where replication occurs mainly in the lower respiratory tract. [Additionally,] viral loads with SARS-CoV1, which are associated with symptom onset, peak a median of 5 days later than viral loads with SARS-CoV-2, which makes symptom-based detection of infection more 11 effective in the case of SARS-CoV-1” (Gandhi et al., 2020, 2158). Additionally, “what has allowed [for the] control of MERS-CoV is a low [reproductive number] (approximately 1), meaning each person with the disease transmit it to only one other person (the SARS-CoV [reproductive number] was approximately 4)” (Guarner, 2020, 420). For an outbreak to grow, the reproductive number must be greater than 1. Fortunately, while SARS-CoV-2 is more infectious than SARS or MERS, it has also had a lower case fatality rate, a ratio of the number of deaths due to a certain disease compared to those who had the disease. In the case of MERS, this number is reported to be 858 deaths ÷ 2,494 cases, or 34.4% while for SARS it was 774 deaths ÷ 8,000 cases, or 9.6%. Afshar et al. note that “one of the biggest differences between SARS and COVID-19 is the speed at which COVID-19 was reported and identified. This led to a quick implementation of screening measurements to restrict the spread of the infection. Using the COVID-19 case data reported above as an estimation, the case fatality rate among all reported global cases would calculate to be approximately 3,000,000 deaths ÷ 160,000,00 cases, or 1.875%. However, this novel coronavirus...is more asymptomatic in the early stage compared to SARS,” meaning that presymptomatic and asymptomatic spread have contributed to the growth of the initial outbreak to the scale of a global pandemic (Afshar et al., 2020, 197). Clearly, the nature of the novel coronavirus SARS-CoV-2, as studied by multiple researchers since it was first documented, has presented particular challenges for detecting cases, containing the spread, and treating its symptoms. Given this background of the similarities and differences between the novel coronavirus and other disease outbreaks in history, a discussion of the virus will follow through a social 12 epidemiological lens, including a further discussion of the disease itself, in addition to the process of tracking the disease in a population, monitoring prevention measures and delivery of care, and implications for public policy related to the virus. 13 COVID-19 & SOCIAL EPIDEMIOLOGY While the focus of this paper is not the medical or biological nature of the virus, nevertheless it is crucial to draw on the field of epidemiology in order to understand how the pandemic has spread. Data is integral to the process of measuring diseases in a population and epidemiologists use a variety of statistics to accomplish their goals. The goals of epidemiology can be defined broadly as: 1. Identify the cause (etiology) of disease and its risk factors 2. Measure the prevalence of a disease in the population 3. Understand the natural history of a disease and its prognosis 4. Evaluate methods of prevention and delivery of care 5. Make recommendations for public policy (Celentano & Szklo, 2019, 2-3) ETIOLOGY: THE ORIGINS OF THE NOVEL CORONAVIRUS Diseases always exist at some level in a population. The normal, expected level of disease is called the endemic level (Celentano & Szklo, 2019, 23). When the prevalence of disease in the population exceeds the endemic level but is restricted to a community, it is called an epidemic; if an epidemic spreads globally, as in the case of the novel coronavirus, it is called a pandemic (Celentano & Szklo, 2019, 24). As Wang et al. note, “the three basic elements required for an infectious disease epidemic are [a] source of...infection, route of transmission, and susceptible hosts (humans)” (Wang et al., 2020, 354). This describes the three elements of the epidemiologic triad, which consists of the host, the surrounding environment, and a disease agent. The route of transmission is described as being either direct human-to-human contact, or indirect through a vehicle or vector (Celentano & Szklo, 2019, 20-1). It is worth noting first, that “pathogens that 14 spread using the respiratory tract...are over-represented among emerging diseases...[and] have fewer barriers moving from one host to another than pathogens spread via other routes” (United Nations Environment Programme, 2020). Especially in the case of a new virus such as SARS-CoV-2, it is important to understand how the disease spreads, including who is susceptible to the disease, what is the nature of the disease agent (virus), how the disease spreads between people (once human-to-human transmission has been established), what vehicles or vectors may be involved, and how the environment affects the dynamics of disease transmission. In reference to the epidemiologic triad, SARSCoV-2 infection, whenever and wherever it occurs, is truly a combination of many factors. Domingo et al. note that “the clinical spectrum of COVID-19 is broad. Not everyone who acquires SARS-CoV-2 becomes sick and the state that emerges after infection can vary among patients or within the same patient over time. Consequently, it is envisaged that virus-dependent, host-dependent, and environment-dependent factors may modify the virus-host interaction explaining not only the individual susceptibility to infection but also the broad scale of damage seen in clinical disease” (emphasis added) (Domingo et al., 2020, 7). Officials of the World Health Organization were first alerted to “cases of pneumonia of unknown etiology” on December 31, 2019 and by January 3, 2020, “44 patients with pneumonia of unknown etiology have been reported to WHO by the national authorities in China. Of the 44 cases reported, 11 are severely ill, while the remaining 33 patients are in stable condition” (World Health Organization, 2020). At that time, “the causal agent [had] not yet been identified or confirmed” with the “clinical signs and symptoms [reported as]...mainly fever, with a few patients having difficulty in 15 breathing, and chest radiographs showing invasive lesions of both lungs” (World Health Organization, 2020). SARS-CoV-2 was isolated by researchers later in January 2020. Zhu et al. described the visible characteristics of the virus based on electron micrographs: Particles were generally spherical with some pleomorphism. Diameter varied from about 60 to 140 nm. Virus particles had quite distinctive spikes, about 9 to 12 nm, and gave virions the appearance of a solar corona. Extracellular free virus particles and inclusion bodies filled with virus particles in membrane-bound vesicles in cytoplasm were found in the human airway epithelial ultrathin sections. This observed morphology is consistent with the Coronaviridae family (Zhu et al., 2020, 730). The isolation of the virus and subsequent analysis pointed researchers toward a new virus as the cause of the mysterious pneumonia cases reported in Wuhan. As Zhu et al. note, “since the sequence identity in conserved replicase domains (ORF 1ab) is less than 90% between 2019-nCoV and other members of betacoronavirus, and 2019-nCoV-- the likely causative agent of the viral pneumonia in Wuhan-- is a novel betacoronavirus belonging to the sarbecovirus subgenus of Coronaviridae family” (emphasis added) (Zhu et al., 2020, 731). Dr. Charles Chiu, a professor and expert in viral genomics at the University of California-San Francisco believes “the group of related coronaviruses giving rise to SARS-CoV-2 has existed for decades in bats and probably originated more than 40 years ago...SARS-CoV-2 shares 96% of its genetic material with a sample of coronavirus taken in 2013 in intermediate horseshoe bats from Yunnan province in China, which suggests the Yunnan virus is its ancestor. [However] how the virus traveled the 1,200 miles from Yunnan to Wuhan remains unknown” (Weise & Weintraub, 2021). Another crucial aspect of identifying the cause of an outbreak, or if it spreads far enough, a pandemic, is examining the physical location in which the initial outbreak began and how the first cases arose. Both of these have been topics of contentious debate 16 since cases began spreading around the globe, with theories ranging from scientific hypothesis to outright conspiracy, and these debates continue to the time of this writing. However, based on the current science, multiple theories have been proposed by the World Health Organization (WHO). WHO expert Peter Ben Embarek discussed these hypotheses in a February 2021 news conference, which can be summarized as follows: 1. An “‘intermediary host species’ likely caused the transmission, meaning some sort of animal gave it to humans” (Scribner, 2021). 2. “there’s a chance the transmission happened through the trade of frozen products” (Scribner, 2021). 3. Another possibility is “a ‘direct zoonotic spillover,’ which suggests the disease transferred from an animal reservoir to a human” (Scribner, 2021). 4. “The final theory is that the virus was created in a lab, but that was the least likely of all four scenarios” (Scribner, 2021). Of these, Embarek notes that “our initial findings suggest that the introduction through an intermediary host species is the most likely pathway and one that will require more studies and more specific targeted research” (Thornton, 2021). While for many areas affected by the COVID-19 pandemic the spillover of a disease from one species to another, such as from a reservoir species to humans, “is sometimes seen as a ‘black swan’--an extremely rare event--they are actually a widely predicted consequence of how people source food, trade animals, and alter environments” (United Nations Environment Programme, 2020). While there are “millions of species of microorganisms on Earth, pathogens (microorganisms that harm the host) are extremely unusual. Only about 1,400 microorganisms are known as potential causes of human 17 infection” (United Nations Environment Programme, 2020). By current estimates, “about 60 percent of human infections are estimated to have an animal origin, and of all new and emerging human infectious diseases, some 75 percent ‘jump species’ from (non-human) animals to people” (United Nations Environment Programme, 2020). In general, there are multiple routes through which such diseases can emerge: “new diseases in humans can emerge either as a result of a change in the nature or behavior of commensal microorganisms that cause disease, or through infection by novel organisms, usually through contact with animals and the environment,” as may have been the case in the COVID-19 pandemic (United Nations Environment Programme, 2020). While it is the most recent, COVID-19 is certainly not the only occurrence of zoonotic transfer of disease to the human population that has caused an outbreak larger than a confined community. Other instances include the “zoonotic bubonic plague or pest (Black Death caused by the bacteria Yersinia pestis) of the mid-fourteenth century [that] killed millions in Eurasia and North Africa, wiping out a third of Europe’s population” or the tuberculosis outbreak in 19th century Europe, as well as the 1918 influenza pandemic (United Nations Environment Programme, 2020). There are multiple conditions that contribute to zoonotic disease emergence, and “many of these drivers are now occurring in the same places amplifying their impact” (United Nations Environment Programme, 2020). Additionally, “certain animals...are more likely to harbour zoonotic or potentially zoonotic pathogens based on their physiological characteristics, ecosystem niche, social behavior and relatedness to humans,” including “livestock, rodents, bats, carnivores, and non-human primates” (United Nations Environment Programme, 2020). In general, the “use and trade in live and dead animals can lead to increased close contact between 18 animals and people throughout the supply chain, which increases the risk of zoonotic diseases emergence” (United Nations Environment Programme, 2020). In other words, in the various conditions in the modern world that make close contact between humans and specific animal species more likely, such as the industrialization of and increased desire for animal protein, changes in food supply chains, or urban development near wildlife and livestock, make zoonotic spillover more likely (United Nations Environment Programme, 2020). Coupled with climate change (since “many zoonoses are climate sensitive and a number of them will thrive in a warmer, wetter, more disaster-prone world foreseen in future scenarios”), the increased availability of travel allowing “diseases…[to] move around the world in periods shorter than their incubation periods (the time between exposure to a pathogen and the first clinical sign of illness)” as well as the significantly larger population (which increased from approx. 1.6 billion in 1900 to approx. 7.8 billion currently), there are a number of conditions that make the possibility of current and future zoonotic transfer concerning and make zoonosis a plausible origin for SARS-CoV-2 (United Nations Environment Programme, 2020). The hypothesis that COVID-19 was created in a laboratory has caught the attention of conspiracy theorists since the pandemic began. Peter Daszak, a BritishAmerican member of the WHO team that traveled to Wuhan commented that “in people’s imaginations there might be this image of one person in a lab in China who drops a petri dish and that somehow leads to a massive outbreak. It’s just not like that. Every year there are millions of people going in bat caves and hunting and eating wildlife. It happens every day. They are being exposed to bat viruses every day. It only takes one of these people to go to a city, cough, and spread the virus” (qtd. in 19 Hjelmgaard, 2021). This theory is centered around the Wuhan Institute of Virology, which has been called “China’s most advanced biosafety lab,” is in close proximity to the wildlife market where some of the first cases originated, and “is known for its work researching coronaviruses in bats. The institute also attracted attention because of China’s initial unwillingness to share information about some aspects of the outbreak.” (Hjelmgaard, 2021). As for the wildlife market in Wuhan that garnered international attention at the beginning of the outbreak, referring to the second hypothesis listed above, “the WHO concluded that the Wuhan wildlife market was an area where the virus began spreading rapidly, but it was unable to determine how it first arrived there” (Hjelmgaard, 2021). Following an international fact-finding expedition conducted by WHO, Embarek commented that “the market probably was a setting where that kind of spread could have happened easily, but that’s not the whole story” (qtd. in Hjelmgaard, 2021). Indeed, WHO is missing crucial information from their fact-finding trip, as “China’s continued resistance to revealing information about the early days of the coronavirus outbreak, scientists say, makes it difficult for them to uncover important clues that could help stop future outbreaks of such dangerous diseases” (Hernández & Gorman, 2021). Such data includes complete patient records, as well as possible suspected cases of “92 people [who] were hospitalized in Wuhan as early as October 2019 with symptoms such as fever and coughing. The Chinese experts said they had found no trace of Covid-19 in those people, but the tests were incomplete. The W.H.O. team members said more research was needed” (Hernández & Gorman, 2021). WHO investigators said “disagreements [with Chinese officials] over patient records and other issues were so tense that they sometimes 20 erupted into shouts among the typically mild-mannered scientists on both sides” (Hernández & Gorman, 2021). Thus, at the time of this writing, the place of COVID-19’s origin remains very much in question. RISK FACTORS Understanding risk factors relates to the characteristics of the host in the epidemiologic triad referenced earlier. Characteristics of the host may contribute to either a host being susceptible to a disease or guarding against it (Celentano & Szklo, 2019, 20). Host characteristics include far more than genetics; age, race, gender, occupation, and lifestyle are all considered characteristics (Celentano & Szklo, 2019, 20). Age is a particularly important predictor of overall mortality, so it will be discussed first. Age has been a somewhat contentious issue regarding risk of both contracting and developing severe symptoms from COVID-19. As with many other diseases, advanced age is considered a risk factor for COVID-19, though the ways in which this is the case are many. First, young people appear to be less likely to develop symptoms of COVID-19 if contracted. As of October 2020, it is reported that “only half of children and teenagers with antibodies against SARS-CoV-2 have experienced symptoms…[though] the reason for the lower burden of symptomatic disease in children is not yet clear” (Snape & Viner, 2020, 287). Second, if symptoms develop, they may be more severe in older adults. In one study, Liu et al. observed 56 patients in Hainan General Hospital between January 1 and February 15, 2020. Overall, the study found that SARS-CoV-2 is “more likely to infect adult men with chronic comorbidities due to their weaker immune function” (Liu et al., 2020, e17). There were also important differences in symptoms between patients aged 60 and above when compared to those below age 60: “in terms of laboratory tests, the 21 proportion of patients with an increased number of white blood cells and neutrophils in elderly patients was significantly higher than that in the young and middle-aged group, suggesting that elderly 2019-nCoV infected patients are more likely to have a bacterial infection…[and] in terms of imaging, the incidence of multilobe lesions in elderly patients is significantly higher than in young and middle-aged patients” (Liu et al., 2020, e17). Overall, Liu et al. note that because “elderly patients are prone to multi-system organ dysfunction and even failure, other systemic complications should be prevented, including gastrointestinal bleeding, renal failure, disseminated intravascular coagulation (DIC) or deep vein thrombosis” (Liu et al., 2020, e17). Third, this means that older adults are more likely to be admitted to a hospital if they contract COVID-19 and require intensive treatment. Compared to adults aged 18-29, the CDC reports that adults aged 6574 years are four times more likely to be hospitalized, adults 75-84 are eight times more likely, and adults 85 and above are thirteen times more likely (Center for Disease Control, 2020). Finally, older adults are at increased risk of death from COVID-19 than younger adults or children. According to the CDC, as of December 2020, 80% of deaths in the United States from COVID-19 were adults age 65 and older (Center for Disease Control, 2020). Further, adults aged 65-74 years are 90 times more likely to die than those aged 18-29 years, adults 75-84 are 220 times more likely, and adults 85 and above are 630 times more likely (Center for Disease Control, 2020). 22 Figure 1. Hospitalization and Death rates by age due to COVID-19. Note. From Older Adults and COVID-19, by the Center for Disease Control, 2020 (https://www.cdc.gov/coronavirus/2019-ncov/need-extra-precautions/olderadults html#rint) However, this is certainly not to say that severe cases of COVID-19 are impossible in young people, as some public figures outside public health have argued. In the first wave of COVID-19, a “rare, but potentially severe, multisystem inflammatory syndrome [was] observed in more than 1000 children and adolescents in multiple countries” characterized by “persistent fever accompanied, to a variable extent, by gastrointestinal symptoms, rash, and conjunctival inflammation” (Snape & Viner, 2020, 287). Further, the long-term effects of this condition, known as Pediatric Inflammatory Multisystem Syndrome, are unknown and “all children and teenagers who experience PIMS-TS require ongoing cardiac review” (Snape & Viner, 2020, 287). While age is certainly an important risk factor to keep in mind, it is by no means the only one. For example, multiple studies have found that male patients are more likely 23 to die from COVID-19 (Noor & Islam, 2020, 1278). Host characteristics also include current and previous diseases and immunologic history. Early studies of SARS-CoV-2 patients investigated such host characteristics and risk factors. In the January 2020 study of 41 patients referenced by Ahmad & Khan, “half of [the patients] had underlying diseases, including diabetes 20%, cardiovascular disease 15%, and hypertension 15%” (Ahmad & Khan, 2020, 1). Indeed, according to Liu et al., those with diabetes, hypertension, cardiovascular disease, and cerebro-vascular disease are more susceptible to COVID-19 (Liu et al., 2020, e14). Noor & Islam conducted a meta-analysis of 58 cohort studies consisting of 122,191 patients across North and South America, Europe, and Asia, examining not only the impact of age and sex on COVID-19 mortality, but also ICU-admissions and 14 comorbidities. While mortality rate reported by each study individually ranged from 0.6 to 61.5% of hospitalized patients, overall mortality among all hospitalized patients in the studies was 18.88% (Noor & Islam, 2020, 1272). However, they also found multiple conditions that contribute to higher risk of death among hospitalized COVID-19 patients. In their analysis, “hypertension, diabetes, cardiovascular disease, cerebrovascular disease, COPD, cancer, coronary heart disease and chronic lung disease were significantly associated with the risk of mortality among the hospitalized COVID-19 patients,” with hypertension and diabetes being most common among these comorbidities (Noor & Islam, 2020, 1278). MEASURING COVID-19 IN POPULATIONS Epidemiologists measure the levels of various diseases in populations on a regular basis. When diseases exceed the expected endemic level, or cases of suspicious illness arise, such surveillance becomes even more important. In public health, surveillance 24 refers to tracking diseases in a population, which may either be conducted passively by clinicians reporting case data, or actively by individuals recruited to identify cases in populations (Celentano & Szklo, 2019, 42). When tracking the prevalence of any disease in a population, it is important to first define what counts as a “case” of the disease. With COVID-19, the use of dichotomous tests (those with a result of either “positive” or “negative”) can confirm the presence or absence of the pathogen SARS-CoV-2. However, it is also important to consider who is able to be tested, who is most likely to be tested (i.e., those without symptoms may be less likely to be tested without confirmed COVID-19 exposure), the probabilities of incorrect results (false positives or false negatives) from the given test, as well as the likelihood that a test will correctly identify those who have the disease and those who do not. The latter of these concepts are encapsulated in the measures of a test’s sensitivity (the likelihood that a test correctly identifies those who have the disease) and specificity (the likelihood that a test correctly identifies those who do not have the disease), while positive and negative predictive value assess false positives and negatives (i.e., if the test result is positive (negative), what is the likelihood that the person actually does (not) have the disease). Determining sensitivity and specificity have been a significant challenge in the case of COVID-19 testing. While tests were developed mere days after the virus genome was released, the lack of a validated comparison test has made it difficult for researchers to determine their accuracy (Watson et al., 2020, 1). As Watson et al. point out, “no test gives a 100% accurate result; tests need to be evaluated to determine their sensitivity and specificity, ideally by comparison with a ‘gold standard,’” which would identify true negatives and true positives for comparison with 25 the new test; however, “the lack of such a clear-cut ‘gold standard’ for covid-19 testing makes evaluation of test accuracy challenging” (Watson et al., 2020, 1). There are also a variety of human factors that may affect the sensitivity and specificity of a test, such as the location of specimen collection and the stage of the disease. For example, one study of sensitivity of RT-PCR (real-time reverse transcription polymerase chain reaction) tests found 93% sensitivity “for broncho-alveolar lavage, 72% for sputum, 63% for nasal swabs, and only 32% for throat swabs” (Watson et al., 2020, 1). The determination of whether or not someone has COVID-19 is what the public typically thinks of with regard to “covid testing.” However, there are many different types of testing programs and depending on the goal of the program, different types of tests may or may not be useful. As Mina & Andersen note, “diagnostics, screening, and surveillance serve different purposes, demand distinct strategies, and require separate approval mechanisms” (Mina & Andersen, 2021, 126). The type of testing to determine if a patient is infected with the virus is called diagnostic testing and “focuses on accurately identifying patients who are infected with SARS-CoV-2 to establish the presence or absence of disease and is performed on symptomatic patients or asymptomatic individuals who are at high risk of infection,” such as those with a known close contact with a confirmed case (Mina & Andersen, 2021, 126). The specificity and sensitivity of the tests are important considerations when deciding what tests are appropriate for a certain testing program, as tests with higher sensitivity and specificity often have tradeoffs in terms of cost, time, availability, or necessary equipment. In diagnostic testing, it is vitally important that the tests used “are highly sensitive so as to not miss COVID-19 patients (false negatives), and specific, so as to not wrongly diagnose SARS- 26 CoV-2-negative individuals as having COVID-19” (Mina & Andersen, 2021, 126). Early in the pandemic, one of the most important actions for government and public health officials to take was increasing the community’s testing capacity. Otherwise, as has been seen in some cases, tests that were designed to return results between 12 and 48 hours instead hit “major bottlenecks…[that] have led to turnaround times exceeding 5 to 10 days in some regions, making such tests useless to prevent transmission” (Mina & Andersen, 2021, 126). In addition to diagnostic testing, surveillance may be used to understand exposure and dynamics of transmission on a large scale. In surveillance testing, “populations can be used both as a tool for understanding historical exposures and as a measure of ongoing community transmission. For the former, serological testing of individuals for the presence of SARS-CoV-2-specific antibodies is used to identify those previously infected. For the latter, surveillance testing can be an effective way to monitor real-time SARS-CoV-2 spread in communities” (emphasis added) (Mina & Andersen, 2021, 126). Serological testing involves detecting immunoglobulins including IgG and IgM (Watson et al., 2020, 3). In contrast to diagnostic testing, for surveillance testing, “the goal is not identification of every case but rather the collection of data from representative samples that accurately measure prevalence and serve to inform public health policy and resource allocation,” therefore a lower level of sensitivity and specificity may be acceptable in this case, especially if the variance is known and can be statistically corrected (Mina & Andersen, 2021, 126). As Larremore et al. argue, another reason it is less important for test sensitivity and specificity to be ideal in COVID-19 surveillance programs is that test frequency and 27 result turnaround time are more important for controlling the spread of the virus. After an individual is infected with SARS-CoV-2, they “undergo a period of incubation during which viral titers are usually too low to detect, followed by an exponential growth of the gurus, leading to a peak viral load and infectiousness, and ending with declining viral levels and clearance” (Larremore et al., 2021, 1). Based on what researchers have found thus far about COVID-19 and the tests that are available to detect the virus, in comparing two tests with different limits of detection (LOD), the real-time quantitative polymerase chain reaction (RT-qPCR) with LOD of 103 cp/ml and point-of-care nucleic acid LAMP and rapid antigen tests with LOD closer to 105 cp/ml, Lerremore et al. note that “since filtered samples collected from patients displaying less than 106 N or E RNA cp/ml contain minimal or no measurable infectious virus, either class of test would detect individuals who are currently infectious” (Larremore et al., 2021, 2). Additionally, “during the exponential growth of the virus, the difference between 103 and 105 cp/ml is short, allowing only a limited window in which only the more sensitive test could diagnose individuals,” which, for other respiratory viruses is approximately a day, and third, “high-sensitivity screening tests, when applied during the viral decline accompanying recovery, are unlikely to substantially impact transmission because such individuals detected have low, if any, infectiousness” (Larremore et al., 2021, 2). In Larremore et al.’s modeling, “testing frequency was found to be the primary driver of population-level epidemic control, with only a small margin of improvement provided by using a more sensitive test,” as delays in reporting could lead individuals who do not yet know they are infected to spread the virus to others due to failure to take necessary precautions (emphasis added) (Larremore et al., 2021, 4). However, it is also important to 28 note that “communities vary in their transmission dynamics, due to difference[s] in rates of imported infections and in the basic reproductive number R0, both of which will influence the frequency and sensitivity with which surveillance testing must occur” (Larremore et al., 2021, 4). In conjunction with other testing modalities, screening may be used to prevent additional spread of the virus. The goal of screening is to “detect people who are likely to be infectious,” and while medical facilities such as outpatient clinics and some schools have implemented screening measures before allowing entry to the building, screening “has been critically underused yet is one of the most promising tools to combat the COVID-19 pandemic” (emphasis added) (Mina & Andersen, 2021, 126). Particularly since asymptomatic cases account for an estimated 20% of all SARS-CoV-2 infections and given that “symptomatic disease is preceded by a presymptomatic incubation period” that lasts for multiple days in the majority of symptomatic infections, “entry screening to detect infectious individuals before accessing facilities (e.g., nursing homes, restaurants, and airports), along with symptom screening and temperature checks, can be beneficial” (Mina & Andersen, 2021, 126). Due to the demands of time, tests used in addition to simple measures like temperature checks require rapid results, and it should be noted that “the required sensitivity and specificity of entry-screening tests are, like all tests, context dependent” (Mina & Andersen, 2021, 126). For example, a facility such as a nursing home would benefit from using tests with higher sensitivity and specificity than a different location with fewer high-risk individuals. While incorrect results may have clinical costs, all tests have economic costs. Brooks & Das sought to estimate the costs of various tests and other surveillance 29 measures. For example, citing a study conducted by Johns Hopkins University, the cost of hiring 100,000 contact tracers was estimated to be $3.6 billion, a cost estimate which can be further specified for local conditions and allocated among cases within the community, while patients and the broader economy also incur costs if a patient is selfisolating (Brooks & Das, 2020, 579). In their analysis of costs, Brooks & Das argue that different test results have different costs, which can be summarized as follows: for true positive tests, “costs include sample testing, contact tracing, and confirmation [of test results] with an orthogonal test,” for true negative tests, the costs include “only testing for molecular tests; testing and confirmation for antigen tests; and testing plus self-isolation for antibody tests” (Brooks & Das, 2020, 581). By contrast, false positive tests “incur costs of testing plus self-isolation and contact tracing for molecular and antigen tests. Cost of false-positive antibody tests include the risk that patients receiving these results may become infected and infect others,” while false negative molecular tests “incur costs of the true-positive result, multiplied by (1 + [R0]) to account for other people infected,” where R0 is the basic reproductive number, “the effective number of people infected by one positive COVID-19 case,” “false negative antigen tests are confirmed with an orthogonal test…[and] false negative antibody tests incur the same costs as true negatives for testing plus self-isolation for antibody tests” (Brooks & Das, 2020, 581-2). While the interpretation of a positive or negative test result may appear simple at first glance, given that no test is 100% accurate, the result should be understood more as a probability in the context of the test’s sensitivity and specificity, rather than as a certain outcome. Watson et al. argue that interpreting a test result should be only one step clinicians engage in, rather than the only step in the process. Instead, clinicians should 30 begin by estimating the pre-test probability of disease “using knowledge of local rates of covid-19 infection from national and regional data [as well as] patients’ symptoms and signs, likelihood of alternative diagnoses, and history of exposure to covid-19” (Watson et al., 2020, 2). This estimate is then adjusted with additional information (such as a test result, while also taking the test’s sensitivity and specificity into account), to arrive at a post-test probability the patient has (or does not have) the disease. Watson et al. describe a numerical way to estimate this post-test probability of disease, by plotting the pre-test estimate on the x-axis of a coordinate grid, construct a leaf plot with curves accounting for the sensitivity and specificity of the test used, drawing a line vertically to meet the line corresponding to the test result received, then horizontally to the y-axis. Figure 2. Example Leaf Plot for COVID-19 using 70% sensitivity and 95% specificity. Note. From “Interpreting a covid-19 test result,” by J. Watson, P. F. Whiting, & J. E. Brush, 2020, BMJ 369, p. 6 (doi: 10.1136/bmj.m1808). 31 It is important to note that in the graph above, “the shift in the probability is asymmetric, with a positive result having a greater impact than a negative test result, owing to the modest sensitivity and negative likelihood ratio of the RNA test” used in the above example (Watson et al., 2020, 2). Many clinicians and public health officials have warned that “while positive tests for covid-19 are clinically useful, negative tests need to be interpreted with caution, taking into account the pre-test probability of disease…[as] false negatives carry substantial risks” of further spread in the population (Watson et al., 2020, 2). With this in mind, Watson et al., along with many medical professionals argue that “patients with a single negative test but strongly suggestive symptoms of covid-19 should be advised to self-isolate in keeping with guidelines for suspected covid-19” in an effort to reduce spread of the virus (Watson et al., 2020, 2). Similarly, with regard to entry screening discussed previously, Mina & Andersen state that the “key to use of tests for entrance screening is that a negative test alone should not be considered sufficient to enter--that should be based on satisfying other requirements, including masks and physical distancing” that reduce the likelihood of virus spread in the case of a false negative, while “a positive test should be sufficient to bar entry in most settings” (Mina & Andersen, 2021, 127). Overall, Mina & Andersen argue that “testing is a central pillar of clinical and public health response to global health emergencies, including the COVID-19 pandemic. Nearly all testing modalities have a role, and the one-size-fits-all approach to testing used by many Western countries has failed” (Mina & Andersen, 2021, 127). Measuring the presence of a disease in a population requires multiple measures. First of these is prevalence, which is defined in epidemiology with the following formula, measured 32 either at a certain time (called point prevalence, which includes all individuals who have the disease at the time of measurement, regardless of the time of diagnosis) or over a period of time (called period prevalence, which includes all individuals who had the disease at any point during the specified time period): 𝑝𝑟𝑒𝑣𝑎𝑙𝑒𝑛𝑐𝑒 = 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐𝑎𝑠𝑒𝑠 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 Incidence, by contrast, only takes new cases into account and is calculated by the formula: 𝑖𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 = 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑛𝑒𝑤 𝑐𝑎𝑠𝑒𝑠 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑎𝑡 𝑟𝑖𝑠𝑘 While the mathematical formula to calculate prevalence in the population is simple, the task of measuring the prevalence of COVID-19 is certainly not. As noted previously, mild or asymptomatic cases have made it much more difficult for epidemiologists to accurately measure the number of cases in the relevant population as these cases are less likely to be detected, particularly early in the pandemic when testing capacity was low (Brüssow, 2020, 4895). The issue of asymptomatic cases will be discussed later. Another important measure to understand the spread of an infectious disease in a population is the basic reproductive number (𝑅0 )which describes the number of secondary infections caused by a primary case (Brüssow, 2020, 4895). As of November 2020, Brüssow reports that “SARS-CoV-2 has an estimated 𝑅0 of 2-3, a higher ‘infectious force’ than influenza virus, but a lower than ‘flying infections’ such as chicken pox or measles” (Brüssow, 2020, 4895). Writing from Belgium, Brüssow also notes that “𝑅0 needs to be reduced from its initial value to about 1.2 to prevent collapse of the healthcare system. Subsequently, social distancing measures must be relaxed 33 gradually in a highly controlled manner over a period of months to years. Even then success is not assured. [Additionally,] a major unknown remains the nature, duration, and effectiveness of immunity acquired by natural infection (or [now] by vaccination)” (Brüssow, 2020, 4895). As vaccines have begun to be administered globally at the time of this writing, the answers to the questions outlined by Brüssow remain unclear. COVID-19 NATURAL HISTORY & PROGNOSIS In epidemiology, the term “natural history” refers to the course of a disease in an individual in the absence of treatment (Centers for Disease Control and Prevention, 2012). In the case of COVID-19, certainly not every case requires treatment. Individuals infected with SARS-CoV-2 may experience no symptoms, mild symptoms, severe symptoms that require treatment, and some ultimately die from the virus. Early efforts to understand the course of infection from SARS-CoV-2 have included measuring, tracking, and estimating the proportions of each of these outcomes, through mortality rates, case fatality rates, and risk factors as described in the previous section. It has also involved developing a list of common symptoms and a general timeline of infectious period, incubation period, and recovery, though it also must be noted that each of these measures may vary between individuals. This section will first describe in cursory detail how researchers believe SARS-CoV infection occurs at a microscopic level, then discuss the wide range of outcomes for COVID-19. As the virus SARS-CoV-2 itself is new, information regarding nearly every aspect of the virus has been evolving, both through formal research and as new cases emerge. One of many topics researchers have sought to investigate is the exact ways in which the virus interacts with the host’s body, including organs, blood, and immune system. The 34 human immune system consists of both innate and acquired systems. The innate system consists of physical barriers, like skin and mucous membranes, as well as cells within the immune system, and acts as the “body’s first line of defence against germs entering the body [and] responds in the same way to all germs and foreign substances” (InformedHealth.org, 2020). The acquired, or adaptive, immune system consists of white blood cells called lymphocytes that mount either antibody responses (B-cells) or cellmediated immune responses (T-cells) (InformedHealth.org, 2020). In contrast to the fastacting, non-specific innate immune response, responses by the adaptive immune system’s responses are “highly specific to the particular pathogen that induced them” but its immunologic memory may provide long-lasting protection against a specific disease (Alberts et al., 2002). When functioning properly, the innate and adaptive immune systems work together to fight against disease. As Subbarao & Mahanty describe in the context of a virus, at a broad level, “the immune system responds to viral infection with cellular and humoral (antibodies, complement, and antimicrobial peptides) responses. These responses are initiated by the innate immune system, which recognizes pathogens and induces production of proinflammatory cytokines and chemokines. This is followed by responses of the adaptive immune system, which consists of T cells, which can directly kill virus-infected cells, and B cells, which produce pathogen-specific antibodies in the serum and at the mucosal surfaces” (Subbarao & Mahanty, 2020, 906). If the immune system is not successful in preventing a virus from invading the body, an infection results. Like many other viruses, SARS-CoV-2 attacks the respiratory tract, and “viral respiratory infections result when a virus infects the cells of the respiratory mucosa; this can occur when virus particles are inhaled or directly contact a 35 mucosal surface on the nose or eyes. Infected individuals shed virus into the environment by coughing or sneezing or even during quiet breathing” (Subbarao & Mahanty, 2020, 905). What happens after this in COVID-19 is less well understood by the public, though researchers have increasingly shed light on these biological processes. As Subbarao & Mahanty describe, The respiratory epithelium is composed of a variety of cells that include ciliated and non-ciliated epithelial cells; goblet cells, which produce mucus that forms the first barrier for an incoming virus; and club cells, which produce proteases. Different respiratory viruses preferentially bind and infect ciliated or non-ciliated epithelial cells of the airways...The presence of specific host cell molecules that are receptors for viral attachment and entry are the main determinants of which cells become infected. Human angiotensin-converting enzyme 2 (ACE2) is the receptor for SARS-CoV and SARS-CoV-2 as well as for human coronavirus NL63...The presence of cells in the respiratory tract expressing the relevant viral receptor is critical for initiation of viral infection, and the clinical presentation depends on where these cells are situated in the respiratory tract. (Subbarao & Mahanty, 2020, 905). Following attachment to the receptor, “the virus gains entry into the cell, and the viral genome is uncoated, releasing the viral genetic material, which is RNA in paramyxoviruses, orthomyxoviruses, and coronaviruses” (Subbarao & Mahanty, 2020, 905). Mokhtari provides additional detail into how this RNA is replicated: “When the virus enters into the cell, it releases the RNA into the cytoplasm, translated into two virus polyproteins and structural proteins. The genome of the virus then begins to replicate and forms a nucleocapsid in combination with genomic RNA and nucleocapsid in combination with genomic RNA and nucleocapsid protein. Then, viral particles use the cellular protein synthesis organelles, such as [the] Endoplasmic reticulum and Golgi complex to be germinated. Finally, vesicles including the viral particles bind to the plasma membrane to be released” and spread within the body (Mokhtari et al., 2020, 614). 36 Inflammation plays a major role in the progression of COVID-19 in severe cases. As La Torre et al. describe, hyperinflammation seen in some cases of COVID-19 is “similar to, but not fully overlapping with other well-known clinical entities, such as macrophage-activated syndrome (MAS), or hemophagocytic lymphohistiocytosis (HLH) and other forms of viral-induced cytokine storm, in that ferritin increase is modest and severe end-organ disease is limited to the lung. Nonetheless, it is becoming more and more evident that it has a central role in disease severity and outcome. Experience from hyperinflammation in HLH, MAS, and cytokine release syndrome suggests that early intervention is essential to avoid irreversible tissue damage” (emphasis added) (La Torre et al., 2020, 77). Subbarao & Mahanty describe how researchers found this “cytokine storm” occurs at a microscopic level: In COVID-19, “A prominent feature of severe SARS-CoV-2 infections is the cytokine release syndrome” (Subbarao & Mahanty, 2020, 907). Cytokines are “small signaling molecules produced by many different immune cells, such as neutrophils (some of the first cells to travel to an infection site), mast cells (responsible for allergic reactions), macrophages, B-cells, and T-cells” that are essential for maintaining healthy immune function in the body’s response to pathogens (Manoylov, 2020). In the case of an invading pathogen, like a virus, “cytokines and organs respond by working together...One of the immune responses cytokines may elicit is inflammation. Cytokines help inflame tissue by directing the cell walls of blood vessels to become more porous by reducing cell-to-cell contact…[a process that] mainly occurs when the body is infected by a pathogen, [though] cytokine-induced inflammatory responses also happen when tissues are physically damaged” (Manoylov, 2020). Another important function of cytokines is that of “direct[ing] brain cells to release chemicals that tell your body that 37 you’re sick, prompting you to rest and avoid activities that could further expose you to pathogens” resulting in feelings of tiredness, achiness, and fatigue (Manoylov, 2020). In the case of SARS-CoV-2, a “cytokine storm” may occur if too many of these cytokines are produced by the following process: “SARS-CoV-2 infection of respiratory epithelial cells activates monocytes, macrophages, and dendritic cells, resulting in secretion of a range of proinflammatory cytokines, including interleukin-6 (IL-6)...Circulating IL-6 and soluble IL-6 receptor complexes indirectly activate many cell types, including endothelial cells, resulting in a flood of systemic cytokine production that contributes to hypotension [low blood pressure] and acute respiratory distress syndrome (ARDS)” (Subbarao & Mahanty, 2020, 907). Other effects of this cytokine storm may include “high levels of fibrinogen and activation of the coagulation cascade on endothelial surfaces of small blood vessels, signaled by very high levels of a fibrin breakdown product called D-dimer. Dysregulation of the [renin-angiotensin-aldosterone system] because of competitive binding of ACE2 by the virus may also induce constriction of blood vessels” (Subbarao & Mahanty, 2020, 907). Subbarao & Mahanty also note what has been observed as a “puzzling phenomenon” seen in COVID-19 patients: “some patients have extremely low blood oxygen levels, but they do not complain of breathlessness. It has been suggested that oxygen uptake in COVID-19 pneumonia is impeded because of clogged and constricted blood vessels in the lungs rather than because of congestion from accumulation of edema fluid in the alveoli, as seen in other viral pneumonias” (Subbarao, & Mahanty, 2020, 907-8). Of importance to social epidemiology is the amount of time after an individual is infected but before they begin to show clinical symptoms, called the incubation period. 38 During this period, which has been estimated to be 4.5-5.8 days in SARS-CoV-2, “the virus attaches to and infects cells, replicates its genome, and spreads to infect adjacent cells...Productive viral infection of respiratory epithelial cells results in clinical symptoms and signs that depend on which part of the respiratory tract is infected. [For example,] infection of the nasal, nasopharyngeal, and oropharyngeal mucosa causes a runny nose, coughing, sneezing, and sore throat…[while] pneumonia occurs when infection and inflammation involve the alveoli and lung parenchyma and its associated with a cough and shortness of breath” (Subbarao & Mahanty, 2020, 906). In COVID-19, following this incubation period, patients then typically begin to experience symptoms, though some individuals never develop symptoms, as will be discussed below. The symptomatic phase of COVID-19 often lasts for 2-4 weeks, with some experiencing long-term symptoms, another outcome of SARS-CoV-2 infection that will be discussed later, with a recovery phase that frequently lasts 2-8 weeks (Subbarao & Mahanty, 2020, 907). Researchers have investigated not only what symptoms occur in COVID-19 patients, but also why they occur and how biological processes might influence the course of the disease, ranging from asymptomatic at one end of the continuum, to severe disease and death on the other. The binding of SARS-CoV-2 to ACE2 receptors is worthy of further discussion due to its influence on the course of COVID-19 symptoms and outcomes, and its typical role in the body. ACE2 is involved in the regulation of blood pressure, which occurs by the following process: “ACE2 metabolizes angiotensin II, a vasoconstrictor, to generate angiotensin 1-7, which is a vasodilator” (Subbarao & Mahanty, 2020, 907). Of relevance for COVID-19, “cells lining the mucosal surfaces of the nose and lungs are endowed with 39 ACE2, which facilitates the infection of the respiratory tract. However, ACE2 is also expressed on cells in many other tissues, including the endothelium, heart, gut, and kidneys, making these organs susceptible by the virus” (Subbarao & Mahanty, 2020, 907). Overall, based on the research of many scientists worldwide, “it is becoming clear that the pathogenic cascade triggered by SARS-CoV-2 infection damages many organs in the body, from the kidneys to the brain. Such pathophysiological changes are rarely seen with other respiratory virus infections” (emphasis added) (Subbarao & Mahanty, 2020, 908). Certainly, SARS-CoV-2 has been found to cause significant damage to the lungs in patients with severe cases of COVID-19. However, the virus has also been found to cause damage to other organs in patients with varying levels of frequency, including the heart, liver, kidneys, in addition to documented effects on the circulatory and neurological systems. Mokhtari et al. hypothesize that in many of these cases, harmful effects might occur either by direct infection of ACE2 receptors by SARS-CoV-2 throughout the body, or by indirect mechanisms, such as the cytokine storm discussed previously. In the case of the lungs, acute respiratory distress syndrome has been documented in many cases of severe COVID-19. As Mokharti et al. describe, “one week after onset of illness, this condition could develop as ADRS in 17% of patients and 65% of these patients died due to multi-organ dysfunction. [However,] the incidence of ARDS was reported 15.6 to 31% higher than that of other organ impairment” (Mokhtari et al., 2020, 615). As noted previously, damage to the lungs may be due to direct infection of ACE2 receptors in the respiratory tract, or by indirect means, such as the “cytokine storm, followed by the attacking immune system via SARS-CoV-2, results in lung 40 failure, multiple-organ dysfunction, and death in severe cases” (Mokhtari et al., 2020, 615). Additional insight has been found from autopsies of COVID-19 patients on the nature and extent of lung damage: “the autopsy evaluations of lung tissue from a COVID19 patient demonstrated the histopathological features, including acute interstitial pneumonia, diffuse alveolar damages with perivascular infiltration of macrophages and T-cells, disrupted cell membrane (direct SARS-CoV-2 infection), intussusceptive angiogenesis, the formation of hyaline membranes, and oedema in the alveolar wall” (Mokhtari et al., 2020, 615). As noted previously, cardiovascular disease is a prominent risk factor for experiencing severe COVID-19; however, some research has indicated that infection with SARS-CoV-2 may result in damage to the heart without a history of cardiovascular diseases. As Mokhtari et al. discuss, as of October 2020, “heart injury has been demonstrated by creatine kinase myocardial band” and one report demonstrated that “12% of COVID-19 patients without a history of [cardiovascular diseases] demonstrated increased levels of troponin...or cardiac arrest during hospitalization period” (Mokhtari et al., 2020, 617). While the exact mechanism(s) of damage to the cardiovascular system remains under investigation at the time of this writing, Mokhtari et al. note that “different pathways [have been] defined for COVID-19 pathogenesis: (1) viremia and direct infection of the lung and heart; (2) recruitment of the innate immune system and macrophages and cytokine storm; (3) adaptive immune system activation, and; (4) death or recovery” (Mokhtari et al., 2020, 617). However, research and reports of COVID-19 cases currently indicate that “COVID-19-associated [cardiovascular diseases] may be induced by the direct SARS-CoV-2 infection and the indirect effects of infection 41 including, cytokine storm, endothelial dysfunction, leucocytes infiltration, and formation of microvascular thrombosis,” the “occlusion of microvessels by fibrin- and/or plateletrich thrombi [blood clots]” (Mokhtari et al., 2020, 618; Pfeiler et al., 2014, S35). In the same vein, SARS-CoV-2 infection has been found to cause a variety of hematological conditions. Indeed, as Mokhtari et al. point out, “SARS-CoV-2 infection significantly affects the haematopoietic system…[and] hematological abnormalities are more frequent in severe cases vs. mild cases” with common hematological conditions being lymphopenia, leukopenia, and thrombocytopenia (Mokhtari et al., 2020, 618). Current research suggests that “lymphopenia seems to be attributed to the ACE2 presence on the lymphocytes and direct infection of these cells, whereas the elevated levels of cytokines may result in… lymphocyte apoptosis [cell death]” (Mokhtari et al., 2020, 619). Acute kidney injury has also been reported in a growing number of severe COVID-19 patients, which was also found to occur in 5-15% of patients with SARSCoV-2 and MERS (Mokhtari et al., 2020, 619). As in the case of other types of organ damage, the frequency of acute kidney injury remains under investigation, though one report found that “the incidence of AKI was reported [to be] 29% among severe cases and reached...69.57% in [the over] 60-year-old age group” (Mokhtari et al., 2020, 619). As is also the case with other types of organ damage discussed here, at the time of this writing, the exact biological mechanism(s) by which acute kidney injury may occur has yet to be determined, though some research has found that “the N-protein of SARS-CoV2 was identified in renal tubules and particles of CoV were identified in podocytes and tubular epithelium, suggesting that this virus can directly invade the renal cells,” though 42 indirect mechanisms may also play a role in “contribut[ing] to the renal injury…[including] hypoxia, hemodynamic instability and shock, sepsis, cytokine storm, and rhabdomyolysis” (Mokhtari et al., 2020, 620). Following the frequency of SARS-CoV-2-induced lung injury, a high frequency of liver injury has also been well-documented. As with some other respiratory viruses, “Acute liver failure (ALF) in COVID-19 patients may result from the virus invasion, which directly infects liver cells,” though other mechanisms may include “systematic inflammation, drug-induced damage, congestion abnormalities, [and] hypoxia-induced damage” (Mokhtari et al., 2020, 621). Some COVID-19 patients have reported diarrhea as one of their symptoms, and “RNA of SARS-CoV-2 [has been] extracted from stool and blood samples, confirming the direct infection of host cells by the virus. [Additionally,] bile duct cells with a critical role in livery recovery and immune reaction, express ACE2 higher than other liver cells, confirming that the infection of bile duct cells and not liver cells may involve the pathology of liver cells” (Mokhtari et al., 2020, 621). Finally, SARS-CoV-2 has been found to affect the nervous system and brain in some cases of COVID-19, with peripheral nervous system manifestations being more common in mild to moderate cases while central nervous system manifestations being reported in severe cases. Documented manifestations include “hypoplasia, neuralgia, fatigue, myalgia, anosmia (or hyposmia), and hypogeusia” in the peripheral nervous system and “head pain, reduced consciousness, dizziness, ataxia, acute cerebrovascular disease, and epilepsy” in the central nervous system (Mokhtari et al., 2020, 622). In the brain, researchers have reported stroke, brain hemorrhage, memory loss, delirium, as well as swelling and inflammation of brain tissues (encephalitis) and demyelination of neurons 43 (Marshall, 2020, 342). Of these, “the most common neurological effects are stroke and encephalitis. The latter can escalate to a severe form called acute disseminated encephalomyelitis, in which both the brain and spinal cord become inflamed and neurons lose their myelin coatings--leading to symptoms resembling multiple sclerosis” (Marshall, 2020, 343). The occurrence of brain and nervous system damage has been particularly interesting and puzzling to researchers for two reasons. First, these outcomes do not only occur in severe cases; young individuals without underlying conditions or similar risk factors have experienced “strokes and...acute chances in mental status that are not otherwise well-explained” (Dr. Benedict Michael, qtd. in Marshall, 2020, 343). Second, severe neurological symptoms have occurred in patients without accompanying severe respiratory symptoms. Indeed,“some of the worst-affected patients [neurologically] had only mild respiratory symptoms, and as Dr. Michael Zandi, a neurologist at the University College of London, explains, in these cases, “the brain being hit [was] their main disease” (Marshall, 2020, 343; Dr. Michael Zandi qtd. in Marshall, 2020, 343). Researchers have similarly sought to understand how these neurological outcomes occur, particularly in an effort to understand if there is direct infection of neurological tissues involved. Dr. Benedict Michael explains why understanding the biological mechanisms is crucial for clinicians: “if this is direct viral infection of the central nervous system, these are the patients we should be targeting for Remdesivir or another antiviral...whereas if the virus is not in the central nervous system, maybe the virus is clear of the body, then we need to treat with anti-inflammatory therapies” (Marshall, 2020, 343). Mokhtari et al. describe multiple hypotheses researchers have posed, 44 including infection of the brain by “transmission of SARS-CoV-2 through neural routes, such as [the] olfactory nerve,” using “neuronal retrograde to infect the [central nervous system], increasing the possibility of viral transmission via olfactory nerve endings in the nasal region,” or “a haematogenous route in which [the] viral agent utilizes the bloodstream to infect the [central nervous system, with]...infected leukocytes [acting] as a vector or infected endothelial cells of blood-brain barrier (BBB) with the ability of ACE2 expression may enhance the chance of CNS infection” (Mokhtari et al., 2020, 622). Each of these hypotheses propose mechanisms by which SARS-CoV-2 might directly infect the brain. Other proposed mechanisms include indirect effects of the cytokine storm, particularly “with BBB disruption, demyelination, and lead to [acute necrotizing encephalopathy] in viral infections. [Additionally,] coagulation abnormalities due to high inflammatory responses [have been] confirmed in patients with stroke” (Mokhtari et al., 2020, 623). SYMPTOMS OF COVID-19 As of January 2020, symptoms of COVID-19 infection were described as by Tang et al. as “very nonspecific and may be very similar to influenza, including fever, cough, fatigue, sore throat, runny nose, headache and shortness of breath, with possible ground glass showing on the chest X-ray” (Tang et al., 2020, 350). Early studies of the novel coronavirus were integral in developing the list of symptoms caused by the virus. One such study of 41 SARS-CoV-2 positive patients in January 2020 found that “their symptoms were mainly fever 98%, cough 76%, and fatigue 44%. The COVID-19 severe complications in such patients included respiratory distress syndrome 29%, RNAaemia 15%, acute cardiac injury 12%, and other secondary infections” (Ahmad & Khan, 2020, 45 1). Researchers also found that symptoms “appear to persist longer in cases of 2019nCoV infection than in most cases of uncomplicated influenza” (Tang et al., 2020, 350). However, it is important to note that in the early months of the pandemic, and continuing to the time of this writing, lists of symptoms have evolved over time and vary by source. Early lists of symptoms circulating around the United States and Utah in the first three months of 2020 focused similarly on the symptoms of cough, fever, fatigue, and shortness of breath. ABC News reported on February 4, 2020, when there were only eleven documented and confirmed cases of SARS-CoV-2 in the United States, that “symptoms of the new coronavirus are similar to pneumonia, and can range from mild symptoms, like a slight cough, to more severe symptoms, including fever and difficulty breathing” (Schumaker, 2020). Characterizations of the symptoms caused by the virus have shifted in the United States, from the initial descriptions focusing on pneumonia in China to referring to them as “flu-like” symptoms. As the United States began its testing protocols, in order to allocate the limited tests to those most likely to have the virus, the Center for Disease Control (CDC) recommended that only those “visiting the doctor with symptoms like fever, cough, or shortness of breath,” (referring to these specifically as “flu-like symptoms”) and “traveled to China within the past 14 days or might have been exposed to the virus” be tested for COVID-19 (Cranley, 2020). MILD & SEVERE SYMPTOMATIC CASES OF COVID-19 Particularly in the absence of standardized and well-researched treatments for COVID-19 infection, one important question researchers have sought to answer is how and why does the virus SARS-CoV-2 cause such different outcomes in different individuals, with some dying from the virus in a matter of days and others never 46 experiencing symptoms. Indeed, as Huand & Pranata note, “while most patients with COVID-19 have mild influenza-like illness and may be asymptomatic, a minority of patients will develop severe pneumonia, acute respiratory distress syndrome (ARDS), multi-organ failure (MOF), and death” (Huang & Pranata, 2020, 2). The symptoms experienced early on in the course of the disease might provide a clue into how the disease will progress in a particular patient. In the meta-analysis conducted by Zheng et al., patients who experienced “shortness of breath/dyspnea were more likely to develop into critical illness or even die, but patients with fever progressed better than those without fever,” while developing a headache, cough, fatigue, diarrhea, nausea, or vomiting did not show statistically significant differences among these patients (Zheng et al., 2020, e20, e19). Additionally, many of the risk factors described previously have been found to affect the progression of COVID-19 and the prognosis if contracted. In their metaanalysis of 13 studies consisting of 3027 Chinese patients, “male, aged over 65, and smoking patients might face a greater risk of developing into the critical or mortal condition and the comorbidities such as hypertension, diabetes, cardiovascular disease, or respiratory diseases could also greatly affect the prognosis of...COVID-19” (Zheng et al., 2020, e20). Song et al. conducted one of the first studies to examine the differences between mild and severe cases of COVID-19 in 41 patients with confirmed SARS-CoV-2 infection in Beijing, China between January and February 2020. While “inflammatory and immune responses are important for the elimination of the infection,...[these responses] may [also] have a significant impact on SARS-CoV-2 pathogenesis, and may play a role in the expression of the clinical spectrum of COVID-19 disease” (Song et al., 47 2020, 2). In their study, Song et al. note that “COVID-19 patients in [the] severe group are characterized by profound lymphopenia, strong T-cell activation, and increased expression of T-cell inhibitory molecules compared to [the] mild group of patients” (Song et al., 2020, 2). Lymphocytes are a crucial part of the immune system that has been found to relate to the severity of illness from COVID-19. Lymphocytes include two main types: Bcells, which “produce antibody molecules that can latch on and destroy invading viruses or bacteria,” or T-cells, which “are direct fighters of foreign invaders and also [produce] cytokines, which are biological substances that help activate other parts of the immune system” (National Human Genome Research Institute, n.d.). Lymphopenia is a condition in which a patient’s number of lymphocytes circulating in the bloodstream is reduced (Merriam-Webster, n.d.). Lymphopenia emerged as an early topic of investigation partly due to its potential to serve as a biomarker to predict severe outcomes in COVID-19 patients, particularly as resources for treatment are limited and healthcare systems are overwhelmed. In their meta-analysis, Huang & Pranata concluded that lymphopenia can serve reliably as a biomarker, as “lower lymphocyte count was associated with increased mortality, [Acute Respiratory Distress Syndrome], need for ICU care, and severe COVID-19,” though they also note that “the association between lymphopenia and severe COVID-19 was stronger in younger patients compared to older patients” (Huang & Pranata, 2020, 5). Indeed, lymphopenia has been found in multiple studies to be a differentiating factor between severe and mild cases of COVID-19. In their metaanalysis, Song et al. found that “in a previous study, mainly in noncritical patients infected with SARS-CoV-2, 35% of patients had only mild lymphopenia. In contrast, in 48 other...reports with patients with severe disease, the incidences of lymphopenia were 63% and 70.3%, respectively. Moreover, lymphopenia occurred in more than 80% of severe COVID-19 patients” (emphasis added) (Song et al., 2020, 5-6). Overall, Song et al. argue that “the degree of lymphopenia may reflect the severity of COVID-19 disease since the lymphocyte counts were slightly decreased in COVID-19 patients with mild disease, and significantly decreased in patients with severe disease” (Song et al., 2020, 6). The impact on lymphocytes is an important difference from other coronaviruses that caused significant outbreaks in the 21st century. Huang & Pranata also note that “although MERS-CoV and SARS-CoV are structurally similar, they bind to different receptors to facilitate entry. SARS-CoV attaches to angiotensin-converting enzyme 1 (ACE2) to enter the host cells, while MERS-CoV attaches to a different receptor, namely dipeptidyl peptidase4 (DPP4). Although the mechanism of significant lymphocyte reduction in severe COVID-19 remains unclear, there are hypothes[es] other than lymphocyte infiltration and sequestration in the lungs, gastrointestinal tracts, and other lymphoid tissues: (1) lymphocytes express the ACE2 receptor and may be a direct target of SARS-CoV-2 infection, and (2) an increase of pro-inflammatory cytokines in COVID19, especially IL-6, could induce further lymphocyte reduction” (Huang & Pranata, 2020, 8). Based on research as of April 2020, Zheng et al. describe the following hypothesis regarding the role of inflammation and the underlying immunological processes in severe COVID-19: Coronavirus is an enveloped, non-segmented, single-stranded RNA virus...SARSCoV-2 attacks the alveolar epithelial cells via angiotensin-converting enzyme 2 (ACE2). ACE2 is the ACE of isozyme, mainly distributed in cardiovascular, kidneys, testes, lung and colon, and other organizations. The main role of ACE2 is to incise Ang II to generate Ang 1-7, which mediates the protective effects of vasodilation, anti-inflammatory and anti-proliferation, to antagonize Ang II- 49 induced vascular smooth muscle contraction, cell proliferation, fibrosis promotion and vascular inflammation. When SARS-CoV-2 binds to ACE2 receptors on the surface of alveolar epithelial cells, the expression of ACE2 in alveolar epithelial cells is down-regulated by mechanisms such as internalization, shedding and viral replication. Then the increased concentration of Ang II leads to inflammatory response, and exudation of neutrophils, macrophages, and fibrinous, resulting in loss of pulmonary ventilation function and difficulty maintaining oxygenation. At the same time, viral infection will cause the imbalance of T helper-1 and T helper-2 responses, and induce an inflammatory storm by increasing the levels of inflammatory factors such as interleukin-4, interleukin-19 and interleukin-6. Inflammatory storm in critical patients releases cytokines, causing systemic immune injury, which may be an important cause of multiple organ failure and even death (emphasis added) (Zheng et al., 2020, e20-e21). ASYMPTOMATIC CASES: THE TWO-EDGED SWORD Asymptomatic cases represent the opposite end of the wide spectrum of COVID19. While asymptomatic cases may appear to be a positive outcome of the virus, and likely are for those who contract the virus but do not develop symptoms, epidemiologically these asymptomatic cases are highly problematic. According to Drs. Isaac Bogoch and Allan Detsky, professors in the Department of Medicine at the University of Toronto, “The amount of harm a virus can inflict on humans depends on two important characteristics: lethality and rate of transmission. Fortunately, viruses that quickly kill their human hosts don’t typically get very far outside of the settings where infected people are cared for (think Ebola). But conversely, viruses that kill only a very small proportion of their human hosts can inflict greater total harm. This is especially true for viruses like Covid-19 that cause the majority of individuals a small amount of damage, if they even present symptoms at all” (Bogoch & Detsky, 2020). One of the reasons COVID-19 is so dangerous is because of this asymptomatic spread, which Bogoch & Detsky characterize as “the ultimate weapon for a virus that ‘wants’ to inflict devastation” (Bogoch & Detsky, 2020). 50 Truly, “asymptomatic” is less a category and more a spectrum of subclinical disease, “ranging from those who are truly asymptomatic to those who have mild symptoms but don’t seek medical support. The latter group is sometimes referred to as ‘paucisymptomatic’ or ‘subclinical’ because they are below the threshold of detection by the healthcare system” (Bogoch & Detsky, 2020). This is to be distinguished from those who are asymptomatic at the time of testing and later develop symptoms, called “presymptomatic” (Bogoch & Detsky, 2020). Some researchers have found that “around 49% of people initially defined as asymptomatic go on to develop symptoms,” though this proportion also varies by study (Pollock & Lancaster, 2020, 371). Additionally, the documentation of asymptomatic cases has complicated efforts to contain the spread of the virus. As of January 2021, studies have found approximately 20% of people who test positive for COVID-19 do not develop symptoms, though the exact number varies by study and attempts to measure this proportion are “limited by heterogeneity in case definitions, incomplete symptom assessment, and inadequate retrospective and prospective follow-up of symptoms” (Pollock & Lancaster, 2020, 371). For example, in a study of 2000 individuals who had come into close contact with COVID-19 confirmed cases, 21% of them developed asymptomatic COVID-19 infection (Brüssow, 2020, 4898). Interestingly, Brüssow notes that despite their lack of symptoms, “abnormal lung radiological findings were detected in 67% of them and a third of them showed elevated C-reactive protein levels, an infection parameter. [Additionally,] viral load was similar in asymptomatic and symptomatic patients, but the asymptomatic subjects remained virus positive in the nasopharynx for longer than the symptomatic patients” (Brüssow, 2020, 4898). 51 Cases that are asymptomatic at the time of testing, either because the individual is a true “asymptomatic case” that remains subclinical throughout the course of infection, or because the cases are “preclinical” and have simply not yet developed symptoms, have contributed significantly to the spread of the novel coronavirus globally. A study among 76 residents of a skilled nursing facility in Washington state provides a window into the prevalence of these outcomes, though the age range of residents and small sample size are also important limitations to consider. In this study, following the initial introduction of the virus to the facility by a symptomatic staff member on March 1, 2020, 48 of the 76 residents (63%) tested positive for the virus in the second and third weeks of March, with “27 (56%) essentially asymptomatic [at the time of testing], although symptoms subsequently developed in 24 of these residents (within a median of 4 days) and they were reclassified as presymptomatic” (Gandhi et al., 2020, 2158). However, despite the presence, absence, or timeline of symptom onset, measurements of “quantitative SARSCoV-2 viral loads were similarly high in the four...groups (residents with typical symptoms, those with atypical symptoms, those who were presymptomatic, and those who remained asymptomatic). It is [also] notable that 17 of the 24 specimens (71%) from presymptomatic persons had viable virus by culture 1 to 6 days before the development of symptoms” (emphasis added) (Gandhi et al., 2020, 2158-9). Moghadas et al. used a low and high estimate of asymptomatic cases to model the contribution of these cases to the overall spread of the virus and provide an overall range of the percentage of overall cases they contribute. If 17.9% of cases are estimated to be asymptomatic at the time of testing, based on their modeling, Moghadas et al. found that “the presymptomatic stage and asymptomatic infections account for 48% and 3.4% of 52 transmission, respectively [but] considering a greater asymptomatic proportion of 30.8% reported in another empirical study, the presymptomatic phase and asymptomatic infections account for 47% and 6.6% of transmission, respectively” (Mohandas et al., 2020, 17513). With either estimate of asymptomatic cases, the results of this study indicate that “silent disease transmission during the presymptomatic and asymptomatic stages are responsible for more than 50% of the overall attack rate in COVID-19 outbreaks. Furthermore, such silent transmission alone can sustain outbreaks even if all symptomatic cases are immediately isolated” (emphasis added) (Mohandas et al., 2020, 17513). Indeed, “even immediate isolation of all symptomatic cases” were both possible and successful, this alone is “insufficient to achieve control” of the spread and these findings highlight the importance of symptomatic testing, but also that “symptom-based surveillance must be supplemented by rapid contact-based surveillance that can identify exposed individuals prior to their infectious period” (Mohandas et al., 2020, 17513). In their study, Johansson et al. similarly found that “across a range of plausible scenarios, at least 50% of transmission was estimated to have occurred from persons without symptoms” by using a different model and drawing from early Chinese studies to make assumptions regarding the average incubation period, infectious period, and infectiousness of asymptomatic individuals (Johansson et al., 2021, 5). Thus, while asymptomatic individuals do not require the same care as severe cases, due to their contribution to overall spread of the virus, these asymptomatic cases nevertheless place a significant burden on healthcare systems globally by the additional cases they cause. THE EMERGENCE OF LONG COVID Another outcome of COVID-19 that gained scientific recognition more recently 53 due to the forceful accounts of patients themselves is that of “long covid” or “longhaulers.” Callard & Peregro describe the evolution of this term and eventual recognition in scientific literature: “The February WHO-China Report stated ‘the median time from onset to clinical recovery for mild cases is approximately 2 weeks and is 3-6 weeks for patients with severe or critical disease…[and] many citizens were told most patients would experience mild illness and rapid recovery” (Callard & Perego, 2020, 2). However, this timeline did not match some patients’ experiences. “In March, patients started sharing experiences on social media, drawing attention to possible Covid-related sequelae. In April, newspapers started publishing first-person accounts documenting how challenging recovery from COVID-19 could be…[and] On May 5, the British Medical Journal (BMJ) published [the] account of [Paul Garner, an infectious diseases professor] suffering seven weeks through a ‘roller coaster of ill health, extreme emotions, and utter exhaustion’” (Callard & Peregro, 2020, 2). While other viruses may also cause long-term illness, many researchers agree that the lingering symptoms caused by SARS-CoV-2 are unique. “For example, infectious mononucleosis caused by Epstein-Barr virus can lead to persistent symptoms, and Guillain-Barré syndrome is a chronic neurological condition that can arise after viral infection. [However] what sets the SARS-CoV-2 virus apart is the sheer number of infected people and the damaging effects of infection on multiple organ systems, including the lungs, liver, brain, kidneys, and heart” (Nature Publishing Group, 2020, 1803). As knowledge of these long-term lingering symptoms of COVID-19 began to emerge primarily from the accounts of patients themselves who posted on social media and formed other groups online, an important question as these accounts began to surface 54 was: how common are these patients’ experience? In other words, what is the prevalence of long-covid? King’s College in London investigated this question through their app-run COVID-19 Symptom Study with the following results: “In May, we learned that ‘about one in 20 Covid patients experience long-term on-off symptoms.’ In June, the study reported ‘one in ten people may still have symptoms after three weeks, and some suffer for months’...Not until 9 July did an article confirm a high incidence of long-term symptoms, showing 87.4% of hospitalized patients reported at least one symptom 60 days after onset” (Callard & Peregro, 2020, 2). Similarly, a study in Rome, Italy found that “87% of 143 patients reported persistence of at least one symptom 60 days after onset of the disease” (Nature Publishing Group, 2020, 1803). Since this time, clinics have emerged dedicated to understanding and treating these COVID-19 “long-haulers.” While it is not yet understood how prevalent long-term COVID-19 symptoms are, one study in Wuhan, China found that of patients in the study who required hospitalization, as many as 76% of patients continued to experience symptoms six months after the onset of symptoms (Waldrop, 2021). Crucially, long-term symptoms are not restricted to patients with severe cases. One recent study of patients who had primarily mild symptoms found that 30% had symptoms that lingered nine months after contracting the virus, while other studies have found a higher percentage (Waldrop, 2021). Importantly, long-term symptoms are also not restricted to those with risk factors and preexisting health conditions. In July 2020, the CDC reported that “one in five patients 18-34 years of age without chronic medical conditions and with a positive outpatient test had not fully recovered by 2-3 weeks after testing” (Nature Publishing Group, 2020, 1803). Additionally, “physicians and therapists say they are treating people 55 of all ages and those who were extremely healthy before they got Covid--including marathon runners, athletes, and trainers” (Waldrop, 2021). Rather, some doctors have observed that “patients who were physically fit before Covid-19 infection--special operations military personnel, airline pilots and runners--tend to have more severe longterm symptoms, another mystery of the disease” (Edwards, 2021). There is also a wide array of symptoms reported by patients, that often includes “fatigue, headaches, brain fog and memory loss, gastrointestinal problems, muscle aches and heart palpitations. Some have even developed diabetes” (Waldrop, 2021). Fernándezde-las-Peñas et al. describe an expanded list of symptoms, including “neurocognitive post-COVID (brain fog, dizziness, loss of attention, confusion), autonomic post-COVID (chest pain, tachycardia, palpitations), gastrointestinal post-COVID (diarrhea, abdominal pain, vomiting), respiratory post-COVID (general fatigue, dyspnea, cough, throat pain), musculoskeletal post-COVID (myalgias, arthralgias), psychological-related post-COVID (post-traumatic stress disorder, anxiety, depression, insomnia), and other manifestations (ageusia, anosmia, parosmia, skin rashes)” (Fernández-de-las-Peñas et al., 2021, 1-2). Even this list is not exhaustive--some reports describe as many as 100 symptoms (Waldrop, 2021). Just as recognition and naming of these “long-haulers” has been driven primarily by the patients who experience lingering symptoms, patients continually add to the growing list of symptoms based on their experiences. For example, one patient contacted an ophthalmologic surgeon, saying “my eyeballs feel like they’re about to explode and I can’t move my eyes up and down…[or to] the sides” following a mild case of COVID-19 (May Kravitz qtd. in Rand & Yamada, 2021). The surgeon this patient contacted, Dr. Jeffrey Dello Russo initially thought the mysterious pain could be an early 56 sign of multiple sclerosis, but “realized later her pain could be from COVID-19. With many people around the U.S. wearing masks, but not goggles or other eye-protective gear: [Dr.] Dello Russo said that the eyes might be an overlooked way in which some people contract COVID-19” because “the whites of people’s eyes is a mucus membrane, just like you have in your mouth [and] nose” (Rand & Yamada, 2021). Epidemiologist Dr. John Brownstein emphasizes, “we know that a range of symptoms can occur with COVID. It’s not necessarily just shortness of breath and fever, especially with milder conditions” (qtd. in Rand & Yamada, 2021). Many patients with persistent symptoms have found them debilitating, leaving them unable to work or resume normal activities. For example, shortness of breath is one commonly reported lingering symptom with some patients being told “they would have to be on oxygen for the rest of their lives” while others may have “treatment that can include respiratory therapy, occupational therapy, mental health check-ins and more” (Associated Press, 2021). Dr. Greg Vanichkachorn, an occupational medicine specialist at Mayo Clinic notes that “patients are reporting that they need about a four- to five-hour nap after doing something as simple as taking their laundry up a flight of steps or taking out the trash...It can be quite severe and debilitating” (Edwards, 2021). Brain fog appears to be one of the most common long-term symptoms, which is as Dr. Allison Navis of the Icahn School of Medicine at Mount Sinai says, is “not a [technical] diagnosis, and it means many different things to many people...Oftentimes it’s a combination of short-term memory issues, concentration, or word-finding speech difficulty” and this symptom has also been observed in children and adolescents who were previously healthy before infection (qtd. in Mascarenhas, 2021). Dr. Navis continues, “for brain fog, we don’t have 57 treatments for cognitive changes, unfortunately, so [for treatments] it’s really addressing any abnormalities in blood work that could be contributing, addressing those other contributing factors like sleep and mood…[and] if attention is a major issue, medications that can help with attention might be needed” (qtd. in Mascarenhas, 2021). At present, there are “no clear diagnos[es] no standard care and no national guidelines for how these patients should be treated” (Edwards, 2021). There is also “no proven cure for long-term COVID problems. But clinics [now] aim to offer relief, not least by giving patients somewhere to turn if their usual doctor can’t help” (Associated Press, 2021). METHODS OF PREVENTION & CARE DELIVERY Previous outbreaks have shown that “combating infectious diseases disseminated by inhalation is very difficult and mainly relies on the use of vaccines” (SzkaradkiewiczKarpińska & Szkaradkiewicz, 2021, 1). However, there are additional means of preventing transmission and controlling population-wide spread that have been demonstrated to be effective prior to the development of a safe and effective vaccine. Indeed, these primary methods of preventing COVID-19 infection and transmission in the public were implemented early on, described by Cohen & Corey as “behavioral changes [that reflect] a rapid but imperfect understanding of the transmission of the virus” (Cohen & Corey, 2020, 551). These measures recommended by medical experts and public health officials are wearing a mask, social distancing, and vigilant handwashing, as well as isolation of confirmed or suspected cases and wide-scale lockdowns in communities. Like every other type of information during the COVID-19 pandemic, information about the proper uses and effectiveness of prevention measures has continued to evolve, as measures of disease prevention draw in large part on existing 58 knowledge of similar diseases (i.e., other respiratory viruses) and the known means of transmission. For example, one study found “viral RNA...from respiratory droplets and aerosols for all three viruses [coronavirus, influenza virus, and rhinovirus], including 30, 26, and 28% of respiratory droplets and 40, 35, and 56% of aerosols collected while not wearing a face mask” (Salter, 2020, 2). Overall, “estimating the risk of transmission between two individuals would be highly complex, because of the large number of variables involved. [Additionally,] such mechanistic modeling is unnecessary, because the goal is not to determine if a given individual will become ill, but to estimate how the risk of transmission may be reduced in a population” (emphasis added) (Salter, 2020, 4). In the case of SARS-CoV-2, with early recognized symptoms being primarily respiratory in nature, face masks were recommended in certain settings to prevent infection among those who may come into contact with COVID-positive individuals, in a similar way that face masks may be worn to prevent spreading influenza. Research has shown that “coughing and sneezing produce the largest respiratory droplets at 10 µm and up to 1,000 µm respectively. Breathing and speaking produce the smallest, in the ranges of 0.8-1 µm and 3.5-5.5 µm respectively” (Salter, 2020, 2). These microscopic particles can have a devastating effect. Given “an average viral load of 7 x 106 per milliliter, [it is estimated] that [only] one minute of loud speaking generates at least 1,000 virion-containing droplet nuclei that remain airborne for more than eight minutes” (Salter, 2020, 2). In the United States, recommendations from the CDC and other public health agencies changed multiple times in the early months of the pandemic, regarding who should wear a mask, what type of masks are effective, and when people should wear them. At this time, with few confirmed cases in the United States prior to March 2020, 59 “Both the World Health Organization and the Centers for Disease Control and Prevention have repeatedly said that ordinary citizens do not need to wear masks unless they are sick and coughing” (Goodnough & Sheikh, 2020). The CDC website similarly recommended in March 2020 that “the only people who need to wear a face mask are those who are sick or are caring for someone who is sick and unable to wear a mask” (Jingnan et al., 2020). This policy was reevaluated and changed weeks later as the CDC reviewed “new data that show[ed] high rates of transmission from people who are infected but show no symptoms,” early reports of asymptomatic cases, and assessed whether “there’s potential additional value for individuals that are infected or individuals that may be asymptomatically infected” (Goodnough & Sheikh, 2020). Some central questions regarding masks have been debated and thoroughly researched since the beginning of the pandemic. First, are masks beneficial for blocking the spread of virus into the surrounding air by infected individuals, both those with symptoms and those without? Second, do masks provide protection for uninfected individuals? Third, if masks are found to be effective, what types of masks are best suited for different populations and settings? Data collected since widespread mask policies have been implemented have shown that even homemade cloth masks, if made with certain materials and containing enough layers of fabric, can prevent many in the general population from becoming infected, though their protection is certainly not equal to higher-grade masks. With the benefit of hindsight and a plethora of research on the effectiveness of facemasks, it is now much easier to say clearly that “the science supports using masks, with recent studies suggesting that they...cut down the chances of both transmitting and catching the 60 coronavirus, and some studies hint that masks might reduce the severity of infection if people do contract the disease” (Peeples, 2020). While important reasons behind the CDC’s and other public health agencies’ early recommendations for the public to not wear masks was both to conserve the limited supply of masks for healthcare workers and other essential workers, as well as the initially low prevalence of cases, research demonstrating the effectiveness of certain types of cloth masks have helped to shift recommendations in favor of widespread mask-wearing. Researchers have broken down the dynamics of viral spread. While “the virus itself is only about 0.1μm in diameter...viruses don’t leave the body on their own, [meaning] a mask doesn’t need to block particles that small to be effective” (Peeples, 2020). Instead, “the moment a person breathes or talks, sneezes or coughs, a fine spray of liquid particles takes flight. Some are large--visible, even--and referred to as droplets; others are microscopic, and categorized as aerosols. Viruses including SARS-CoV-2 hitch rides on these particles [and] their size dictates their behavior” (emphasis added) (Peeples, 2020). Droplets and aerosols can range in size from 0.2μm to “hundreds of micrometers across. The majority are 1-10μm in diameter and can linger in the air a long time” (Peeples, 2020). Droplets have been the focus of most policies, as they “can shoot through the air and land on a nearby person’s eyes, nose or mouth to cause infection. But gravity quickly pulls them down. Aerosols, by contrast, can float in the air for minutes to hours, spreading through an unventilated room” (Peeples, 2020). “The evaporation of respiratory droplets is one reason that loosefitting masks reduce the risk of infection for others more than for the wearer: the concentration and diameter of respiration droplets are at a maximum as they are expelled” from an individual coughing, sneezing, breathing, or speaking (Salter, 2020, 3). 61 Thus, “it is...easier to reduce respiration droplets at their source” with mask-wearing “than to filter out their smaller and more diffuse residual droplet nuclei [from the environment] later” (Salter, 2020, 3). When discussing face masks, it is important to note that a variety of masks have been used to prevent COVID-19 infection. N95 and respirator masks used in healthcare and other industries typically provide a tight fit for wearers, while surgical masks generally fit more loosely, and cloth masks fit the loosest of all. The range in masks and face coverings that have become available during 2020, as well as the range of conditions have made face mask policies, if in place, confusing for the public to understand and difficult for medical and public health officials to clarify. For example, observational studies have found surgical and cloth masks to be 67% effective in protecting the wearer, while another study found that masks with layers of different materials--such as cotton and silk--could catch aerosols more efficiently than those made from a single material” with more layers generally being better (Peeples, 2020). Eric Westman, a clinical researcher at Duke University School of Medicine contributed to a separate study testing the effectiveness of 14 types of cloth and surgical face coverings in blocking droplets while a person spoke. “‘I was reassured that a lot of the masks we use did work,’ he says, referring to the performance of cloth and surgical masks. But thin polyester-and-spandex neck gaiters--stretchable scarves that can be pulled up over the mouth and nose--seemed to actually reduce the size of droplets being released. ‘That could be worse than wearing nothing at all,’ Westman says” (Peeples, 2020). However, it is difficult to extend the results of findings from various studies to the conditions people may experience in everyday life. Thus, “being more definitive about how well [masks] work or when to use 62 them [quickly] gets complicated. There are many types of mask, worn in a variety of environments. There are questions about people’s willingness to wear them, or wear them properly. Even the question of what kinds of study would provide definitive proof that they work is hard to answer” (Peeples, 2020). During a pandemic of an infectious respiratory virus, measures that disinfect surfaces and hands become increasingly important. What is of issue in the case of a novel virus like SARS-CoV-2 is what cleaning methods and products are effective, based on knowledge of environmental conditions, chemicals and materials, as well as the virus itself. COVID-19 “is an enveloped, positive-sense single-stranded RNA virus with the largest genomic configuration of 26-32 Kb amongst all RNA viruses….Coronavirus infection is a surface to surface, airborne, contagious disease” which may be contracted by virus in either the air or on infected surfaces; thus, both routes must be addressed (Pradhan et al., 2020, 365). The fact that the virus causing this most recent pandemic is an RNA virus is important for its ability to be transmitted between people, replicate, and evolve into distinct variants. In general, “RNA viruses mostly lack the ‘proofreading’ mechanisms of DNA viruses and hence develop many more mutations as they evolve, some of which may make the virus better able to infect a new host” (United Nations Environment Programme, 2020). While many variants of the novel coronavirus have already caused concern at the time of this writing, it has been found that “SARS-CoV-2 has a proofreading mechanism, which results in a low mutation rate compared to influenza” (Manzanares-Meza & Medina-Contreras, 2020, 262). Additionally, research has found that “depending upon the nature of the surface, pH, temperature and relative humidity of the surrounding [area], virus persistence time varies from 1-9d,” for example, 63 the virus may persist for 1 day on aluminum or surgical gloves, 3-4 days on wood, glass, and plastic, and over 5 days on ceramics, teflon, silicon, and paper (Pradhan et al., 2020, 366). For disinfectants, “in general, quaternary ammonium compounds, hydrogen peroxide, alcohol (ethanol, isopropyl alcohol, phenol), aldehyde, hypochlorous acid, octanoic acid, citric acid conjugate with silver ions, sodium hypochlorite, sodium bicarbonate, etc. are the key ingredient[s] responsible for virucidal activity. Alcohols (7895%) and isopropanol (70-100%) have been used as an effective disinfectant [against SARS-CoV-2] as they show potent virucidal activity with negligible toxic effect on human skin. All lipid enveloped virus gets inactivated within 2 minutes” (Pradhan et al., 2020, 366). Pradhan et al. also describe the intricacies of selecting a soap that is effective against SARS-CoV-2: Soap is a salt of fatty acid composed of solid fat (olive oil, coconut oil, palm oil. Rice bran, sunflower seed oils, and lard), emollient and texture enhanced (glycerine, sorbitol), surfactant (sodium lauryl sulfate alkyl benzene sulfonate), water softener (penta sodium penetrate, tetrasodium etidronate, and tetrasodium EDTA). Each soap molecule has a long tail of a hydrocarbon chain with an anionic carboxylated head. The non-polar head of the surfactant tightly binds with the phospholipid layer of the viral envelope...The cleansing property of the soap [depends] on the hydrocarbon chain length, degree of unsaturation of the fatty acid. Hydrocarbons with shorter chain length possess good lathering profiling due to enhanced water solubility. However, hydrocarbon chain length less than 10 (<C10:0) shows poor lathering profiling with objectionable odor and skin irritation. Conversely, fatty acids with longer chain length (C16: - C18:0) enhance the cleansing property with reduced lathering ability due to poor water solubility. More interestingly, fatty acids like palmitic acid, stearic acid, oleic acids, etc. provide a greater extent of protection against viruses. (Pradhan et al., 2020, 36970). The point here is not to detail the chemical properties of soap. Rather it is simply to emphasize that while cleaning and handwashing may be common sanitary measures in everyday life under normal conditions, even these activities have become more complicated during the COVID-19 pandemic, with nuances that the public may not 64 adequately understand or implement. Another important means of prevention that has been widely recommended and implemented is social distancing, maintaining a distance of 6ft (approx. 2m) from other individuals. As Salter describes, “distance is used as a proxy for time: the time for larger droplets to reach the ground or floor in the spaces among individuals, the time for smaller droplet nuclei to disperse in air currents, and the time for virus particles in droplet nuclei to become inactive” (Salter, 2020, 3). However, while 6ft social distancing has been implemented as almost a “hard-and-fast” rule across a variety of circumstances, what is less understood by the public is that “ambient conditions vary so widely that time cannot reliably be represented by distance. [For example,] aerosol droplet nuclei produced by individuals disperse in air, so their concentration decreases with distance. In most buildings, however, ventilation systems remove air contaminants relatively slowly” (Salter, 2020, 3). Indeed, in certain circumstances, 6ft social distancing may not be sufficient, as “a SARS-CoV-2 carrier person talking, sneezing, or coughing at distance of 2m can still provide pathogenic bioaerosol load with submicron particles that remain viable in air for up to three hours for exposure of healthy persons and far from the source in a stagnant environment” (Salter, 2020, 3-4). Thus, “medical science and physics strongly suggest social distancing [alone] is not a reliable barrier to transmission of COVID-19” and is best utilized in conjunction with other primary prevention measures (Salter, 2020, 4). RECOMMENDATIONS FOR PUBLIC POLICY Governments at every level have attempted to implement these measures with varying degrees of success and resistance, coming both from members of the public or 65 perhaps government leaders themselves. Certainly, while wide-scale lockdowns demonstrated benefits and reduced the spread of the virus particularly at a time when both knowledge of the virus and protective equipment necessary to protect healthcare workers were limited, such measures can only be implemented for a limited amount of time before economic and other negative impacts become devastating for individuals and communities. Thus, “behavioral changes,” such as mask-wearing, social distancing, and vigilant cleaning protocols, “to reduce SARS-CoV-2 spread must be accepted as the ‘new normal.’ The COVID-19 toolbox must include safe and effective interventions whose values have been proven through robust scientific methods honed over decades. Ongoing research in each prevention domain must [also] be sustained. We simply cannot depend on any single ‘magic bullet’” (Cohen & Corey, 2020, 551). Indeed, the measures previously discussed are most effective when used in concert with one another. The task of determining what measures to implement, and when and how to implement them is the task of public health officials making recommendations for public policy. Asymptomatic spread has been a particular issue for public health officials’ making policy recommendations. While measures such as widespread lockdowns have become politicized and have “had negative effects on the economy and many social activities,” even among experts, the best course of action and “the relative importance of mitigation measures that prevent transmission from persons without symptoms has been disputed” (Johansson et al., 2021, 2). However, as multiple studies have found that over 50% of transmission may be due to individuals who do not have symptoms, Johansson et al. note that with regard to the public policy implications of asymptomatic spread, “successful control of SARS-CoV-2 cannot rely solely identifying and isolating 66 symptomatic cases; even if implemented effectively, this strategy would be insufficient. These findings suggest that effective control also requires reducing the transmission from people with infection who do not have symptoms” (emphasis added) (Johansson et al., 2021, 7). However, quantifying the effect of asymptomatic spread within various populations remains an open question. Mask-wearing has also been an especially difficult policy to implement for a variety of reasons, ranging from resistance to mask mandates and confusion over what types of masks to wear and when to wear them to the time it naturally takes for research to find sufficient converging evidence of their effectiveness. As mentioned previously, in February and early March 2020, when cases were far less prevalent in Utah and elsewhere in the United States, leading public health authorities in these locations did not recommend widespread mask wearing among members of the public. Reasons for this included the low prevalence of cases and lack of sufficient evidence regarding the nuances of mask-wearing in public, but also attempting to save the limited supplies of personal protective equipment for healthcare and other essential workers most at risk for contracting the virus. Clearly, “questions about masks [and mask policies] go beyond biology, epidemiology, and physics” (Peeples, 2020). The Institute for Health Metrics and Evaluation at the University of Washington reported, as an article published in October 2020, that “across the United States, mask use has held steady around 50% since late July. This is a substantial increase from the 20% usage seen in March and April [2020]” (Peeples, 2020). This has been the result of evolving research and policies regarding mask wearing in the months since SARS-CoV-2 began spreading across the United States. However, this has taken significant time with 67 inconsistent messages sent by various researchers, physicians, public health officials, and government representatives. For example, “one study in April found masks to be ineffective but was retracted in July. Another, published in June, supported the use of masks before dozens of scientists wrote a letter attacking its methods. The authors [as of October 2020,] are pushing back against calls for retraction. Meanwhile, the World Health Organization (WHO) and the US Centers for Disease Control and Prevention (CDC) initially refrained from recommending widespread mask usage, in part because of some hesitancy about depleting supplies for healthcare workers. In April [2020], the CDC recommended that masks be worn when physical distancing isn’t an option; the WHO followed suit [but not until] June” (Peeples, 2020). With regard to research studies, it is important to note that the time necessary to find converging evidence across studies for something like mask wearing and all its relevant variables, sufficient to make widespread recommendations across a population may take longer than an urgent pandemic allows. At the beginning of the pandemic, “medical experts lacked good evidence on how SARS-CoV-2 spreads, and they didn’t know enough to make strong public-health recommendations about masks,” with the main question being at that time, “should members of the public both wearing basic surgical or cloth masks? If so, under what conditions? ‘Those are the things we normally [sort out] in clinical trials,’ says Kate Grabowski, an infectious-disease epidemiologist at Johns Hopkins School of Medicine in Baltimore, Maryland. ‘But we just didn’t have time for that” (Peeples, 2020). A particular sticking point among researchers, physicians, and public health officials has been the possible transmission of SARS-CoV-2 in aerosols, which, as mentioned previously, are smaller than droplets. One hypothesis argues that 68 since asymptomatic individuals have been found to spread SARS-CoV-2, this “would suggest that viruses aren’t typically [or necessarily] riding out on coughs and sneezes...So, it is worth looking at which masks can stop aerosols” (Peeples, 2020). In July 2020, aerosol scientist Donald Milton, “supported by an international group of 237 other clinicians, infectious-disease physicians, epidemiologists, engineers, and aerosol scientists, published a commentary in the journal Clinical Infectious Diseases that urges the medical community and public-health authorities to acknowledge the potential of airborne transmission. They also call for preventive measures to reduce this type of risk” (Lewis, 2020). The WHO subsequently responded with a press conference, stating, “We have to be open to this evidence and understand its implications regarding the models of transmission, and also regarding the precautions that need to be taken,” describing aerosol transmission as a possibility but calling for additional research (qtd. in Lewis, 2020). The idea of airborne transmission remains a contentious topic, partly due to the lack of research on this with other viruses. As Linsey Marr of Virginia Tech, one of the few researchers who studied airborne virus transmission prior to the COVID-19 pandemic, describes, “‘There’s this very entrenched idea’...that viruses mostly spread through droplets (short-range globs of snot and spit) rather than aerosols (smaller, dustlike flecks that travel farther). That idea dates back to the 1930s, when scientists were upending outdated notions that disease was caused by ‘bad air,’ or miasma. But the evidence that SARS-CoV-2 can spread through aerosols ‘is now overwhelming’” (Yong, 2020). Any members of the public paying attention to the evolving research on aerosol transmission have received inconsistent messages, as “for months, the WHO had steadfastly pushed back against the idea that there is a significant threat of coronavirus 69 being transmitted by aerosols that can accumulate in poorly ventilated venues and be carried on by air currents. The agency [as of July 2020] maintains that the virus is spread mainly by contaminated surfaces and by droplets bigger than aerosols that are generated by coughing, sneezing, and talking. These are thought to travel relatively short distances and drop quickly from the air” (Lewis, 2020). Indeed, the idea that SARS-CoV-2 could be spread by aerosols “is not popular with some experts because it goes against decades of thinking about respiratory infections. Since the 1930s, public-health researchers and officials have generally discounted the importance of aerosols--droplets less than 5 micrometres in diameter--in respiratory diseases such as influenza. Instead, the dominant view is that respiratory viruses are transmitted by the larger droplets, or through contact with droplets that fall on surfaces or are transferred by people’s hands. When SARSCoV-2 emerged at the end of 2019, the assumption was that it spread in the same way as other respiratory viruses and that airborne transmission was not important” (Lewis, 2020). Public health officials initially made recommendations accordingly, including frequent handwashing, physical distancing, and other means to prevent transmission through droplets, measures that have been supported by contact tracing and other studies. Studies in the virus first emerged have supported the proposition of recommending measures to prevent possible aerosol transmission. For example, one study found that “people infected with SARS-CoV-2 exhaled 1,000-100,000 copies per minute of viral RNA, a marker of the pathogen’s presence. Because the [study participants] simply breathed out, the viral RNA was probably carried in aerosols rather than in the large droplets produced during coughing, sneezing, or speaking” (Lewis, 2020). Another study found that “aerosols of SARS-CoV-2 remain infectious for longer 70 than do aerosols of some related respiratory viruses. When researchers created aerosols of the new coronavirus, the aerosols remained infectious for at least 16 hours, and had greater infectivity than did those of the coronaviruses SARS-CoV and MERS-CoV” (Lewis, 2020). Ultimately, researchers are still unsure how many virus particles are needed to trigger an infection, and under what conditions the necessary transmission might occur, contributing to inconsistent advice from experts. Overall, there are many important considerations when public health officials attempt to apply research findings to make recommendations to the public. Certainly the issue of supplies is important. Issues of health literacy, information processing, risk, and other aspects of the population itself are also important and will be discussed more fully below. However, it must also be noted that scientists operate using the scientific method, including hypothesizing, launching studies to test hypotheses, and attempting to find converging evidence, all of which takes time and is open to criticism, debate among scholars, and studies with different findings. While this may be viewed by the public as inconsistency leading them to doubt the evidence undergirding public health recommendations, this is an important part of the scientific process. Additionally, the standard of evidence to support a conclusion in science may be different than the evidence required for asking or requiring members of the public to change their behavior, and the amount and type of evidence required for a public health official or agency to recommend certain measures is open to the subjective judgment of the individuals in positions of authority. For example, Julian Tang, a virologist at the University of Leicester, United Kingdom, who contributed to the July 2020 commentary in Clinical Infectious Diseases, “says that the bar of proof is too high [required by the WHO] 71 regarding airborne transmission. ‘[The WHO] ask for proof to show it’s airborne, knowing that it’s very hard to get proof that it’s airborne...In fact, the airbornetransmission evidence is so good now, it’s much better than contact or droplet evidence for which they’re saying wash [your] hands to everybody” (qtd. in Lewis, 2020). With this in mind, some important conclusions must be drawn regarding public health policies. First, there is a wide variety of figures and agencies that might be regarded by the public as equally authoritative. While inconsistencies and disagreements between them may be viewed negatively and unpersuasive by members of the public, it is nevertheless the basis of the scientific process through which they work. Second, the standard of evidence for supporting a conclusion in the scientific community differs from the standard of evidence public health officials may require to recommend a policy or behavior change, the latter of which may be influenced by the subjective judgment of public health officials based on their individual education and background. 72 WHAT MAKES THE COVID-19 PANDEMIC DIFFERENT Although the COVID-19 pandemic is one of many notable epidemiological events in the course of human history, one element that makes this pandemic different is the availability and prevalence of modern technology in every domain including medicine, entertainment, communication, and many others. Such technology has allowed for largescale closures of communities without completely halting all activities, like education and businesses, to the extent they are able to conduct business activities using technology. As is the focus of this paper, modern technology has notably created new modes of communication, including social media. Government and public health officials have utilized popular and well-established social media platforms such as Twitter, Facebook, and Instagram to inform audiences and seek to increase compliance with recommended health protective behaviors. In contrast to traditional forms of communication however, social media is characterized by its interactive features, allowing for a system of nearly instantaneous response and feedback from viewers and followers. According to Chen et al., “social media, due to its openness, dialogism, and participatory nature offers significant benefits in delivering synchronous and interactive communication between governments and citizens, bringing new impetus to citizen engagement” (Chen et al., 2020, 1). Thus, social media’s capabilities span far beyond simple, one-sided information dissemination and government and public health officials can therefore use social media to not only inform citizens, but also engage with them in an interactive format that traditional channels of broadcast do not allow. 73 COVID-19 ON TWITTER: WHAT RESEARCH HAS FOUND SO FAR “We’re not just fighting an epidemic; we’re fighting an infodemic.” ~Tedros Adhanom Ghebreyesus (Munich Security Council, February 15, 2020), Director General of the World Health Organization (United Nations Department of Global Communications, 2020) Researchers have already begun to address COVID-19 in fields beyond medicine and the role of social media has been a popular topic of investigation. While the COVID19 pandemic is the first worldwide health crisis of its kind to occur since the advent of the Internet, other smaller-scale outbreaks have also shown an alarming trend with regard to social media. Shahi et al. note that “Misinformation on COVID-19 appears to be spreading rapidly on social media. Similar trends were seen during other epidemics, such as the recent Ebola, yellow fever, and Zika outbreaks” (Shahi et al., 2020, 1). Additionally, Nguyen & Catalan-Matamoros argue that “digital platforms--with their omnipresent algorithm and ability to afford emotional support and bias confirmation-make it [especially] easy for mis/disinformation to travel and to engender ill-formed public debates and dangerous situations” (Nguyen & Catalan-Matamoros, 2020, 324). Misinformation about the COVID-19 pandemic was already a significant problem as early as February 2020, when Tedros Adhanom Ghebreyesus, Director General of the World Health Organization stated at the Munich Security Council, “We’re not just fighting an epidemic; we’re fighting an infodemic” (United Nations Department of Global Communications, 2020). One approach researchers have taken to understand the spread of COVID-19 information on social media is to use the same models used to track the spread of the disease itself. Using this method, researchers found “that the basic reproductive number 74 R0, i.e. the number of infections due to one infected individual for a given period, is between 4.0 to 5.1 on Twitter, indicating a high level of ‘virality’ of COVID-19 information in general” (Shahi et al., 2020, 3). While information about COVID-19 spreads quickly on Twitter in general, the spread of partially or completely false information must be of particular concern to government and public health officials. As will be discussed below, for a variety of reasons, information and misinformation has the potential to shape attitudes, beliefs, mental models, and behavior with respect to COVID19. In their analysis Shahi et al. found that “the speed of propagation is higher for the false category and it was the highest during the peak time of tweet (time duration from the beginning to the day tweet not getting [a] new retweet),” that English language tweets constituted nearly 50% of tweets with misinformation, and that completely false tweets saw a large increase in the middle of March 2020 (Shahi et al., 2020, 10, 7, 10). Their analysis also revealed important trends in the nature of COVID-19 tweets containing information: First, we find that misinformation more often concerns discrediting information circulating on social media...Second, compared to general COVID-19 tweets, completely false misinformation more often mentions governing bodies related to health (‘world health organisation’ and ‘ministry of health’) and their communication with the outside world (‘medium briefing’). Conversely, partially false misinformation appears more concerned with human-to-human transmission, mortality rates and running updates on a situation (‘latest information’ and ‘situation report’) than the average COVID-19 tweet (Shahi et al., 2020, 12). Clearly, the efforts of government and public health officials to inform the public and encourage compliance with health protective behaviors are under attack from the mass of tweets containing entirely or partially false information, whether or not that information is spread with malicious intent. The consequences of widespread misinformation in the case of a pandemic can be 75 dire, especially since, as Shahi et al. argue, “the actions of individual citizens guided by the quality of the information they have at hand are critical to the global response to this public crisis” (Shahi et al., 2020, 1). Indeed, researchers have already described COVID19 misinformation on social media as a “tipping point” in the trend of general misinformation and conspiracy theories promoted by social media (Nguyen & CatalanMatamoros, 2020, 325). Nguyen & Catalan-Matamoros provide a telling example of the extent to which misinformation has already become firmly-held attitudes and influenced behavior: Anyone with basic school education and in their right mind would be able to laugh at the bizarre idea of a biological virus spreading through mobile phone networks. Things might become a little more complicated with the claim that radiation from such networks suppresses the immune system against the virus, but it takes only a few clicks to find a reputable health advice source to refute it. Yet, as the novel coronavirus takes hold and wreaks havoc across the world, these two unfounded claims have been able to convince many people to break lockdown rules, pouring onto the street to smash and torch hundreds of 5G phone masts in many countries--from Australia and New Zealand to the UK, Ireland, Finland, Sweden, Belgium, the Netherlands and Italy (Nguyen & Catalan-Matamoros, 2020, 323). While this is an extreme example, it nevertheless demonstrates the power and effects misinformation perpetuated online can have, and how such misinformation can persist even in the face of credible scientific information that refutes it. Further, these conspiracy theories are essentially “recycled” and placed in the context of the COVID-19 pandemic. Nguyen & Catalan-Matamoros note that “the 5G mast attacks, for instance, are just the latest escalation of the anti-5G activist movement that has been spearheaded by Stop 5G groups around the world [and] the immune-system suppression claim that leads to recent vandalism is just an extension of the basic theory that anti-5G groups have promoted for years” (Nguyen & Catalan-Matamoros, 2020, 324). However, “despite being repeatedly 76 discredited and dismissed by national and international health authorities, such claims have featured in every recent international outbreak--such as SARS (2002-2004), H1N1 (2009-2010), MERS (2012-2013), Ebola (2014-2015) and Zika (2015)--and have shown no sign of stopping their contagion soon” (Nguyen & Catalan-Matamoros, 2020, 324). Therefore, understanding the nature of social media, its users, and content, as well as the psychological processes that underlie the use of social media and the processing of information, is crucial for government and public health officials to make their communication efforts more effective. Understanding the nature of the platform alone is insufficient in combating ‘infodemics’ in the future, as “it would be vastly oversimplified...to attribute everything to digital technologies...The fundamental issue remains that many people are still willing to believe things that, by normal intellectual standards, are unmistakably unscientific or counterintuitive” (Nguyen & CatalanMatamoros, 2020, 324). In this vein, the following discussion will seek to elucidate some of the reasons misinformation on social media can be so damaging to the efforts of government and public health officials in light of the COVID-19 pandemic. After this discussion, analysis of the Twitter accounts of government and public health officials in Utah will follow. 77 PSYCHOLOGICAL CONCEPTS RELEVANT TO THE PROCESSING OF INFORMATION Throughout the COVID-19 pandemic response, government and public health officials have continually asked the public to comply with health protective behaviors as recommended by medical experts and emerging research, all to varying levels of success. This clearly shows that while communication has provided abundant information about the desired and effective health protective behaviors that can reduce risk of infection for healthy individuals and transmission from those already infected with the virus, further evaluation of how this communication has taken place is warranted. Recommended health protective behaviors include wearing a face covering, maintaining “physical distancing,” and increasing handwashing, among others. Because this is a significant deviation from many citizens’ typical behaviors prior to the COVID-19 pandemic, and because there has been significant resistance toward these measures, communication between government and public health officials and the general public must, at least in part, seek to persuade the public to engage in these behaviors. While government and public health officials use social media to communicate with the public, they certainly do not control the discourse on public health topics. Making public health recommendations requires officials to understand a wealth of scientific literature and research, which may be continuously evolving as in the case of the novel coronavirus, then communicate portions of that research and its implications to a population with varying levels of scientific literacy and a range of personal views. Further, as Endsley notes, “in addition to the benign misinformation that litters the Internet, there has been a growth of deliberate information attacks” (Endsley, 2018, 1081). While Endsley’s analysis focuses on the political sphere, the social media 78 landscape can easily become just as contentious towards public health recommendations that members of the population find disagreeable or conflict with their own views. In addition to the accounts of human Twitter users, as Broniatowski et al. argue, on a topic such as vaccination, public health officials must also combat a range of bots, trolls, content polluters, and other accounts promoting malicious agendas using a range of tactics (Broniatowski et al., 2018). Based on their analysis, Broniatowski et al. found that, instead of simply flooding Twitter with polarizing messages against vaccination, “Russian trolls and sophisticated bots promote both pro- and anti-vaccination narratives. This behavior is consistent with a strategy of promoting political discord” (Broniatowski et al., 2018, 1383). This strategy has the potential to undermine efforts by [US] government and public health officials as these tweets “may lead the public to question long-standing scientific consensus regarding vaccine efficacy” (Broniatowski et al., 2018, 1382). Content polluters, by contrast, more typically employ a strategy of flooding hashtags and topics with targeted messages. These accounts were found to tweet messages in opposition to vaccination 75% more often than average non-bot accounts (Broniatowski et al., 2018, 1382). Finally, a separate group of accounts these researchers were unable to definitively identify as bot- or troll-operated were found to tweet “content that was both more polarized and more opposed to vaccination than is that of the average non-bot account” which may suggest to other Twitter users “the false impression of grassroots debate regarding vaccine efficacy--a technique known as ‘astroturfing’” (Broniatowski et al., 2018, 1382). These tactics create a significant obstacle for government and public health officials that is unique to social media’s interactive nature both because of how malicious 79 accounts tailor their content and because of how social media users consume information and use the platform. First, Broniatowski et al. note that much of the malicious activity by these types of accounts uses increasingly complicated tactics to both avoid detection and by including “both neutral and divisive content…[the accounts attract] followers as they present messages that are perceived to be fair and confirm their beliefs” (Sutton, 2018, 1282). As these accounts remain on Twitter promoting their content, their sophisticated appeals are also designed to avoid detection by human users consuming their content. A 2018 study estimated that 9%-15% of Twitter accounts are bots, excluding trolls and other accounts operated primarily by humans (Endsley, 2018, 1084). With the plethora of information available through technology today, many humans have great difficulty in detecting accurate, factual information, due in part to errors in judging the reliability of the source of the information. According to one 2017 poll, “only 27% of Americans are very confident that they can tell when a news source is reporting factual news versus commentary or opinion” (Endsley, 2018, 1086). Additionally, as human users fail to detect malicious accounts, this content has the potential to spread widely within social networks, despite, and perhaps because of, its falsity. Based on a 2016 study, “false news stories are accessed through social media 42% of the time whereas actual news sites are accessed only 10% of the time,” 23% of individuals surveyed reported sharing false news, either knowingly or unknowingly, and that “false news stories are 70% more likely to be retweeted by people on Twitter than regular news stories” (Endsley, 2018, 1084). Second, with regard to human users’ tendencies on social media, as Sutton notes, “Many individuals rely on their social media connections to curate content by identifying and editorializing newsworthy information that conforms to their own personal biases [and] 80 in doing so, [they] eliminate dissenting voices and insulate [them]selves from recognizing when a filter bubble might have become infected” (Sutton, 2018, 1282). Clearly, rumors, misinformation, and conspiracy theories pose a significant obstacle to government and public health officials wishing to inform the public. They must not only persuade a large population to practice behaviors that are outside their normal behavior, but do so in the face of increasingly sophisticated attacks on their efforts orchestrated by both humans and computers. It is therefore useful to examine the psychological processes that underlie social media use and consumption of information. To do so, Endley identifies six major cognitive processes that intersect with social media use: attention, anchoring, confirmation bias, perception and confidence level, comprehension and mental models, and projection, goal, and motivation. These and other relevant concepts are discussed in the following sections. ATTENTION Before information can be processed, an individual must first allocate attention to it. In the internet age, information is abundant, requiring individuals to select some information over others, due to the limits in both human attention and time. As Van der Meer et al. explain, “the ubiquity of news offers audiences the opportunity to actively personalize and shape their individual information environment” (Van der Meer et al., 2020, 937). In a world with so much information available for consumption, false information may have an advantage over factual news when it comes to attention, as “online information attacks often utilize salient cues, such as outrageous or incendiary headlines, as ‘click bait’” (Endsley, 2018, 1084). Attention may also become selective, based on preexisting attitudes and biases. Multiple studies have found that many 81 consumers of media “primarily expose themselves to news and sources that reinforce existing...beliefs to exclude attitude-discrepant messages, rely on channels that are part of their habitual media diet, or self-select into information that is merely entertaining” (Van der Meer et al., 2020, 938). Attention is therefore a crucial consideration for government and public health officials wishing to use social media to communicate with the public because attention has a great effect on the information social media users choose to expose themselves to at the beginning of information processing. If members of the population selectively attend to messages that affirm their preexisting attitudes and these attitudes conflict with the messages of authorities in the community, they may refuse to attend to messages from these authorities in the future. Fundamentally, if the messages of government and public health officials never reach the intended recipients, they cannot be effective. The nature of many social media platforms, beyond simply the vast amount of content available, also has significant impacts on attention and sharing behavior. Drawing on the inattention-based account of misinformation on social media, Pennycook et al. argue that “people generally wish to avoid spreading misinformation...however, they nonetheless share false and misleading content because the social media context focuses their attention on factors other than accuracy...As a result, they get distracted from even considering accuracy when deciding whether to share news” (Pennycook et al., 2020, 771). In a study on COVID-19 misinformation conducted in March 2020, Pennycook et al. found that by directing attention with subtle primes about accuracy, participants were better able to detect true, factual news stories. In one of their experiments, researchers asked participants a question about the accuracy of an unrelated story prior to being 82 presented with a COVID-19-related news story: “This minimal, content-neutral intervention nearly tripled participants’ level of discernment between sharing true and sharing false headlines” (Pennycook et al., 2020, 777). Based on these results and findings from other research, Pennycook et al. argue that “it seems likely that people are being distracted from accuracy by more fundamental aspects of the social media context…[for example,] attention may by default be focused on other factors, such as concerns about social validation and reinforcement rather than accuracy. Another possibility is that because news content is intermixed with content in which accuracy is not relevant (e.g. baby photos, animal videos), people may habituate [or, become used to] a lower level of accuracy consideration when in the social media context” (Pennycook et al., 2020, 777). Despite these obstacles, as Pennycook et al. demonstrated, guiding social media users’ attention to consider accuracy when sharing COVID-19-related messages may be an effective way of reducing the spread of misinformation. ANCHORING Anchoring is a well-documented heuristic in psychology in which individuals attempting to evaluate information or make a decision begin with an initial piece of information, the “anchor,” with the intention of adjusting as new information is gathered (Baumeister & Vohs, 2007, 35). However, as has also been documented in a variety of studies, “people typically fail to adjust sufficiently. That is, the initial value” or information “exerts some ‘drag’ on the final estimate, systematically biasing the result” (Baumeister & Vohs, 2007, 35). In other words, as Endsley describes, information encountered early can influence the development of mental models, a person’s representation of a phenomenon in the surrounding world, which can later “have a 83 significant effect on what information is later attended to or believed [and] if the early information is erroneous, this can have a significant negative effect on later information processing” (Endsley, 2018, 1085). Anchoring requires that government and public health officials carefully consider the messages they send in the beginning of new and evolving situations like the COVID-19 pandemic, as these messages may influence the formation of mental models for members of the population to a much greater extent than later messages. CONFIRMATION BIAS While anchoring and selective attention have specific application to the earliest messages on a topic, confirmation bias must remain a consideration for government and public health officials throughout their communication efforts. In social psychology, confirmation bias is defined as “processing information by looking for, or interpreting, information that is consistent with one’s existing beliefs [resulting in a]...biased approach to decision making [that] is largely unintentional and often results in ignoring inconsistent information” (Baumeister & Vohs, 2007, 162). Confirmation bias is therefore a driving force behind the search for, selection, and integration of additional information into existing beliefs and mental models. Meppelink et al. argue that confirmation bias presents a unique challenge in the Internet age because “Online information is not centrally controlled and therefore differs from many other information sources...everyone on the internet can be an author, with or without medical qualifications, which means that lay-people can also easily spread information on a very large scale” (Meppelink et al., 2019, 129). With regard to evaluating information, many studies have found that “people are not always motivated to critically verify the reliability 84 and accuracy of online information. Instead, people rely on subjective characteristics to determine whether a certain [source] provides reliable information. One important evaluation [criterion] involves the degree to which the information confirms people’s existing ideas or beliefs” (Meppelink et al., 2019, 130). In their 2017 study of parents’ attitudes towards childhood vaccination, for example, Meppelink et al. found that when presented with a series of messages, people “tend to selectively expose themselves to information that confirms their prior beliefs…[and] this confirmation bias was found not only with respect to message selection but also with respect to message evaluation. People [also] consistently perceived belief-confirming information as being more credible, useful, and convincing,” regardless of which side of the issue they supported (emphasis added) (Meppelink et al., 2019, 135). Thus, confirmation bias may act as not only a filtering mechanism to influence what information an individual chooses to consume, but also as a critical evaluation criterion to determine the extent to which the information should be integrated into existing attitudes, beliefs, and models. With health information specifically, the effect of confirmation bias may be influenced by the degree of an individual’s health literacy, defined as “the ability to obtain, process, understand, and communicate about health-related information needed to make informed health decisions” (Berkman, Davis, & McCormack, 2010 qtd. in Meppelink et al., 2019, 130). The same 2017 study also found that “confirmation bias in message selection was stronger among parents with higher [health literacy]” (Meppelink et al., 2019, 135). PERCEIVED CONFIDENCE While the degree to which information confirms an individual’s existing beliefs may be one criterion for evaluating information, the perceived level of confidence in the 85 source of the information is another. As Endsley notes, perceived confidence in a source is generally determined by “the reliability of its source, the presence of incongruent or conflicting data, missing information, the timeliness or latency associated with the data, and the presence of noisy or ambiguous signals” (Endsley, 2018, 1085). Smith et al. demonstrated that people evaluating persuasive messages in a consumer context also make judgments on source characteristics such as credibility, expertise, and trustworthiness that affect both their explicit, or stated, and implicit, “evaluations [that] are assumed to reflect associations that are activated in memory at any point in time” (Smith et al., 2012, 200). In other areas, ranging from eyewitness testimony to financial advising, the effect of this “confidence heuristic is so powerful that it can overwhelm the role of judgmental accuracy” (Karmarkar & Tormala, 2010, 1034). Across their five studies, Smith et al. found that “implicit evaluations were more strongly affected by a source high in credibility as operationalized through expertise” when participants were presented with identical messages (Smith et al., 2012, 200). However, Karmarkar & Tormala demonstrate that positive source characteristics do not necessarily mean greater persuasion in every situation. They found that participants were more persuaded by a message “when an expert source expressed low rather than high certainty in his recommendation” and similarly, “when a nonexpert source expressed high rather than low certainty in his recommendation.” (emphasis added) (Karmarkar & Tormala, 2010, 1041). Source characteristics are likely not evaluated in isolation, however. Karmarkar & Tormala found in a separate study “a three-way interaction among source expertise, source certainty, and argument quality on attitudes and intentions as well as though 86 favorability” (Karmarkar & Tormala, 2010, 1044). In this instance, researchers believe that, in line with the informational incongruity perspective, “incongruent expertise and certainty information would violate [audience] expectancies, stimulate [audience] involvement, and foster persuasion as long as message arguments were strong...when message arguments were weak, this hypothesis suggested that incongruity could either undermine or reverse the impact of a message on persuasion” (emphasis added) (Karmarkar & Tormala, 2010, 1044). In other words, unless the persuasive environment is carefully considered, positive source characteristics such as expertise and confidence may backfire. It is important to highlight that “source incongruity does not invariably increase persuasion. Rather, it increases elaboration, which can boost or undermine persuasion depending on argument quality” (emphasis added) (Karmarkar & Tormala, 2010, 1045). Thus, understanding how both implicit and explicit attitudes may be influenced by characteristics of the source of a persuasive message, and how evaluation of those source characteristics interacts with the environment, is vital, as both implicit and explicit attitudes “predict variation in behavior that is not accounted for by the other” (Smith et al., 2012, 194). The notion of perceived level of confidence extends beyond confidence in the source, as people generate thoughts when confronted with a persuasive message and assess their level of confidence in the thoughts themselves. These thoughts may then influence their perception of both the source and the message, thereby impacting the overall effectiveness of the persuasive attempt. Multiple studies have found that “thoughts held with high confidence” by the recipient or audience of the message “have a strong impact on attitudes, whereas thoughts held with low confidence do not” (Tormala 87 et al., 2006, 685). In essence, “thought confidence can increase or decrease persuasion depending on what thoughts are elicited by a message” (Tormala et al., 2006, 685). As in the case of the source characteristics described previously, the patterns of effective persuasion depend on multiple factors working together. With regard to the thoughts the recipient of the message generates, “when thoughts are primarily favorable, increasing confidence in their validity increases persuasion and increasing doubt about their validity decreases persuasion. When thoughts are primarily unfavorable, however, increasing confidence in their validity decreases persuasion and increasing doubt about their validity increases persuasion” (Tormala et al., 2006, 685). Source credibility also intersects with thought confidence and thought valence. In a 2002 study, Briñol et al. found that “source credibility only affects thought confidence when people have already processed the message and generated their thoughts” (Tormala et al., 2006, 685). Additionally, the ability to elaborate, or deeply think about and process the message beyond the information provided, also interacts with these factors. People differ in their natural inclination to elaborate when presented with persuasive messages, called need for cognition. According to a separate study by Briñol et al., “thought confidence explained source credibility effects among high but not low need for cognition individuals. Thus, source effects are only guided by thought confidence when people have the motivation and ability to think about their thoughts and gauge their confidence in them” (emphasis added) (Tormala et al., 2006, 685). MENTAL MODELS As has been well-documented in cognitive psychology, humans do not typically process new information in isolation. Rather, “individuals must interpret new information 88 together with other preexisting knowledge to form an internally coherent understanding of an issue,” concept, or idea (Endsley, 2018, 1086). While the concept of mental models extends to fields ranging from psychology, system dynamics, cognitive science, and others, definitions across multiple fields agree that a mental model is “a concentrated, personally constructed, internal conception of external phenomena (historical, existing or projected), or experience that affects how a person acts” (Rook, 2013, 42). Crucially, a mental model is therefore constructed by an individual, thereby subject to the individual’s experiences, prior knowledge, existing attitudes, and biases (Rook, 2013, 40). Scholars also agree that mental models are constructed to represent phenomena in the surrounding world with the intent of guiding behavior, an intervening force between stimulus and response (Rook, 2013, 41). While early information forms the foundation of a mental model, new information may also be integrated, though this process is often influenced by existing beliefs. Because mental models affect both an individual’s understanding of the world and their behavior within it, it is important for government and public health officials to consider how their actions contribute to the construction of mental models for members of the population. Additionally, many members of the general public are unfamiliar with scientific and medical concepts, such as the dynamics of infectious diseases or the calculation of epidemiological measures, so public health officials have an opportunity to guide the population’s construction of mental models for relevant concepts. When new information “is in agreement with already existing mental models, this is fairly effortless. When it is in conflict [with existing mental models], people are more likely to discount the new information or explain it away in order to retain the existing 89 (although possibly incorrect) mental model” (Endsley, 2018, 1086). Misinformation, particularly if presented early on, may have a significant impact on the formation of mental models, whether that information is presented with malicious intent, as in the case of social media accounts run by bots or trolls, or the simple case of emerging research. In the case of malicious misinformation, presenting information early on may create an advantage because “by creating preexisting, but incorrect, mental models of a subject, it is possible to induce people to ignore conflicting information and seek out confirmatory information based on anchoring and confirmation biases” (Endsley, 2018, 1086). In other words, “people find ways to explain away conflicting data that disagrees with their existing mental model, either consciously or unconsciously. [Further,] this cognitive dissonance can be quite strong and difficult to overcome” (Endsley, 2018, 1086). Given that medical information ranges in its complexity, public health officials must be able to communicate the essential information, provide the necessary context, and support the development of sound mental models in populations with varying levels of education and health literacy. While literacy is certainly related to intelligence and education, health literacy is a distinct construct, describing “a constellation of abilities…[including], in addition to literacy and numeracy, rhetorical discourse (effective speaking, listening, and writing), the ability to use technology (particularly the Web), motivation, cognitive ability, and networking and social skills” (Berkman et al., 2010, 12). Based on their review of the evolution of the concept of health literacy, Berkman et al. define health literacy as “the degree to which individuals can obtain, process, understand, and communicate about health-related information needed to make informed health decisions” (Berkman et al., 2010, 16). Despite the fact that illiteracy is 90 relatively uncommon in the United States, in 1993 and again in 2005, the National Center for Education Statistics found that “approximately half of US adults lack the literacy skills needed for full participation in the current economy and for the demands of everyday life in an industrialized nation” (Rudd, 2007, 8). With regard to health literacy in particular, the National Center for Education Statistics announced in 2006 that “53% of US adults have health literacy scores in the intermediate range, a category indicating needed skill building. Fourteen percent of adults scored at below basic level; an additional 22% scored at basic level; and 12% scored in the proficient level” (Rudd, 2007, 16). Despite this, “findings reveal that health materials are generally written at levels of complexity far beyond the reading skills of average high-school graduates…[A strong] body of literature firmly establishes that access to health information is hampered by unnecessary complexities such as unusual words, long sentences, poor organization, or didactic style” (Rudd, 2010, 2283). Another branch of research hints that some of these difficulties in communication stem from the “faulty assumptions about the public’s skills and the expectations health systems have of their users” (Rudd, 2010, 2283). Therefore, it is essential, especially for public health officials and any clinicians or scientists communicating with the population, to remain aware of the limitations of health literacy and public health literacy among members of the population. Rudd argues that “adults trying to apply health information would benefit from clear and straightforward written and oral communication and from improvement in the design of charts and graphs...Materials [should be] designed from the perspective of the user” (Rudd, 2007, 17). While the term “health literacy” is generally thought to be an individual-level 91 construct, a related term, “public health literacy,” describes “the knowledge, skills, and engagement that groups of individuals have to address the public health of their community” (Berkman et al., 2010, 15). Health literacy is increasingly considered a dynamic, rather than static, construct. That is, “individuals’ health literacy can change as they gain experience with the various health circumstances and choices that they face” (Berkman et al., 2010, 17). The dynamic nature of health literacy may provide public health officials an opportunity to communicate more effectively with the general public over a period of time, as in the case of the COVID-19 pandemic, as members of the population become better acquainted with the relevant medical terms, definitions, and concepts. However, the population can only become knowledgeable about such information if they process the messages of public health officials. Thus, public health officials must keep the psychological biases and heuristics in mind when crafting their public messages. Both individual and population-level health literacy are important considerations for government and public health officials during a pandemic of a highlyinfectious virus such as COVID-19 because the actions of an individual inevitably affect not only themselves, but also their family, friends, coworkers, acquaintances, as well as strangers throughout the population. Both constructs are essential and interrelated: “public health literacy is complementary to individual health literacy and outcomes include a community’s understanding of public health messages as well as having the skills to evaluate and participate in civic action related to health care issues” (Berkman et al., 2010, 15). PROJECTION & PREDICTION Making predictions about the future is an important task under normal conditions 92 and is made more difficult under conditions of uncertainty, such as in the case of the COVID-19 pandemic. According to Bar, “Rather than passively ‘waiting’ to be activated by sensations...the human brain [may be] busy generating predictions that approximate the relevant future” (Bar, 2007, 280). Bar’s view links three concepts that have been thoroughly studied in isolation in both psychology and neuroscience: “The first is associations, which are formed by a lifetime of extracting repeating patterns and statistical regularities from our environment and storing them in memory. The second is the concept of analogies, whereby we seek correspondence between a novel input and existing representations in memory (e.g. ‘what does this look like?’). Finally, these analogies activate associated representations that translate into predictions” (emphasis added) (Bar, 2007, 280). This process is enabled by the fact that these types of memories are stored as a network related not through chronology but through the meaning of related information: “for example, objects that tend to appear together are linked on some level, and these representations include properties that are inherent to and typical of that same experience,” a concept called “context frames” (Bar, 2007, 280). Additionally, when a concept in this network is activated, that activation spreads to related concepts, which Bar argues, are used in the process of making predictions: “the structure of these context frames enables co-activations that prime our subsequent perception, cognition, and action by remaining ‘on-line’ and making available predictions of what to expect in the immediate environment” (Bar, 2007, 280). Based on studies of brain networks that are activated when participants are not engaged in a specific task, Bar notes that this “associative activation is an integral process of the brain’s mental ‘default’ mode” (Bar, 2007, 281). Following activation, analogies “map novel inputs to representations in 93 memory that most resemble this input. Subsequently, information associated with these representations is activated to provide predictions about what else might be expected in the same situation. [However,] by taking context into account in this associative activation, only the most relevant predictions are generated” (Bar, 2007, 283). Bar argues that this process uses essentially the same mechanisms regardless of whether the experience to be predicted is relatively simple or extremely complex: “one can anticipate complicated experiences, plan far ahead, or mentally ‘travel’ in time to the future, based on simulations and memory…[and] complex executive predictions result from the integration of multiple smaller analogies and ‘atomic’ associations” rather than a single analogy used for simpler experiences (Bar, 2007, 285). Of course, predictions based on associative activation and analogies from past experiences are not always correct, and these “errors” made in prediction are also taken into account, both in order to adjust to the current context and modify the framework to make more accurate predictions in the future (Bar, 2007, 285). The process of prediction relates to information processing in important ways. If the information is erroneous, as Endsley notes, “information attacks can...be directed at creating inaccurate projections of the likely future” (Endsley, 2018, 1086). However, regardless of the validity of the information, humans generally “seem to minimize processing of incoming information when this information is predictable (e.g. habituation, repetition blindness, change blindness, inhibition of return), and in parallel encourage the allocation of mental resources to unexpected and/or novel information (e.g. orienting response toward novel and unexpected stimuli” (Bar, 2007, 287). 94 REACTANCE Psychological reactance theory provides important insight into how an audience might respond to a persuasive message. According to this theory, “when faced with messages that limit their freedom, people experience anger and negative cognitions that motivate them to restore that freedom…[and] research indicates that people primarily attempt to restore freedom by participating in the discouraged act” (Richards & Banas, 2015, 451-2). Reactance is of particular concern in health communication and promotion due to the nature of most messages: “campaigns often directly discourage unhealthful behaviors, and those that encourage healthful behaviors, by implication, simultaneously discourage unhealthful ones. This discouragement may be perceived as freedom threatening, thereby eliciting unhealthful behavior aimed toward freedom restoration” (Richards & Banas, 2015, 451). As a result, Richards & Banas argue that “many public health campaigns have been ineffective in causing change that results in healthful behaviors [and] in some instances, health messages even cause the behaviors they are trying to prevent” (Richards & Banas, 2015, 451). However, research also indicates that when health promotion campaigns spark reactance in an audience, “it is [often] not a product of a poorly constructed public health campaign. It is [instead] plausible that individuals cognitively and affectively process similar messages in different ways. That is, a message that motivates some people to respond in ways beneficial to their health may motivate others to react in ways that are harmful to their health” (Richards & Banas, 2015, 452). While the concept of reactance tends to focus on reactions to persuasive messages, Richards & Banas note that health behaviors are a topic particularly vulnerable to the threat of reactance: “all promotional health messages, even those that do not overly 95 threaten freedom, are capable of causing reactance that results in unhealthful behaviors aimed toward the restoration of freedom” (emphasis added) Richards & Banas, 2015, 452). To better understand reactance, previous studies have attempted to identify factors and individual characteristics that predict the experience of reactance. As Miller et al. report, “the experience of reactance...depends upon an individual’s self-perceived competence at exercising a freedom, as well as the belief that he or she merits that particular freedom. Research has [also] shown that reactance is positively correlated with self-esteem and efficacy, such that individuals tend to be particularly sensitive regarding freedoms they believe themselves to be both worthy of having and capable of sustaining” (Miller et al., 2007, 221). Based on this logic, it must be emphasized that many of the behaviors recommended by public health officials to combat COVID-19 are simple behaviors, such as wearing a mask, washing hands, and regulating the distance people stand between one another, as well as altering the practice of everyday behaviors like socializing. It is therefore likely that these sorts of simple behaviors have a greater potential to spark reactance among members of an audience precisely because these behaviors are so simple, as people may be more likely to feel that they both ought to retain control over these behaviors and are easily capable of sustaining this control. Additionally, Miller et al. note that “while it seems clear that certain kinds of persuasive messages can generate reactance in most individuals, it is now believed that reactance is a personality trait with levels of reactance varying from individual to individual, but holding steady, following a trimodal trajectory throughout the lifespan. Individuals, particularly during transitional stages (e.g., the “terrible twos,” adolescence, 96 or late in life…), tend to ardently feel their behaviors are solely their own business and are often strongly inclined to reject persuasive appeals perceived as attempting to control them in any way” (Miller et al., 2007, 221). Two important considerations arise from this discussion. First, reactance can be thought of as a personality trait with variation between individuals, which underscores the fact that people in any population will not interpret the same message in the same way. Indeed, “although a health message source may have good intentions, the person whose freedoms are threatened or removed is unlikely to see it that way” (Miller et al., 2007, 222). Second, there are predictable patterns public health officials must keep in mind that increase the potential for igniting reactance: certainly, the three age groups throughout the lifespan as mentioned by Miller et al., but also other potential periods of transition, such as young parents who feel public health guidelines are infringing on their parenting rights. Public health officials must remain acutely aware of the potential for reactance because the negative effects of sparking reactance in an audience can be severely damaging and may last long into the future, even if the message source has only good intentions. As Miller et al. argue, “an unintended consequence of psychological reactance--one perhaps [even] more damaging than message rejection or boomerang effects [discussed previously]--is source derogation, where reactance may be followed by aggression or hostility aimed at the threatening agent” (Miller et al., 2007, 222). Source derogation has particularly dire ramifications as it affects not only the message, but the underlying relationship between the source and the audience. Miller et al. explain that “source derogation has long-term implications for ongoing influence attempts, because the sources of a reactance-producing message may lose referent power and credibility and 97 thus suffer diminished future influence over their reactant audiences” (Miller et al., 2007, 222). Because the likelihood of reactance cannot be changed by public health officials, they must be extremely mindful of the language they use to communicate public health messages, especially those providing guidelines for behavior. Miller et al. explain the task of public health officials nicely: health promotion messages must be “both explicit in their intent and capable of attracting attention, yet compatible with their audience’s [range of] sensibilities, and tailored to what their targets are willing to consider,” while avoiding the effects of reactance that can damage health promotion efforts (Miller et al., 2007, 219). One important aspect of a message’s language that can be modified, then, is the intensity of the language, because “the strength or intensity of language used in a persuasive message is known to affect whether an individual will respond favorably toward an advocated behavior” (Miller et al., 2007, 222). According to Miller et al., “health messages often tend to be explicitly directive in nature” and doing so may make messages clearer and more understandable for the audience (Miller et al., 2007, 222). Such messages often share these characteristics: “explicit messages tend to convey a single meaning and leave little doubt as to the source’s intentions. Explicit commands tell a person clearly and directly what to do, frequently [though not always] using forceful adverbs such as ‘ought,’ ‘must,’ and ‘should’...Controlling language is characterized by increased use of imperatives as opposed to propositions or indirect suggestions” (Miller et al., 2007, 223). Additionally, “following the maxim of manner, they tend to adhere to Grice’s (1975) cooperative principle by being task-efficient, clear, unambiguous, and brief” (Miller et al., 2007, 98 223). At the opposite end of the continuum is low-controlling language, or what some researchers call autonomy-supporting language. This type of language “tend[s] to be less forceful and more polite, leaving the source’s intention less obvious, but at the expense of increased ambiguity. Unlike high-controlling language, low-controlling language...implicitly emphasizes self-initiation and choice. Because autonomy-supportive language often uses qualifiers such as ‘perhaps,’ ‘possibly,’ and ‘maybe,’ they tend to be less detailed, less precise, and more open to multiple interpretations” (Miller et al., 2007, 223). It seems, then, that there is a tradeoff between messages that promote autonomy and the level of ambiguity, which is an extremely difficult line for public health officials to walk. In the context of the COVID-19 pandemic, messages promoting the recommended behaviors are much more prevalent, and perhaps more direct, than other health promotion campaigns, such as those designed to reduce smoking or encourage seat belt use due to the increased urgency of the situation. The nature of COVID-19, the prevalence of the disease in communities, and the responsibility placed on individuals to comply with health recommendations for the public good make the possibility of reactance particularly dangerous and consequences may be long-term. Therefore, it is absolutely essential for government and public health officials to remain acutely aware of the possibility of reactance when promoting their messages. A NOTE ABOUT ADOLESCENT PSYCHOLOGY Teenagers and young adults have presented a particular challenge for any attempt to guide behavior, and COVID-19 behavior modifications are no exception. These groups, which will be referred to collectively as adolescents, are particularly prone to 99 reactance and rebelling against restrictions. For example, between three and ten thousand adolescents in Utah attended an unauthorized “rave-like Halloween event that was advertised as a ‘protest’ against coronavirus restrictions,” with organizers “The Tribe Utah” and “Utah Tonight” telling attendees that they would “keep the faces of everyone at the Protest on Halloween confidential so you have nothing to worry about!” (Fitzsimons, 2020). Adolescents in other areas have engaged in similar activities, including a fraternity party at the University of New Hampshire held shortly after the school opened for in-person classes that has since been linked to an outbreak on campus, with UNH President James Dean responding by stating, “this is reckless behavior and the kind of behavior that undermines our planning and will lead us to switching to a fully remote mode. The August 29 party is reprehensible and will not be tolerated” (qtd. in Beer, 2020). Students at the University of North Georgia similarly “decided the best way to start the first semester of the coronavirus pandemic was to throw a wild party” on August 15th, with the university responding by saying “we are disappointed that many of our students chose to ignore COVID-19 public health guidance by congregating in a large group without social distancing or face coverings” (New York Post, 2020). Another party forced the Rutgers football team into a two-week quarantine after fifteen players tested positive for COVID-19 in July 2020 (Rittenberg, 2020). At East Carolina University, police shut down not one but 20 parties during the college’s opening weekend, with officials fearing “they [wouldn’t] be able to stop rowdy college students from partying amid the pandemic” and calling plans to reopen for in-person classes a “recipe for disaster” (McEnvoy, 2020). Students at Penn State University did not wait for classes to start before holding large parties, “as hundreds of...students gathered outside of the 100 freshman dorms [on Aug. 19th]--many without masks--to dance, twerk, and party amid the COVID-19 pandemic, enraging community members and concerning plenty of other students” (Moyer, 2020). While these are only some of many examples, parties and other large gatherings are not necessary to drive the spread of the virus. Analysis by The Texas Tribune found in September 2020, “in counties where four-year college students make up at least 10% of the population, cases have grown 34% since Aug. 19 [2020]” only weeks after many colleges begin their academic year (Platoff et al., 2020). Indeed, “around the county, college towns…[emerged at the beginning of the fall 2020 semester] as new hot spots for the coronavirus, with cases surging among student populations and administrators scrambling to keep infections from reaching the broader population” (Platoff et al., 2020). While this is not to say that all adolescents are similarly reckless or rebellious towards public health guidelines, in the case of an infectious disease like COVID-19, it is not necessary for a large number of people disregarding public health guidelines to drive virus spread and endanger others. Colleges across the United States opened to in-person classes, activities, and on-campus housing to thousands of students in the fall of 2020, some with less success than others. One striking example is that of the University of North Carolina at Chapel Hill, which was “the first major college to pivot to online classes after reopening in person. The reversal took one week,” during which the campus reported 130 student cases of COVID-19 (Quintana, 2020). Barbara Rimer, the Dean of the School of Public Health at UNC Chapel Hill called for the switch to remote learning, saying, “the number of clusters is growing and soon could become out of control...It is time for an off-ramp. We have tried to make this work, but it’s not working” (qtd. in 101 Quintana, 2020). One major reason reopening plans were unsuccessful at universities is because these plans assumed that students would follow the rules and protocols put in place, though based on the examples cited above, this is clearly a dangerous and incorrect assumption to make regarding adolescents in particular. For example, as of Aug. 24th, 2020, the University of Toledo’s attempt to address large student gatherings simply began “asking students to self-quarantine if they’ve attended recent large off-campus parties,” with Toledo-Lucas County Health Commissioner Eric Zgodzinski noting that “the plan is only going to be as good as the students’ willingness to follow it” (Burke, 2020). Indeed, “among universities that have brought students back [on campus], large parties have quickly become the kryptonite of college administrators and their reopening plans, with several outbreaks tracing back to the gatherings” (Burke, 2020). Recent research in adolescent psychology would argue that such behavior is not surprising. In developmental psychology, adolescence is defined as “the period after the initial stages of pubertal maturation have begun but before young people have fully adjusted to the rapid developments in their bodies and before they have been accorded full adult status by society,” with adolescence spanning from as early as age 13 or 14 to the early- or even mid-20s in the United States (Yeager et al., 2018, 102). Adolescence is also “a dynamic period of learning and change,” with important cognitive changes occurring during this time in addition to, and alongside, the physical ones (Yeager et al., 2018, 104). During this time, adolescents typically experience a change in motivation and emotionality. As Yeager et al. describe, “compared with younger individuals, middle adolescents show a greater sensitivity to status and respect, resulting from pubertal maturation (e.g., changes in hormones), changes in social context (e.g., school 102 transitions) and social-cognitive developments while the cognitive developments associated with maturation of the prefrontal cortex with its heightened abilities, [such as] planning, decision-making, goal-setting, and metacognition have not yet occurred (Yeager et al., 2018, 104). Many previous attempts to guide adolescent behavior toward healthier or more constructive alternatives with regard to bullying, drug and alcohol use, and many others have been largely unsuccessful, and in some cases, even encourage the behavior the interventions are intended to prevent. Yeager et al. propose, however, that it is not impossible to design interventions that are effective in guiding adolescents toward a desired behavior. They argue that “compared with children, adolescents are more sensitive to whether they are being treated with respect and accorded high status. Traditional programs might work against this sensitivity, but effective adolescent interventions allow young people to make choices that benefit their long-term future while also feeling that they are respected and have high status in the short term” (emphasis added) (Yeager et al., 2018, 102). Yeager et al. define sensitivity to status and respect as “a readiness to alight attention, motivation, and behavior with the potentially rewarding feelings that come from attaining status or being respected. In turn, status is defined as one’s relative rank in a social hierarchy. Individuals discern their status in part on the basis of how others treat them, and in particular whether others treat them with respect. Respect is a complex, gestalt social judgment that hinges on whether one is being granted the rights one expects to be granted in one’s role in society” (Yeager et al., 2018, 104). Anthropological research has found that “individuals feel respected and that they have high status when they are treated as though they are competent, have agency and 103 autonomy, and are of potential value to the group” (Yeager et al., 2018, 104). Status and respect can be especially powerful for adolescents because “status and respect-relevant experiences can be highly rewarding; they elicit social emotions such as pride and admiration, which makes them motivationally salient” (Yeager et al., 2018, 104). This may explain why so many adolescents have disregarded public health guidelines outlined by university and public health officials and why the statements from universities regarding reckless behavior are no more effective: they do not take these fundamental adolescent motivations into account. Rather, as with the approaches used in similarly ineffective interventions, simply telling adolescents what they “can” or “cannot” do denies their sense of autonomy, status, and respect. Indeed research has found that as early as middle adolescence, individuals “come to perceive adult authorities’ efforts to influence their behavior, even when seemingly benign, as a sign that they are being disrespected or deprived of full adult status (Yeager et al., 2018, 105). It must also be noted that, due to their stage in biological and cognitive development, adolescents may react more strongly to perceived threats to their sense of status. This is because testosterone levels “increase dramatically after the onset of puberty in both boys and girls”; however, as noted previously, the cognitive processes associated with logical thinking, planning, and impulse control, have not yet fully matured (Yeager et al., 2018, 104). Testosterone “is often stereotyped as an ‘aggression’ or ‘sex’ hormone, but a growing line of research in both humans and animals suggests that it increases the motivation to search for, learn about, and maintain status in one’s social environment…[while] at an attentional level, endogenous levels of testosterone predict greater reactivity to status-related emotional stimuli” (Yeager et al., 2018, 104). 104 Importantly, typical approaches aimed at modifying adolescent behavior frequently “focus on providing knowledge or self-regulation skills with the intent of suppressing short-term desires for the sake of long-term goals. In doing so, these interventions may ignore or fight against the powerful reasons why adolescents are engaging in the ‘problem’ behavior in the first place” (Yeager et al., 2018, 105). One example of such an effort by government officials is the official statement released by a spokesperson for then-Governor Gary Herbert regarding the Halloween gathering in Utah, which stated that “Covid-19 is mainly spread through social gatherings and the virus is ‘all too real and terrifying for the medical professionals working overtime in our packed ICUs. We must decide, and show by our actions that the lives of everyone around us matter more to us than parties. If we do not, we will have a difficult time beating COVID-19 as a society’” (Fitzsimons, 2020). However, as Yeager et al. explain, appealing to logical arguments or relying on adolescents’ ability to self-regulate and delay gratification simply target the undeveloped, logical cognitive processes of the prefrontal cortex while ignoring the cognitive processes that are developed and the motivations that catalyze their behavior. This is exactly why the statements condemning the gatherings described above have little to no effect on changing the “problem” behaviors. Thus, any attempt to modify adolescent behavior, as in the case of COVID-19 protective behaviors, must be specific to adolescent psychology, with officials being especially mindful of what is being said, as well as how and by whom the message is communicated. RISK PERCEPTION An important consideration when making predictions about the future, 105 particularly in the case of a pandemic or other hazards, is the perception of one’s level of risk. Based on their meta-analysis of vaccination studies, Brewer et al. found “a high degree of consistency and a strength of association between risk perceptions and behavior that is larger than had been suggested by prior meta-analyses” by dividing the notion of health-related risk into three, rather than two categories (Brewer et al., 2007, 142). Thus, it is vitally important for government and public health officials to understand how members of the public understand risk because of its close relationship with behavior. When describing health threats, “nearly all theories focus on only two [dimensions]: the likelihood of harm if no action is taken and the severity of harm if no action is taken”; however, Brewer et al. argue that likelihood and susceptibility, while they are related concepts, are “logically distinct” and should therefore be examined separately (Brewer et al., 2007, 137). In their view, likelihood describes “one’s probability of being harmed by a hazard under certain behavior conditions” while susceptibility describes “individual resistance or constitutional vulnerability” (Brewer et al., 2007, 137). Brewer et al. note of the relationship between susceptibility and probability that “susceptibility to a disease should influence the likelihood of developing that disease, but being susceptible to an illness does not necessarily mean that the absolute probability of that illness is large” (Brewer et al., 2007, 137). The third concept examined is severity, defined as “the extent of harm a hazard would cause” (Brewer et al., 2007, 137). The Health Belief Model is a simple and well-supported model that seeks to explain the relationship between an individual’s understanding of a disease and the actions (or lack thereof) taken in response. The Health Belief Model has four basic elements: “perceived susceptibility (likelihood of getting the disease), perceived severity 106 (perception of how serious an outcome or consequence is from the disease), perceived benefits (efficacy of preventive action undertaken) and perceived barriers (time, effort, money, inconvenience, pain, side effects of preventive action)” (Bond & Nolan, 2011, 2). In other words, these elements can be divided into two categories: susceptibility and severity comprising “threat perception” and benefits and barriers comprising “behavioral evaluation” (Ayers et al., 2007). It is important to note that each element is perceived, and therefore a subjective judgment made by an individual rather than an objective measure. Additionally, the Health Belief Model proposed that “cues to action can activate health behavior when appropriate beliefs are held. These ‘cues’ include a diverse range of triggers including individual perceptions of symptoms, social influence and health education campaigns” (Ayers et al., 2007). Some later conceptions of the Health Belief Model also include “an individual’s general health motivation or ‘readiness to be concerned about health matters’” Ayers et al., 2007). Much research on the Health Belief Model has been supportive of the model. Quantitative reviews have revealed that “the susceptibility, severity, benefits, and barriers constructs...are very often found to be significant predictors of health-related behaviors but that their effects are small. However,...[some] caveats are important. [Including that] the effects of individual health beliefs should be combined and the combined effect may be greater than the sum of individual effects” (Ayers et al., 2007). The Health Belief Model “has been applied to an impressively broad range of health behaviors among a wide range of populations…[including] (a) preventive health behaviors, which include health-promoting (e.g. diet, exercise) and health-risk (e.g. smoking) behaviors as well as vaccination and contraceptive practices; (b) sick role behaviors, particularly adherence to 107 recommended medical regimens; and (c) clinic use, which includes physician visits for a variety of reasons” (Ayers et al., 2007). The Health Belief Model may therefore serve as a useful framework to evaluate not only how individuals adhere to behaviors recommended to prevent COVID-19, as well as to understand their behaviors if they do contract the virus and their use of medical facilities, such as clinics and testing. One reason the Health Belief Model was initially developed was to provide a framework for health behavior modification using the beliefs individuals hold about health behaviors specifically. Indeed, the Health Belief Model’s constructs are “correlated with a range of health-related behaviors and changing these beliefs may prompt behavior change” (Ayers et al., 2007). In their study of adults in Macao, China in April 2020 in the absence of legal requirements to follow health recommendations, Tong et al. found that, “adherence to different types of precautionary measures was correlated with four [Health Belief Model] factors (i.e. perceived severity, perceived benefit, perceived barrier, and cue-to-action)” (Tong et al., 2020, 1216). When seeking to apply these factors to health promotion efforts, Tong et al. echo other researchers in arguing that “the Health Belief Model as a whole may improve adherence [to recommended behaviors], but the specific [Health Belief Model] factors that work best may vary across behaviors” (Tong et al., 2020, 1216). For example, cue-to-action, defined as the intensity of the cue catalyzing the behavior, “was found to be positively associated with adherence to COVID-19 precautionary measures,” as well as “perceived benefit, with positive valence, and perceived barrier, with negative valence” Tong et al., 2020, 1206, 1217). However, it is important to note that while this and other research has led to many consistent findings, it is often with small effect sizes, implying that “interventions based 108 purely on…[the Health Belief Model] alone may be insufficient to substantially influence the adherence to precautionary measures” (Tong et al., 2020, 1218). Thus, while useful, the Health Belief Model alone is not enough to explain decisions people make regarding health protective behaviors, in part because this model assumes people make rational decisions after weighing their perceptions of disease risks and the effectiveness of protective action. Bond & Nolan examine risk in the context of the Health Belief Model, as well as decision-making and subjective risk perception. As has been well-documented across a range of studies, however, people frequently do not make the “rational” decisions these types of models would predict (Bond & Nolan, 2011, 2). Bond & Nolan extend their analysis beyond the Health Belief Model in isolation and add important contributions from decision-making and subjective perceptions of risk under uncertainty: “both approaches focus on risk as a subjective, rather than an objective concept, and both involve social and psychological aspects that impact on the individual cognitive structure of risk perception” (Bond & Nolan, 2011, 2). Indeed, “to a scientist who conducts risk assessments, the definition of risk is ‘hazard times exposure equals consequence,’...but to the average person, the definition of risk is ‘the probability of something bad happening.’ And risk communication does not always account for the subjectivity of ‘something bad’” (Brown, 2014, A277). In the case of the COVID-19 pandemic, public health officials similarly quantify risk using epidemiologic measures that track the progression and spread of the disease among the population. Thus, while members of the general public tend to treat risk as individual and subjective, experts typically treat and communicate risk in the same way they measure it: with objective and detailed statistics that apply to a population. 109 Individual calculations of risk are often influenced by mental processes that may remain beneath the level of consciousness. The brain may use “mental shortcuts to quickly make sense of partial information” such as to “map partial information against patterns of what we already know…[For example,] if our associations are negative, we will react fearfully…’and if our associations are not negative, we might not react with as much caution as we should’” (Brown, 2014, A278). Research has found a group of factors that frequently contribute to subjective perceptions of risk, all of which are important considerations in a public health setting, offering a window into this complex, individual process of calculating risk. In sum, research has consistently found that risks are perceived more negatively if exposure to the hazard is involuntary, people perceive they have little personal control over outcomes and there is uncertainty about the consequences of the outcome(s), the hazard is unfamiliar, the effects of the hazard are delayed, the hazard has catastrophic potential, the benefits are not immediately apparent and the hazard is caused by human rather than natural causes. These have been summarized by two factors labelled ‘Dread’ (uncontrollable, feared, involuntary exposure, inequitable distribution of risk, not easily reduced, catastrophic, risk increasing, fatal consequences, risk to future generations) and the ‘Unknown’ (not observable, risk unknown to science, delayed effect, new risk” (Bond & Nolan, 2011, 3). While people do weigh these factors to calculate individual levels of risk, research has also found that people tend to underestimate their personal risk relative to that of others: “People...demonstrate a tendency to believe that their own risks are less than others, particularly if they believe that their exposure to risk is in some way under their control. The impact of this tendency, described as unrealistic optimism and/or the illusion of control, is to reduce the perceived need to take protective measures” (Bond & Nolan, 2011, 3). However, emphasizing to individuals that they may be at higher risk for contracting a disease or suffering from severe effects of it may not be effective. In Bond 110 & Nolan’s study of parents’ attitudes toward vaccinations, even when scenarios “were written specifically so that those being interviewed fitted the ‘at risk group’, the participants [still] did not identify with this group. They believed that people who suffered serious consequences of [the] flu were different from themselves. When they heard similar items on the news they assumed the people who were badly affected were old, frail, sick, had not been eating well, had poor immune systems, low resistance, or were people who did not look after their health” (Bond & Nolan, 2011, 10). Further, having a friend, family member, or other known contact who had suffered severe consequences of disease may not necessarily lead an individual to adjust their calculation of their own risk. When asked, participants responded “that it would depend on the state of the friend’s health and their habits: whether they were unhealthy or stressed, or careless of their health. While it would be more concerning to hear of a friend who was ill, most did not think it would mean they themselves were more at risk” (emphasis added) (Bond & Nolan, 2011, 10). Indeed, “congruent with other primary studies...experience of measles [also] often rendered it a less threatening disease,” as “what is familiar is not [feared]” (emphasis added) (Bond & Nolan, 2011, 12, 7). Additionally, Bond & Nolan found that parents weighed the likelihood of severe outcomes of disease and vaccine side effects differently. Among vaccinators: “even though most considered it unlikely that their children would contract these diseases, it was easy to imagine that if contracted, the worst was likely to happen...The risks associated with vaccination were also perceived as being rare but rather than imagining the worst in this instance, they believed that one would be unlucky to have severe reactions” (emphasis added) (Bond & Nolan, 2011, 6-7). In contrast, non-immunizers 111 “dreaded the unknown or uncertain outcomes of the vaccines with major fears being for invisible/undetectable/distant problems...On the other hand, severe outcomes of the diseases were believed to be rare or only a problem for children with poor nutrition, poor sanitation, and compromised immune systems. Non-[immunizers] believed it was unlikely that their children would suffer serious complications if they contracted the diseases because they had healthy immune systems” (emphasis added) (Bond & Nolan, 2011, 7). In other words, “perception of risk may be influenced by an unrealistic optimism about one’s own risks or unrealistic perception of control over one’s life” (Bond & Nolan, 2011, 10). In addition to the mere use of statistics appearing to be incompatible with the way most members of the general public understand and calculate risk, the format government and public health officials use to present statistics may cause further damage to their efforts. The term “equivalency framing” is used broadly to describe “alternative messages in which logically equivalent information is portrayed in different ways, and [such changes in] framing has been shown to sway people’s responses,” including mathematically equivalent representations of numbers, such as “a policy framed as reducing the unemployment rate from 10% to 5%, as opposed to one framed as increasing the employment rate from 90% to 95%” (Lee et al., 2019, 345-6). Lee et al. state that framing of a particular message consists of multiple levels: concept frames refers to the labels that journalists choose to describe certain objects or ideas...which signal different interpretations to the audience (also known as cuing effects. By selectively assembling words into sentences, journalists make statements that further transmit preferred meanings, known as assertion frames. These assertions are then logically organized to form arguments, known as thematic framing, which are then assembled into a news story. These choices, therefore, that journalists make in constructing new stories can deliver meaning through each of these levels of the news story’s text (Lee et al., 2019, 112 347). While Lee et al. focus on the effects of framing in news stories written by journalists, government and public health officials must also be aware of these effects for two main reasons. First, in the midst of a public health crisis, such as the COVID-19 pandemic, the general public, as well as the journalists who write news stories, more routinely look to government and public health officials for this information; the way they present this information matters immensely to the way the information is understood and received. Second, as journalists write their articles, government and public health officials should understand how information is conveyed to the public in this format, both to determine if the information presented in news stories is accurate and portrayed as such, and to better understand the wide array of messages the general public receives. The format of numbers as a part of message framing has particular relevance for perceptions of risk and assertions about the degree of safety in the surrounding environment. Previous research has found that “displaying statistics in different formats, such as absolute frequencies (e.g., X number of Americans), simple fractions (e.g., X out of Y Americans), and percentages (e.g., X% of Americans), can have divergent effects on how people perceive an assertion. This is because factors such as the perceived complexity of given numerical information and the group sizes used to normalize raw statistics differ across these formats” (Lee et al., 2019, 347). Thus, one crucial factor that must be considered when presenting numerical information is the natural frame of reference the number format provides. For example, in the fraction format, the reference population would be the denominator; for the percentage format, 100% is the natural comparison. However, the absolute frequency format lacks a well-defined and quantified 113 reference population, thereby increasing the general public’s difficulty in understanding the true meaning of the value. Overall, “information presented [as] complex statistics is likely to be inaccurately perceived, or even ignored, when it requires considerable mental effort to process” (Lee et al., 2019, 348). In their studies, Lee et al. examined the effects of various number formats on the perception of risk and found that “the lack of understanding that results from using large numbers can hinder people’s ability to estimate the implications of prevalence information for their personal safety,” thereby reducing the level of perceived risk (emphasis added) (Lee et al., 2019, 348). Thus, “according to this perspective, formats easier to understand, such as probabilities, are more likely to invite the receiver to consider the likelihood of being affected or unaffected by a particular event than are absolute frequencies. Consequently, people...can become more attentive to the risk when it is displayed in probabilities rather than absolute frequencies” (Lee et al., 2019, 348). While their study focused on representations of violent crime rates, Lee et al. found that the probability format can be especially powerful and may aid in producing emotional reactions: “what appeared to be the sole effect of switching [assertion frames] was in fact a result of assertions accompanied by numerical information in probability formats...[and thereby suggesting] that the effects of assertion frames on emotions are mediated by probability formats” (Lee et al., 2019, 358). With regard to number formats, the intuitive approach is therefore likely not the most effective way to present numerical information to the public. Lee et al. note that “it may be tempting to assume that the large numbers used...as the absolute frequencies draw more attention due to their size, thereby creating a greater impression. Our results show, however, that probability formats, though small in absolute size, more strongly 114 accentuate assertions in generating emotions than do absolute frequency formats” (Lee et al., 2019, 358). Perceptions of risk may have different influences as well as different results depending on the health behavior in question. Brewer et al. argue that “the importance of risk perceptions to health behaviors undoubtedly varies across behaviors. Risk perceptions are probably more important for behaviors, such as sunscreen use, that are intended to reduce a specific health threat and are probably less important for behaviors, such as exercise and diet, that have a wide range of health and nonhealth consequences. Risk perceptions are probably more important when people make individual decisions about a behavior with relatively diffuse external influences, as in sunscreen use, than when strong external influences are present, as with physician recommendations for cancer screening tests (Brewer et al., 2007, 138). In the case of the COVID-19 pandemic, the health behaviors recommended, such as mask-wearing, physical distancing, and frequent hand-washing, are all intended to reduce a specific threat; however, especially in light of widespread social separations and shutdowns, the extent to which social media, traditional media, and other electronic technologies for connection and information gathering act as an external influence is an important, and likely individual, consideration based on technology use. Of particular relevance to the Health Belief Model, Brewer et al. also argue that “when it is easy to carry out the health behavior, there is likely to be a stronger association between perceptions and behavior than when it is difficult to carry out the behavior” (Brewer et al., 2007, 138). Unrealistic optimism is a complex phenomenon that is crucial to consider when seeking to understand how people calculate their personal level of risk. According to 115 Shepperd et al., “Researchers have documented...unrealistic optimism in over a thousand studies and for a diverse array of undesirable events, including diseases, natural disasters, and a host of other events” (Shepperd et al., 2015, 232). Unrealistic optimism may be one of two types: “The first type is unrealistic absolute optimism, which refers to an unjustified belief that a personal outcome will be more favorable than the outcomes indicated by an objective standard…The second type of unrealistic optimism is unrealistic comparative optimism, which refers to the erroneous estimate that one’s personal outcomes will be more favorable than the outcomes of one’s peers” (Shepperd et al., 2015, 232-3). Shepperd et al. divide the underlying motivations behind unrealistic optimism into three categories: first that people are generally “motivated to believe, or to have others believe, that they are unlikely to experience unfavorable outcomes…,that people process different (and often more) information about themselves than about the average person…[and] finally, unrealistic optimism can be a natural consequence of the way people process information” (Shepperd et al., 2015, 233-4). It is important to note that, while unrealistic optimism is common across a wide range of phenomena, there are also well-documented situations in which people do not exhibit unrealistic optimism. For example, “people often show less unrealistic comparative optimism when estimating their chances of experiencing negative events that occur frequently in the population…[or] for events that they perceive to be beyond their control” (Shepperd et al., 2015, 233). Government and public health officials must therefore be careful about promoting notions of control, due to the risk of increasing unrealistic optimism along with it. Experience also plays a role in unrealistic optimism, as “people who have previously experienced an unfavorable outcome are less optimistic about their chances of avoiding a 116 recurrence” (Shepperd et al., 2015, 233). The possibility of recurrence is especially relevant to the COVID-19 pandemic as emerging research indicates that individuals can be reinfected with the virus. The true impact of unrealistic optimism lies not only in its impact on people’s perceptions of risk, but also in its impact on behavior. Shepperd et al. note that as of now, “we have a limited understanding of when unrealistic optimism is beneficial versus problematic, but we suspect that it may be beneficial for people who are already actively coping with a problem and for events that are temporally distal rather than proximal” (emphasis added) (Shepperd et al., 2015, 524). For example, one study found that “people who were unrealistically optimistic about avoiding the H1N1 virus reported lower intentions to wash their hands and use hand sanitizers” (Shepperd et al., 2015, 524). Indeed, unrealistic optimism “has been found in relation to the probability of experiencing various diseases, such as alcoholism and heart attack, breast cancer in women, and prostate cancer in men” (Dolinski et al., 2020, 2) and recent research has found that the case of the COVID-19 pandemic is no different. In their study of Polish university students, for example, Dolinski et al. found “a fairly general occurrence of unrealistic optimism (especially in men as it appeared in all three measures [during the month of March], but also in women in the last two measures)” (Dolinski et al., 2020, 6). A similar study conducted with a representative sample of citizens in the United Kingdom found that “UK adults...display [unrealistic] optimism concerning many aspects of COVID-19…[for example,] compared to people of their age and gender, they were somewhat or extremely unlikely to have accidentally infected people with COVID-19 in the past and to infect others or get infected themselves in the next month. They were also 117 [unrealistically] optimistic, but to a lesser extent, about their likelihood of getting hospitalized due to COVID-19, finding themselves in an ICU, being ventilated, and making a full recovery” (Asimakopoulou et al., 2020, 1507-8). Unrealistic optimism may present a particular challenge to government and public health officials for a number of reasons. First, unrealistic optimism may be resistant to factual information contesting those beliefs. Dolinski et al. note that “the dominant pattern of the results in our study is that unrealistic optimism…[often] persists in spite of further alarming or horrifying information” (Dolinski et al., 2020, 7). Second, unrealistic optimism has been found to be sensitive to perceptions of controllability of the event, as well as the timescale in which the event may occur. Previous studies conducted similarly soon after an earthquake and the Chernobyl nuclear accident found unrealistic optimism, but a clear effect of unrealistic pessimism regarding negative consequences of the specific hazard, earthquakes or radioactive material respectively (Dolinski et al., 2020, 6). In contrast to these events, however, Dolinski et al. argue that “the studies on unrealistic optimism have [shown] that an important factor influence[s] the magnitude of this effect is the belief in the controllability of a given event. The effect of unrealistic optimism is clearly stronger in relation to events that people think they have influence over” (emphasis added) (Dolinski et al., 2020, 6). Additionally, while Asimakopoulou et al. found that adults in the United Kingdom exhibited unrealistic optimism about a variety of events related to COVID-19, including experiencing severe symptoms, being hospitalized or admitted to an ICU, and spreading the virus to others, “participants showed [unrealistic] pessimism about COVID-19 infections in the more distant future. As compared to the average person of their age and gender they felt likely to get infected by 118 COVID-19 in the next year and to develop COVID-19-related symptoms” (Asimakopoulou et al., 2020, 1508). Asimakopoulou et al. postulate that this may be due to the fact that in the United Kingdom, as has been the case in the United States, “controlling the pandemic was very much placed in the hands of individuals restricting their lives,” such as complying with lockdowns, working from home, and restricting social gatherings, which may grow more difficult as the pandemic progresses (Asimakopoulou et al., 2020, 1508). Finally, the effects of unrealistic optimism in the short run may be appealing to many individuals. Scholars have argued that “the consequences of unrealistic optimism are likely innocuous or even positive in the short run (e.g., reduced anxiety, persistence toward goals) but negative in the long run (e.g., increased likelihood of negative outcomes due to increased risk behavior, failure to take precautions, or insufficient preparation)” (Shepperd et al., 2015, 524). In addition to the use of heuristics, culture has also been found to influence risk assessments and affect how heuristics are used. Brown notes that “people tend to shape their view so they match those in the groups with which they most closely relate, a concept known as cultural cognition” (Brown, 2014, A278). According to the cultural theory of risk, “individuals selectively attend to risks and related facts in a way that reflects and reinforces their ‘cultural worldviews,’ or preferences about how society should be organized” (emphasis added) (Kahan et al., 2010, 502). Group membership may also explain, at least in part, differences between how individuals perceive their own risks: “a large portion of the variance in risk perception coheres with membership in groups integral to personal identity, such as race, gender, political party membership, and religious affiliation” (Kahan et al., 2010, 503). Both are important considerations for 119 government and public health officials communicating with the general population because membership in groups central to one’s identity, and specifically the values that group holds, influence many of the subconscious calculations people make when estimating their subjective level of risk. As Kahan et al. note, heuristic mechanisms interact with cultural values: People notice, assign significance to, and recall the instances of misfortune that fit their values; they trust the experts whose cultural outlooks match their own; they define the contingencies that make them worse off, or count as losses, with reference to culturally valued states of affairs; they react affectively toward risk on the basis of emotions that are themselves conditioned by cultural appraisals--and so forth (Kahan et al., 2010, 503). With regard to differences between individuals, taking culture cognition and group membership into account is crucial, as these concepts help to “complete the heuristic model by showing how once and the same heuristic process (whether availability, credibility, loss aversion, or affect) can generate different perceptions of risk in people with opposing outlooks” (Kahan et al., 2010, 503). As noted previously, existing views have a large effect on the perception of risk, specifically, people tend to “selectively credit and dismiss information in a manner that confirms their prior beliefs,” a phenomenon known as biased assimilation (Kahan et al., 2010, 504). However, “the cultural cognition thesis implies that this dynamic will interact with culture, That is, people will selectively credit or discredit information on risk in a manner that fits their cultural predispositions toward them, and thus polarize along cultural lines” (Kahan et al., 2010, 504). Additionally, the perceived credibility of a source, including knowledge, impartiality, and honesty, is an important, if complicated, consideration for individuals evaluating information about risk (Kahan et al., 2010, 504). Taking cultural cognition into account, “it is also well established that individuals tend to impute these qualities to, 120 and hence be more persuaded by, information sources with whom they have some ‘ingroup’ connection, and to deny the same to ‘outgroup’ information sources” (emphasis added) (Kahan et al., 2010, 504). In other words, “the cultural cognition thesis suggests that the operation of the credibility heuristic will be highly sensitive to the source’s and target’s cultural worldviews. On this account, cultural affinity and cultural difference supply the relevant in-group/out-groups references that in turn determine whom people see as knowledgeable, honest, and unbiased, and thus worthy of being credited in debates about risk” (Kahan et al., 2010, 504). Norms are an important concept in many disciplines, and while there may not be a singular definition, scholars generally agree that “there are shared societal expectations that impact on behavior…[and] that people occupy different social positions and belong to different groups [that] do not necessarily agree on what is (in)appropriate behavior” (Reynolds et al., 2015, 46-7). While social norms are considered an external force influencing behavior, the norms themselves are “upheld through internal forces, [and] a common emphasis in most definitions [of social norms] concerns the degree to which norms, as rules about human social behavior, are internalised as part of a person’s sense of self-definition (‘who am I,’ ‘who we are’) or are forces external to the person that have impact through peer pressure and conformity (e.g. fear of social sanctions)” (Reynolds et al., 2015, 47). Thus, social norms may be maintained by placing an importance on group membership as central to one’s identity or due to fear of repercussions from the other members of the group. Social psychologists Fishbein & Ajzen define norms simply as “what other people think about the behavior” and argue that norms are “important considerations for predicting an individual’s behavior” (Petty & Cacioppo, 1996, 28). 121 Membership within a group, called the “ingroup” is particularly important, because “when people come to categorise themselves as similar to others (ingroup, weself), the characteristics that define the group become self-defining [for the individual] and shape behavior” (Reynolds et al., 2015, 48). This also relates to how the “social others,” called the “outgroup,” are defined: “when people come to categorise themselves as similar to others,...they internalise the characteristics that they have in common and define the group as being distinct from comparison outgroups. The dimensions that come to define the ingroup shape behavior because they become self-relevant and selfdefining” (Reynolds et al., 2015, 49). Reynolds et al. propose that not only are social norms “a key element of many behavior change interventions, especially when the goal is en masse, sustained behavior change,” they may also be used as “a crucial component of motivation and behavior and thus a key to behavioral influence and change” (Reynolds et al., 2015, 45). This idea is based in the Theory of Planned Behavior, a theory which states that “intention to perform a behavior derives behavior [and] intention is governed by the relationship between people’s attitudes toward the behavior (favourable or not), their perceived behavioral control (his or her confidence in performing the behavior and certainty it will produce certain outcomes) and normative beliefs,...[of which] how likely ‘important referent’ others are to (dis)approve of a particular behavior” is an important component (Reynolds et al., 2015, 49). According to Reynolds et al., social norms may change behavior in two ways: first, “if people’s group memberships and associated social identities change, so too can behavior” and second, “it is also possible that the meaning of existing groups…-what ‘we’ value and believe in--is redefined, requiring a change in expected and accepted 122 (i.e. normative) patterns of behavior” (Reynolds et al., 2015, 51). In other words, it may be possible to change either an individual’s “ingroups,” or change the norms within the groups they already consider to be their “ingroups.” In the area of health behavior change specifically, one previous study found that “the relationship between attitudes and behavioral intentions can be strengthened through the use of ‘ingroup norms.’ When an ingroup message was congruent (rather than incongruent) with the initial attitude of the participant, the relationship was stronger, and this was particularly so when identification with the ingroup was high compared to low” (Reynolds et al., 2015, 50). Norms have been shown to have a powerful effect on behavior even when people are not aware of the influence of norms on behavior, and have been used successfully to increase compliance with a desired behavior or decrease the occurrence of an unwanted behavior. As Griskevicius et al. argue, if descriptive social norms, norms that refer to the “perception of what is commonly done in a situation,” are used unwisely or ignored by leaders and institutions hoping to change people’s behavior, they might increase the very behavior they are trying to discourage, as in the case of the Petrified Forest National Park in Arizona. According to Griskevicius et al., “each month the park loses more than a ton of petrified wood because of theft” and in an attempt to curb such behavior, the park posted a sign stating that “Your heritage is being vandalized every day by theft losses of petrified wood of 14 tons a year, mostly a small piece at a time”; however, by highlighting the prevalence of the unwanted behavior, park officials inadvertently presented theft as the norm (i.e., what “most people do” in the particular situation), and thereby “stimulated the precise behavior they had hoped to suppress by making thievery appear commonplace” (Griskevicius et al., 2008, 8). A simple experiment by 123 Griskevicius et al. testing the type of sign visitors to the park encountered demonstrates the power of social norms. In one condition, visitors encountered a sign that “urged visitors not to take wood, and it depicted a scene showing several different thieves in action, highlighting the problematic prevalence of this behavior” while the second sign “also urged visitors not to take wood, but it depicted only a lone thief. Visitors who passed the first type of sign--the sign containing the type of normative information that the actual Petrified Forest contains--were more than twice as likely to steal the precious wood than those who had passed the second type of sign. Thus, by failing to take into account the raw force of descriptive social norms on their guests’ behavior, park administrators produced the opposite of what they intended. More generally, because people of all sorts may under-appreciate the strength of social norms, they can engage in persuasion tactics that are not only ineffective but markedly counterproductive” (Griskevicius et al., 2008, 9). In terms of using social norms to encourage a desired behavior, Griskevicius et al. make two additional points: “first, social norms have particularly strong impact on recipients under conditions of uncertainty” and second, that people tend to be “especially influenced by those others who are similar to them[selves]...[and] when the decision maker is connected to others by a similar circumstance” (Griskevicius et al., 2008, 11). Of course, any message consists of far more than the simple meaning of the words it contains. The language used to express a message may have a sizable effect on how the source is perceived by the audience, even if neither party is consciously aware of this influence. According to the Linguistic Category Model, “one way by which a communicator conveys an implicit message is via variations in the degree to which that 124 communicator’s language is concrete versus abstract” and the patterns how each is used (Porter et al., 2015, 94). Additionally, “although communicators may be unaware of these features of their language, research suggests that levels of abstraction are used systematically to convey information across a variety of contexts, audiences, and desired goals...Likewise, although audiences may not recognize the use of different linguistic categories per se, there is evidence that they are [nevertheless] influenced by this linguistic variation” (Porter et al., 2015, 94-5). This linguistic variation consists of a range, with “descriptive action verbs [being] at the most concrete end of the continuum, followed by interpretive action verbs, state verbs, and finally, nouns or adjectives. Compared with more abstract language, descriptive and interpretive actions verbs (e.g., ‘Sam hit her friend’) indicate[s] that a target’s behavior is more discrete and less characteristic. By contrast, adjectives and nouns (e.g., ‘Sam is violent’) suggest[s] that a target’s behavior is more stable and lasting” (emphasis added) (Porter et al., 2015, 94). As Porter et al. state, previous research has found predictable patterns in how communicators tend to use abstract and concrete language, depending on group membership: “individuals tend to use abstract language to describe in-group members’ desirable behaviors and concrete language to describe their undesirable behaviors--a pattern we refer to as the favorable linguistic intergroup bias...In contrast, individuals tend to use abstract language to describe out-group members’ undesirable behaviors and concrete language to describe their desirable behaviors--the unfavorable [linguistic intergroup bias]” (Porter et al., 2015, 95). Abstract language, which describes states that are more stable, are therefore used more commonly to describe ingroup members’ desirable behavior and outgroup members’ undesirable behavior, thereby implying a 125 level of ingroup superiority that is stable and enduring. Similarly, concrete language, which describes states that are more transient, are used more commonly to describe ingroup members’ undesirable behaviors and outgroup members’ desirable behaviors, as if to subtly state that undesirable behavior by the ingroup or desirable behavior of the outgroup are not reflective of who they are, but rather, such actions are exceptions to their normal behavior. This tendency may influence how the source, and therefore the message, are received by an audience. For example, Porter et al. note that “communicators are evaluated more positively when they use either component of the favorable [linguistic intergroup bias] to describe ingroup members, rather than using the unfavorable [linguistic intergroup bias]. Likewise, communicators using more abstract language to describe a target are perceived as holding more biased attitudes” (Porter et al., 2015, 95). Porter et al. argue that this linguistic pattern may also provide a subtle, but detectable, clue regarding the source’s own group memberships, and thereby, their attitudes and beliefs. Porter et al. found in their studies first that “individuals can infer a communicator’s social identity from his or her language, regardless of their own social identity,” (called a social category inference), and second that “irrespective of both their own social identity and that of the target, participants were more likely to believe that the target and communicator shared a potential identity when the communicator used a favorable (rather than unfavorable) [linguistic intergroup bias]; however, participants’ own group membership moderated their evaluative assessment of the communicator” (Porter et al., 2015, 96, 100). In other words, “social category inferences were not dependent on the participants’ own group membership…[but] relevant evaluative 126 judgments--unlike social category inferences--were shaped by both [linguistic intergroup bias] and participants’ own social affiliations” (emphasis added) (Porter et al., 2015, 100). Government and public health officials must therefore seek to increase their awareness of the linguistic patterns they use to communicate with the general public because, “given that persuasion is shaped in part by social categorization, use of [a certain linguistic intergroup bias] could undermine or augment persuasive communications” (Porter et al., 2015, 100). A major challenge for public health officials, then, is to communicate with the general public in ways that support the development of more accurate perceptions of risk for a large number of individuals who understand risk as subjective and whose natural cognitive tendencies work against this goal. In Bond & Nolan’s study, researchers asked participants, in considering news reports, “how many would ‘several deaths’ [need to] be to cause them to worry about the risks of the disease” and while participants found it difficult to quantify and provided a wide range of responses, the question “provoked participants to define the type of information they wanted in order to make sense of...reports” (Bond & Nolan, 2011, 11). Based on this study, public health officials must carefully consider the familiarity of the disease to members of the public, as “if [the disease] was familiar (flu) [participants] wanted to know the details of who it was who had died or suffered severe complications. Thus, it was not the statistics that were important for deciding on risk, but the characteristics of those who had the disease and the familiarity or unfamiliarity of the disease” (emphasis added) (Bond & Nolan, 2011, 11). Public health officials crafting messages for the population must therefore keep in mind that “when listening to reports of epidemics, it is not the number of people who are 127 affected but the familiarity or unfamiliarity of the disease and the characteristics of those who had the disease that,” in this study, “caused parents to worry about taking preventive action” (Bond & Nolan, 2011, 12). Given that many social media users tend to use social media in a way that confirms their existing attitudes and beliefs, rather than seeking the most correct information, government and public health officials must keep these psychological principles in mind in order to use social media to its greatest degree of effectiveness in promoting population health. Sutton argues that this social media landscape necessitates a range of actions by government and public health officials on their social media accounts, including “actively engaging with the online public through routine, daily communication, as well as planned information campaigns that target specific topics such as mitigation and preparedness for seasonal events...and unplanned ones, such as an outbreak of measles” (Sutton, 2018, 1281). She also argues that authorities in the communities should monitor the discourse on relevant topics, and advocates for new, more effective communication strategies (Sutton, 2018, 1281). Clearly, while the intuitive approach to health communication would be simply to disseminate facts, this approach is not sufficient to face the obstacles of how many people tend to select, evaluate, and integrate information, and particularly information found on the Internet. AN OVERARCHING MODEL OF PERSUASION While scholars have proposed many models seeking to explain various aspects of persuasion and attitude change, this section will instead highlight a broader model proposed by Richard E. Petty and John T. Cacioppo called the Elaboration Likelihood Model (ELM). This model identifies two “routes” to persuasion: the peripheral route and 128 the central route. In the peripheral route, “attitude change is determined by such factors as the rewards or punishments with which the message is associated, or the judgmental distortions that take place in perceiving the message, or the simple inferences that a person draws about what a speaker advocates about a certain position,” or a number of other “peripheral” cues not relating to the content of the message itself (Petty & Cacioppo, 1996, 256). The central route, by contrast, “emphasizes the information that a person has about the attitude object or issue under consideration...The message recipient attends to the message arguments, understands them, and evaluates them. Some arguments lead to favorable thoughts, whereas others lead to counterarguments. The person then integrates all of this information into a coherent and reasoned position” (Petty & Cacioppo, 1996, 255-6). As implied by the array of biases and heuristics used in information processing discussed previously, “it is very important to note that the difference between the central and peripheral routes to attitude change is not that the former actually is rational and logical whereas the latter is not...the favorable thoughts and counterarguments that a person generates in response to a message need not be logical or rational at all. They only have to make sense to the person who generates them” (Petty & Cacioppo, 1996, 256). Rather, the difference between the two routes “has to do with the extent to which the attitude change that results from a message is due to active thinking about either the issue or the object-relevant information provided by the message” (Petty & Cacioppo, 1996, 256). Thus, peripheral cues may be any “factors or motives inherent in the persuasion setting that are sufficient to produce an initial attitude change without any active thinking about the attributes of the issue or the object under consideration” (Petty & Cacioppo, 1996, 256). Because the central route requires the 129 additional effort of object-relevant thinking, attitude change by this path is more difficult, but also more enduring (Petty & Cacioppo, 1996, 263). There are two main factors that determine whether the central or peripheral route is used. First, a message recipient must be motivated to process the message rather than rely on peripheral cues. Personal relevance or responsibility towards the issue, “need for cognition” (the desire to think about a message), and a number of other variables contribute to an individual’s motivation to process the message (Petty & Cacioppo, 1996, 263). For example, previous studies have found that “people could become more motivated to think about the content of a message if they were told that they were going to be subsequently interviewed about the issue” or that “people would sometimes be more motivated to think about incongruent information than congruent information” (Petty & Cacioppo, 1996, 263). However, “people become less motivated to think about a message when many other people are also evaluating it” (Petty & Cacioppo, 1996, 263). 130 Figure 3. Elaboration Likelihood Model. From “The Elaboration Likelihood Model of Persuasion,” by R. E. Petty & J. T. Cacioppo, 1986, Advances in Experimental Social Psychology 19, p. 126. 131 Second, a message recipient must be capable of understanding and processing the message. As with motivation, the ability to process the message is affected by multiple variables. For example, “the more a message is repeated...the greater the opportunity the person has to think about the message content,” but “the more distracted a person is, the less thinking about the message that can occur” (Petty & Cacioppo, 1986, 265). The ability to process a message may also be impacted by the message itself, as well as characteristics of the person and the persuasive environment: “written messages provide people with greater opportunity for elaboration than audio messages because people can process written messages at their own pace” and “if the message is incomprehensible, or if the person has no schema or framework for relating the message to his or her existing beliefs, then no processing can occur, even if sufficient motivation is present” (Petty & Cacioppo, 1996, 265). If the message recipient does not have both the motivation and the ability to process the message, the ELM states that they will rely on peripheral cues. If persuasive peripheral cues are not present, the person will retain their initial attitude, but if there are persuasive peripheral cues, attitude shift will take place, though it must be noted that this shift is temporary (see Figure 3). If motivation and ability to process the message, elaboration, or thinking, about the message will occur by the following process: The nature of that elaboration will be determined primarily by the subjective quality of the arguments presented in the communication, so our next question becomes What is the nature of the arguments in the message? If the person perceives the message to contain strong, compelling arguments, then thinking about the arguments will cause favorable thoughts to be rehearsed--and enduring persuasion will result; but, if the person perceives the message to contain weak arguments, thinking about the arguments will cause counterarguments to be rehearsed, and it is possible for the person to move in a direction away from that advocated in the communication (boomerang) (Petty & Cacioppo, 1996, 265). 132 The Elaboration Likelihood Model provides a useful framework for crafting and evaluating communications intended to persuade audiences that may be applied to public health. Inherent in this model is the consideration of the four aspects of persuasive environments identified in the Yale Attitude Change Approach proposed by Carl Hovland and colleagues: the message, the source of the message, the recipient of the message, and the channel through which the message is sent (Petty & Cacioppo, 1996, 61). However, it must be noted that while the ELM proposes a uniform framework applicable to a variety of settings, messages, and individuals, there is nevertheless subjectivity throughout the persuasive process it outlines. As discussed previously, for example, an individual must have both the motivation and ability to process the message, though these each may be affected by subjective evaluations of whether the issue is personally relevant or whether the message is understandable. The peripheral route clearly relies on subjective evaluations of cues within the persuasive environment, but the central route also relies on subjective evaluations of message-relevant information, such as the strength of the argument. Due to both the subjectivity inherent in persuasive attempts by both routes and the individual differences within populations, the ELM must be used as a general framework for government and public health officials to understand the dynamics of the persuasive environment, rather than as a clear formula for persuading a large population. 133 MODERN COMMUNICATION TECHNOLOGY IN PUBLIC HEALTH The COVID-19 pandemic is the first pandemic to occur since the advent of modern technology. Especially since the start of the 21st century, “there is a growing sentiment that governmental public health authorities should adopt and apply...Internet technologies to assess, protect, and promote public health” (Kass-Hout & Alhinnawi, 2013, 7). This outlines three separate, plausible uses of data in a public health setting: assessing (including tracking and measuring) levels of disease in a population, protecting the community, and promoting public health. However, there has thus far been little research to guide public health officials in using these technologies effectively. This paper will conduct exploratory analysis of how government and public health leaders in Utah have used modern communication technology to communicate with the public and subsequently make propositions regarding the use of these technologies in a public health setting. SOCIAL MEDIA Social media may be defined broadly as any “‘activities, practices, and behaviors among communities of people who gather online to share information, knowledge, and opinions using conversational media.’ Social media applications are broadly categorized as forums and message boards, review and opinion sites, social networks, blogging and microblogging, bookmarking, and media sharing” (Thackeray et al., 2012, 1). In 2012, when public health departments were beginning to adopt common social media technologies, Thackeray et al. found that state health departments had an “overwhelming preference for Twitter” in a possible attempt to use this medium to keep “the public upto-date with [state health department]-related news”; however, even in 2021, only 21% of 134 adults in the United States use Twitter, a “percentage that has stayed relatively steady over the past few years, with a bit of a dip in 2019, to 19%” (Thackeray et al., 2012, 5; Newberry, 2021). Twitter will be the focus of social media analysis for the purposes of this paper with characteristics of the audience and the use of Twitter will be discussed in detail below. Assessing disease within a population is the cornerstone of epidemiology, an activity that relies on collecting, analyzing, and reporting high-quality data. However, as discussed previously, there are multiple limitations with traditional methods of disease surveillance, ranging from the cost of conducting active surveillance to the possibility of inaccurate results from dichotomous testing, and many others. Timeliness is a particular problem with traditional methods of surveillance with significant consequences, as “the timeliness of health data limits the availability of actionable public health information as the traditional route for data moves from self-report to a physician, through diagnostic confirmations, and then from a physician or laboratory facility to a public health authority” (Kass-Hout & Alhinnawi, 2013, 8). Thus, if social media may be used to close this time gap and provide information to public health officials more quickly, it is possible to take action to address outbreaks before spread gets out of control. Previous attempts to do exactly this have shown “that [social media] can be useful when seeking timely and reliable data on the spread or severity of influenza, west nile virus, and meningitis” (Kass-Hout & Alhinnawi, 2013, 9). At present, however, many scholars argue that it is best to use social media in conjunction with, rather than in place of, traditional methods of disease surveillance. Indeed, “influenza surveillance may include laboratory-based virological surveillance, sentinel syndromic surveillance (e.g. tracking 135 emergency room visits, school-based absenteeism reports, etc.) and evaluation of mortality trends of pneumonia and influenza, which taken together may provide a more complete picture of disease risk or impact” (Kass-Hout & Alhinnawi, 2013, 9). This is because a balance must be met between the timeliness of the data and the quality of the data that drives action. As Kass-Hout & Alhinnawi explain, “health data found on social media circumvent the traditional route by removing the ‘middle man.’ However, the middle man plays a critical role. They validate and can tell authorities something about the population they serve where generalizations of this information can be made with a level of confidence. That confidence is crucial in triggering a governmental action or intervention” (emphasis added) (Kass-Hout & Alhinnawi, 2013, 8). While additional work is necessary to fine-tune the use of social media data for disease-tracking activities, at present, the nature of social media lends itself well to efforts intended to protect and promote public health. Traditional means of communication, ranging from interviews and advertisements on broadcast media to information pamphlets, websites, or broader campaigns, the direction of communication is almost inherently one-sided, going from public health officials to the public, with little ability to adapt communication efforts for different populations or trends. However, because social media is designed to promote interactions, the use of social media allows for two-way communication between the public health officials and members of the public. The ability to interact with the public rather than simply broadcast messages may have multiple benefits in a public health setting. First, as Thackeray et al. argue, “social media is more than another communication channel...if utilized effectively social media has the potential to improve the way public health agencies engage, interact and 136 communicate with various audiences. Specifically, social media are technologies that facilitate opportunities for engaging with the audience and for creating and maintaining relationships. If public health agencies can use social media to engage their audiences and create relationships, something that has previously been hindered by time and distance restrictions, they are one step closer to establishing true community-based partnerships to address public health problems” (Thackeray et al., 2012, 5). Many health departments at the state level have historically found this difficult to accomplish. In their analysis of state health departments in the United States in 2012, Thackeray et al. found that state health departments that had a Twitter account at the time posted, on average, once per day with 22.5% being retweets, and only 1.5% of tweets made being responses to followers’ posts (Thackeray et al., 2012, 3). Meanwhile, on state health departments’ Facebook pages surveyed, “the majority (86%) of Facebook posts received no comments and 45.1% of Facebook posts had no likes”; though videos and photos were rarely used in Facebook posts, “over three-fourths of videos (78.3%) received no likes and 70.7% of videos received no comments” (Thackeray et al., 2012, 3, 5). Heldman et al. similarly note at the time of their writing that “there is very little evidence to indicate that social media are being adequately used by public health organizations in ways that leverage the ability to have meaningful conversations with [their] audiences” and argue that “public health practitioners, it would seem, must embrace the unique characteristics and functionality of social media to engage members of digital communities in interactive conversations about health” (Heldman et al., 2013, 2). Additionally, “social media...has [a] tremendous potential in assessing health behaviors and health sentiments or rumors,” something the CDC attempted to implement 137 in 2009, prior to the COVID-19 pandemic: “when [the] CDC responded to queries from Facebook and Twitter on health concerns regarding contracting H1N1 virus from eating pork, [the] CDC was able to immediately answer questions and clear any confusion, rumors, or misinformation” (Kass-Hout & Alhinnawi, 2013, 9). During this time, using Facebook allowed the CDC to “educate the public about the disease and the importance of vaccination. This provided [the] CDC the opportunity to engage with the public through a dialogue, spread public health messages and quickly correct information” (Kass-Hout & Alhinnawi, 2013, 9). Correcting misinformation early is especially important as members of the public begin to form mental models and initial attitudes regarding a new threat or behavior, as was the case with the novel coronavirus SARSCoV-2. After the formation of initial attitudes and mental models has taken place, social media may allow public health officials a valuable window through which to continually monitor attitudes and compliance with public health behaviors. While many people in the modern world appear eager to share their thoughts and behaviors, the reliability of such information and the potential gap between comments posted publicly and those undertaken in everyday life remain important considerations as social media is integrated into public health practice. Indeed research in the field of business shows that “if a person likes a product page, they are more likely to buy the brand, recommend the brand to others and share branded content. However, liking the page does not [necessarily] result in purchasing the product. This may be true for public health as well. Liking a page or post may not equal following behavioral recommendations or participation in public health programs’ (Thackeray et al., 2012, 5). Regardless, social media has already shown much promise in this area, as one recent study “assessed the spread of vaccination 138 sentiments from person to person during the unfolding of the H1N1 pandemic…[and] found that anti-vaccination sentiments could reliably be assessed across time and space, and that those sentiments seemed to cluster in certain parts of the online networks” (KassHout & Alhinnawi, 2013, 10). In sum, “as behavioral interventions are becoming increasingly important in public health, the potential of social media to study person-toperson transmission of health behaviors and sentiments in very large populations is unparalleled, and offers a clear benefit that traditional sources rarely provide” (Kass-Hout & Alhinnawi, 2013, 10). Because social media is still in the early stages of use in public health, however, it is worth understanding in greater detail the importance of engaging with audiences within this field and why public health officials have had difficulty engaging on social media in the past. One challenge noted by Heldman et al. is ubiquitous across any truly “social” use of social media, that is, “potential risks of and fears related to social media engagement includ[ing] loss of control of the message...concerns about negative comments misdirecting and reshaping conversation, or ‘online’ incivility” all of which are valid concerns and should be “factored into the management of an organization’s social media engagement strategy” (Heldman et al., 2013, 10). It must also be noted that, though many common social media platforms allow users to create accounts for free, managing large-scale social engagement well is resource-intensive, in order to “maintain a consistent, coordinated engagement approach. Organizations must be willing to commit the time to provide appropriate oversight of the effort, routinely monitor and respond to feedback and questions received, identify engagement opportunities, review data from engagement efforts, and adjust strategies accordingly” (Heldman et al., 2013, 11). While 139 social media accounts associated and operated by an organization likely have greater access to the needed resources, individuals operating their own account are left to expend the time and effort of engaging with their audiences themselves. Heldman et al. note an additional challenge in public health: “our default public health perspective; in public health we are trained to look at populations in the aggregate. We are not always comfortable in thinking about one-on-one interaction on the individual level. How to mitigate these risks and concerns, while also leveraging social media engagement, is an important area of inquiry for the field to move forward,” though a shift in thinking to allow for individual- as well as population-level initiatives will be needed (Heldman et al., 2013, 10). While social media has the potential to build relationships and engage with users, this has clearly been a challenge for health departments. Regardless, it must become an important focus for social media use in public health, as this capability for interaction is what sets social media apart from other modes of communication and where the greatest opportunities for social media use in public health lie in the near future. It is indeed vitally important for public health organizations to invest the necessary resources in developing a strong, engaging social media strategy for a variety of reasons. First and foremost, Heldman et al. note that “conversations in social media about public health topics are happening whether we participate or not” (Heldman et al., 2013, 12). Second, social media engagement is an important way of fulfilling broader public health goals that have remained important in the field for decades: “engagement is not a strategy for public health social media communication work. It’s an organization’s core value, philosophy, and culture. It’s an acknowledgement and a commitment to our users--the 140 people in ‘public’ health--that they matter, that we care and are listening, and that their opinion matters” (Heldman et al., 2013, 11-2). Third, engaging with users rather than simply broadcasting messages can enhance the persuasiveness of messages for a variety of audiences, as “engagement keeps us from being perceived as tone deaf or insensitive to the concerns and needs of our audiences--if we are engaged, truly engaged, we are constantly listening, responding, interacting, learning, and growing” (Heldman et al., 2013,12). In the modern world, there is a widespread “expectation of direct relationships with organizations and individuals through social media” (Heldman et al., 2013, 9). The absence of engagement may render social media useless for public health officials, for, as Hedlman et al. argue, “we risk losing our audience if we do not engage [while] others will step in to fill the information gap” (Heldman et al., 2013, 12). While it is clearly important to engage with audiences, how, exactly, to do so in a public health setting remains the topic of much research and debate. As of yet, Heldman et al. propose a model consisting of seven “levels” of engagement. At a low level, “one of the most basic forms of engagement is to use social media to identify the health information needs of users” through the use of social media monitoring tools (Heldman et al., 2013, 6). They also recommend adopting a practice that has been widely used in commercial social media policies: working with influencers. Influencers on social media “can include both organizations and individuals and exhibit characteristics of credibility, persistence in convincing others, and ability to drive conversations so that others take notice of the topic or idea and show support” (Heldman et al., 2013, 7). While this may be a “simple and powerful way to increase the reach of messaging and align it with other public health influencers, [nevertheless,] caution is advised...when reaching out to and/or 141 partnering with social media influencers as they often achieve such a status because they are perceived as being independent and trustworthy. Public health organizations must be sensitive to the possible risk of influencers becoming seen as ‘spokespersons’ for their organizations” (Heldman et al., 2013, 7). A mid-level of engagement is perhaps the most obvious form of engagement, responding to questions and comments made by users. In order to be effective, however, this must “include responding to health-related questions and comments--both negative and positive--received through organizational channels,” which Heldman et al. argue can be accomplished by “establishing a nimble, responsive infrastructure [that] allows an organization to thoughtfully and efficiently provide feedback to such social media inquiries” (Heldman et al., 2013, 7). Negative comments made by users must be a central concern of public health officials’ and organizations’ use of social media because they have been found to have a greater impact, on average, than neutral or positive comments. Drawing on the psychological literature of persuasion discussed previously, as well as additional studies conducted in business contexts, there are ways to mitigate the effects of negative comments made by users on social media. As noted previously, people employ a variety of biases and heuristics, consciously or subconsciously, when evaluating information on social media. One of these biases is “negativity bias,” a well-documented phenomenon in which negative information bears more persuasive weight than positive information (Rim & Song, 2016, 478). Further, the Edelman Trust barometer survey consistently finds that the majority of respondents view comments made by other users, what the survey calls “a person like you,” to be highly credible and trusted sources of information, with negative peer comments thereby having a greater capacity to influence users’ evaluations of products, companies, and campaigns 142 in a business setting, as well as public health organizations and their efforts to protect the health of a population (Rim & Song, 2016, 478). The most widely-accepted explanation for negativity bias currently is that “negative information is generally perceived as more diagnostic than positive information in categorizations because negative information is perceived to be more informative and less ambiguous. Thus, [in a business setting,] when reading a negative comment about a product or company, people tend to attribute the comments to the product itself rather than the reviewer and are more likely to be influenced by the negative comments” (Rim & Song, 2016, 479). In response to negative comments, multiple studies have found that two-sided messages (those that contain both positive and negative information) are more effective than one-sided messages containing only positive information for a variety of reasons, including “boosting perceived altruistic motives...reducing perceived negativity in the public’s comments, and eliciting favorable attitudes” (Rim & Song, 2016, 490). Two-sided messages are also considered a stronger form of engagement, signaling to audiences, in the business context of Rim & Song’s analysis, “that the company listens and responds to what people say, and that it meets the public’s expectations” (Rim & Song, 2016, 492). Finally, two-sided messages have also been found to increase the persuasiveness of messages, shifting users from the peripheral route of persuasion in the Elaboration Likelihood Model discussed previously, to the central route: “negative information [contained in a two-sided message] is more likely to motivate individuals to assess information with a greater scrutiny than positive information, thus it may increase the individual’s elaboration efforts. A two-sided message that contains both positive and negative information requires more cognitive resources than a one-sided message, and therefore, yields better persuasion” (Rim & 143 Song, 2016, 490). At higher levels of engagement, public health organizations can work actively to “identify opportunities to connect directly with users, and to facilitate discussions between users. These ‘small acts of support’ may have a ‘ripple effect’ across social media sites, leading to increased engagement among other users” while additional, highlevel engagement strategies may include soliciting user-generated content and perhaps integrating online and offline engagement (Heldman et al., 2013, 7-8). Other organizations that have utilized this last strategy include NASA, which “provides inperson opportunities for users who currently engage on NASA social media accounts, such as meet and greet sessions and behind the scenes events to connect with NASA scientists” while the American Red Cross “offers training to individuals using social media on behalf of the Red Cross” (Heldman et al., 2013, 8). More work must be devoted to testing and developing evidence-based strategies, though the recommendations above provide a valuable starting point. DIVING INTO SOCIAL MEDIA DATA While modern communication technologies present a range of possibilities for public health activities, it is not only the lack of research that has left public health officials without clear guidelines for its optimal use. For example, the nature of data created by these technologies may be difficult to handle, as “in the 21st century, data are not just numbers; it is YouTube videos, Twitter posts (or ‘tweets’), crowdsourcing information (which is engaging large groups of people to perform a task)” and many others, creating so much data that, as of the time of Kass-Hout & Alhinnawi’s writing, an “estimated...90% of the data stored in the world today has been created in the past 2 144 years” (Kass-Hout & Alhinnawi, 2013, 7). Clearly, there is a large and ever-increasing volume of data, but it must be noted that “the validation of large, noisy data sets poses enormous challenges” (Kass-Hout & Alhinnawi, 2013, 10). However, this also leads to one of the primary advantages of this type of data: “one of the key advantages of online social media data, apart from the increasingly large data volumes, is that they are highly contextual and networked” (emphasis added) (Kass-Hout & Alhinnawi, 2013, 7). These two characteristics in particular make social media a promising way to engage the public and “social media is providing hope to answer some fundamental questions in the public health arena, including the identification of non-cooperative disease carriers…adaptive vaccination policies, augmenting public health surveillance for early disease detection and creating disease situation awareness picture, updating or enhancing our understanding of the emergence of global epidemics from day-to-day interpersonal interactions, while engaging the public and communicating key public health messages” (Kass-Hout & Alhinnawi, 2013, 7). NETWORK EFFECTS On social media, networks are constructed with both users and information. Because the psychology behind the ways in which people interact on social media has been discussed previously, this section will instead focus on the arrangement of social media data in networks and the implications for social media use in public health. The design of social media is intended to promote connections between posts as well as people. Social media, and many communication efforts on such platforms, benefit from network effects, defined as a situation in which the “payoff depends positively on the number of other people who consume the product” or participate in the activity (Qiu 145 et al., 2015, 82). In other words, “the product or service becomes more valuable as more people use them” (Qiu et al., 2015, 82). While these definitions focus on a commercial context, the same logic may be applied to public health recommendations during the COVID-19 pandemic. If epidemiologic measures such as the case positivity rate, hospitalization and ICU capacity, case counts, etc. are considered the measures of success, then public health recommendations to prevent the spread of COVID-19, such as face masks, handwashing, and social distancing, become more successful as more people engage in these behaviors, due to the infectious nature of the disease and primary mode of transmission between people. One powerful way to form networks on social media is through the use of hashtags, a practice that began on Twitter but has since expanded far beyond this platform and even outside the digital world. A hashtag is simply “a keyword or phrase preceded by a hash symbol (#) that people include in their social media posts. Essentially, it makes the content of [the] post accessible to all people with similar interests,” beyond those who follow the account (Knapp & Baum, 2015, 131). While it may be possible to search using hashtags on social media platforms, most users do not engage with them in this way. Rather, “hashtags [often] appear as clickable links within a post and, at the click of a mouse, will display a real-time live feed of every other post tagged with the same hashtag” (Knapp & Baum, 2015, 131). The initial “idea of hashtagging was to add a # symbol before a relevant keyword, which allows users to link their postings to other content around the same topic,” for purposes of content organization and reach and these remain important uses of hashtags on many social media platforms (Rauschnabel et al., 2019, 473). As of 2018, the Twitter support page affirmed these intentions with their 146 explanation of hashtags, saying that they “permit users to ‘index keywords or topics’ of their own content and to ‘easily follow topics they are interested in.’ Thus, Twitter recommends that people use the [hashtag] ‘before a relevant keyword or phrase in their Tweet to categorize those tweets and help them show more easily in Twitter search’” (Rauschnabel et al., 2019, 474). However, hashtags are used for other reasons beyond simply searching for similar content and “understanding…[a] target group’s underlying motivations of hashtagging is crucial for effective campaign management,” though research in this area as late as 2019 was scarce (Rauschnabel et al., 2019, 473). In addition to organizing content and reaching broader audiences, hashtags also “connect interested people and network them into one discussion through social media networks” and fulfill the purposes of “marketing, attracting, organizing, retrieving, grouping, finding people of the same interest and communicating content” (Buarki & Alkhateeb, 2017, 301). As with nearly every other element of social media, the use of hashtags is also greatly influenced by social norms and the behaviors of others. As Buarki & Alkhateeb found in their study of hashtag use on Instagram, “the use of a hashtag depends on its popularity through posting, on the followers of the account(s) and survival time” results that were similar to another study that found “the content of an image and social cues ([through] additional features, such as ‘likes,’ ‘comments,’ hashtags and mentions) contribute to the popularity of an image, thus strengthening social/personal relations among users” (Buarki & Alkhateeb, 2017, 301). Here, then, there are two sets of network effects at play: first, the use of hashtags, as the hashtag becomes more popular and benefits greatly when more people use it; second the network effects among users, as users also gain social capital 147 with greater numbers of followers and other forms of digital “support.” Rauschnabel et al. encompass these ideas and more by identifying a total of ten uses for hashtags on social media including organizing and reaching broad audiences, many of which are heavily influenced by psychological concepts discussed previously. The first of these is for “amusing...which reflects the desire to be funny, [entertain] followers, and [make] them smile,” with “hashtaggers motivated to amuse others tend[ing] to use funny words as hashtags but make these decisions rather spontaneously” (Rauschnabel et al., 2019, 483). Second is organizing content, which Rauschnabel et al. argue applies not only to social media posts, but also “encompasses the broader understanding of organizing one’s own social media participation, including showing and linking other related content areas and organizing own postings around topics,” a motivation they argue is stronger on Twitter than on other platforms like Instagram or Facebook (Rauschnabel et al., 2019, 483-4). Third is the activity of “designing, [which] refers to the desire to create unique and creative postings” and may derive from “the need for self-presentation, which previous research suggests is one of two basic needs behind the broader motivation for using social media platforms” (Rauschnabel et al., 2019, 484). The fourth motivation is especially influenced by social norms and is highly relevant on Twitter: the desire to confirm. Here, “confirming” refers to “the desire to meet the conventions of either...friend groups or the respective social media platform in general” (Rauschnabel et al., 2019, 484). Fifth is what Rauschnabel et al. call “trendgaging...a portmanteau derived from the words ‘trends’ and ‘engaging’ that describes the motivation to engage in and be associated with popular conversations and trendy topics” (Rauschnabel et al., 2019, 484). Rauschnabel et al. argue this desire may stem from “a 148 deeper need for belonging...which drives social media users in general” and “is represented significantly stronger on Twitter than on Facebook or Instagram” (Rauschnabel et al., 2019, 484). The sixth motivation is “bonding, [which] is more intimate in nature than trendgaging and describes the desire to show that one belongs to an inner circle of friends and connects with them by using hashtags with an insider content” (Rauschnabel et al., 2019, 484). Clearly, network effects are valuable in a variety of contexts when there is a desire for content to spread across wide audiences; however, network effects may be particularly valuable in public health. Prior research has shown that “risk factors--such as drug abuse, smoking, poor diet and exercise--and the associated diseases are often found to be clustered in a population” (Kass-Hout & Alhinnawi, 2013, 7). Indeed, this is supported by research in social psychology regarding norms within groups discussed previously. Not only do people tend to surround themselves, in everyday life or in online communities, with people who are similar to themselves in some way, they also follow norms of acceptable behavior established within those groups, whether they be behaviors that enhance or destroy health. Further, with regard to public health promotion, social norms “a key element of many behavior change interventions, especially when the goal is en masse, sustained behavior change,” and they may also be used as “a crucial component of motivation and behavior and thus a key to behavioral influence and change” (Reynolds et al., 2015, 45). Seventh is the desire to inspire, reflecting “the aim to encourage or stimulate others to think about the meanings of one’s post[s]” or engage in a desired behavior (Rauschnabel et al., 2019, 485). Eighth, Rauschnabel et al. discuss the motivation for greater reach, “the goal of bringing one’s message or opinion to a broader 149 audience and address[ing] people interested in [that] topic,” but note that “this motivation is part of the more general motivation to develop a sense of empowerment, which includes leading and influencing social opinion” (Rauschnabel et al., 2019, 485). Ninth is the use of hashtags for summarizing, “the desire to recap and reflect on the meaning of one’s post[s] and to emphasize the main message. The hashtagging activity resulting from this motivation might serve users in two ways: (a) by fulfilling their cognitive needs for structure...and (b) by fulfilling their need to be understood by their audiences” (Rauschnabel et al., 2019, 485). Summarizing may be especially useful on Twitter, a platform that limits posts to a small number of characters. Finally, the motivation to endorse “other people, brands, or events or championing other things or topics that one finds interesting,” noting that, yet again, “one driving force behind the endorsing motivation comes from established social norms, such as responsibility (helping those who need help), justice (helping those who earned help), and reciprocity (helping those who helped you)” (Rauschnabel et al., 2019, 485). Hashtagging is clearly much more than a widely-used system of categorizing and accessing data--the multiple uses of hashtags and the role of network effects overall must be understood by government and public health officials if they are to use this valuable tool wisely. CONTEXTUALIZED DATA For social media, context may exist at multiple “levels,” including the words within the message itself (i.e. what words are used to frame and present the overall message, outside the central points), within the thread (i.e. what post a user is responding to, what comments they make, and how other users have responded), as well as within the users’ overall activity (i.e. the user’s overall posting and responding activity and other 150 users’ responses). First, while the context within a message or comment may provide useful insights for public health officials, they must also be aware that users’ ability to shape a message around a piece of data also allows such users to spread false information in a highly-persuasive way. As Marin describes, “contextual [mis- and dis-information (MDI)] may appear when placing genuine information in a fabricated setting...a large extent of COVID-19 related MDI shared on social media was contextual or, as some have put it, reconfigured” with one study finding that “most (59%) of the misinformation in [their] sample involves various forms of reconfiguration, where existing and often true information is spun, twisted, recontextualized, or reworked. Less misinformation (38%) was completely fabricated” (Marin, 2020, 2). Drawing on the Elaboration Likelihood Model discussed previously, social media is therefore prime territory for shifting users toward the peripheral route of message processing because, in addition to the large volume of data available and accessible on social media, data on social media is presented with context that may overload users, rather than as a single datapoint. Indeed, “studies in the psychology of social media have shown that, under conditions of information overload, users will revert to using mental shortcuts or heuristics for assessing new information...Taking these mental shortcuts occurs ‘in conditions of low motivation and limited ability to process the incoming information,’” the two elements required for elaboration to occur (Marin, 2020, 1). According to Marin, there was not one, but two separate issues: “the rapid propagation of misinformation and disinformation (MDI) and the so-called infodemic. Both problems concerned how information travels in an online social network medium, but only one of them was tackled with some degree of efficiency” (Marin, 2020, 2). In 151 response to the spread of mis- and dis-information, the standard response has thus far been largely by “social media platforms [that] rapidly stepped up with pre-existing measures of dealing with MDI and targeted specifically the COVID-19 related misleading information...In the wake of the COVID-19 pandemic these efforts for factchecking were accelerated to an impressive extent: ‘the number of English-language factchecks rose more than 900% from January to March [2020]’” (Marin, 2020, 2). Meanwhile, the infodemic, defined by the WHO as “an overabundance of information-some accurate and some not--that makes it hard for people to find trustworthy sources and reliable guidance when they need it...was seen as a side-effect of the intensification of user interactions of social media, some informational noise that accompanied the humming of online communications,” and has thus far failed to be addressed effectively (Marin, 2020, 1-2). Marin also identifies three levels of implied or unstated contexts that have been largely ignored in public health uses of social media, rendering attempts to use social media to prevent the spread of false information only partially successful during the COVID-19 pandemic. The first of these levels is the emotional context surrounding the user’s interaction with a message. Prior to the COVID-19 pandemic, “it was already shown that MDI propagates on social media platforms by playing on the emotional reactions of the online audience” (Marin, 2020, 3). However, emotional processing is also built in to the structure of common social media platforms themselves: “users of social media platforms are allowed a palette of actions and reactions: some are seemingly neutral (commenting, sharing and posting) while others have a clear emotional valence: liking and using other emoticons to endorse or dislike a post. These emotionally charged reactions are easier to perform than the neutral ones: it takes a split second to click ‘like’ 152 on a post, but some more time to comment on it or even share it. Most of these emotional reactions have dedicated buttons which can be clicked mindlessly, yielding the interaction seamless” (Marin, 2020, 3-4). The second context is what Marin calls a “weak epistemic context in which information is not always shared [in order] to inform” (Marin, 2020, 4). There are indeed a variety of reasons users may share information on social media besides informing others and these motivations may be affected by changing circumstances. For example, one study comparing the use of social media between teenagers and adults found that “teens...tended to post personal topics and focused on self-expression. They used social media as a ‘conversation space’ because they were more aware than adults of using hashtags” (Buarki & Alkhateeb, 2017, 290). However, “in a crisis situation, users tend to change how they use the platform and [shift] towards the communication of vital information such as imminent risks or their location and also [seek] to be informed by latest developments from people from the local site of the disaster. The entertainment function tends to become secondary in emergencies” (Marin, 2020, 4). The COVID-19 pandemic clearly does not meet this definition of an emergency, though, because “the crisis [is] global and...the duration [has been] rather long. This time, the uncertainty that accompanies a crisis situation was extended over months. As epistemic agents, online users tried to make sense of what was going on with them, what they could expect and to assess the personal risks, over a longer period” (Marin, 2020, 4). Marin argues that this weak epistemic context is yet another reason public health organizations’ use of social media as a platform primarily to broadcast messages is ineffective. In the face of COVID-19, some users have a desire “to be an expert so that they could at least 153 understand what was happening to them. People did not want to be experts in epidemiology, quarantine measures, and home remedies for viruses because of a sudden intellectual curiosity. They needed a way of coping that was also understandable to them. Meanwhile, the official discourse of ‘trust the experts’ and ‘please don’t share information you do not understand’ incapacitated them as epistemic agents” (Marin, 2020, 5). This conflict may then spark reactance, as, “requiring users to do nothing and just comply [with public health guidelines] went against the general desire to do something, as a way to take control” (emphasis added) (Marin, 2020, 5). The last level of social media context discussed by Marin is a strong normative context. While social media may provide varying levels of anonymity for users, social norms and acceptance among a perceived group, as discussed previously, remain strong influences on social media behavior. Marin argues this strong normative context is another reason why previous attempts to curb the spread of false information have been ineffective. Typical false claims “are not merely descriptive claims of a state of affairs in the world, but often embedded in a normative context be those prescriptive or evaluative claims, both types are meant to change attitudes of the online users. [Mis- and disinformation] was shared because it prescribed actions or led to evaluations of the state of affairs which users already agreed with. Hence debunking the facts would have solved only half of the puzzle, since the user’s motivation to believe these normative claims would have not been dealt with” (emphasis added) (Marin, 2020, 5). Indeed, “even if the scientific claims of regular users are checked,” which would be incredibly difficult due to the ever-increasing volume of posts, “[users’] aim remains to prescribe actions for others and to evaluate the world in a way that will be endorsed by one’s community of friends... 154 Content-based approaches are then ineffective against this strong desire for social media users to emit evaluative or prescriptive claims about the world and strategically use science-looking sources to back those up. One should address the very desire of regular users to evaluate the world from the little soap-boxes that social media affords” (Marin, 2020, 6). Because social media data is highly contextualized, it may provide not only information on a user’s or event’s time and place, but also additional context into the lifestyles and attitudes of individuals that may affect their health. This may be a valuable window into social determinants of health that may be otherwise difficult to assess with traditional methods. Indeed, “as behavioral intentions are becoming increasingly important in public health, the potential of social media to study person-to-person transmission of health behaviors and sentiments in very large populations is unparalleled, and offers a clear benefit that traditional sources rarely provide” (Kass-Hout & Alhinnawi, 2013, 10). Internet technologies move data quickly, meaning that if public health officials can harness this data, they can gather insight and act much more quickly than traditional data sources allow. Indeed, “the timeliness of health data limits the availability of actionable public health information as the traditional route for the data moves from patient self-report to a physician, through diagnostic confirmations, and then from a physician laboratory to a public health authority” (Kass-Hout & Alhinnawi, 2013, 8). While “health data found on social media circumvent the traditional route by removing the ‘middleman,’” there are also significant tradeoffs to the timeliness of such data, namely that information is not validated or confirmed (Kass-Hout & Alhinnawi, 2013, 8). As Kass-Hout & Alhinnawi describe, “the ‘middle man’ plays a critical role. 155 They validate and can tell authorities something about the population they serve where generalizations of this information can be made with a level of confidence. That confidence is crucial in triggering a governmental action or intervention” (emphasis added) Kass-Hout & Alhinnawi, 2013, 8). Thus, what social media data may add in timeliness, it may lack in quality. As such, “one [recommended] preliminary way of analyzing social media data is through cross validation with other sources, which may help distill rumors early” (Kass-Hout & Alhinnawi, 2013, 10). 156 ANALYSIS OF TWEETS: GOVERNMENT AND PUBLIC HEALTH IN UTAH The following analysis contains analysis of Twitter accounts important to the management of the COVID-19 pandemic in Utah. Accounts surveyed include Governors Gary Herbert (governor during the beginning of the pandemic) and Spencer J. Cox (Lieutenant Governor under Herbert before being elected as his successor in 2020), as well as State Epidemiologist Dr. Angela C. Dunn. All of these accounts are associated with public figures with posts being publicly viewable. The Twitter account created for the Utah government’s response to COVID-19 was also included. The accounts of healthcare institutions that operate hospitals recognized by the Utah Hospital Association were also included where a Twitter account could be identified: Intermountain Healthcare, University of Utah Health, MountainStar Health, and Steward Health. Finally, the Utah State Health Department as well as county health departments for all 29 counties in Utah were included, though it must be noted that many of these county health departments in fact serve more than one county. Tweets were accessed and downloaded using the Twitter Developer API and the Python 3 programming language. All tweets posted between March 1, 2020, and March 1, 2021, were collected. In cases where the Twitter account exceeded the maximum 3,200 tweets allowed by the available Twitter API access, the most recent 3,200 of the tweets posted during that time were included. All tweets were collected May 6, 2021. A total of 16,986 tweets were collected and analyzed, including 14,080 from notable people, healthcare systems, and health departments in Utah, and 2,906 from the Utah Coronavirus Info account. The following analysis is purely exploratory for the purposes of proposing hypotheses, as no experimental controls were set in place. Additional processing and all graphs were 157 constructed using the R programming language. ANALYSIS OF TWEET CONTENT Tweets were collected from three types of accounts: people (3 accounts, described above), healthcare systems (4 accounts: Intermountain Healthcare, University of Utah Health, MountainStar Health, and Steward Health), and health departments (14 accounts: Utah Department of Health and 13 accounts serving all 29 counties). Terms relating to COVID-19 were searched in the text of each tweet posted during the one-year range (see Table 1, Figure 4). Dividing lines in Table 1 indicate the terms included in each category, described in greater detail below. The descriptive statistics below clearly indicate there is a wide range of tweeting behavior by different accounts. The type of account with the greatest average number of tweets was “people,” though it must be noted that this is an underestimation because the account of Governor Cox had posted tweets during this period in excess of the limit allowed by the Twitter API. Healthcare systems were the next most frequent with an average of 781, though this average was also affected by a noticeable outlier, Steward Health, which posted only four tweets. The average number of tweets for departments was the lowest with 467.5, but this value was greatly impacted by outliers, ranging from 18 tweets to well over 1,000. By far, “COVID” was the most frequently used term, followed by #COVID, which was searched for separately in the hashtag information collected by the Twitter API. Healthcare systems used both “COVID” and #COVID most frequently, with an average of 222.75 and 122.5 uses respectively, followed by the “people” accounts, which averaged 196.3 uses of “COVID” and 61.3 uses of #COVID. The “people” accounts more frequently referred to the virus as “coronavirus,” though usage was much lower 158 than COVID at an average of 55 uses. Tweets posted beginning in March 1, 2020 were collected, which was relatively early in the pandemic and it was more common to refer to the virus as the “novel coronavirus.” Table 1 Descriptive Statistics for Terms Found in Tweets (averages across accounts) Account Type Term Health Department Healthcare System People COVID 158.6429 222.75 196.3333 coronavirus 24.2143 38.25 55 pandemic 9.3571 34.75 34.6667 cough 2.1429 4.5 2 short breath 2.2857 4.25 3.3333 fever 1.9286 4.5 1 taste loss 1.0714 0.25 0.3333 smell loss 1.0717 1.25 1 mask 23.4286 68.25 86 face covering 8.357 4.5 6.6667 distancing 20.9286 23.75 35.6667 six feet 6.2857 2 3 stay home 5.8571 6.5 11.6667 quarantine 3.2143 4.25 7.6667 handwashing 37.3571 11 9.3333 caregiver 1.0714 28.5 1 hero 1.0000 14.75 8 front line 0.8571 12.75 2.3333 ICU 5.3571 13.5 28.6667 hospitalization 1.9286 4.75 14.3333 #COVID 0 122.5 61.3333 #AllIn 0 1.75 1 #MaskUp 0.0714 29.25 6.3333 #Hero 0 3.5 0.3333 #Frontline 0 0.25 0 5 781 1775 586 1469.6667 2926 Minimum total tweets 18 Average total tweets 467.5 Maximum Total Tweets 1411 159 Figure 4. Graph of the average frequency of terms found in tweets by account type. The terms “coronavirus” and “pandemic” were used somewhat frequently, though “COVID” was found in the tweet body over four times more frequently on average across all accounts. Evidently, these accounts frequently referred to SARS-CoV-2 most commonly using the word “COVID,” which is more commonly used across members of the general public. Terms referring to protective behaviors (mask, face covering, distancing, “six feet,” and handwashing) were used much more frequently than terms describing symptoms (cough, shortness of breath, taste loss, and smell loss) across all account types. This seems to indicate that regardless of whether posts are made by a health department, a healthcare system, or a notable government or public health leader in Utah, tweets were made more commonly to encourage appropriate behavior change than to educate the public about symptoms. The term “mask” was used most frequently among these behaviors, followed by the term “distancing” across all types of accounts. Both “mask” 160 and “distancing” were used most frequently by “people” accounts, followed by healthcare systems, while health departments used both terms an average of less than 25 times. After the term “COVID,” the terms most frequently used by health departments were “handwashing” and “mask,” though each was used an average of less than 50 times. Perhaps unsurprisingly, healthcare systems seemed to have a particular focus on those working on the front lines, as these accounts were the most frequent users of the terms “caregiver,” “hero” and “frontline,” though each was used, on average, less than 30 times with “caregiver” being the most frequent among these. Across all account types, the term “ICU” was referred to more frequently than the term “hospitalization” implying a greater focus on those more severe cases of COVID-19. Given the nature of Twitter and social media more broadly, it is worth examining hashtag use as separate from language used in the body of the tweet. As noted previously, hashtags can serve a variety of powerful purposes and may therefore provide multiple benefits if used properly. However, in this data, hashtags were used by health departments extremely rarely: the only one of the five hashtags that was used was #MaskUp, with an average of only 0.0714 uses while all other hashtags had an average of 0. Just as the term “COVID” was used most frequently in the body of tweets, #COVID was the most frequent hashtag used, averaging 122.5 among healthcare systems and 61.3 among the “people” accounts collected. However, all accounts used #COVID much less frequently than the term “COVID” itself, meaning that there are clearly tweets referring to COVID-19 in which this popular hashtag was not used. The term “mask” was the second most frequently used term in the body of tweet text, and #MaskUp was accordingly the second most popular hashtag used, averaging 29.25 among healthcare 161 systems and 6.3 among “people” accounts. Across all hashtags surveyed, healthcare systems used them more frequently than the people surveyed. However, given the average total number of tweets and hashtag use in Table 1 above, these hashtags appear to be used in less than 20% of tweets, meaning that in addition to the tweets these accounts post that are about other topics, these accounts are missing the benefits of popular hashtag use in a large number of COVID-19 related tweets. Correlation analysis was then conducted to analyze the relationships between terms used in the body of tweets across all accounts. As described previously, all tweets within the one-year period were collected and the terms were searched in the body of each tweet. Then, the total frequency of term use was calculated for each account. For the below correlation analysis, each account constituted a row of data with the total number of times each term was used. A correlation matrix was then calculated with the results shown in Figure 5. It must be noted that this correlation analysis is not stating that the terms that are highly and significantly correlated are used frequently together within the same tweets. Rather, this correlation analysis takes a larger view of the terms that are used in a consistent pattern of frequency over the entire one year period. Figure 5 (below) shows both which terms have a significant correlation and those that do not, at the .95 significance level. Perhaps unsurprisingly given the terms searched, nearly all terms are positively correlated, with the exception of the term “taste loss” with terms related to healthcare workers: “caregiver” (r = -.13), “hero” (r = -.08), and “frontline” (r = -.14)-- though all of these correlations are statistically significant, they are weak correlations. Strong positive correlations that are statistically significant include “face covering,” “distancing” and “six feet,” all protective behaviors that are more 162 effective when properly practiced together. However, terms encouraging people who are infected with COVID-19 to “stay home” or “quarantine” had fewer statistically significant correlations. “Stay home” was only significantly correlated with “hospitalization” (r = .48) while “quarantine” was only significantly correlated with the terms “coronavirus” (r = .73), “pandemic” (r = .54), and “mask” (r = .68). In contrast, terms relating to healthcare workers, “caregiver,” “hero,” and “frontline,” had multiple statistically significant correlations, not only among these three terms but also terms relating to some symptoms (“fever,” “cough”), hospitals (all three healthcare worker terms were significantly correlated with the term “ICU,” while only “hero” was significantly correlated with the term “hospitalization”). Symptoms were also significantly correlated, though “smell loss” was only significantly correlated with “COVID” (r = .77) and “taste loss” (r = .74). 163 Figure 5. Correlation coefficients among all text terms and across all account types significant at .95 level. Interestingly, while “COVID” was the most frequently used term, it had few statistically significant correlations, with “pandemic” (r = .70), “short breath” (r = .85) and “smell loss” (r = .77). While a less frequently used term to describe the virus, “coronavirus” had more statistically significant correlations, including “pandemic” (r = .69), “short breath” (r = .48), “mask” (r = .79), “distancing” (r = .73), “quarantine” (r = .73), “hero” (r = .64), and “ICU” (r = .47). This may indicate that while the term “COVID” is used most frequently across all these accounts, there is less consistency regarding which terms are also used frequently by these accounts. 164 Figure 6. Network plot showing communities of term frequency across all accounts. By contrast, the less frequently used term “coronavirus” shows a more consistent pattern of use on these accounts with the terms described. Figure 6 (above) depicts the same terms but arranged in an adjacency matrix and identifies communities of terms used in patterns of frequency across all types of accounts. This network plot provides a visual representation of trends in term use and identifies six separate clusters. Interestingly, none of the clusters with more than 3 terms are comprised of a single category of terms (i.e., terms relating to symptoms, healthcare workers, protective behaviors, or general COVID-19 terms). The three smallest clusters are comprised of one or two terms: “handwashing,” “stay home,” “coronavirus” and “quarantine.” “Handwashing” was the most frequently used term after “COVID” by health departments, but this term is not part of any clusters of terms, nor is it linked by a significant connection to a larger community 165 or the overall network. This implies that one of the terms used most frequently by health departments in Utah lacks a consistent pattern of use with other important terms related to COVID-19. There are three large clusters identified, with each consisting of five or more terms. One group contains the terms: “caregiver,” “frontline,” “hero,” “fever,” “cough,” and “short breath.” Cough, fever, and shortness of breath are three of the most common and earliest-recognized symptoms of COVID-19, while the other symptoms included in this search, “taste loss” and “smell loss” are in a separate group on the opposite side of the overall network. This cluster contains the terms “smell loss,” “taste loss,” “distancing,” “face covering,” and “six feet.” In the middle of these two communities is one community that contains some of the most frequently used terms: “COVID,” “pandemic,” and “mask,” along with the two terms relating to healthcare settings, “hospitalization” and “ICU.” Based on this network plot, it appears that there are distinct sets of terms used to tweet about the COVID-19 pandemic, which may indicate inconsistency in language across accounts the public would consider experts. For comparison and to better understand how these terms are used by the accounts included in this study, network analysis was repeated for each type of account, people, healthcare systems, and health departments, with the results shown in the Figures below. 166 Figure 7. Network plot showing communities of term frequency in “people” accounts. Figure 8. Network plot showing communities of term frequency in healthcare system accounts. 167 Figure 9. Network plot showing communities of term frequency in health department accounts. There are many clear differences in these three network plots in terms of how many communities are identified, how large the communities are, and what terms are included. For example, the health department accounts and healthcare systems accounts both identify at least one community consisting of a single node while the “people” account network plot does not. In all three plots, the most popular term, “COVID” appears in the central and largest community, but the terms also included in that community, as well as those excluded, differ. In the case of healthcare systems, the central community contains all terms except “quarantine,” “taste loss,” “smell loss,” “face covering,” “pandemic,” “caregiver,” and “stay home.” The “people” account network plot identifies three distinct communities: one consisting of “coronavirus,” “face covering,” and “distancing,” one consisting of “mask,” “six feet,” “short breath,” “frontline,” and “hospitalization,” and the third containing the rest. The health department account network plot identifies four connected communities with the single term “front line” being unconnected to the 168 network. It therefore seems that these accounts have different tendencies and patterns in the use of these common terms relating to COVID-19. However, all of the accounts included may reasonably be thought of as “experts” or “authorities” during the COVID19 pandemic. If the posts containing these terms have different emphases or central messages, there is the potential for the audience to perceive inconsistency or even disagreement across experts. Consistency across multiple “expert” accounts is important in communicating messages clearly with an audience because, based on previous research on health promotion messages, members of a group, even a small group, may have different understandings of the same terms. In their study of terms relating to healthy diets and food choice, a “word-association exercise produced 1,386 individual responses across [only] four prompt-terms, 260 of which were unique,” which were later categorized into 16 separate themes (Buckton et al., 2015, 4). Thus, from a small number of terms, a group of individuals can derive a wide range of meanings. These meanings, in turn, also affected associations and behavioral intentions. In Buckton et al.’s study, the four prompt terms all “elicited a different set of responses, in both the word association exercise and...focus groups. None of the terms fully conveyed the sense of eating a nutritionally balanced diet which supports health and fitness, reduces the risk of [non-communicable diseases] and may even have beneficial or therapeutic effects. [However,] none of the focus group participants were able to suggest a term or language that would be more understandable, largely because of the confusion over what is meant by food and dietary behaviors underpinning lifelong health, a complex multifactorial concept” (Buckton et al., 2015, 8). In the context of the COVID-19 pandemic, there is likely to be a similarly 169 wide range of meanings associated with the common terms searched in the tweets above. For example, the term “face covering” is much more broad than the term “mask” and may include articles like neck gaiters and bandanas shown to have little protective effect against the virus. However, the term “mask” is not without its own level of ambiguity as there are also many different types of masks and recommendations regarding who should wear masks, when and where to wear them, and what types are effective, have changed multiple times during the course of the pandemic. Many other COVID-19-related terms could be considered similarly ambiguous. The argument here is that consistency, both within a single account and across multiple accounts the audience may perceive as expert sources, is vital because of the link to behavior and behavioral intentions, as “misperceptions reduce both the opportunity and the motivation for behavior change” (Buckton et al., 2015, 11). In their study, Buckton et al. found that “public perceptions of what it means to eat for health are influenced by the language used to communicate the concept and how it is communicated, resulting in confusion and misconceptions. Repeated use of non-specific messages and communication channels has resulted in a wide scale [desensitization] and antipathy among [the public]” (Buckton et al., 2015, 12). From this logic, the question then becomes: have the COVID-19-related terms searched in the tweets above been used in a nonspecific manner thereby contributing to difficulties in this communication process? Certainly, not every tweet contains the same level of specificity, nor is it necessary for all tweets to contain ambiguous language. Over time, however, as these terms have become commonplace, it becomes easier to assume that the public has an understanding of a certain term and subsequently fail to communicate with a level of 170 specificity sufficient to account for the evolving case counts and guidelines. This tweet, posted by the Utah Department of Health, illustrates this point: Example 1. Tweet by the Utah Department of Health. August 3, 2020. Though this tweet includes a video and a text description, little guidance is provided regarding what constitutes a mask, how to properly wear one, and when it is necessary to do so with the wide variety of circumstances one might encounter during everyday activities (though some example settings are described). This tweet was posted on August 3, 2020, nearly five months after the first lockdowns in March, so it may have been assumed that the public had already gained this understanding. While the tweet below 171 posted by the Salt Lake County Health Department, posted much earlier in the pandemic, provides more guidance, there is still a significant potential for confusion over what constitutes a mask among members of the public. Example 2. Tweet by the Salt Lake County Health Department. April 22, 2020. This tweet includes a graphic depicting a process to make a face mask as well as text both on and above the image pointing the audience to the CDC website containing further information. A view of this web page on the date this tweet was posted is included in Appendix A. This information, however, not only contains vague language regarding the masks, but also inconsistencies between the graphic posted in the tweet above and what is found on the CDC page linked in the tweet. For example, the graphic above shows a simple procedure for making a face mask out of a t-shirt, but fails to mention how many 172 layers the mask should be, how to ensure a “snug fit” or what constitutes such a fit. The CDC provides some additional guidance, noting that masks should “include multiple layers of fabric” but fails to mention what type(s) of fabric are most effective, research that was detailed previously in this paper (Centers for Disease Control and Prevention, 2020). CDC recommendations are similarly vague in describing how frequently cloth masks should be washed, noting only that “they should be routinely washed depending on the frequency of use” (Centers for Disease Control and Prevention, 2020). A primary inconsistency in this content is that the graphic posted by the Salt Lake County Health Department says to “remember to maintain 6 feet of social distance whenever possible!” while the CDC web page linked in this twitter post advises the public to wear “cloth face coverings in public settings [only] where other social distancing measures are difficult to maintain” (Centers for Disease Control and Prevention, 2020). If an individual were to read both of these sources, they might be left with the logical questions of whether a mask is necessary when social distancing is possible, whether they are necessary if in an outdoor setting, what and how many people constitute a “public setting,” and other similar questions. It must be noted that especially in April 2020, much research to answer many of these questions had not yet occurred; however, the purpose of this section is to highlight how the knowledge that did exist at the time was communicated to the public and the effects of these methods on behavior. ANALYSIS OF TWEET SENTIMENT: GENERAL COVID-19 PANDEMIC TERMS The TextBlob Python library was used to conduct a basic sentiment analysis on all tweets collected, with a particular focus on the subjectivity, polarity, and overall analysis of the tweets. Polarity is a value in the range [-1, 1] calculated for each tweet in 173 which 1 is a positive statement and -1 is a negative statement (Shah, 2020). Subjectivity is a value that is similarly calculated for each tweet in the range [0, 1] that “quantifies the amount of personal opinion and factual information contained in the text. The higher subjectivity means that the text contains personal information rather than factual information” (Shah, 2020). The overall analysis was determined using a method in which a polarity score of [-1, 0) was considered “negative,” a score of exactly 0 was considered “neutral,” and a score of (0, 1] was considered “positive” and all tweets were labeled accordingly. In Figure 10 below, all collected tweets were combined into a single analysis with a total of 4,279 neutral tweets, 1,919 negative tweets, and 7,882 positive tweets. All tweets were separated into two categories: those containing the terms “COVID,” “coronavirus,” or “pandemic” and all others. The frequency of each sentiment (positive, negative, and neutral) in these two categories of tweets are shown below. 174 Figure 10. Bar chart showing frequency of tweet sentiment across all accounts. Both tweets containing the words “COVID,” “coronavirus” and “pandemic” and tweets not containing these words show a similar pattern of sentiment distribution. The majority of tweets convey a positive sentiment, approximately half as many tweets are considered neutral, and approximately half of that number convey a negative sentiment. Because the analysis of the tweet as positive, negative, or neutral is derived from the score the TextBlob library calculates for polarity, it is worth further examining trends in polarity. The relative proportion of tweets conveying each sentiment shows the same pattern in both COVID- and non-COVID tweets, implying that there has not been an intentional effort to communicate about COVID-19 in a pattern of tone that deviates from that used to discuss other topics. 175 Figure 11. Scatter plot showing tweet polarity and subjectivity scores for each tweet collected across all accounts. Figure 11 above shows all tweets collected separated into the same “COVID” and “non-COVID” categories discussed previously. As noted previously, tweets with a polarity score in the range [0, 1] were considered either neutral or positive. As Figure 11 shows, the majority of negative tweets, those with a polarity score in the range [-1, 0), also contain some level of subjectivity, at least 0.25 as measured by TextBlob’s [0, 1] scale. Sentiment is an interesting and important consideration for anyone attempting to craft an effective, persuasive message and research has shown mixed results regarding the impact of message sentiment on persuasion. While some researchers and advertisers have argued “that negative emotions undermine persuasion,” other studies have found that positive message framing may not be more effective than negative framing under all 176 circumstances (Block & Keller, 1995, 193-4). One of these studies demonstrated that “in effortful processing, negative frames are more effective than positive frames, because negative information is perceived to be more informative than comparable positive information” while another study found “that with a lower level of efficacy, that is, when it is uncertain that the recommendations will lead to the desired outcome, negative frames are more effective than positive frames” (emphasis added) (Block & Keller, 1995, 193). A separate study “sheds some light [on positive message framing, showing]...that positive frames may be more effective than negative frames when subjects are not motivated to process the message” (emphasis added) (Block & Keller, 1995, 193). Thus, the element of message processing, or elaboration as it is called in the Elaboration Likelihood Model, is also important in assessing the effect of message sentiment on persuasive effectiveness. With this in mind, it is worth examining the likelihood and implications of elaboration on health messages in the social media context. As noted previously, many social media platforms, and Twitter especially, are designed to promote emotional, rather than effortful, processing of information. Additionally, research has found that “low motivation to process the message often prevails in health contexts because people without apparent health problems typically engage in defensive tendencies to avoid the message” (Block & Keller, 1995, 193). Additionally, “when presented with an uncertain problem, decision makers [frequently] comprehensively analyze the situation and search widely for information; in contrast, decision makers faced with a more certain opportunity will engage in little or no analysis of the situation” (Block & Keller, 1995, 194). This would imply that members of the public were more likely to effortfully process the central message of COVID-19 information earlier in the pandemic when there 177 was less research and more uncertainty. Further, in their study of behavior efficacy on elaboration likelihood in a health context, Block & Keller found, in support of previous research, that “the extent to which people process a health-related message varies as a function of the perceived efficacy...in the case of low perceived efficacy, when performance of the recommendations is uncertain to lead to the desired outcome, people are forced to evaluate tradeoffs of compliance versus noncompliance, and, therefore, engage in effortful processing of the message” (Block & Keller, 1995, 198). Thus, “a greater extent of message processing occurs under conditions of lower perceived efficacy. Consequently, negative frames are more effective than positive frames” under these circumstances (Block & Keller, 1995, 198). In contrast, under conditions of higher perceived efficacy, “attitudes and behavioral intentions toward the recommendation [were] equally favorable for both positively framed and negatively framed messages,” meaning that message framing may have less importance under conditions of high efficacy (Block & Keller, 1995, 198). While it is true that elaboration likelihood depends, in part, on individual differences in need for cognition, there are clearly multiple factors encouraging non-effortful processing of tweets. As time has passed in the COVID-19 pandemic, two important processes have occurred in this regard: first, the evolving research showing that proper face masks and other measures are, indeed, effective in slowing the spread of the virus; second, the virus has become familiar to the public, reducing the degree of uncertainty. Based on the research, both of these processes may reduce the likelihood of elaboration, shifting the Twitter audience towards the peripheral route to persuasion. On this route, peripheral cues take precedence, such as message framing, although there are certainly many others. 178 The studies cited previously indicate that when it is unlikely that an audience will process a message effortfully, a positive message frame is typically more effective. By this logic, then, the positive sentiment conveyed by the majority of the tweets collected in this study are following the more persuasive strategy. However, it is also true that the behavior modifications recommended during the COVID-19 pandemic must be undertaken for a long period of time, extending past 12 continuous months for many countries at the time of this writing, thereby requiring more persistent, sustained attitude change. Going back to the previous discussion of the Elaboration Likelihood Model, the peripheral route to persuasion, at best, can only lead to temporary attitude change that Petty & Cacioppo describe as “susceptible and unpredictive of behavior” (Petty & Cacioppo, 1986, see Figure 3). Going a step further, it is important to understand not only that the majority of tweets convey some sentiment (primarily positive), but the extent to which they do so. In the analysis conducted using TextBlob, such a concept is represented by the polarity of each tweet, with the sign of the number denoting the sentiment, and the value denoting the extent of that sentiment. In Figure 11, the extremity of the language is therefore denoted by the y-value of each point on the scatterplot, with more extreme language being more polar, or further away from the horizontal axis (y = 0) in either direction. Across all accounts, it appears that not only is there a greater number of positive tweets, but those tweets also seem to use more extreme language. There are many tweets that cluster around 0.1 to 0.5 (out of 1) in the positive direction for polarity, while for negative tweets (below the horizontal axis), the values for polarity are more frequently below -0.25 (out of -1, with the negative sign denoting the negative sentiment) for all 179 levels of subjectivity. Figure 12 follows the same premise, plotting the subjectivity and polarity, but does so within each account category (people, health departments, and healthcare systems). Within each category of tweets, there is a similar range of typical subjectivity and polarity scores, with polarity being typically between -0.2 and 0.5 and subjectivity being primarily under 0.6. Figure 12. Scatter plot showing tweet polarity and subjectivity scores for each tweet collected by account type. Despite this similar range, the trendlines indicate different patterns in the relationship between subjectivity and polarity scores across tweets. For example, the accounts of 180 individual people have a sharper increase in polarity as sentiment increases for tweets using the words “COVID,” “coronavirus,” and “pandemic,” and while the same is true for tweets not using these words, the relationship is less linear. By contrast, healthcare systems show the opposite tendency, with a linear relationship among non-COVID tweets and a more gradual increase for COVID tweets. Health Departments show none of these trends, with both COVID and non-COVID tweets increasing in polarity as subjectivity increases until 0.5, a decrease in polarity for subjectivity scores between 0.5 and 0.65, another increase in polarity for subjectivity scores between 0.65 and 0.75, and finally, another decrease until the maximum subjectivity score of 1.0. Additionally, the polarity score for subjectivity scores of 1.0 on the trendlines for both COVID and non-COVID tweets is only slightly higher than for subjectivity scores of 0. Based on these graphs, some important conclusions can be drawn. Each account included in this study can be reasonably thought of as an expert in the COVID-19 response in Utah. However, as was the case in the content of the tweets as indicated by the patterns of term use discussed previously, there are also wide variations in the degree of subjectivity and strength of sentiment conveyed to the public. The accounts of individual people seem to be, on average, both more subjective (containing less objective, factual language) and more strongly positive in sentiment, though there are noticeably more strongly negative polarity scores as well, much more so than health departments or healthcare systems. By contrast, healthcare systems and, to an even greater extent, health departments, it appears that there are much smaller increases in polarity scores for COVID tweets as subjectivity scores increase. The extent to which this trend is an intentional effort by these organizations to primarily communicate about COVID-19 using slightly, but not strongly, 181 positive language is beyond the scope of this paper; however, these trends are apparent, with the presence of some outliers above 0.5 polarity score being visible on both graphs. Where the trends in these two types of accounts primarily differ is the trends in tweets not using the terms “COVID,” “coronavirus,” and “pandemic.” Where healthcare systems show a sharp increase in polarity scores, on average, as subjectivity scores increase, this increase in polarity scores is absent in health department tweets, where the non-COVID tweets tend to follow the same pattern of increase and decrease as the COVID tweets. Figure 13. Scatter plot showing tweet polarity and subjectivity scores for each tweet collected by account type. Figure 13 presents this data in a different way, further illuminating trends in subjectivity, polarity, and account type. This Figure clearly shows trends in not only what 182 accounts used the terms “COVID,” “coronavirus,” and “pandemic,” most frequently, but also the degree of subjectivity, sentiment, and extremity of language used to convey such messages. The majority of tweets collected do not use these terms, despite the fact that “COVID” was the most frequently used term and hashtag. However, despite the fact that health departments had the fewest average tweets and the lowest frequency of use across all terms and hashtags, a clear cluster of COVID tweets posted by health departments is nevertheless visible in a certain range with polarity scores lower than the “people” or healthcare system accounts. Particularly between 0.25 and 0.75 subjectivity scores, these health department tweets ranged more frequently between -0.3 and 0.2 in polarity, indicating not only a more negative tone than the other types of accounts when using these terms, but also a tendency to use less extreme language in either direction when using these terms. Overall, Figures 12 and 13 do not indicate a large difference in subjectivity scores across multiple tweets, though the “people” accounts may be slightly higher on average. In terms of polarity, there are clear differences between each type of account in both frequency of use and the extremity and tone of the language when using them. Health departments were, by far, the most consistent in their level of subjectivity and polarity when using these terms, frequently scoring close to a neutral tone and between 0.25 and 0.75 in subjectivity, though there are some more extreme values. Craig & Blankenship discuss the impact of the extremity of language in persuasive attempts. Linguistic extremity may be thought of as “a set of stylistic markers that increase the perceived extremity of a message’s position,” or “the degree of the message’s valence,” while message intensity is defined as “language markers that influence the extent to which the message differs from a neutral position” (Craig & 183 Blankenship, 2011, 291). Linguistic extremity is therefore a specific component of linguistic intensity. Within the context of the present study, intensity describes whether a message is positive, negative, or neutral (the overall sentiment calculated above); extremity describes the extent to which that message is positive or negative (the polarity score in this analysis). While distinct constructs, both message intensity and extremity are important concepts because they may impact the effectiveness of persuasive attempts. Linguistic (message) intensity, for example, “has been considered an important characteristic in the persuasion setting, associated with increases in perceived credibility, message discrepancy, persuasion, and resistance to persuasion” (Craig & Blankenship, 2011, 292). Additionally, multiple models, including the Elaboration Likelihood Model, predict that “linguistic [or message] extremity may increase message processing, and this increase is likely due to extremity markers increasing position discrepancy. In other words...strong arguments may become stronger (i.e., assimilation toward a message’s position and weak arguments become weaker (i.e., contrast from a message’s position) when paired with [these] markers” (Craig & Blankenship, 2011, 293). In their studies using messages on different topics, Craig & Blankenship report that “this hypothesis received partial support, such that weak arguments paired with the extremity markers led to less favorable attitudes than in the control condition…[while] in the strong argument conditions, the pattern was less clear. Strong arguments paired with the extremity markers did not yield more favorable attitudes...but the pattern was in the expected direction. [Nevertheless,] this overall pattern [does suggest] that linguistic extremity may interact with argument quality such that weak arguments are perceived as weaker and strong arguments are perceived as stronger,” while an effect was observed on behavioral 184 intentions in their second experiment (Craig & Blankenship, 2011, 299). In short, “linguistic extremity influenced message processing (i.e., elaboration), which in turn influenced behavioral intentions…[as] extreme language use in a message leads to increased intentions to comply with a message when the message is compelling rather than weak,” (Craig & Blankenship, 2011, 304). In addition to an effect on behavioral intentions, language extremity may also affect attitude strength. As Craig & Blankenship describe, “a number of [other] language variables have been found to influence attitudes and behavior by influencing the extent to which the persuasive message is processed...Attitudes based on high rather than low levels of information processing are more likely to guide future thinking and behavior” (Craig & Blankenship, 2011, 304). This is the attitude change by the central route to persuasion discussed previously. The fact that attitudes changed or formed in this way are more predictive of behaviors and more resistant to change make the central route the primary target of persuasive attempts intending to encourage compliance with public health recommendations over a long period of time. However, doing so is certainly not as easy as simply using strong language to describe strong arguments. Strong or extreme language must be used with caution, as there are other implications of the role of language extremity. Specifically, as is also outlined in the Elaboration Likelihood Model, simply processing information on the central route to persuasion does not necessarily mean that an individual will experience the intended attitude change. What matters next is the nature of cognitive processing and whether favorable, unfavorable, or neutral thoughts predominate in response to the message, which requires taking the initial attitude into account. Some researchers discuss this as message discrepancy, which is defined as “the difference 185 between [individuals’] initial attitudes and the perceived position of the persuasive message” (Craig & Blankenship, 2011, 305). While initial attitudes were not measured in the studies conducted by Craig & Blankenship, they predict that “as initial favorability of the topic decreases, we should see diminishing effects of extremity on persuasion. The less favorable initial attitudes are, the less persuasive a message that contains linguistic extremity markers would be. [Therefore,] the use of extreme language may lead to counterarguing of the message, especially when involvement is high” (Craig & Blankenship, 2011, 305). ANALYSIS OF TWEET SENTIMENT: COVID-19 PROTECTIVE BEHAVIORS In a similar manner as the above analysis of general COVID-19 pandemic terms, the terms relating to recommended behaviors were searched within the text of each tweet. Tweets mentioning any of these terms were grouped as “behavior” tweets while all others were considered “non-behavior” tweets. Analysis of tweet sentiment in terms of subjectivity and polarity were then conducted using these categories of tweets. Tweets mentioning COVID-19 protective behaviors make up a small proportion of all the tweets posted by these accounts during the year. However, it must be noted that this analysis only includes searches within the text of tweets and therefore does not include media that may have been used to convey messages. Figure 14 shows that tweets, when separated based on those that describe COVID-19 protective behaviors and those that do not, mirror the pattern of sentiment when separated by general COVID-19 pandemic terms: the majority of tweets convey a positive sentiment, approximately half of that number convey a neutral sentiment, and approximately half of that number conveys a negative sentiment. 186 Figure 14. Bar chart comparing sentiment of tweets mentioning COVID-19 protective behaviors and all other tweets across all accounts. 187 Figure 15. Scatter plot showing tweet polarity and subjectivity scores for each tweet collected across all accounts, separated by mention of protective behaviors. As was also the case in the previous analysis of general COVID-19 pandemic terms, the majority of tweets mentioning protective behaviors and those not using those terms both convey some sentiment, either positive or negative, though a significant number also convey a neutral tone. The relative proportions of each sentiment are mirrored in tweets mentioning the COVID-19 protective behaviors and those not mentioning those behaviors, implying that there is likely not an intentional effort to deviate from the tone used to convey information about other topics. Figure 15 shows a 188 general trend of increasing polarity as subjectivity increases for both tweets mentioning behaviors and those not using these terms. While the trendline for behavior tweets appears to be linear, there is also a wide range of both subjectivity and polarity scores. Polarity most typically ranges from -0.15 to 0.5, indicating not only do a small number of these tweets convey a negative overall sentiment, they also tend to avoid strong negative language in doing so. By contrast, positive tweets frequently use moderately strong language, near 0.5 polarity score. Interestingly, of all the tweets shown above, only one tweet mentioning COVID-19 protective behaviors is at the highest degree of subjectivity. Tweets mentioning these behaviors tend to score below 0.75 on subjectivity, indicating an avoidance of communicating about these behaviors using entirely subjective opinion. Figure 16. Subjectivity and polarity scores of all tweets by account type and mention of COVID19 protective behaviors. 189 Figure 16 shows clear differences in the language used to describe COVID-19 protective behaviors across these account types. The “people” accounts and the healthcare systems both show a sizable increase in polarity with increasing subjectivity scores; for health departments, while the direction of this relationship is also found, polarity scores do not increase to the same extent as the other account types. Thus, it appears that health department tweets mentioning COVID-19 protective behaviors tended to avoid highly subjective or strong language in either direction, with a few noticeable outliers present. Figure 17. Scatter plot showing tweet polarity and subjectivity by account type. Figure 17 similarly evaluates the subjectivity and polarity of tweets, categorizes 190 them by mention of COVID-19 protective behaviors, and compares trends between account types. Clearly, there are far more posts not mentioning these terms in the main text of the tweet. However, examining this graph more closely, it is also clear that the three types of accounts tend to follow a similar pattern of increasing polarity for subjectivity scores less than 0.75--beyond this point, health departments and healthcare systems seem to maintain their marginal increase in polarity score, with health departments scoring lower on polarity, while notable people see a sizable increase in polarity score. This is different from the trend seen in tweets not mentioning these COVID-19 protective behaviors. While a general upward trend in polarity is observed as subjectivity increases, healthcare systems instead see the largest increase, with health departments remaining the least polar. It is also important to note that there does not appear to be a pattern in a type of account posting tweets with more polar or subjective language. There also does not appear to be an observable pattern in typical ranges of subjectivity and polarity for any account type whether the tweets are categorized by those mentioning general COVID-19 pandemic terms or protective behaviors. In contrast to the previous analysis of general COVID-19 pandemic terms, there does not appear to be an account type that tends to cluster sentiment or subjectivity in a certain range. Rather, all accounts seem to discuss COVID-19 protective behaviors with a moderate level of subjectivity, frequently between 0.25 and 0.75, and between a slightly negative and moderately positive overall tone, frequently between -0.2 and 0.5 in polarity. The implications of using different sentiments or tones to convey a message were discussed previously in reference to the analysis categorizing tweets by mentions of the general COVID-19 pandemic terms. This discussion will add to those ideas already 191 mentioned by focusing on behavior appeals and the effectiveness of behavior change campaigns. Previous research found that “in the United States, health communication campaigns that include use of the mass media and avoid coercion have an average effect size of about 5 percentage points...Thus, if 60% of people were doing the target behavior before the campaign, about 65% can be predicted to do the health behavior after the campaign” (Snyder, 2007, 33). However, certain differences must be noted between the communication landscape described by Snyder in 2007 and the current situation. Of course, this includes the development of social media and its integration into everyday life. Additionally, “the level of effectiveness of health campaigns that include some form of media depends in part on the specific behavior that is promoted. For example, seatbelt campaigns (r = .15), dental care (r = .13), and adult alcohol reduction (r = .11) have had the greatest success rates, whereas youth drug and marijuana campaigns have had the least success (r = 0.1-0.2),” though one reason for small effect sizes may be due in part to the fact that “the ability to get movement on attitudes, beliefs, and knowledge depends in part on whether the messages were designed to affect those specific components, and it is often unclear from published accounts whether the messages were so designed” (Snyder, 2007, 33-4). For new health topics, “it is possible to anticipate the relative success rates by examining the characteristics of the behavior” (Snyder, 2007, 33). This may, in turn, be affected by the framing of a message about behavior: “across health issues, campaigns promoting the adoption of a behavior that is new to the individual or replacement of an old behavior with a new one have a greater success rate than campaigns aiming to cease an unhealthy behavior people are already doing or prevent commencement of a risky behavior” (Snyder, 2007, 33). By this logic, a campaign introducing the use of face 192 masks or social distancing would be more successful in the beginning of the COVID-19 pandemic than would a campaign seeking to prevent people from engaging in the risky behaviors of rejecting these protective behaviors. This leads to another important distinction: as Snyder notes, “habits...can be harder to change than adopting new behaviors that only need to be performed once or twice” (Snyder, 2007, 34). Thus, a situation like the COVID-19 pandemic in which the recommended behaviors become familiar over time, must be performed repeatedly over time, and people form habits regarding the compliance or disregard of these behaviors, there are multiple characteristics of the behavior itself that create significant obstacles in behavior change efforts. ANALYSIS OF ENGAGEMENT In order to use social media to its full potential, as discussed previously, it is crucial to engage intentionally and strategically with audiences. This section will seek to evaluate the available measures of audience engagement in the tweets collected. Engagement on social media can be thought of broadly as the extent to which an audience engages with the source and the source’s content as well as the extent to which the source engages with the audience. While there are many ways engagement might be measured, for the purposes of this analysis, engagement will focus on common measures, such as tweet frequency, the number of comments and retweets, as well as the number of followers and observations of the source’s responses to audience behaviors. To introduce analysis of engagement efforts made by these accounts, this study will begin by examining the frequency of tweets posted, as other aspects of the tweets were analyzed in detail previously. This section will then move to analyzing other aspects of the social 193 media platform. One of many capabilities afforded to Fusers of all levels is the ability to post frequent, even “real-time” information in a variety of forms. Users and analysts alike consider “the frequency with which an organization sends out tweets… [to be a measure of] how active an organization is” (Lovejoy et al., 2012, 315). There are, in fact, higher expectations for organizational social media accounts, as many “followers expect organizations to be much more active than individual users” (Lovejoy et al., 2012, 315). In their study of nonprofit organizations’ Twitter accounts, Lovejoy et al. found that the majority of these organizations were able to meet this expectation, using the definition of an “active” account being one that posts “at least three times per week,” though the study also looked at data from 2009, when the social media landscape was much different (Lovejoy et al., 2012, 315). This will serve as a point of comparison for the accounts examined in this study, though the different time periods between this study and the study conducted by Lovejoy et al. should be noted. As shown in Table 1 and in contrast to the expectation of social media followers described by Lovejoy et al., the accounts run by individual people posted far more tweets than the organizational accounts, with an average of 1,469.67 during the one-year period (approx. 28 tweets per week) while healthcare systems posted an average of 781 (approx. 15 tweets per week) and finally, health departments (approx. 9 tweets per week). However, for each type of account, this is an average and each average was impacted by the presence of outliers. As such, Table 2 shows the number of tweets posted between March 1, 2020, and March 1, 2021, as well as the average number of tweets per week for each account. In contrast to the expectations of followers in Lovejoy et al.’s comments, the data 194 in Table 2 show that individual people, rather than organizations, were the most frequent tweeters--Governor Spencer Cox tweeted over 2,900 times during this interval, with an average of nearly 57 tweets per week, more than any other account included in analysis. Even former Governor Gary Herbert and former State Epidemiologist Dr. Angela Dunn tweeted more than many health departments, with 898 (17.27 tweet/week) and 586 (11.27 tweets/week) tweets respectively. Health departments and healthcare systems both showed a wide range in the number of tweets posted by the accounts in that category. For healthcare systems, the least frequent tweeter was Steward Health, with only 5 tweets (0.10 tweets/week) during this interval, while University of Utah Health posted 1,775 tweets (34.13 tweets/week) during this interval on their main Twitter account. For county health departments, the number of tweets posted ranged from Southeast Utah’s 18 (0.35 tweets/week) to Utah County Health Department’s 1,411 (27.13 tweets/week). Many health departments posted somewhere in the hundreds of tweets during the year. 195 Table 2 Number of tweets posted by each account between March 1, 2020 and March 1, 2021 Account Total number of tweets Tweets/week Bear River Health Department 92 1.77 Central Utah Health Department 112 2.15 Davis County Health Department 642 12.35 San Juan County Health Department 194 3.73 Southeast Utah Health Department 18 0.35 Southwest Utah Health Department 1270 24.42 Salt Lake County Health Department 814 15.65 Summit County Health Department 203 3.90 Tooele County Health Department 167 3.21 Tri-County Health Department 705 13.56 Utah County Health Department 1411 27.13 Wasatch County Health Department 23 0.44 Weber-Morgan Health Department 558 10.73 Utah Department of Health 336 4.46 Governor Spencer Cox 2926 56.27 Governor Gary Herbert 898 17.27 State Epidemiologist Angela Dunn 586 11.27 Steward Health 5 0.10 Mountain Star Health 427 8.21 Intermountain Healthcare 919 17.67 University of Utah Health 1775 34.13 While there is no exact number of tweets an account must post each week to be considered “active,” there is clearly a wide range of activity for the accounts above. Some accounts posted only a handful of tweets during the year while others posted over 1,000. By the standard used in Lovejoy et al.’s study, the majority of accounts surveyed above would be considered active, but this standard must be adjusted due to the significant differences in social media adoption since 2009. The ideal number and frequency of tweets depends on a number of factors, including the type of account, the number, type, 196 and information need of followers, as a guiding principle, government and public health officials should keep in mind that “while sending to many tweets may clutter followers’ Twitter feeds, an organization that sends too few messages risks having those messages lost among other messages” (Lovejoy et al., 2012, 315). In other words, it seems it is best to keep the message accessible and salient, but not excessively so. The idea of involvement, which is closely related to engagement, “has a rich history in communication research. It is widely used to explain the influence of media on audiences and is usually defined as the degree of a person’s psychological response to a mediated message or persona” (Hu et al., 2020, 2). Traditional communication literature includes multiple forms: “transportation refers to getting lost in a story during consumption of media,” “worship is extensively conceptualized as an obsessive preoccupation with a media persona,” “parasocial relationships are based on the closeness that an audience feels with a media persona, while identification refers to the process of conforming to the perceived identify of the mediated persona” (Hu et al., 2020, 2-3). Hu et al. argue that the final two components, parasocial relationships and identification, are the two that apply to a social media context, with the difference between these components being that “audience[s] parasocially interact with media personae as if media personae were their friends. Comparatively, they identify with media personae by changing attitudes, values, aspirations and other characteristics to match those of a model” (Hu et al., 2020, 3). Research has repeatedly found that the emotional and psychological processes of audiences matter greatly to their behavior. Identification and parasocial relationships provide one example of the importance of psychological processes: “in terms of the functions of parasocial relationships, previous studies have 197 demonstrated that they provide an audience with emotional and social gratification, such as belongingness and companionship. Identification is thought to meet one or more of an audience’s self-definitional needs, such as self-verification, self-distinctiveness, and selfenhancement. Therefore, identification is a more powerful form of audience involvement than parasocial relationships” (Hu et al., 2020, 3). Engagement, as measured by the traditional metrics described above, seems to be an extension of the concept of involvement in that it includes an element of action beyond following, such as liking, commenting, or sharing content. Drawing on prior research, Azer et al. argue that “being engaged ‘is to be involved, occupied, and interested in something.’ Engagement goes beyond mere participation and involvement, as it encompasses an interactive relationship with an engagement object. Engagement is considered a multidimensional concept comprising cognitive, emotional, and behavioral investment in specific interactions” (Azer et al., 2021, 101). Of course, engagement varies from low to high, though scholars disagree on exactly how to segment this continuum. Kaye notes that the majority of research distinguishes between “active” and “passive” use of social media: “here, ‘active’ use focuses on social media activities such as posting information, uploading photos, talking to others, liking, [and] commenting [while] ‘passive’ use [instead denotes] time spent browsing or observing content created by others” (Kaye, 2021, 2). Engagement may be segmented further. Some define the subcategories “‘interactive’ (in which people are engaged in bidirectional interactions with others) and ‘reactive’ (in which people may be responding to others but is not necessarily bidirectional” (Kaye, 2021, 2). On a social media platform like Facebook or Twitter, other researchers “have suggested [that]...’active’ and ‘passive’ are perhaps best broken 198 down into: active social use, active non-social use, and passive use. Here, active social use would refer to direct communication with people (e.g. commenting on posts). Active non-social may include direct communications but when no written content is used (e.g., likes, RSVP to events). Passive would be consuming content but not interaction with others” (Kaye, 2021, 2). While there is disagreement regarding how to divide the continuum of user social media engagement, there is an underlying understanding that engagement behaviors vary in important characteristics, functions, and motivations. Of particular note to this paper and to any organization tasked with managing social media engagement efforts, “what we currently know about socialness of social media use is mostly in respect to personal use”; however, as Kaye argues, “this highlights why additional focus is needed which takes a user-centered approach to ask questions about ‘why’ we use social media, so we are better positioned to theorize how social media behaviors may vary between personal [and organizational] use, and also the differential psychosocial impacts these may bring” (Kaye, 2021, 4). Previous research has “identified different motives to engage in online contexts. These include concern for others, selfenhancement, advice-seeking, realizing social/economic/hedonic benefits, social/personal integration, helping the [organization], utilitarian motive[s], the pleasure derived from sharing information, a desire to help others, social interaction, information seeking, pass[ing] the time, entertainment, relaxation, communicatory utility, [and] convenience utility...However, these are all motives to engage in or adopt social media platforms. They are unique, individually-based motives to generally engage in online communities rather than drivers that may drive social media users to engage in a specific form of engagement behavior” (emphasis added) (Azer et al., 2021, 102). Lovejoy et al. note that 199 while “organizational research specific to Twitter is scarce...evidence points to continued findings of lost opportunities,” particularly where engagement is concerned. Across many types of organizations surveyed in previous research studies, “one-way communication is still the most pronounced form of messaging strategy used by organizations on Twitter” (Lovejoy et al., 2012, 314). Thus, while further research is needed in a variety of related areas, engagement is clearly important to social media, both for individual users and for larger organizations. It is not only important to understand what engagement means and how users engage, but also the individuals and organizations with whom they choose to engage. As Kaye notes, “research has shown that [the] size and structure of users’ networks on social networking sites is important, and are influential to social media use and social interactions” (Kaye, 2021, 4). Table 3 below provides descriptive statistics for the number of followers for each account type. There is clearly a wide range in the number of followers, both within and between account types. Health departments, for example, ranged between 96 followers for San Juan County Health Department up to approx. 26,600 for the Utah State Department of Health, which, it must be noted, is a state department while all others are at the county level (the county with the greatest number of followers when the data were collected was the Salt Lake County Health Department, with approx. 15,400). Healthcare Systems showed a similarly large range, between 813 for Mountain Star Health up to approx. 31,900 for Intermountain Healthcare. Accounts for notable people in the state ranged from Governor Spencer Cox’s 7,706 followers to Governor Gary Herbert’s 65,600 followers. Figure 18 provides a visual representation of this data. 200 Table 3 Number of Followers by Account Type Account Type Health Department Healthcare System People Minimum Followers 96 813 7,706 1st Quartile 344 1,470 9,553 Median 1,648 9,644 11,400 3rd Quartile 3,175 21,175 38,500 Maximum Followers 26,600 31,900 65,600 Average Followers 4,267.79 13,000 28,235 Standard Deviation 7526 14,744 32,411 Figure 18. Histogram of number of followers by account type. 201 While more health departments were surveyed than the other account types, the histogram in Figure 18 shows that health departments have, on average, a lower number of followers than the other two account types. With the exception of the Salt Lake County Health Department, no county health department had greater than 4,000 followers. Meanwhile, notable individuals in the state could easily surpass 7,000 followers while the two major healthcare systems each exceeded 17,000. This leads to the questions: What is the importance of the number of followers for these accounts? Is it necessary to have a large number of followers for a social media account to be effective? Are there additional considerations that must be taken into account when using social media in a public health context? Finally, what motivates followers to remain followers of an account? The obvious conclusion regarding the importance of followers is simply increased visibility and reach of content--generally, the greater number of followers an account has, the greater their reach is likely to be, as followers provide a somewhat reliable audience for that account’s content. However, the importance of followers to the success of a social media account extends far beyond this. As discussed previously, social media is prime ground for peripheral cues and heuristics to influence information processing. In their experiments, Cherry & Tessitore found that “merely increasing the number of followers of a social media account that discusses healthy food can positively influence people’s intentions to eat more healthily” (Charry & Tessitore, 2021, 4). Healthy eating is a behavior that bears some important similarities to the behaviors recommended to prevent COVID-19; namely, both healthy eating and COVID-19 protective behaviors must be sustained long term, and both have important social influences. On the latter point, Charry & Tessitore observe that “while communication strategies promoting 202 unhealthy food tend to focus on the social dimension (emphasizing the social benefits of consuming junk food, e.g., pleasure of sharing), when healthy food is promoted, its functional or nutritional value is usually emphasized” leaving the social motivators unaddressed (Charry & Tessitore, 2021, 2). Example 3. Three-tweet thread posted by Intermountain Healthcare, January 4, 2021. The above thread posted by Intermountain Healthcare on January 4, 2021 (example 3) provides such an example--while there are clear appeals to the audience to engage in the recommended health protective behaviors, the only reasons cited are rooted in science and epidemiology. Certainly, it is necessary to communicate about the evolving science 203 and epidemiological measures and this is not to suggest otherwise. Rather, focusing solely on such reasons to engage in recommended behaviors and hoping the public will do so fails to address the range of social and psychological influences that affect attitudes and decision-making surrounding such decisions. Beyond supporting the argument that the number of social media followers serves as a peripheral cue to audiences, Charry & Tessitore further investigate the underlying mechanisms to understand how this occurs. They find that the number of followers an account has on Twitter frequently serves as a cue for audiences that “indicate[s] how much social influence a person has on others” and has sufficient influence to steer individuals toward a desired behavior (Charry & Tessitore, 2021, 2). More specifically, “a socially relevant source (i.e., a source with many followers) on social media appears to create the perception that many other people are influenced by this source. This presumption of social influence increases the perceived social value of [messages] presented by that source, which in turn leads to a more positive evaluation...and eventually to [behavioral] intentions” (emphasis added) (Charry & Tessitore, 2021, 6). Beyond simply serving as a peripheral cue and leaving audiences to make this perceptual leap, Charry & Tessitore find that the number of followers can influence behavior by being what researchers call a “nudge.” Put simply, “nudges are simple interventions that change individuals’ choice architecture and lead them to act in a predetermined direction without limiting their freedom of choice” (emphasis added) (Charry & Tessitore, 2021, 1). The final part of this excerpt is vitally important in reducing the reactance potential among members of the audience. Charry & Tessitore’s findings have a number of important implications. First, and specifically referencing the context of healthy eating, 204 “the scarce nudging research focusing on the dimension of healthy food relies on real-life settings, that is, contexts in which people physically interact with the nudge such as plates in cafeterias or products packaging, limiting per se the dissemination to the physical properties of the nudge” (Charry & Tessitore, 2021, 6). In the context of COVID-19, this might equate to nudges encouraging mask wearing at the entrance to a building or nudges detailing proper hand-washing in public restrooms, or any other nudge in the physical environment. Charry & Tessitore “use the social media context to broaden the reach offered by traditional settings in which nudges are usually implemented. [They] show that the effectiveness of nudges may be expanded to virtual settings, thereby enhancing its potential dissemination” (Charry & Tessitore, 2021, 6). Next, it is important for each account to understand the types of people and accounts that are in this group of followers. Fung et al. and Haman both demonstrate the importance of understanding the types of followers an account has in the context of public health and government leaders during the COVID-19 pandemic. First, in their 2018 study of one Twitter account run by the CDC (for Advanced Molecular Detection, @CDC_AMD), Fung et al. find that of the 809 followers, 26.0% were individual healthcare professionals, 11.6% were nongovernmental organizations, 3.3% were government agencies, 3.3% were media organizations and journalists, and 0.9% were academic journals, while 54.9% did not fit into categories specified by the researchers (Fung et al., 2018). Based on the focus of the CDC’s Twitter account under investigation in Fung et al.’s study, it is highly likely that the followers would have varying levels of background knowledge and scientific literacy skills which would then lead to different education needs, even without additional information about the 54.9% of followers that 205 did not fit into the researchers’ categories. For example, it is likely that a physician or other healthcare provider following this account would have a greater amount of background medical knowledge than a government agency, a news organization, or journalist. Meanwhile, a government agency following an account would likely have different needs and motivations than an individual following the same account. The same principle applies in a broader context for the Twitter accounts surveyed for government and public health officials in Utah. While data was not collected in this study regarding the types of followers each account has, it is important for each account to assess their followers and seek to better understand their experiences, information needs, and social and psychological motivations in order to better tailor a social media strategy that fits these audience characteristics. By investigating the Twitter accounts of Canadian politicians during the COVID-19 pandemic, Haman dives deeper into this issue of follower motivations and their impacts on social media behaviors. He first notes that “there are a number of possible motives behind Twitter use…[including] convenience, entertainment, self-expression, guidance, information-seeking and social utility” (Haman, 2021, 136). Based on his analysis, there was an overall increase in followers for the accounts surveyed following the first case of community spread of COVID-19 in Canada, with “the sharpest increase appear[ing] approximately two weeks after the first case of community transmission. In one week [for example], from March 16 to 23, 2020, [Prime Minister of Canada] @JustinTrudeau gained more than 70,000 followers, @GovCanHealth gained almost 25,000 followers, [and] @CPHO_Canada gained more than 40,000 followers” (Haman, 2021, 140). Haman further surveyed over 350,000 accounts and identified the new followers that created their accounts during the 206 pandemic, finding that “accounts created during the COVID-19 pandemic are, for all analyzed accounts, significantly overrepresented. Twitter users who created accounts between March 16 and March 22, 2020, make up more than 10 percent of all analyzed accounts” following the accounts of focus (Haman, 2021, 142-3). With regard to motivations and Twitter use, Haman argues that “people who joined Twitter during the COVID-19 pandemic could be expected to have different behavior on Twitter, since they were primarily information-seeking users and less interested in social interaction; this assumption is based on the fact that they were not sufficiently interested in Twitter before the COVID-19 pandemic to create an account” (Haman, 2021, 143). As predicted, “Twitter users who created accounts a few days before and after COVID-19 was declared a pandemic by the World Health Organization...have significantly fewer tweets per month. For @GovCanHealth followers who created accounts during the week of March 9, 2020, and during several of the weeks following, the median number of tweets is around 0.25. This suggests that users who created a Twitter account during the pandemic period used Twitter mainly as a source of information rather than as a platform for sending tweets” (Haman, 2021, 144). These information-seeking users show similar trends with other measures of engagement, such as likes: “The median of liked/favorite tweets ranges from 2 to 8, depending on the week, for accounts created before the COVID-19 pandemic, while the median of liked/favorite tweets drops to just 1 for accounts created during the COVID-19 pandemic” (Haman, 2021, 144-5). It must be noted that Haman’s findings “do not imply that the interests of people who joined Twitter during the COVID-19 pandemic are significantly different than other Twitter users,” just that their behaviors while using Twitter do significantly differ (Haman, 2021, 146). Thus, 207 while Haman was only able to analyze publicly-viewable accounts, his analysis nevertheless underscores the importance of understanding followers’ needs and motivations and adjusting content and communication strategies accordingly. However, it must also be noted that simply attempting to increase the number of followers is not enough to increase engagement, particularly if, as has been found to be common amongst organizational Twitter accounts, organizations attempt “to secure followers [only] to receive the one-way messages” it posts (Lovejoy et al., 2012, 314). With this background on the importance of social media followers in mind, what can government and public health officials do to entice more followers and keep them long-term? As discussed previously, social media users may be motivated by multiple different factors. One notable difference is the extent to which a user engages in social media content to fulfill social and psychological goals or to acquire information. A single individual or organization included in this analysis is likely to have a range of each type of user in their audience or among their followers. The overall guiding principle should be to fulfill the needs of the users, which is, of course, much easier said than done. For both types of users, source characteristics like credibility and trust play an important role, as users are unlikely to heed or interact with content posted by sources lacking these characteristics in a public health setting. For information-seeking users, government and public health officials should seek to post frequent, useful, and timely information that satisfies the needs of users. The frequency and type of information must depend, not only on the account and response received from audience members, but also on the event itself. During the COVID-19 pandemic, information was much more crucial in the beginning when little was known 208 about the virus and its transmission and no vaccine was available to reduce the risk of infection. As the pandemic has continued, updates regarding significant policy changes, the emergence of dangerous variants, or news regarding vaccine development and distribution have likely been of greater importance to members of the public than the daily or weekly case count data as the former is the type of information that would more directly affect individuals’ lives. However, these are assumptions that must be tested by government and public health officials on separate accounts, as each account has different audiences and followers. Furthermore, social media communication for information-seeking audiences must be thought of as an iterative process of publishing information, gauging the reaction from audience members, and adjusting to better fit their needs. This is not to say that some information that may be deemed less important or relevant for the audience should not be available in public health settings or in a pandemic more specifically. This is merely to say that government and public health officials must make careful decisions regarding what information to publish in what medium and what information is best suited for a social media audience, a process in which audience members’ information needs must be taken into account. For users who instead use social media to fulfill social and psychological goals, there are additional considerations for government and public health officials to take into account. Kim et al. examine follower behavior in a commercial business context. In line with research mentioned previously, they highlight the importance of a users’ perception of identification with the organization through a concept called “brand identification,” denied as “the extent to which the consumer sees his or her own self-image as overlapping with the [organization’s] image. This can [then] be used to construct the self, 209 as a reference point for distinguishing oneself from non-brand users, as well as to present the concept of self to others” (Kim et al., 2014, 20). This is a particularly important application of the social psychological principles discussed previously. If social and psychological motivations are at play with the ability to influence social media behavior, both conceptions of the self (i.e. as a member of a certain group) may enable the formation of communities--the ingroup, or a supportive group of followers, while others are considered the out-group. In social media environments, individuals may then “display themselves publicly to others” by engaging with content through likes, comments, or retweets (Kim et al., 2014, 20). UNDERSTANDING COMMENTS Comments have become increasingly popular across online platforms, and social media are, in fact, fundamentally built around interaction. Wallsten & Tarsi, referencing a study conducted by the Pew Research Center in 2010 note that “there is certainly evidence that internet users take advantage of [these] interactive tools...and view them as an essential part of their online experience” (Wallsten & Tarsi, 2015, 1023). In that study, a quarter of internet users reported commenting on news stories or blog posts while 37 percent reported that “the opportunity to post comments is an important feature in selecting which news websites to visit,” with social media clearly being in that category (Wallsten & Tarsi, 2015, 1024). Comments on social media provide much more than a one-way feedback loop, in which a source can broadcast information and view users’ responses. Indeed, perhaps one of the greatest opportunities for government and public health officials to use social media effectively for these users is through the ability to interact directly with users, something that may be unfamiliar in public health due to the 210 nature of the field. In line with other research, the results of Kim et al.’s study “highlight the role of Twitter as a relationship management tool to help [organizations] communicate with their [audiences] and further cultivate relationships,” noting that “from a public relations perspective, trust and commitment are the key components of developing...successful interpersonal relationships” with audience members (emphasis added) (Kim et al., 2014, 22). Cultivating relationships requires much more time, effort, and resources than simply broadcasting messages on social media. In a business context, companies that use such a strategy frequently monitor their social media pages and the ways in which their audience members engage with the brand and the content, respond to individual customers, and manage the overall brand image on social media. Because managing social media accounts requires a significant effort of both time and resources, it is worth asking why government and public health officials should do so, particularly while managing a public health crisis. There are, in fact, many sound reasons. First and foremost, relationship management has been found to be a driving factor behind users’ inventions and desires to remain members of an online community: “a consumer’s intention to remain engaged with a brand community depends, to some extent, on [their] satisfaction with [their] relationship to the brand” (Kim et al., 2014, 21). As Kim et al. argue, “this is drawn from a general altruistic motive to patronize a company or brand, following consumers’ positive experiences with the employees and their responses to problems”; however, there are certainly ways in which government and public health officials can invest in audience relationships and create similarly positive experiences (Kim et al., 2014, 21). As a simple example, public health officials managing the social media accounts of health departments may take time to frequently monitor and respond to 211 individual users’ comments made in response to posts, as manual analysis indicates this is currently extremely rare among the health department Twitter accounts analyzed. Additionally, the number of followers may serve as a heuristic cue that presents the account “as a reliable source of information” (Lahuerta-Otero et al., 2018, 567). Indeed, though “consultants have stressed that organizations develop a social media strategy to grow virtual communities with stakeholders,” Lovejoy et al. and other studies have found that “organizations are continuing to use social media [primarily] as they would a traditional information subsidy” (Lovejoy et al., 2012, 315). The current analysis indicates this may still be the case. While comments may be less frequent, on average, on public health posts when compared to other topics (e.g., popular products, sports, or politics), and while not every tweet receives comments, engaging with users who do post comments and responding to their questions or misconceptions in an effective way is important to building and maintaining relationships. Example 4 illustrates one comments thread responding to a post by the Utah Department of Health. In the above example, the only time the Utah Department of Health replied on this comments thread is to its own initial post. Meanwhile, multiple individuals posted negative thoughts while the Utah Department of Health made no response. While comments were unable to be collected for analysis in this study, implications will be inferred from relevant research, particularly with regard to the effect comments may have on both the source of a message and the message content. First, Wallsten & Tarsi investigate the effects of anonymous comments on a message source in the context of news organizations’ websites. White Wallsten & Tarsi’s study focuses on a different context than the present study, there is an argument to be 212 made regarding the level of anonymity users may be able to maintain on a social media platform like Twitter. While many of the most popular social media platforms require users to enter a certain amount of personal information when creating their accounts, as is the case for Facebook, Instagram, and Twitter, such information may not be visible to other users. Additionally, while it is common for such social media platforms to require users enter their name and some form of contact information, such as an email address or phone number, such platforms have little means to verify the veracity of the informations users provide, or to link the information to a person, rather than simply tying the information to an account. As a result, users may create multiple accounts, and may be easily able to conceal their names or pose under a false one, even if their username and profile picture is posted alongside online comments. It is therefore useful to consider what research has found regarding the impact of anonymous comments in an online, information-disseminating environment. Many researchers and public relations practitioners have claimed that anonymous posts pose a danger to the online environment, specifically in that they “shape the attitudes of internet users,” which may lead them to become “more negative towards the media, in general, and the reporting outlet they were reading, in particular” (Wallsten & Tarsi, 2015, 1020). As Wallsten & Tarsi point out, “because comments sections appear directly underneath or alongside news reports, they provide easy access to the views of fellow audience members,” which is also the case for many social media platforms (Wallsten & Tarsi, 2015, 1023). If comments do not align with the perspective of the story or the message content, “the juxtaposition of a news story and readers’ unfiltered reactions to that story represents an unprecedented intersection between two traditionally separate processes, mass and 213 interpersonal communication,” two processes that can be a challenge to navigate simultaneously (Wallsten & Tarsi, 2015, 1023). In their study, Wallsten & Tarsi demonstrate how difficult a task this is with regard to maintaining source reputation. In their study, results showed that “exposure to any media criticism in anonymous comments made respondents feel more coldly toward the media” in general (Wallsten & Tarsi, 2015, 1028). What was surprising, however, was that they “discovered a similar pattern of results among those exposed to an article with praise for the media in the anonymous comments section...complimentary comments did not persuade internet users to view the news media more positively” (Wallsten & Tarsi, 2015, 1028). In other words, “exposure to anonymous comments--regardless of their content--reduce warmth [toward] the news media” (Wallsten & Tarsi, 2015, 1029). Further, Wallsten & Tarsi’s regression analysis demonstrated that “the influence of anonymous comments is not conditional upon an individual’s prior beliefs about the media...Taken together, these findings suggest that all respondents, regardless of their preexisting feelings about the media, react the same way to anonymous comments” (Wallsten & Tarsi, 2015, 1029). Wallsten & Tarsi’s findings paint a grim picture for the possibility that government and public health officials might be able to leverage the comment capability of social media platforms to fulfill their goals. However, additional research suggests that managing comments sections on social media may be important for another reason, namely the impact user comments may have on perceptions of the message content. Certainly, responding to comments made in clear opposition to the message the source attempts to communicate is a difficult task that requires a delicate balance between conveying the necessary information and avoiding 214 sparking reactance. However, if no attempt is made to do so, there can be severe consequences for persuasive attempts. In short, “exposure to crisis-related information [through] social media...affects public perception of [the] crisis…[and] it is difficult for crisis managers to alter user-generated crisis-related messages,” which Hong & Cameron call “unrefined information” that can create an “organizational vulnerability” (Hong & Cameron, 2018, 173). In previous studies, scholars have tested the influence of users’ comments on readers’ perceptions of public opinion and personal opinions. For instance, “an experimental study compared participants who only read online news with other participants who were exposed to online comments after reading the same news. Their findings showed those who were exposed to online comments inferred public opinions based on what they read from comments. As a result, the effects of online comments mitigated the impact of news articles on readers’ perceptions” (emphasis added) (Hong & Cameron, 2018, 174). Another previous study found, using an experimental design, that “exposure to critical comments not only shaped how people felt about the issue being reported on, but also led them to perceive public opinion as more discrepant from the position articulated in the news report” while a third study found “strong evidence for a ‘nasty effect’ that resulted from exposure to rude comments; namely that incivility polarized readers’ views on [the topic] and changed how readers interpreted the basic facts of the news stories they read” (emphasis added) (Wallsten & Tarsi, 2015, 1024). In other words, members of the audience, whether or not they engage directly with the comments, may arrive at an estimate of public opinion by reading comments. 215 Example 4. Comments posted in response to a tweet made by the Utah Department of Health. December 31, 2020. (Identifiers for private individuals are omitted). 216 In the example tweet above, this means that users who view comments posted in response to the tweet may infer from the comments that public opinion is against the measures and caution advised by the Utah Department of Health, especially in the absence of a response from the UDOH. Hong & Cameron continue: “user-generated comments affect public opinion, given that public perceptions toward the media can be amplified or attenuated depending on what comments they read” (Hong & Cameron, 2018, 174). Scholars have “termed the...phenomenon as the bandwagon effect/heuristic…[in which] individuals are likely to assume that something is correct if many others think that is the case” (Hong & Cameron, 2018, 176). It is important to understand the effects of the tone of comments relative to the tone of the original message--for example, if the comments are in conflict as in the example tweet above. A previous study found that “online news readers who [were] exposed to positive news stories and positive news comments perceived the issue more positively compared with those who [were] exposed to comments that conflict[ed] with the tone of the news story. When participants read comments that contradicted the news story’s tone, they exhibited changes in judgment [regarding] the news stories” (Hong & Cameron, 2018, 175). This effect may influence both central and peripheral routes of information processing, as the tone of comments may provide an estimate public opinion that can be subsequently used as a heuristic, while other studies have found that “negative social network comments harmed the [central] argument” of the message and that “comments with a tone that opposed the [message] content weakened...readers’ attitudes toward the [message] as comments refuting the contents of news story had strong persuasive effects” (Hong & Cameron, 2018, 179). Because comments are less frequent 217 in public health than on other topics, it would be less difficult for a user to read all of the comments, making it potentially more likely that users will do so, thereby making each comment have a greater relative impact on the perception of public opinion. However, as Lee notes, a large number of comments is not necessary to have a significant effect on users’ attitudes, since “individuals tend to regard [even] a few others’ visible reactions as a snapshot of the typical or predominant opinion” (Lee, 2014, 572). Many researchers believe an important reason comments can have such an impact on audience perceptions, and potentially attitudes, is because “the public perceives user-generated comments as more trustworthy than mainstream media content, and online comments are perceived to be unbiased” however untrue that perception may actually be (Hong & Cameron, 2018, 174). While this comment is made in reference to news media, some of the same perceptions may also apply to health departments or government officials and the messages they convey. The behavior of commenting has also been found to impact audience perceptions. In one analysis of 129 health news articles, researchers found that the frames journalists used to convey the message did not “necessarily result in similar frames in corresponding comments. Simply put, frames used in news stories were not always positively associated with online comments…[Rather,] online newsreaders who commented on the websites did not perceive stories according to the journalists’ intentions. They held different views and opinions (represented by their comments)” (Hong & Cameron, 2018, 174). USE OF HASHTAGS As noted previously, using hashtags can provide a number of benefits to accounts and the content they post which range from organizing content and making tweets more 218 easily accessible to those outside an account’s followers, to aligning content with a deeper sense of belonging among social groups that research has found many social media users crave. However, it must also be noted that “hashtags reduce the number of characters [available] in a tweet, which means that less information is provided,” particularly if the hashtag is not used as part of a sentence (Lahuerta-Otero et al., 2018, 573). Because of this tradeoff, it is useful to more closely examine hashtags and the circumstances in which they are most useful. Erz et al., quoting a previous study, argues that Instagram, and it can be inferred, other similar social media platforms, are “a medieval battlefield...the social media pro, [is] the gallant knight and the hashtag is [a] trusted sword” (Aynsley 2016 qtd. in Erz et al., 2018, 48). The specific dynamics, including the massive amounts of content and the potentially contentious interactions between users, as have already been described, contribute to this sense. However, the main conclusion from the previous quote is that hashtags “are no longer simply ‘trending’ but [are] a ubiquitous and seemingly permanent and power feature” of social media--content on social media is nearly always accompanied by hashtags, and approximately 125 million hashtags are shared each day on Twitter alone” (Erz et al., 2018, 48). As is readily apparent by the data in Figure 4, included again below for reference, this is certainly not the case with the accounts surveyed in this study. Hashtag use among all three categories is strikingly low, with health departments using hashtags least frequently (only one hashtag searched, #MaskUp, was used by health departments, an average of 0.0714 times across all health departments surveyed). While healthcare systems were the most frequent users of these common hashtags, even their use fell far below an average of one per tweet. Erz et al. note that 219 “extant research has found hashtags to be a crucial currency for all users, primarily fulfilling a broadcasting function to increase the visibility of content” (Erz et al., 2018, 48). Indeed, their study comparing influencers to “potential influencers” and followers found that “influencers use more hashtags in a post, again corroborating the role of hashtags as a broadcasting medium” (Erz et al., 2018, 58). Thus, given the previous analysis of the accounts surveyed in this study, which found that they most typically use Twitter as a broadcasting medium, these same accounts are neglecting perhaps the most powerful tool available to them on social media platforms to accomplish that goal. Erz et al. continue with useful descriptions of hashtagging behavior that have important implications for the current study. First, they identify six primary motives driving hashtag use: “self-presentation, inventiveness, chronicling, information-seeking, venting, and etiquette” (Erz et al., 2018, 57). It must be noted that Erz et al.’s study focused on individual influencers or followers, and thus, not all of these motivations are applicable in an organizational, public-health setting. It is likely that the two most applicable motives identified by Erz et al. would be relevant to information-seeking, defined in the study as “finding information and inspiration through clicking hashtags,” and perhaps chronicling, “including documenting and contextualizing experiences through adding hashtags” (Erz et al., 2018, 52). Still, because the accounts surveyed in the present study are all recognized leaders in Utah, these accounts would likely not be the information seekers, but rather the information propagators, conveying information to a social media audience that may include individuals with information-seeking motives, as discussed previously. Perhaps, then, there are other relevant motivations for using hashtags in a public health setting that have yet to be uncovered through additional 220 research. Each of these six motivations were found to relate to different aspects of hashtagging behavior. Specifically, “self-presentation, inventiveness, and chronicling particularly express the media-producing role” while only “one factor, informationseeking, pertains to the consumption of media content” (emphasis added) (Erz et al., 2018, 57). In other words, the motivation most likely to be relevant for these accounts in a public health context is more related to consuming information than producing and spreading it. Thus, it becomes highly problematic for users with information-seeking needs if the content posted by important, reputable accounts, like those surveyed in this study, do not use hashtags that make the content more readily accessible. Overall, the accounts included in the present study appear to exhibit behavior more typical of followers than of social media leaders, based on the two crucial characteristics of influencer accounts Erz et al. identify. For the time being, many of Erz. et al.’s conclusions do apply to the present context, particularly with regards to understanding the hashtagging behavior of the most influential social media accounts. Influencers and potential influencers were identified by exhibiting two types of characteristics and behavior that differed from that of followers. First, these accounts had a “more favorable followers/followings ratio, that is, they follow fewer accounts than they are followed by,” indicating not only that followers are an important and useful measure of an account’s influence, but also that the ratio is important for accounts seeking to increase their influence (Erz et al., 2018, 56). The second indicator “was the number of hashtags used per post...influencers [indeed] used more hashtags per post than followers did” while “potential influencers were more engaged in clicking than followers” (Erz et al., 2018, 56). Both of these behaviors are 221 important for accounts that seek to be leaders on social media. Importantly, Erz et al. also found that “the number of hashtags [in a post] and their degree of genericness may be a message [to users] in itself” (Erz et al., 2018, 53). While the number of hashtags may be limited by various social media platforms, a 2016 study found that “nine hashtags appear to be optimal in increasing reach,” a number that lands far above the number of hashtags used by accounts in this study (Erz et al., 2018, 56). If used properly in the ways described, hashtagging can be an effective and efficient way to target audiences and add to a broader conversation. Two general strategies may be used with regard to hashtags: “on the one hand, many hashtags can help in contextualizing an event as detailed as possible, for example, by indicating many related topics” (Erz et al., 2018, 53). In this strategy, using “multiple hashtags allow users to associate…[a single post] with different meanings and topic streams, thereby directing their communication toward multiple readers” (Erz et al., 2018, 56). The other strategy, “as some interviewees [in Erz et al.’s study] elaborated, in some instances users might refrain from using many hashtags and instead use only a few, quite distinct hashtags to document events and direct their communication towards a limited group of receivers” (Erz et al., 2018, 53). In any case, it is important to use hashtags strategically to increase reach, target categories of users, and guide message processing without appearing to do so, as “appearing to be reach-driven is socially unattractive” (Erz et al., 2018, 57). Additionally, regardless of the strategy selected, it is important to not only use hashtags, but to follow their activity. As Erz et al. note, “clicking hashtags helps [accounts] to reach or maintain their influencer status, as it enables them to catch up on 222 trends and use the ‘right’ hashtags to become or stay part of the conversation” (Erz et al., 2018, 58). However, due to the nature of hashtags on Twitter and many other mainstream social media platforms, any user can use any hashtag for their posts, regardless of knowledge, expertise, attitude, intent, or content topic. In their study of three popular health-related hashtags, (#lets talk for World Health Day 2017, #HIV prevention for World AIDS day 2016, and #No tobacco for World No Tobacco Day 2017), less than two-thirds (62.4%) of the total social media posts collected across Instagram, Facebook, and Twitter were related to the study, noting of the 37.56% of unrelated posts that “the unrelatedness of [the] post[s] is due to the fact that, when a particular hashtag is trending in social media, people use it to gain attention and popularity” (George et al., 2018). Beyond simply posting content that is off-topic, the problem may go much deeper, and it is vital that government and public health officials using hashtags are aware of the potential influence of posts made by the public. Indeed, there may be active attempts by members of the public to take over, or “hijack” the hashtag, a phenomenon in which a group of social media users flood the platform with content using the hashtag that deviates from the initial intent of the hashtag creator. While off-topic posts are an important and related problem, the greater danger comes from this latter phenomenon. The underlying problem here may be understood as a battle to shape the narrative behind a hashtag, which can have a powerful and widespread effect. Eyre & Littleton explain this well: “we live in an age of dialogue, not monologue, a time when many-tomany modes of communication constitute...the zeitgeist” (Eyre & Littleton, 2012, 183). By contrast, many in government, business, or other public-facing professions are still “operating under the auspices of the traditional influence model of communication, 223 whereby one-to-many message projection information transfer controlled by the sender, whose messages are assumed to act as persuasive sources of motivation on passive audiences” (Deutch qtd. in Eyre & Littleton, 2012, 183). This, however, is not the case in a participatory social media environment where users are granted such freedom in content generation and potential in influence. Eyre & Littleton continue, “this is not merely a matter of ‘public opinion’ (as a set of ‘values’ or ‘attitudes’) but of understanding narrative structures, tropes and metaphors, allusions, rhetoric and references--the rich texture of meaningful communications” (Eyre & Littleton, 2012, 184). Narratives may have an especially strong influence in times of uncertainty. As Eyre & Littleton explain, “during ‘settled times’ social reality is reproduced through institutionalized socal orders. Meaning becomes habit, ‘this is the way we do things; this is the way we are’” (Eyre & Littleton, 2012, 182). Social norms and other values shared among social groups are especially powerful during such times. By contrast, when conditions are less certain, “everything (it seems) is in flux; social institutions (family, community, tribal, religious, economic) are disrupted...and vast tracts of social reality are now subject to contention, doubt, and reinterpretation. Taken for granted reality becomes fragmented as we negotiate new identities, construct new narratives, recalibrate our perceptions of interest, and move in different networks in response to the changes that occur around us. As a consequence, conflicting versions of reality emerge, responding to, and creating, social tensions that often spill over into conflict” (Eyre & Littleton, 2012, 182). Many of these conditions apply to the beginning of the COVID-19 pandemic--social structures such as families and workplaces were separated during lockdowns. Meanwhile, new information has emerged nearly continuously since the beginning of the pandemic (of course, some of 224 which is false), drawing the attention of different individuals and shaping different narratives over time until a “new normal” of living with the virus emerged. Some have come to view the virus as a significant threat, particularly as concerning variants have spread around the globe, while others have instead ignored or actively fought against the calls of government and public health officials to either abide by mandates or exercise caution of their own free will. In other words, the COVID-19 pandemic created conditions of uncertainty, but as the virus has become a part of life, the conflicting versions of reality can also be observed. Hashtags may then be “hijacked” by these different narratives. Jackson & Welles analyzed one powerful example of hashtag hijacking from 2014 regarding a hashtag started by the New York Police Department. As they describe, “what started as a public relations campaign quickly turned into an online protest as thousands of citizens appropriated the #myNYPD hashtag to highlight instances of police brutality, abuse, and racial profiling” (Jackson & Welles, 2015, 932). The speed with which this occurred is staggering: “in total over 100,000 #myNYPD messages were tweeted between 22 and 24 April 2014. The vast majority of these messages propagated counternarratives that directly challenged the New York City Police Department’s goals...The hijacking of [this hashtag] helped to popularize [these] counternarratives about police-citizen interactions, setting the tone of discussions on social media and in mainstream media” (Jackson & Welles, 2015, 932). Along with many other obstacles in the process of disseminating factual information, social media is also a prime breeding ground for the popularization of counternarratives and counterpublics that “explicitly and strategically seek to challenge the ‘dominant knowledge’ inherent to the mainstream 225 public sphere” (Jackson & Welles, 2015, 933). Indeed, “the internet, and social media in particular, enables a ‘mass amateurization’ of the media, including a shift from professionally produced news toward citizen journalism, and an overall reduction in the coordination costs involved in collective action” (Jackson & Welles, 2015, 933). While some of the causes and effects of this democratization of media have been discussed previously, the danger with regard to hashtag hijacking is that “social media...does not require unified messages or legitimizing groups to influence the terms of mainstream political dialogue, with online groups frequently rejecting traditional structures of leadership altogether. This opens up the possibility of counterpublics to leverage the architecture of the social web to advance their causes” (emphasis added) (Jackson & Welles, 2015, 933). In other words, “because of social media’s connectedness and...the ‘technological architecture’ of platforms like Twitter,” as well as the information sharing environment and the psychology of users, “members of [counterpublic] groups whose worldviews are not traditionally reflected in the mainstream may have more power to rewrite dominant narratives than ever before” (Jackson & Welles, 2015, 933). The social media environment allows counternarratives to gain both credibility, legitimacy, and popularity in powerful ways never before seen in the media landscape. As Jackson & Welles point out, one major reason counternarratives can become so powerful so quickly is because “outrage is one of the only forms of citizen affect that mainstream media rewards with attention” (Jackson & Welles, 2015, 943). Indeed, “the success of [these] organizations and social movements is inherently linked to masses of people, from multiple networks, sharing a sense of outrage” thereby causing “individual fear and outrage [to become] mass outrage as citizens share their experiences” and 226 counternarratives (Jackson & Welles, 2015, 943). Importantly, a large preexisting network of users is not needed--a single individual can have a significant effect in reshaping the narrative behind a hashtag. In the example of the #myNYPD, Jackson & Welles found that “individual citizens used their voices to hijack #myNYPD in ways that may have been restricted in more organizational settings” and that “diverse individuals with relatively few followers were elevated to a similar level of visibility in the hijacking of #myNYPD as organizations with far more followers” (Jackson & Welles, 2015, 940). Hashtag hijacking constitutes a unique problem in that resistance to the trending counternarratives is largely unsuccessful, either from individuals or mainstream media. In Jackson & Welles’ analysis, they found that “the only traditionally elite account among the top 12 most retweeted using #myNYPD was the New York Police Department--unsurprising given their initiation of the hashtag and attempts to maintain control of it as others subverted their message. Notably, @NYPDNews reached elite status as a direct result of counterpublic efforts; the vast majority of retweets and replies to @NYPDNews were users responding with ire and criticism” (Jackson & Welles, 2015, 940). Based on their network analysis, Jackson & Welles found that “any online backlash to the #myNYPD hijacking either failed to be retweeted with any significance, failed to gain notice among the core discussion network, or simply failed to exist altogether” (Jackson & Welles, 2015, 939). Hashtag hijacking is therefore a pressing problem for hashtags referring to any potentially contentious topic and one that all government and public health officials must keep at the forefront of planning social media campaigns. Nevertheless, further research will be needed to identify strategies that are successful in preventing this phenomenon from occurring or 227 preventing its far reaching effects. At present, it is vital to continually monitor the narrative being conveyed by other accounts using a specific hashtag. RETWEETS AND SHARING BEHAVIORS Retweets are a popular and encouraged behavior on Twitter, akin to sharing content on other social media platforms and carry benefits for both the message source and the user sharing the message. For the message source, “it is reported that the dissemination of a viral message through [social networking sites] can bring a range of benefits for marketers, including brand awareness, improved brand advocacy, and increased sales at a lower cost” (Shi et al., 2017, 844). While public health has different goals that are not profit-driven, the same general benefits might translate to a public health setting. For example, in place of brand awareness, public health might benefit instead from a greater presence in online communities and increased awareness among community members of their services; instead of brand advocacy, there might be more dialogues on important public and personal health issues; instead of increased sales, public health promotion campaigns conducted through social media might be less costly than traditional forms of media. Malhotra et al. also argue that “even a few additional retweets can increase a tweet’s reach by the thousands,” that “retweets add credibility to the tweet [and] thus [increase] its effectiveness by making it more likely to be read by those who are familiar with the retweeter” and potentially, that “those who are not following the [account] as yet may begin following...thus further increasing its network, and, hence, the reach of its tweets” (in Jalali & Papatla, 2019, 663). For the user, retweeting and sharing content provides a quick and convenient way of sharing information the user deems important, interesting, or otherwise worthy of sharing with 228 others. Retweeting “costs little [effort] to construct a message, which was believed to be valuable enough to share with others, compared to having to formulate a new message by oneself. The cost-efficiency might be an explanation for [the] retweet function’s popularity” (Lee et al., 2015, 190). As much as “60% of people have reported frequently sharing online content with others” (Shi et al., 2017, 844). There have been many studies investigating whether or how the number of retweets can serve as a cue. Haim et al. conducted a review of 47 studies of popularity cues, which have been defined by many scholars as having three characteristics: first, popularity cues (PCs) “represent meta-information about the popularity of an entity (e.g., a product, social media post, or news article). By itself, meta-information is neither inherent to nor entirely dependent on an entity’s manifest characteristics. From a general perspective, the informational value of PCs merely can be seen as a cue for further interpretation” (Haim et al., 2018, 190). Second, popularity cues “reveal either intended user-generated information (e.g., ratings) or unintended (observed) user-generated information (e.g., number of clicks). Yet, in reality, PCs do not necessarily rely on or reveal this [distinction]. Thus, their value could be user-generated, observed, or a (nontransparent) combination of the two” (Haim et al., 2018, 190). Finally, “PCs depict metrics rather than qualitative data (e.g., comments). That said, PCs are not necessarily represented as plain numbers. Instead, they might also be illustrated, for instance as an icon or as a graphic image” (Haim et al., 2018, 190). Haim et al. specifically define four types of relevance cues: internal relevance cues designated by the originator of a message include all kinds of signals that are intentionally included in an entity (e.g., in a news article or a product description) to indicate importance, such as highlighted news values...or linguistic features. Second, external relevance cues designated by the 229 originator of a message include signals that are intentionally attached to an entity and indicate importance relative to other entities. Such cues include labels...layouts, or an item’s ranking on a website. Third, external relevance cues designated by intermediaries depict intentionally attached signals to an entity by a third party that is neither the originator nor the user of a message. Examples include algorithmically derived rankings or personalization features which present information because they supposedly fit users’ preferences. Fourth, external relevance cues designated by users are signals attached to an entity that are intentionally or unintentionally produced and curated by recipients or consumers. In contrast to relevance cues designated by either the originator or a message or intermediaries, relevance cues designated by users indicate a level of popularity among those users...While these might include live reactions on TV or radio, such as applause, individual close-up reactions, our more general understanding also includes online reactions, such as comments or metric information about users’ behavior or their evaluations of entities (Haim et al., 2018, 191). Popularity cues, such as the number of likes or retweets, are clearly included in this fourth type of relevance cues, as they are separate from the message itself but are metrics generated based on user activity. While many scholars agree that retweets (and likes) can serve as external cues, there has been disagreement as to what effects they might have. In their review, Haim et al. find that 13 of the studies resulted in no effect, 16 showed nuanced effects, and 18 showed positive effects of popularity cues (Haim et al., 2018, 202). Much of the discrepancy between studies, they argue, can be attributed to the nature of the studies as well as the different operationalizations of popularity cues. As they explain, “many studies refer to PCs as a central feature of social network sites, news aggregation, and e-commerce. Yet, this strong dependence on context results in the fragmentation of conceptual assumptions, thus hampering a comprehensive perspective. In those studies, the equivocal variety of conceptualizations and operationalizations allows for the cherry-picking of suitable findings. It does not facilitate a systematic overview of the possible effects of PCs” (Haim et al., 2018, 189). Haim et al. assert that “context and implicit or explicit reference points are necessary for PCs to be effective” 230 and that “the effectiveness of PCs depends on the general traits or situational interests and characteristics of the user” (Haim et al., 2018, 203-4). However, more research is ultimately needed to understand popularity cues in general and the effects they might have on users. In Haim et al.’s words, “to move forward in PC research, it is necessary to develop a comprehensive theoretical framework that is open to emerging and evolving online environments” and continues to evolve in the rapidly-changing technological world. However, there is an obvious benefit to increasing the number of retweets--that is, the content is likely to spread more widely, meaning that it is still useful to understand why users retweet and what types of content are shared more frequently. This will be the focus of the remaining discussion on sharing behaviors. Users may retweet content for a variety of reasons, many of which have been identified and analyzed by researchers. As Lee et al. describe, “Twitter users value retweeting as a tool to pass along information and to reach out to [the] larger public. A user [may] retweet a tweet if it is perceived as valuable information worth sharing with others. Studies [have] indeed found that Twitter users frequently applied [the] retweeting to spread and share information with others” (Lee et al., 2015, 190). However, there are a range of motivations underlying this sharing behavior, that is, why a user might view a post as worth sharing with others. In one study of 768,000 tweets testing four models of reasons a user might retweet a message (communication, content, profile, or random occurrence), “all [four] models turned out to be significant...at certain times, which implie[s] users’ retweeting behaviors [are] drawn out for a range of different reasons” (Lee et al., 2015, 189-90). Lee et al. further describe that “several studies [have] investigated...reasons for retweeting and found them to be for: communication 231 purpose[s], social relationship maintenance, self-expression, obtaining/updating information, and others” (Lee et al., 2015, 190). Studies conducted using survey data have found similar results. Specifically, these motivations were described as “a way to start a conversation,” “social relationship building/maintenance…[in that] users retweeted a message for more social interactions and more intimacy with people in their network and for more influence on their followers,” as well as “information sharing (i.e. let other users know important information), responsiveness (i.e., be able to have immediate feedback), and emotional propagation (i.e., share public indignation)” (Lee et al., 2015, 191). Finally, “users retweeted a message in their social network, and by doing so, the users could gather social resources they could rely on. In particular, retweeting was perceived as a more efficient way to develop social capital since retweeting cost less for obtaining valuable information in comparison to searching for new ones from scratch” (Lee et al., 2015, 191). In addition to this wide range of motivations, Lee et al. add support in their study for an impact of some prosocial motivations. Specifically, “the path from altruistic motivation to behavioral intention was found to be significant, and behavioral intention was also positively related to actual behavior. [This means] that the more altruistically motivated the person was, the greater [the] behavioral intention to retweet was, and finally, more actual retweeting. Meanwhile, reciprocity motivation predicted behavioral intention in a negative way. In other words, the reciprocity motivation does not encourage people to retweet a message but suppresses their intention. The relationships in this model, however, [were] differentiated by the number of followers and followees. A user with more followers is prompted by altruistic motivation and less motivated by [the] principle of reciprocity. [By contrast,] a user with [fewer] 232 followers is driven by altruistic motivation but is resistant to [the] reciprocity principle” (Lee et al., 2015, 199). Features of the message may also influence user sharing behavior. Shi et al. found that “the more URLS or hashtags a tweet contains, the more likely it will be retweeted by the individual, all other things being equal…[whereas] length has very little influence,” which has also been found in other studies (Shi et al., 2017, 849). Clearly, there are many possible factors that may influence a user’s choice to retweet or share content on social media platforms and such a list may make it difficult for government or public health officials to use in their communications strategies and attempts to craft content that will be shared widely. Where Shi et al. most contribute to this research is in investigating which factors are most influential across a variety of users. Their study found that “all of the major components involved in a social communication process have impacts on individual retweeting behavior. Among them, topical relevance and social tie strength are the most important factors, followed by information richness of the stimuli and the bandwagon effect. The information source and sentiment of the tweet, however, have trivial impacts on retweeting behavior...[with this] ranking of features...confirmed again using multiple feature selection methods” (Shi et al., 2017, 855). The features identified by Shi et al. have important links to the psychological principles discussed previously. For example, as was previously discussed, many users do not process a message deeply, especially in a social media environment, which has significant implications for government and public health officials attempting to publish messages that will be retweeted and shared widely. First, as Jalali & Papatla note, “findings in the literature...suggest that followers are more likely to retweet [a message] 233 only if they recognize that the tweet’s topic fits their tweeting interests,” a finding that makes intuitive sense (Jalali & Papatla, 2019, 663). However, research has also found that “people [often] spend only a few seconds reading a tweet,” for a variety of reasons; for example, a user that spends only a few minutes on social media each day or a user that scrolls quickly through large amounts of content (Jalali & Papatla, 2019, 663). As a result, a user may make only a cursory glance over the text of a tweet before moving to the next. Indeed, heuristic processing is supported by the fact that reading itself is a cognitively intensive task, involving not only short fixations on each word (called saccades), but also “multiple sub-stages [of processing] like parsing, understanding the syntax and context, interpreting the totality of the text, and reanalyzing the text if necessary” (Jalali & Papatla, 2019, 650). Many users do not engage in this full extent of cognitive effort in a social media environment because, in such an environment, “the goals of interacting with tweets are...likely to be modest rather than highly significant, [meaning that] reading them may [therefore] be based on...heuristics” (Jalali & Papatla, 2019, 650). In deciding how to present a message in a tweet, Jalali & Papatla argue that “if words related to tweeting interest occur more often in the tweet, therefore, the likelihood of [users] locating them as they scan it, fixating on them, and using them to comprehend the message, should also increase. This should increase the likelihood of the tweet’s topic being recognized even if it is just scanned and thus increase retweets by those interested in the topic” (Jalali & Papatla, 2019, 650). Similarly, they state that “earlier location of topic-related words is also likely to increase the likelihood of fixations on subsequently occurring topic-related words” (Jalali & Papatla, 2019, 650). Thus, the following three strategies may be useful, echoing the findings and 234 recommendations from Jalali & Papatla’s study. First, “social media managers should start by identifying topic-related words for the topics of their tweets…[and] place several topic-related words as close to the start as feasible” (Jackson & Welles, 2015, 664). Second “add a hashtag in front of each topic-related word to draw attention to the topic,” as research has found not only the benefits of hashtags discussed previously, but also that tweets containing hashtags tend to be retweeted more frequently (Jalali & Papatla, 2019, 664). Third, “compose the rest of the tweet by filling it with symbols and words as needed, based on the following priority: (a) RT_IF[a symbol asking users to retweet] and HTTP: the significant positive effects of RT_IF and HTTP imply that they should also be included in the tweet. (b) Other words, punctuations, and symbols relevant to the message. For instance, a tweet about a new product may include ‘New,’ the name of the new product, and when the product would be available [components that should be adapted to a public health context]. (c) Words, punctuations and symbols necessary to complete the composition. For instance, verbs, pronouns, and periods have to be included as needed” (Jalali & Papatla, 2019, 664). Lin & Lee-Won demonstrate that there are also benefits for organizations that retweet others’ tweets, in yet another example of the value in using the tools social media affords beyond treating the platforms as a one-way mode of broadcasting communication. Their study found that “an organization’s retweets of user mentions addressed to the organization, which were termed dialogic retweets, resulted in a higher level of perceived social presence than did the organization’s monologic [one-way] tweets. The results further demonstrate that social presence played a key role in the mechanism through which the organization’s Twitter practice exerted persuasive influence on the audience. 235 That is, dialogic retweets (vs. monologic tweets) from the organization increased participants’ intention to adopt the behavior promoted in the messages through two significant indirect paths: (a) serially through social presence and subjective norms associated with the target behavior and (b) serially through social presence, subjective norms associated with the target behavior, and attitudes toward the behavior” (Lin & LeeWon, 2017, 430). Lin & Lee-Won continue, explaining that an organization’s retweeting of individual user mentions addressing the organization led audience members to feel greater social presence--the sense of being with other people and being engaged in non-mediated conversations with them--when compared to the organization’s regular, monologic tweets. The results suggested that social presence further influenced subjective norms, attitudes toward the target behavior, and the intention to perform the target behavior...In our study, mere retweets of user mentions addressing the organization--[even] without the involvement of any direct responses or comments added to the original tweets--still evoked greater feelings of social presence than did the monologic tweets of identical content. The simple action of retweeting user mentions addressing the organization, signaling reciprocity and dialogic intent to reach out to the public might have served as a cue of perceived interactivity. It is also possible that the dialogic retweets amplified the illusory sense of being together and interacting with others for the viewers by engaging the viewers as co-listeners of the conversation and as (vicarious) participants in the organization’s dialogic loop (Lin & Lee-Won, 2017, 430-1). Thus, if used strategically, retweeting and sharing behaviors can be powerful tools for government and public health officials both for encouraging users to spread their content more widely and for building relationships, shaping norms, and influencing attitudes that can lead to behavior change. Given both the motivations behind sharing behaviors and the potential benefits, it is therefore important to evaluate the tweets collected in the current study and identify any trends in content that received more retweets. First, Figure 19 below shows the number of retweets by month separated by whether the tweet contained words specifically referring to the COVID-19 pandemic (Covid, coronavirus, or pandemic). 236 Based on this graph, there is a typical range of retweets most tweets receive, regardless of whether the tweet contains the words described above or not. This range appears to be between 0 and 15 retweets, certainly a low number indicating limited reach from firstdegree retweets. However, it is also clear from the graph above that nearly every instance in which a tweet receives a greater number of retweets between March 1, 2020 and March 1, 2021, the tweet includes at least one of the three terms described above, classified as a “COVID tweet.” There are many tweets in each month that fit this description, with some tweets receiving as many as 100 or more retweets. It is also clear that the month with the most tweets that received above average numbers of retweets was March 2020, the first month of the pandemic in Utah (as March 2021 was excluded from this date range). This includes both tweets that were original content as well as tweets that were retweeted from other accounts on the accounts surveyed in the present study. Figure 19. Distribution of retweets for COVID and non-COVID tweets by month across all accounts. Range was restricted to 0-200 on the x-axis to enhance visibility, though the general trends described are 237 not affected by this restricted range. For example, a tweet originally posted by the Utah Department of Health on March 14, 2020 stating “MEDIA ALERT: First case of #COVID19 community transmission in #Utah” along with a link to additional information was retweeted by another account in this dataset. The tweet itself received a total of 108 retweets. Tweets that were retweeted from larger accounts appear to carry the retweet total of the original tweet, such as a tweet originally posted by the CDC and was retweeted on March 16, 2020, with the text “New: Starting immediately for the next 8 weeks, CDC recommends cancelling all events of 50 or more people” had a total of 9300 tweets, far above the typical range of retweets for original content. Indeed, all of the most retweeted posts during March 2020 were retweets from other accounts, most notably the CDC. While there are cases in which an original tweet received a high number of retweets, such as one posted by Dr. Angela Dunn on July 2, 2020 (Example 5) stating, “We had a #COVID19 surge after Memorial Day. Our hospitals can’t handle another one. We are spending the holiday weekend apart from others, outside in the fresh air, and wearing masks IF we have to go anywhere else. Please do the same” which received 406 retweets, manual analysis indicates this is an infrequent occurrence for original posts created by the accounts included in this study. This observation, in conjunction with the relatively low numbers of tweets that received large amounts of retweets (as indicated by the typical range for each month shown in Figure 19), indicates that there is room for growth in regularly crafting original content that users will retweet widely, particularly in the case of local health departments. 238 Example 5. Tweet by (former) State Epidemiologist Dr. Angela Dunn. Figure 20 shows the number of retweets received by COVID and non-COVID tweets but separated instead by the sentiment the text of the tweet conveys (the process by which sentiment is calculated using the TextBlob Python library is described previously). Based on the graphs below, tweets conveying a neutral sentiment appear, on average, to have an extremely limited appeal for users to retweet--with the exception of two small peaks for COVID tweets at approximately 75 and 125 retweets, all of the tweets above the range of 0-15 convey either a negative or positive sentiment. While positive tweets seem to be more popular than neutral tweets, on average, the data in Figure 20 ultimately seem to confirm what has been described in other studies--that negative content is generally more popular on Twitter and receives more attention. For both COVID and non-COVID tweets, the only tweets receiving above 300 retweets conveyed a negative sentiment. While not every tweet should convey a negative 239 sentiment, this is nevertheless useful information for government and public health officials attempting to post information that will spread widely. It also seems that COVID tweets have a wider range for the typical number of retweets--there is a large number of tweets that received fewer than 25 retweets, there are also many tweets that received up to 125 retweets, while these peaks in the ridges plots above are much smaller for nonCOVID tweets. Figure 20. Distributions of retweets by sentiment across all accounts. Range was restricted to 01000 on the x-axis to enhance visibility, though the general trends described are not affected by this restricted range. The previous analysis was repeated with tweets instead being categorized by whether the tweet mentioned a COVID-19 protective behavior, with the results shown in Figures 21 and 22. 240 Figure 21. Distribution of retweets by month for tweets mentioning COVID-19 protective behaviors across all accounts. Range was restricted to 0-750, though no trends described were impacted by this restricted range. Figure 21 shows clear trends in the types of content that users retweeted more frequently, despite the fact that just over 11,000 tweets were included in this analysis, a relatively limited dataset for the computing capabilities available. While tweets not mentioning COVID-19 protective behaviors (“non-behavior tweets”) seem to have an extremely limited appeal for users as gauged by the number of retweets, again falling below 15 retweets, in every case in which there is a tweet that received above this amount, regardless of month, that tweet mentioned COVID-19 protective behaviors. Of course, the month with the largest peak is March 2020, with a sizable number of tweets receiving just under 125 retweets, but there are many other months, such as June or July, in which 241 many tweets received far above 15 retweets. These tweets seem to have an additional appeal for users to retweet widely, sharing this information with their social networks, in contrast to tweets not mentioning COVID-19 protective behaviors. Perhaps one reason might be because these tweets specifically mention actions that leaders in the state encouraged people to practice, such as masking and social distancing. Users might want to make sure their network is informed of updates in policies or practices to keep their friends, family, and followers safe; conversely, they might share news about these behaviors out of anger that they wish to share with their network. The exact reason a user chooses to retweet such content will require both additional research from scholars, as well as closer attention from the account administrators posting content. However, the data above show a clear trend that tweets mentioning COVID-19 protective behaviors received more retweets than those not mentioning behaviors regardless of month. Figure 22 similarly compares the tweets mentioning COVID-19 protective behaviors but separates tweets by sentiment conveyed by the text of the tweet. This Figure amplifies the trends described in Figure 21--that tweets mentioning COVID-19 protective behaviors receive more retweets than those that do not. In the graphs above, the typical range for tweets not mentioning behaviors appears to be, on average, below 25. In contrast, there is much more variability in the number of retweets the behavior tweets receive. The typical range is more difficult to discern from visual inspection, but there are a large number of tweets that receive up to 100 retweets, with smaller peaks visible beyond that point up to approx. 600 retweets. However, unlike the analysis comparing tweets mentioning the terms “COVID,” “coronavirus,” or “pandemic,” there does not appear to be a clear trend in which negative tweets receive more retweets than 242 positive or neutral tweets. While in both cases neutral tweets received the least retweets, there are both negative and positive tweets that received above the typical range of retweets, with both red and Figure 22. Distribution of retweets mentioning protective behaviors by sentiment. Range was restricted to 0-750, though no trends described were impacted by this restricted range. green peaks clearly visible above 100 retweets, though negative tweets do seem to appear more frequently on the higher end of the x-axis. While more research will be needed to investigate this finding beyond an exploratory level, it nevertheless implies that government and public health officials ought to pay close attention to the sentiment they convey as well as the topic of a tweet, as they might affect the number of retweets, and similarly spread, of the content. 243 LIKES, FAVORITES, AND REACTIONS “Liking” or “favoriting” a post is one common reaction afforded to users on a range of social media platforms. These platforms may define the “like” or “favorite,” button using descriptions like the following: “Facebook (2014) defines their Like feature as, ‘an easy way to let people know that you enjoy [a post] without leaving a comment” and Twitter (2014) espouses their Favorite as a way to, ‘let the original poster know that you liked their Tweet’” (Hayes et al., 2016, 175). Hayes et al. make some important observations about these features, first classifying them as “paralinguistic digital affordances (PDAs): cues in social media that facilitate communication and interaction without specific language associated with their messages” (Hayes et al., 2016, 172-3). They also note that these PDAs “explicate their intended purpose and most of these explications center around the expression of positive emotion toward a post” and that “the designers of these tools intended phatic uses,” defined as “‘language used in free, aimless, social intercourse’ that displays sociability, but is comprised of limited, or no, real information” (Hayes et al., 2016, 175, 173). These quotations explain the primary use and goal behind including these features as part of social media platforms, and, as Hayes et al. note, “PDAs [may be] faithfully adopted when users literally enjoy the content and seek to display positive socioemotional attributions directly toward the post...without inferring attributions about the poster, the context of the post, or the cultural or societal meanings that may be related to but beyond the scope of the post” (Hayes et al., 2016, 175). Yet, likes and favorites have become far more complicated and have many more meanings and uses than designers originally intended. In other words, “perhaps a Like 244 does not [necessarily] mean that one literally felt an affinity toward the content, nor does Favoriting a post indicate the post is preferred over others, and users instead have adopted their own meaning and use of these tools specific to the media, their social connections, and their communicative goals” (Hayes et al., 2016, 175). Because users can attribute different meanings to likes and favorites depending on their individual circumstances and needs, as Lahuerta-Otero et al. note, “getting a like on a tweet is a complicated task” (Lahuerta-Otero et al., 2018, 572). For example, in addition to denoting a positive reaction toward the content, “liking a tweet is a special action on the part of the user (content that they wish to save for the future or a message that they agree with implicitly) but does not necessarily imply that it is shared with others” (LahuertaOtero et al., 2018, 572). Liking, then, while indicative of a generally positive response to the content, is different from other actions on Twitter because “when an individual considers a tweet to be important or interesting, they are more likely to share it with their peers and followers as a form of solidarity and generation of information flows,” which describes the behavior of retweeting or sharing, discussed in greater detail below; by contrast, “just liking a post does not imply its diffusion to other users” (Lahuerta-Otero et al., 2018, 573). While research is scant on these other meanings users ascribe to likes and favorites on various social media platforms, Hayes et al. conducted interviews with small focus groups in an attempt to understand users’ behavior. In their study, “one of the first questions asked participants to describe whether Likes, Favorites, and Upvotes meant the same thing to them. The majority of respondents indicated the meaning of these cues were not synonymous...Participants repeatedly noted that when scrolling through their 245 feeds, their [liking] behavior on Facebook is more reactionary than on other platforms, based somewhat on the poster rather than the posted content...On Facebook, the act of aimlessly providing Likes in reaction to seeking their friends, rather than actually processing the content, seemed to manifest itself in participants’ expectations for a threshold of Likes a post needed to receive to be ‘good’” (Hayes et al., 2016, 177). Meanwhile, “this was not the case on Twitter: ‘Lots of posts go by with no favorites. It isn’t weird,’ [one participant noted.]...On Twitter, the content had to be ‘good’ to get favorited, though what ‘good’ meant seemed to vary, with criteria including content perceived as funny, thought provoking, or ironic” (Hayes et al., 2016, 177-8). Thus, user expectations and behaviors may vary between what would otherwise appear to be equivalent features on different social media platforms. Additional research into the social norms and expectations regarding user liking or favoriting behavior on different social media platforms would be useful knowledge for government and public health officials seeking to maximize their use of one, or especially multiple, platforms. Hayes et al. do note that “it was clear different meanings were attributed to [likes] sent and received across social media platforms, differences most manifest in a threshold level of [likes], varying both by users and platform, for users to be confident in the quality of posted content” (Hayes et al., 2016, 178). Users may consciously or subconsciously use the number of likes a post receives as a cue related to the quality of the post, though whether this is the case, the circumstances under which it might or might not be true, and any other factors affecting the relationship must be investigated with more robust studies. As noted previously, some users do use likes and favorites to convey the meaning intended by designers of the features, “as an evaluation of the content to which they are 246 responding,” though in Hayes et al.’s interviews, “certain types of content seemed to lend themselves to literal Likes, including celebrity news, pets, humor, and new product releases. Those who interpreted the Like literally refused to provide the [like] if they disliked the content” (Hayes et al., 2016, 178). Where health information falls, both that communicated under normal circumstances and under crisis conditions, is an important topic that is yet to be researched fully. For the time being, it is emphasized, here as well, the importance of government and public health officials paying close attention to the engagement behaviors of users in response to their posts. In their study, Hayes et al. note that users reported using liking and favoriting features for purposes such as “sav[ing] broadcast content for later reference,” or, particularly on Facebook, “a way to [simply] acknowledge [that] they had seen a post,” which some participants described as “‘a subtle recognition,’ [or] ‘affirmation of someone’s post,’ and ‘OK, will do’ when there was a request from them” (Hayes et al., 2016, 178). Some participants pointed to perceived social aspects of likes and favorites that clearly go beyond the original meaning: “several participants talked about how significant life events or achievements, such as weight loss, always warrant using [likes] to show their support…[though] this social support also had an aspect of social grooming, as participants would respond to others’ content through [likes] because they felt obliged to do so as part of relationship maintenance. Several participants discussed that they had to ‘Like back’ or reciprocate to preserve the relationship” (Hayes et al., 2016, 179). While the participants here described liking or favoriting behaviors in reference to other individuals or social contacts, this is nevertheless useful knowledge for government and public health officials, both because it provides a window into how networks or groups on social media may function and 247 because liking individual users’ content (in addition to the content of other leaders or related agencies) might be a way to create and preserve the relationships with members of the public that are so important to maintain. The analysis above was repeated for the number of favorites (“likes”) each tweet received. Figure 23 shows the number of favorites received by tweets mentioning the terms “COVID,” “coronavirus,” or “pandemic” separated by month. The graphs appear to follow the same general trend as for the number of retweets--that tweets mentioning these terms are the only ones receiving more likes above the typical range regardless of month, but the trend is not nearly as pronounced as it was for retweets. Those not mentioning COVID-19 pandemic terms similarly appear to have an extremely limited appeal for users, as they typically received fewer than 50 tweets in each month, as denoted by the blue ridges. Interestingly, the two months that appear to have the most tweets with large numbers of favorites (above 200) were July and October while March 2020, the first month of the pandemic in Utah, saw few tweets with more than 50 favorites. These graphs lead to concerns regarding the popularity of tweets published under normal circumstances, as well as the high degree of variability regarding COVID tweets. If users do, in fact, use the number of likes as a cue indicating the quality of a tweet’s content, a low number of likes may be damaging to users’ perceptions. 248 Figure 23. Distribution of favorites (“likes”) by month for COVID and non-COVID tweets across all accounts. Range was restricted to 0-500, though this restricted range did not impact described trends. Figure 24 presents the same data separated instead by the sentiment conveyed in the text of the tweet. Here as well, there appears to be an extremely limited appeal for tweets conveying a neutral sentiment if favorites are used as a measure. The typical number of likes a neutral tweet received appears to be less than 20, though there are a small number of neutral COVID tweets that received as many as 150 likes. 249 Figure 24. Distribution of favorites (“likes”) by sentiment for COVID and non-COVID tweets across all accounts. Each of the tweets that received a large number of likes conveyed either a positive or negative sentiment, though again, there does not appear to be more negative tweets in this range than positive. In fact, for COVID tweets, it seems there are more positive tweets that received over 250 likes than negative tweets. Perhaps the clearest trend in Figure 24 is that many more COVID tweets received larger numbers of likes than non-COVID tweets. Indeed, very few non-COVID tweets received more than 200 tweets. 250 The analysis was repeated a final time to compare the number of likes received by Figure 25. Distribution of favorites (“likes”) by month for tweets mentioning COVID-19 protective behaviors and those not mentioning such behaviors. tweets mentioning COVID-19 protective behaviors and those that did not. Figure 25 shows the same pattern as the number of retweets for tweets mentioning COVID-19 protective behaviors compared to those not mentioning such behaviors. Here, there is a limited range of likes for tweets not mentioning these behaviors, typically receiving fewer than 20 likes. By contrast, tweets mentioning COVID-19 protective behaviors appear to be much more popular as measured by the number of likes. While some months have a significant peak extending beyond the typical range of the non-behavior tweets, such as March, May, and September, there are also many instances in which behavior tweets received well over 100 likes. Interestingly, the months in which the most tweets 251 received high numbers of tweets were April, July, and October, while March did not have any tweets receive more than 150 likes. This might be explained by the fact that in March 2020, the pandemic was still new in the United States and it took time for experts at every level to begin making recommendations for behavior change among the general public. However, the primary takeaway here is that tweets mentioning behaviors appear to be much more popular, both as measured by likes and retweets, than those not mentioning behaviors. Additional analysis, such as thoroughly examining the users that like and retweet these posts, as well as the comments made in response to the post, may further aid government and public health officials in crafting their message. Further, research is needed to investigate whether tweets mentioning actionable behaviors in this and other Figure 26. Distribution of favorites (“likes”) by sentiment for tweets mentioning COVID-19 protective behaviors and those not mentioning such behaviors. 252 public health contexts do, in fact, see more user engagement than tweets covering other topics. Figure 26 again amplifies the pattern observed for the number of retweets in behavior and non-behavior tweets. Specifically, neutral tweets have a limited appeal for users regarding liking the posts, while there is a mix of positive and negative tweets receiving above the typical number of likes (above 50 likes). The graph of non-behavior tweets is a striking contrast to the behavior tweets in that almost none of the non-behavior tweets received more than 100 likes while there were many behavior tweets, both positive and negative, that received over 100 likes. Analysis of both likes and retweets was similarly conducted for @UtahCoronavirusInfo, an account set up by state officials early in the pandemic to communicate important information, and is included in Appendix B. USE OF MEDIA As discussed previously, characteristics of the message and the way in which it is presented can have a significant impact on the effectiveness of a message in achieving its goal. In addition to the text functionalities of popular social media platforms (which may or may not be limited to a certain number of characters) social media platforms like Twitter, Facebook, Instagram, and others allow users to post images or videos either with or without a text description. While there is very much a need for additional research regarding the use of images and videos on social media in a public health context, previous research will be discussed briefly. Using the example of encouraging exercise among users, Johnston & Davis used a within-subjects experimental design to test the responses of participants in four conditions (individual posts with and without images as well as corporate posts with and without images) adding to the theoretical and non-experimental research. These four conditions were tested based on two areas of previous research. First, many studies 253 suggest that “content with images may be more likely to influence exercise motivation than content without images on social media. [For example,] in [one] study analyzing social media posts from the National Cancer Institute’s Facebook page, social media posts with images were more engaging and received significantly more likes, comments, and shares than posts without images” (Johnston & Davis, 2019, 120). Second, previous studies have also found that “while social media content provides many beneficial opportunities for corporations, its effectiveness can be limited if social media users recognize corporate social media content as an attempt to influence, persuade, or manipulate them” (Johnston & Davis, 2019, 120). Specifically, “research on sponsorship disclosures has demonstrated that when media content is identified as advertising prior to or during exposure, people process the information more critically and are more resistant to persuasion. Similarly, people find content to be less persuasive if they perceive it to be using manipulative tactics like deception” (Johnston & Davis, 2019, 120). Meanwhile, “research [also] suggests that user-generated content actually has a stronger influence on purchasing behavior than marketer generated content in brand communities on social media. The persuasiveness of user-generated content may be explained in part by individuals recognizing the marketer-generated content as advertising and evaluating it more critically...However, this does not mean that content from individuals is always seen as more trustworthy than corporate content,” and in fact, attitudes toward these tweets can vary widely (Johnston & Davis, 2019, 120-1). Nevertheless, individuals being wary of organizational tweets that are perceived to have the intent and persuasion is one important consistency that can be applied to a public health context. Individuals are far less likely to be responsive to a message posted by an organization if the organization is 254 perceived as attempting to persuade or use manipulative tactics. In Johnston & Davis’ experiment, they found a number of important results that may be of use to government and public health officials in their individual and organizational social media accounts. Overall, “stimuli with images were rated as significantly more motivating than stimuli without images” though there was an interaction between the presence of images and the type of account (Johnston & Davis, 2019, 123). Specifically, “stimuli with images from individual accounts were more motivating than stimuli with images from corporate accounts, and stimuli without images from corporate accounts were more motivating than stimuli without images from individual accounts” (Johnston & Davis, 2019, 123). Johnston & Davis also investigated participants’ perceptions of the source’s intent in posting the content, namely whether the motive was intrinsic (self-motivated) or extrinsic (linked to external rewards), which would typically make users more wary of the content and persuasive attempts. In their study, “participants clearly identified corporate posts as having been posted for extrinsic reasons than individual posts [though] stimuli with images were also rated as having been posted for more extrinsic reasons than stimuli without images” (Johnston & Davis, 2019, 123). However, as was the case with motivation, “these main effects were qualified by a significant interaction between content type and account type. Ratings of stimuli with and without images did not differ for individual accounts, whereas stimuli with images were rated as having been posted for more extrinsic reasons than stimuli without images from corporate accounts” (Johnston & Davis, 2019, 123). Regarding these findings, Johnston & Davis provide the following possible explanations. First, regarding the greater effectiveness of individual posts with images, 255 they argue that “research has...shown that social comparisons are more informative and influential when you are comparing yourself to similar others...The relative similarity between users and people portrayed in the images may help explain why content with images from individuals more effectively promoted exercise motivation than content with images from corporations” (Johnston & Davis, 2019, 124). Next, regarding the greater effectiveness of corporate content without images than for similar posts made by individuals, Johnston & Davis note that “the corporate posts without images in this study focused almost exclusively on motivating the individual reading the post, whereas posts without images from individuals more often focused on the achievements of the person posting the message. By focusing on a motivational message without an image drawing attention to the brand or product, corporate posts without images may also be less likely to be seen as advertisements or as being manipulative compared to corporate posts with images. Consistent with this possibility, ratings of ‘being posted for extrinsic reasons’ were lower for corporate posts without images than corporate posts with images. In contrast, the individual posts may have been at greater risk for being dismissed as a form of bragging” (Johnston & Davis, 2019, 124). While this study focused on a public health issue (exercise), it compared individuals with corporate social media posts. Where, exactly, public health organizational accounts fit in must be the topic of additional research, as they are clearly not individual accounts, but they share some features with corporate accounts and do not share others. For example, they may similarly be perceived as posting for extrinsic reasons, using tactics that users may perceive as manipulative or trying to persuade; however, the goal behind public health organizations is not to maximize profit as it is for businesses. Until there is further research regarding the use of 256 images on social media in a public health context, government and public health officials should seek feedback from users and remain aware of how they and their messages might be perceived. It may be more effective, and perhaps motivating, for notable individuals to post images than for organizations to do so, though again, feedback and data would be needed to evaluate the circumstances in which this might be the case. Though their research was conducted more than 15 years ago, Houts et al. provide useful insights regarding the use of images in health education more broadly, albeit not on social media. Using a literature review approach, they evaluate the effectiveness of images on four specific steps in information processing: attention, comprehension, recall, and adherence. First, they find that “patients receiving handouts with pictures were significantly more likely to read the handouts, and among those who read the handouts, patients, receiving...illustrated versions were significantly more likely to remember what they read and to follow the instructions than those who read just text,” though Houts et al. were able to locate only one study on this topic (Houts et al., 2006, 175-6). In short, they argue that “one contribution to health education is to attract the attention of patients and families and to stimulate them to attend to the information” (Houts et al., 2006, 175). To assess whether and when this may be true on social media, government and public health officials should seek to monitor their individual and organizational accounts, both for the engagement metrics discussed above, as well as post or page views where possible. Regarding comprehension, Houts et al. note that in addition to the lack of literacy and especially health literacy among many members of the public, two major problems regarding communication between physicians and scientists and the public are the use of jargon or technical terms, and the tendency for “health professionals [to] qualify 257 statements and speak in broad generalizations to patients who want specific information that applies to them, personally” (Houts et al., 2006, 177). While potential legal issues with providing individual recommendations over social media ought to be avoided, there are still many ways public health officials in particular can use social media to bridge this gap. For example, they might respond to a specific user’s comment asking if a certain activity is considered risky during the COVID-19 pandemic, or assist users in connecting with a physician, even through telehealth visits. They might also seek ways to make images used on social media feel more personally applicable to the audience, though this is another area in which additional research is needed. However, many studies have found that pictures can “aid comprehension by providing context for organizing information in the text” and that “pictures can sometimes be effective substitutes for words when the information to be conveyed is primarily visual” (Houts et al., 2006, 180). In fact, Houts et al. argue that pictures are “almost always helpful” except when the concepts being conveyed are beyond the audience’s ability to comprehend (Houts et al., 2006, 180). In general, research has found that “line drawings seem to facilitate comprehension more than do shaded drawings or photographs and color photographs seem to have a greater effect than black and white pictures’” (Moore qtd. in Houts 2006, 180). Such line drawings are illustrated in Hout et al.’s study as simple cartoon images, though a more modern, computer-based parallel might be infographics (if designed for simplicity). Additional research provides ways that government and public health officials can seek to make images feel more personally applicable to members of the audience, that is, through images that are locally- and culturally-developed, again emphasizing the value in using social media as a way to interact with many individuals, 258 rather than as a single group. For example, Houts et al. cite one study in South Africa which showed “significantly higher comprehension of the locally developed pictures in comparison to the standardized pictures. They also reported a strong preference for the African-based pictures” (Houts et al., 2006, 180). Even if the changes between versions of the same image appear to be small to those who post or create them, Houts et al. note that “these small differences [are] important to the people viewing [them]” (Houts et al., 2006, 180). Studies on both free recall and cued recall have found positive effects from the use of images. Houts et al. located five studies evaluating free recall with three finding greater recall when pictures were used in conjunction with text, though one found no effect and one had conflicting results (Houts et al., 2006, 184). The study that saw no effect surveyed 40 mothers in rural Kenya with limited reading skills and “while they found no differences in recall between text groups, they did find that mothers recalled information in the pictures at a much higher rate than information in either form of text” (Houts et al., 2006, 184). Additional research is needed to confirm these findings and evaluate them in a social media context, as well as to determine if certain types of images have different effects on memory. Finally, Houts et al. reviewed research addressing if pictures affected behavioral intentions. While some studies had results that were difficult to interpret, such as one that used icons and graphs with numerical information that the control group did not have, in the few studies they did locate, pictures appear to have a positive effect on behavioral intentions. In one study of 234 emergency room patients, “not only were instructions with pictures more likely to be read and remembered, they also found that people who 259 received the illustrated instructions were more likely to do what was recommended in the instructions (77% versus 54%),” a difference that was statistically significant (Houts et al., 2006, 186). Another study specifically used images that were designed to include people that resembled the participants and living in similar circumstances to the participants. Here, “results showed that patients in the picture group took, on average, 90% of the pills prescribed for the time period as compared to 78% for the control group,” which was also statistically significant (Houts et al., 2006, 186). These researchers concluded that “for pictures to be effective, they must be meaningful to the people in the audience” (emphasis added) (Houts et al., 2006, 186). Based on the review by Houts et al., there may be additional reasons to use images in social media posts, such as attracting the attention of users or making messages more memorable, that cannot be measured by engagement metrics such as likes, comments, or retweets. In the absence of strong evidence supporting specific strategies, however, for now, government and public health officials must rely on these metrics, in part, to understand their audience and their thoughts, attitudes, and behaviors. Engagement metrics do not only measure engagement from users, they may be used as one way to gather feedback from users, a powerful advantage social media platforms hold over traditional media sources that were examined by Houts et al. and the studies they reviewed. 260 CONCLUSIONS AND PROPOSITIONS “Clearly, pandemics are more than just health crises. They produce concerns about the mentality of a nation as much as they do about the microbes of the people.” ~Nancy Bristow in “Reconsidering the 1918 Influenza Pandemic in the Age of COVID-19” (2020) As the quote above describes, the primary purpose of this paper is to emphasize the importance of human thoughts and behaviors in the spread of a disease like COVID19 and to reframe thinking about public health promotion. The scientific community has conducted numerous studies throughout the COVID-19 pandemic to better understand the nature of the virus as well as the interventions that are most effective in preventing infection, creating information that has been used to inform public health policy in the United States and around the world. However, a public health recommendation can only be as good as the public’s compliance with the recommended behavior, however scientifically well-informed a policy may be. When behaviors must be sustained over a long period of time as required by a global pandemic, social behavior presents additional challenges. Rypdal et al. constructed a model of first and second waves of COVID-19 based on the SIR (susceptible, infected, recovered) model in which epidemic evolution, as measured by the basic reproductive number, “depends on three parameters: (i) the relaxation rate when incidence is low, (ii) the intervention rate when incidence is high, and (iii) [in intervention] fatigue rate that gradually weakens the effect of interventions over time” (emphasis added) (Rypdal et al., 2020, 2). Using data from countries around the world, Rypdal et al. found that “a necessary condition for the development of a strong second wave is the absence of a resolute response when it becomes clear to everybody that the reproduction number is rising well beyond 1. With no intervention fatigue, our 261 model cannot produce a second wave more severe than the first, even with a very high relaxation rate [such as that] resulting from summer holidays in the Northern hemisphere” (emphasis added) (Rypdal et al., 2020, 15). In a country like the United States, compliance with public health policies relies on individuals choosing their own behaviors. Information is also available to the public from a variety of sources, some more factually correct than others. These factors must be taken into account by public health officials in the United States because the communication context is therefore far more complicated than simply conveying facts and recommendations to a population and expecting everyone to understand the messages and make the rational choice to engage in the recommended health protective behavior. This concluding section will synthesize the previous discussions and make propositions regarding the nature of public health communication with the goal of improving the effectiveness of public health promotion in the future. These recommendations break the public health communication environment down into the four elements identified by the Yale Attitude Approach discussed previously (message, source, audience, and channel). 1. A relationship exists between the source of a message and an audience outside of any communications. This is to say, for example, that due to the public nature of positions and institutions who are public health leaders in a community, the audience forms an evaluation of these authorities whether or not they receive or send any messages to these authorities. Evaluations may include a perception of the authority’s level of expertise. However, in the modern world, the relationship between source and audience is two-way, consisting of both perceptions of the other party and feedback, as well as any messages sent between them. This 262 relationship impacts the reception of any messages sent from the source to the audience and cannot be ignored if public health communications are to be maximally effective. 2. While the underlying cognitive mechanisms behind information processing are the same across audience members, the processing and understanding of the same core message will not be the same across all members of the audience due to the inherent variations in audience members’ characteristics and the diversity of their experiences. 3. A “one-size-fits-all” approach does not maximize the effectiveness of public health communications because it fails to take into account the many factors documented by psychology that can affect an audience’s perception and receipt of a message. Messages must be adjusted and presented differently for different subgroups within the population. As Snyder notes, “one of the basic tenets of campaigns is to specify fairly homogeneous target groups for the campaign and to create messages designed for each group” (Snyder, 2007, 35). Buckton et al. echo this message: “there is no such thing as a typical consumer and consequently ‘one size fits all’...health promotion messages may remain ineffective, their constant repetition resulting in…[desensitization] and negativity….It should be feasible to [categorize] groups of consumers, as markers do, and tailor...health promotion messages and their delivery accordingly” (Buckton et al., 2015, 11). 4. For public health communications to be truly effective, the audience cannot be thought of as one singular group, but rather as many subgroups in much the same way epidemiologists already segment populations by relevant factors (such as 263 age, race, pre-existing conditions or risk factors). However, for health communication, relevant factors would include those discussed previously that may impact how a message is processed, such as health literacy, risk perception, personality, media (channel) preference, pre-existing attitudes, social group membership, and need for cognition (from the Elaboration Likelihood Model), as well as age with a particular focus on adolescents and the unique psychology that applies during this time of life. Snyder agrees that “although targeting is often based on demographic categories, it can be [more] advantageous to target theoretically meaningful segments of the population. Campaign designers should consider targeting by behavior, actual and perceived risk, misinformation and beliefs, environmental barriers, and communication patterns, because these factors affect the nature of the messages produced for the campaign and the ways to communicate them” (Snyder, 2007, 35). 5. The channel places important limits on messages. For example, the character limit on Twitter does not exist on a health department’s website, meaning that the messages shared on these two channels must be, and ought to be, different. The channel must be selected first and a message adjusted to fit the constraints, as well as the strengths, of the channel. 6. In the modern world, every element of this model (message, source, audience, and channel) is, or can be, affected by outside sources. The ways in which an audience or source may be affected by outside influences are perhaps more obvious, such as media reports or individuals outside the source-audience relationship posting public comments. The message may be conveyed, reframed, or falsified through 264 media organizations or by other people. The channel, while the least obvious of these, is also not immune to outside influences, particularly changing policies, practices, or platform updates implemented by the company that offers the technology (such as Twitter, Facebook, YouTube, hosting companies for websites). 7. In a public health setting, communications intending to promote healthy behaviors that are outside an individual’s normal activities of life must be considered, at least partially, persuasive communications. 8. Due to the nature of public health and its goal of promoting and protecting health, the messages communicated inherently involve the concept of risk, which is measured on a population scale in an objective way from the perspective of the source, but calculated and understood in an individual, subjective way by the audience. Risk is a greater consideration in urgent situations such as a pandemic. 9. While risk perception is a necessary part of health promotion, attempting to change behavior or report on levels of disease in a population, the concept of unrealistic optimism may or may not come into play. This depends not only on the message communicated but also on the characteristics of the audience. 10. It is important for government and public health officials to closely monitor comments made in response to their posts and engage in strategic ways. Part of this strategy must be to pay close attention to the content of comments, such as whether the comment refers to the source of the message (i.e. the individual or entity behind the account) or the message itself. Wallsten & Tarsi’s work suggests that the source of a message should not expect to enhance its reputation through 265 comments on social media, monitoring comments is important because users frequently use comments posted by other users as an estimation of public opinion on a certain topic. Social norms may then be inferred based on the users’ analysis of comments, which may prove powerful in influencing their attitudes and subsequent behavior. 11. In the modern world, there is an internal loop within audience subgroups in which the meaning of the message may be amplified, contested, or mutated outside the control of the source. This may take many forms, such as the formation of echo chambers, defined as an environment in which “one is only presented with information they already agree with, thereby reinforcing one’s confirmation bias” (Jiang et al., 2021). Many social media platforms are prime environments for the formation of echo chambers, which “is due [in part] to a conscious decision made by users when choosing who or what to follow, selectively exposing themselves to [content] they already agree with; but this may also be a consequence of the algorithms social media platforms use to attract users. [Indeed,] numerous studies have shown that echo chambers are prevalent on Twitter” (Jiang et al., 2021). Flaxman et al. echo the importance of algorithms in addition to users’ selfselection: “search engines, news aggregators, and social networks are increasingly personalizing content through machine-learning models, potentially creating ‘filter bubbles’ in which algorithms inadvertently amplify ideological segregation by automatically recommending content an individual is likely to agree with” (Flaxman et al., 2016, 299). Another important mechanism in audience interactions is what Buder et al. call “cross-cutting adversarial debates” (Buder et 266 al., 2020, 6). These are all crucial considerations the source must take into account when crafting messages because each process can lead to attitude polarization. It is crucial that those creating messages for the public acknowledge what recent research on attitude polarization has found: while “early theories on group polarization have argued that group members process discussion arguments in a rational manner so that groups will ultimately converge on the attitude position that has the highest number of sound arguments in its favor...more recent work on polarization…[has] observ[ed] that individuals or groups rarely engage in such kind of rational and unbiased information processing of argument quality” (Buder et al., 2020, 2). 12. Language sentiment or tone is an important consideration due to the effects it may have on how a message might be perceived and how the source might be perceived, as well as the extent to which a persuasive attempt might succeed or cause reactance. Sentiment may also affect how and to what extent users engage with a message or share it with their social networks. Government and public health officials must carefully monitor the tone they use in any text, including in a format like an infographic, and seek to understand how users might perceive the message. These efforts should include both seeking feedback from users and using the engagement features built into social media platforms (likes, comments, retweets, etc.). Ultimately, it is important to seek to match the tone of a message with both the intent of the message source and other elements of the communication environment, such as matching the tone with the nature of the message content, the norms of the social media platform, and the intended 267 audience. 13. Consistency is a vital consideration in successful persuasive attempts. Consistency may be thought of as a source being consistent in the message it conveys over time; however, it is also far more complex. Consistency may be extended to include other similar sources as perceived by the audience (i.e. consistency across messages sent by a state and local health department). One definition of consistency that is frequently overlooked by the source of a persuasive message is the perceived consistency in an audience’s view of the source that may serve as cues for the audience’s evaluation of a message. This might take the form of an audience perceiving some positive characteristics and some negative characteristics in the same source, such as an expert source (positive characteristic) perceived to be dishonest, unlikable, or even unattractive (negative characteristic) (Ziegler et al., 2002, 501). It is vitally important for a source to consider how they are perceived by an audience with respect to common source characteristics because previous research has suggested that “different combinations of source characteristics affect the amount of message scrutiny,” or elaboration undertaken by an audience (Ziegler et al., 2002, 501). In other words, the perceived characteristics of a source affect the audience’s extent of engagement with and evaluation of the core of a message. While it may seem counterintuitive, message elaboration has been shown to increase when source characteristics are perceived to be inconsistent. As Ziegler et al. describe, in their experiment, “strong arguments led to more agreement with the communicator’s position either when he was expert with respect to the message topic [positive 268 characteristic] but dislikable [negative characteristic] or when he was nonexpert [negative characteristic] but likable [positive characteristic]” (Ziegler et al., 2002, 501). In contrast, “argument quality had no effect on recipients’ attitudes and message-related thinking in the case of a likable expert [both positive characteristics] as well as a dislikable nonexpert [both negative characteristics]” (Ziegler et al., 2002, 501). However, it must also be noted that the need for cognition, a characteristic and difference between members of the audience, play a role as well. In short, the logic is that “high [need for cognition] individuals would show evidence of message elaboration regardless of the combination of source characteristics. In contrast, for individuals low in [need for cognition], message elaboration should be heightened in inconsistent (vs. consistent) source conditions” (emphasis added) (Ziegler et al., 2002, 502). 14. Because source characteristics are assessed from the perspective of the audience, the source has limited ability to control them. However, to the extent possible, it is important that public health officials and other scientific and medical experts maintain their standing as experts, with the intention that the audience will perceive them as such. It would clearly be counterproductive to the field of public health if public health officials presented themselves as non-experts in general health matters. The same is the case for honesty as it would clearly go against professional ethics to be dishonest. Therefore, if public health officials wish to use the strategy of creating inconsistency in source cues to encourage elaboration in those with a characteristically low need for cognition, then modifications must be primarily made in source characteristics other than expertise or honesty. 269 15. Source consistency must be thought of both within one source and spanning multiple individuals and organizations that communicate relevant information to the public. For example, in the COVID-19 pandemic, the audience received messages from health departments at local and state levels, national organizations like the CDC, and international organizations like the WHO, in addition to scientists, researchers, clinicians, and government and public health officials. When their messages conflict, there is a great potential for confusion among the public, as well as the potential to increase doubt in the source and resistance to persuasive attempts by these sources. This did, in fact, occur during the COVID19 pandemic: because of the speed with which many researchers shifted to studying the virus, as well as the nature of academia, some “flawed research made the pandemic more confusing, influencing misguided policies. Clinicians wasted millions of dollars on trials that were so sloppy as to be pointless. Overconfident poseurs published misleading work on topics in which they had no expertise” (Yong, 2020). This is certainly not to say that all, or even the majority, of research and practices undertaken during the COVID-19 pandemic meet this description. However, even a small amount of such work may lead to unintended, negative perceptions amongst members of the audience. 16. In a situation like the COVID-19 pandemic, in which the health threat is new, messages presented to audiences in the beginning have an inordinate influence on the thoughts, and potentially behaviors, of the audience. This is due to two psychological phenomena discussed previously: anchoring and mental models. As noted previously, anchoring occurs when an individual attempting to evaluate 270 information or make a decision begins with an initial piece of information but fails to adequately adjust to new information. Gathering information about a new subject, such as the novel coronavirus, may lead to the development of a mental model, an individual’s understanding of some phenomenon based on their experiences, prior knowledge, existing attitudes, and biases. In the case of SARSCoV-2, this might be an individual’s understanding of how the virus is transmitted or how the recommended behaviors protect against the virus (or, conversely, beliefs that the recommended behaviors do not do so). Indeed, survey data found as early as April 2020 that previous influenza vaccine behavior was reliably predictive of intentions to receive a vaccine for COVID-19, though the vaccine would not be widely-available in the United States until the following year and despite the fact that “influenza and COVID-19 stem from distinct viruses” (Southwell et al., 2020, 1708). Other studies have found similar results with other viruses, such as one study that found that participants in Guatemala primarily based their knowledge of the Zika virus on other mosquito-borne illnesses, dengue and chikungunya, despite the fact that Zika may also be transmitted between humans and those infected do not always show visible symptoms (Southwell et al., 2020, 1708). Mental models among members of the public linking SARS-CoV-2 and influenza were likely formed, encouraged, and solidified by the scores of news and media reports, as well as medical experts, government, and public health officials explicitly describing common COVID-19 symptoms as “flu-like.” For example, one report from March 2020 describes the first COVID-19 symptoms a person may experience as follows: “the first 271 symptoms of COVID-19 are pretty common with respiratory illnesses--fever, a dry cough and shortness of breath, says Dr. Carlos del Rio, a professor of medicine and global health at Emory University....’some people also get a headache [or] sore throat,’ he says. Fatigue has also been reported--and less commonly, diarrhea. It may feel as if you have a cold. Or you may feel that flulike feeling of being hit by a train” (emphasis added) (Godoy, 2020). While the report goes on to explain how these symptoms can turn into a deadly disease, it nevertheless begins by telling readers that, “according to the World Health Organization, the disease is relatively mild in about 80% of cases,” thereby presenting the subsequent discussion of severe cases as an unlikely possibility, rather than truly emphasizing the importance of the differences between COVID19 and a cold or the seasonal flu (Godoy, 2020). In short, “past consideration of previous diseases sets the stage for at least initial consideration of novel diseases. Prior experiences afford some opportunities for public communication, but also can leave us vulnerable to blind spots regarding unique aspects of emerging diseases and can post unappreciated challenges for communication campaign design” (Southwell et al., 2020, 1708). 17. Regardless of whether a health threat is new, as was the case with SARS-CoV-2, the public cannot be thought of as “blank slates” (Southwell et al., 2020, 1708). As Southwell et al. argue, “it is unlikely we can successfully introduce wholesale new concepts about infectious disease to audiences without any reference to their past experiences, but it should also be possible to build upon and add to existing conceptualization[s] in productive ways” (Southwell et al., 2020, 1709). Two 272 courses of action are then recommended: first, simply recognizing and “acknowledging existing biases, awareness, and even misunderstanding held by audiences will improve [the] ability to craft communication efforts that respectfully introduce the unique aspects of new public health threats” (Southwell et al., 2020, 1709). Second, Southwell et al. propose “scanning the information environment for evidence of how past news coverage and popular media content [have] framed other diseases and accounting for existing public perception data on related diseases” (Southwell et al., 2020, 1708). 18. As was also discussed previously, additional biases, such as confirmation bias, also prevent the intake of new information that might correct information publicized early that has since been proven false. In the case of COVID-19, this means that information circulated early by public figures, such as the idea that children are immune to the virus, may persist months later, even in the face of sound disconfirming evidence. 19. Reactance can have severe and long-term consequences and therefore ought to be avoided to the extent possible. While public health officials are unable to modify the potential for reactance that is due to personal characteristics among members of the audience, they may do so by modifying the language they use to communicate messages, particularly those designed to provide behavior guidelines. In addition to changing the intensity of the language, Miller et al. suggest that autonomy-restoring language may be used--indeed this may be a good strategy under certain circumstances. They note that restoration may take many forms: “the most basic form of restoration occurs when the target actually 273 performs the behavior threatened in the persuasive appeal, thereby directly restoring his or her perceived freedom to act autonomously. There are also possibilities of vicarious restoration, where another similar person performs the threatened behavior, or indirect restoration, whereby the target performs a different freedom as an alternative to the one threatened. Finally, restoration can also occur by derogating the source of the message” (Miller et al., 2007, 224). When the behaviors are designed to protect the health of both the individual and the larger community, however, this is not how a sense of freedom should be restored and are precisely the reasons reactance should be avoided. Miller et al., however, present an alternative: “any persuasive communication has the potential to elicit reactance to the extent a receiver perceives that the communicator has the intent to influence. However, when an attempt to persuade is disguised within ‘overheard communication,’ or by a distraction, or by the unexpectedness of its presentation, the explicit nature of its intent is disguised and the possibility for reactance is reduced” (Miller et al., 2007, 225). By this logic, Miller et al. “believe it should be possible to disguise the overt nature of a persuasive message and/or immediately restore a threatened freedom by attaching a short postscript message to the end of the main persuasive message--one reminding the receiver that he or she ultimately has the final choice in how to behave” (Miller et al., 2007, 225). While potentially viable, public health officials must use this strategy carefully, particularly given the fact that under the dire circumstances a pandemic presents, the behavior may not always be the choice of the public. Ultimately, further research is needed to investigate the possibility of reducing reactance in 274 this way, as well as other strategies that may mitigate the effects of reactance. However, it is generally advisable that public health officials avoid sparking reactance among members of the audience based on current knowledge of reactance and remain apprised of new research in this area. 20. As a potentially viable strategy for avoiding reactance in audiences, Charry & Tessitore propose nudges as “a concrete way to stimulate behavior changes in the health domain” and specifically point to the number of followers as a readily available nudge on social media platforms (Charry & Tessitore, 2021, 6). Nudges may be of particular use in mitigating the impact of reactance as they can “lead [people] to act in a predetermined direction without limiting their freedom of choice,” which is the fundamental condition that ignites reactance (emphasis added) (Charry & Tessitore, 2021, 1). Specifically, “popular influencers may be solicited to support such causes, notably because people naturally turn to ‘peers’ to validate opinions in the health domain. Inventive tactics to gain followers on a social media account (such as organizing contests, etc.) may also be implemented” (Charry & Tessitore, 2021, 6). While additional research is needed to investigate the effectiveness of such strategy, nudges may provide a starting point for finding concrete strategies to reduce the potential for, and impact of, reactance among audience members. 21. This impact of early information presents an important conflict in social media communications in the field of public health especially. Public health is a field that is grounded in science--a discipline that uses a systematic method of hypothesis generation, experimentation, and theory modification to come 275 progressively closer to the truth over multiple experiments and longer periods of time. Social media, by contrast, thrives on spreading information quickly. Additionally, while science strives to be an objective, logical field, social media is intentionally designed as an emotional space. Indeed, emotional processing is also built in to the very structure of many social media platforms: “users of social media platforms are allowed a palette of actions and reactions: some are seemingly neutral (commenting, sharing and posting) while others have a clear emotional valence: liking and using other emoticons to endorse or dislike a post. These emotionally charged reactions are easier to perform than the neutral ones: it takes a split second to click ‘like’ on a post, but some more time to comment on it or even share it. Most of these emotional reactions have dedicated buttons which can be clicked mindlessly, yielding the interaction seamless” (Marin, 2020, 3-4). Social media platforms also encourage emotional expression, such that “the common assumption is that Twitter [for example], by its very nature, is a space where irrationality, aggression, and ignorance are markers of communication. This does not mean such communication is always the case; rather, the shared appraisal of Twitter as fostering this kind of interaction means users are largely socialized to expect to experience with or about another user in this way” (Duncombe, 2019, 417). Some also point out that “Twitter ‘valorizes pithy retorts…[and] makes substantive engagement difficult and snark very easy” (Drezner qtd. in Duncombe, 2019, 417). Additionally, “the sheer volume of information and the speed at which it arrives suggest Twitter use runs counter to prudential reasoning, which requires more time in its development…[while] 276 ‘surface’ emotions such as anger--easily accessible and socially acceptable to express...are more frequently expressed outside of contained deliberations. Posting emotionally charged tweets is a way to ensure they get attention, particularly when repeating monosyllabic words like good, bad, and sad. The affective dispositions of anger, disgust, and fear represented through informal Twitter statements are much more likely to receive attention online as ‘negative sentiment’ is the key to popularity on Twitter” (Duncombe, 2019, 419). Emotions can also spread across an online environment, either consciously or subconsciously, in a similar way to what has been observed in other conditions. Duncombe explains this phenomenon as follows: “if we see others display an emotion, this experience can engender an emotional response within ourselves that shares, at the very least, ‘threads’ of that initial, witnessed emotional display...This is not to say that individuals explicitly replicate the feelings circulated within the broader social sphere; as Andrew Ross argues, ‘those exposed to emotion[al] contagion do not somehow become affective carbon copies’ (Ross 2014, 32). There is always an element of interpretation that feeds into the transmission of emotion between one person and another or from one group to another” (Duncombe, 2019, 415). However, research has found that “even just imagining how oneself and another might feel in a given circumstance activated similar areas of the brain and produced comparable psychological responses…The social transmission of emotion is therefore possible via online social media platforms such as Twitter. Users can represent emotions via their tweets, and this can also provoke an emotional response” in other users 277 (Duncombe, 2019, 416). Therefore, in order to use social media to communicate with an audience, public health officials therefore have the unique challenge of navigating these central conflicts associated with communicating complicated, logical, and continuously-evolving information (some of which may conflict as experiments do not always replicate results) with a public on an emotional and time-sensitive platform, in essence, a conflict between the nature of the message and the channel. 22. Heuristic processing is likely in both the beginning and end of a crisis situation like the COVID-19 pandemic. This is due not only to the nature of social media as a platform, but also due to characteristics of the users and the situation. Users, “according to the least information principle...would prefer to process messages heuristically as they prefer to do less cognitive effort unless it is necessary,” though there are certainly individual differences regarding need for cognition that influence how much information the user considers necessary. Additionally, regarding the environment, “internet users employ heuristic cues (rather than systematic) to cope with uncertainty,” which is likely prevalent at the beginning of a crisis “and [to cope with] information overload in web environments,” which evolves over time as people post news, stories, research, opinions, and reactions in the online environment (Lahuerta-Otero et al., 2018, 565). 23. Social norms may be especially powerful in a public health context, but public health officials must be aware that they may either promote the intended behavior or the discouraged behavior among the audience and audience subgroups. Prior research has shown that “risk factors--such as drug abuse, smoking, poor diet and 278 exercise--and the associated diseases are often found to be clustered in a population” (Kass-Hout & Alhinnawi, 2013, 7). This knowledge may help public health officials better understand beliefs and behaviors within the population (audience), which may lead to more targeted, effective interventions and messages directed at these subgroups. Further, social norms have been shown to translate powerfully to an online context. As Lim & Lee-Won explain, “perceptions of social norms and the influence of subjective norms on behavioral intentions (and actual behaviors) amplify when referent others are physically present. Our findings suggest that technology-evoked social presence may function in a similar manner to physical presence of referent others” (Lin & LeeWon, 2017, 431). If public health officials fail to properly recognize the power of social norms and the specific norms at play within an audience, a situation may arise in which messages intended to discourage an undesired behavior or promote a healthy behavior in fact produce the opposite effect (Griskevicius et al., 2008, 9). 24. It is necessary to shift the thinking in public health communications efforts. With the immense power social media can wield on both attitudes and behaviors, public health communications efforts must go beyond simply providing the public with facts or recommended actions backed by scientific research, as this is based on the false assumption that the members of the public have the ability, motivation, and psychological conditions to evaluate the information in a logical manner and reach the conclusions intended by public health officials. Rather, as Eyre & Littleton [powerfully] explain, “it is not a case of us ‘talking to them’ (which is 279 the basic premise of commercial advertising and political communications), but of influencing how ‘they’ perceive and talk to/with each other,” influencing thoughts and behavior on a daily basis, both within and outside the social media platform (Eyre & Littleton, 2012, 184). Messages, regardless of the intent of the source, “are not [in themselves] effects--what we communicate is not the same thing as the cognitive, emotional, behavioral, or socio-political changes we seek to achieve. Successful messaging [therefore] is not done by telling...but by moving individuals--both cognitively and emotionally--and by reframing, shifting the discourse to terms of reference that are favorable to our objectives” (Eyre & Littleton, 2012, 184). The ultimate goal is, of course, to increase voluntary compliance with public health recommendations in order to promote health across the population. The question then becomes: how does all of this relate to behavior? The Theory of Reasoned Action, first proposed by Martin Fishbein and Icek Ajzen, provides a useful framework for linking the primary ideas presented in this paper to the desired behavioral goals. 280 Figure 27. Theory of Reasoned Action. From “Predicting and Changing Behavior: A Reasoned Action Approach” by I. Ajzen & D. Albarracín, 2007, Prediction and Change of Health Behavior: Applying the Reasoned Action Approach, p. 6. The Theory of Reasoned Action, now frequently called the Theory of Planned Behavior, seeks to understand the underlying mechanisms that lead to behaviors in a different way than many other theories. Rather than “studying the role of self-esteem, prejudice, internal-external locus of control, or some other global disposition, [this theory] suggest[s] that we direct our attention to the particular behavior of interest and try to identify its determinants. Much prior theory and research had focused on one or another global disposition that might serve as an overarching causal agent and then tried to rely on this disposition to account for many different types of behavior in the disposition’s domain of application. By contrast, Fishbein and Ajzen proposed that we identify a particular behavior and then look for antecedents that can help to predict and 281 explain the behavior of interest, and thus potentially provide a basis for interventions designed to modify it” (Ajzen et al., 2007, 4). There are an infinite number of behaviors one might wish to analyze meaning that it would be impossible to examine each one individually to identify the unique set of antecedents that influence them. To address this problem, the Theory of Planned Behavior instead focuses on “a small set of causal factors that should permit explanation and prediction of most human social behaviors” (Ajzen et al., 2007, 4-5). As shown in the Figure above, these antecedents are defined broadly as follows: A central determinant of behavior is the individual’s intention to perform the behavior in question. As they formulate their intentions, people are assumed to take into account three conceptually independent types of considerations. The first are readily accessible or salient beliefs about the likely consequences of a contemplated course of action, beliefs which, in their aggregate, result in favorable or unfavorable attitude toward the behavior. A second type of consideration has to do with the perceived normative expectations of relevant referent groups or individuals. Such salient normative beliefs lead to the formation of a subjective norm--the perceived social pressure to perform or not perform the behavior. Finally, people are assumed to take into factors that may further or hinder their ability to perform the behavior, and these salient control beliefs lead to the formation of perceived behavioral control, which refers to the perceived capability of performing the behavior. As a general rule, the more favorable the attitude and subjective norm with respect to a behavior, and the greater the perceived behavioral control, the stronger should be an individual’s intention to perform the behavior under consideration. Finally, given a sufficient degree of actual control over the behavior, people are expected to carry out their intentions when the opportunity arises (Ajzen et al., 2007, 5). It is important to note that “the three major determinants in the theory of planned behavior--attitudes toward the behavior, subjective norms, and perceptions of behavioral control--are traced to corresponding sets of behavior-related beliefs. The relation between beliefs and overall evaluative attitude is embodied in...the expectancy-value model...In this theory, people’s evaluations of, or attitudes toward, an object has a certain attribute. The terms object and attribute are used in the generic sense and they refer to any 282 discriminable aspect of an individual’s world. When applied to attitudes toward a behavior, the object of interest is a particular action and the action’s anticipated outcomes. For example, a person may believe that physical exercise (the attitude object) reduces the risk of heart disease (the attribute)” (Ajzen et al., 2007, 5). Drawing further on the expectancy-value model, Ajzen et al. argue that “a person’s overall attitude toward performing a behavior is determined by the subjective values or evaluations of the outcomes associated with the behavior and by the strength of the associations. Specifically, the evaluation of each outcome contributes to the attitude in direct proportion to the person’s subjective probability that the behavior will lead to the outcome in question” (Ajzen et al., 2007, 5-6). The connection between beliefs and the associated antecedent to behavioral intention can be conceptualized in an equation. As Ajzen et al. describe, “the basic structure of the model is shown in the [following] equation…[in which] AB is the attitude toward the behavior, bi is the strength of the belief that the behavior will lead to the outcome i, ei is the evaluation of outcome i, and the sum is over all salient outcomes: AB∝ ∑biei” (Ajzen et al., 2007, 6). The same idea also applies to subjective norms and behavioral control. In the case of subjective norms “normative beliefs--in combination with the motivation to comply with the different referents...determine the prevailing subjective norm regarding behavior” while “control beliefs have to do with the perceived presence of factors that can facilitate or impede performance of a behavior” (Ajzen et al., 2007, 6). It is important to note that each of these antecedents to behavior begin with a belief, open to a high degree of subjectivity, bias, and individual difference. These beliefs may “originate in a variety of sources: personal experiences, formal education, radio, newspapers, TV, the 283 Internet and other media, interactions with family and friends, and so forth. No matter how beliefs were acquired, they are assumed to produce attitudes, subjective norms, and perceptions of control with regard to behavior, and thus guide the formation of behavioral intentions and actual performance of the behavior” (Ajzen et al., 2007, 6-7). Subjective norms were discussed in detail previously and therefore will not be discussed further here. Great attention has also been given to the element of attitude formation, particularly with reference to the Elaboration Likelihood Model proposed by Petty & Cacioppo. However, this paper traces the process of information processing as well, beginning with the initial attention given to a message, precisely because, as Ajzen et al. point out, information processing is essential to the formation of the relevant beliefs. There is nothing in this description to say that the beliefs are grounded in factual information. Indeed, Ajzen et al. explicitly make a distinction between reasoned and rational behavior: “In actuality, all the theory assumes is that behavioral intentions follow reasonably from beliefs about performing a behavior. People may hold few or many beliefs. Some beliefs persist over time, some are forgotten, and new beliefs are formed. However, there is no assumption in the reasoned action models that these beliefs are veridical. On the contrary, the theory recognizes that beliefs...may derive from invalid or selective information, be self-serving, or otherwise fail to correspond to reality. However, once a set of beliefs is formed it provides the cognitive foundation from which attitudes, perceived social norms, and perceptions of control--and ultimately intentions-are assumed to follow in a reasonable and consistent fashion” no matter how factual or erroneous the basis of those beliefs may be (Ajzen et al., 2007, 8). Thus, the discussion of information processing previously is highly relevant to understanding belief formation. 284 Concepts that are especially important when communicating messages over time (such as anchoring and mental models) as well as communicating within a public health context (such as reactance, risk perception, and unrealistic optimism) were also included in this crucial discussion of belief formation and information processing. The third antecedent to behavioral intention, perceived behavioral control, was primarily discussed in the context of the Health Belief Model. The Theory of Planned Behavior, then, provides the crucial link between the analysis of messages posted on Twitter, the psychological concepts discussed previously, and the ultimate goal of driving behavior. However, the story is also not so simple. The psychological concepts discussed previously relating to information processing, if used wisely, may be highly effective in leading the public to engage in the desired health protective behaviors. The key in the case of COVID-19 protective behaviors is maintaining that change over time. This leads to the final topic of discussion in this paper, what researchers call the intention-behavior gap. As Van Cappellen et al. explain, one of the main limitations to theories of behavior change like the Theory of Planned Behavior is “the common focus on change in behavioral intentions. The assumption is that people with the right intentions will be able to summon the willpower to enact their intended behavioral changes. People’s conscious behavioral intentions, however, often do not align with their actual behavioral engagement, a mismatch termed the intention-behavior gap. That is, many people, despite having the support, the means, and the beliefs that they can initiate positive health behaviors, [nevertheless] fail to adopt or sustain such healthy habits” (Van Cappellen et al., 2018, 86). It must be noted that this statement was made within an article that primarily focused on health behaviors like meditation and healthy eating. These are 285 behaviors that many people wish to engage in on their own accord, due to the potential for noticeable improvements in personal health, but they may fail to maintain these behaviors over time despite their intention. By nature, preventive or protective behaviors are different. If health protective behaviors are effective in preventing a health threat from affecting an individual, the individual would instead experience a lack of harm, which is unnoticeable to the individual because the harmful event would not occur. In short, an individual likely notices the benefits of some health behaviors, but cannot notice the lack of a threat occurring. Indeed, the affect individuals experience with reference to the health behavior has been found to be not only one important factor, but under some circumstances, the sole psychological predictor of behavior maintenance at follow-up for periods as long as a year (Van Cappellen et al., 2018, 78-9). “Generally speaking,” Van Cappellen et al. explain, “when people associate enjoyment with the thought of engaging in a health behavior, they are more likely both to intend to engage in that health behavior, and to actually engage in it” (Van Cappellen et al., 2018, 78). A meta-analysis of 82 studies on physical activity, for example, found “a medium to large effect size between people’s positive affective judgment about the experience of physical activity and their overall physical activity, which exceeds effect sizes for other predictors of physical activity that receive greater empirical attention” (Van Cappellen et al., 2018, 78). Studies on meditation have found similar results, including one that found “the extent of people’s early positive affective reactivity to meditation was the sole psychological predictor of whether, more than a year later, they voluntarily chose to continue meditating as a regular habit” (Van Cappellen et al., 2018, 79). This finding appears quite intuitive, as in short, “actions that are rewarding or 286 satisfying are more likely to be maintained”; however, the link between affect and behavior maintenance is more complicated (Van Cappellen et al., 2018, 79). According to the Upward Spiral Theory of Lifestyle Change, the reward system is broken into “separate ‘liking,’ ‘wanting’ and ‘learning’ systems. Specifically, the incentive salience theory of addiction holds that dopaminergic activity does not account for reward in a general sense, but rather only the subcomponent of wanting. The subcomponent of liking, by contrast, is underpinned by the brain's release of neurochemicals...Over time, associations between pleasantness (liking) and cues predictive of it endow those cues with incentive salience, making them more likely to capture attention in the future. When those cues are subsequently encountered, their heightened salience triggers dopaminergic wanting and seeking behaviors” (Van Cappellen et al., 2018, 79). Importantly, “when that pleasant affect is experienced is critical to forecasting subsequent behavioral engagement. Affective boosts experienced during [the] activity appear to be especially important...a recent systematic analysis of 24 studies concluded that pleasant affect experienced during physical activity forecasts people’s future physical activity, whereas pleasant affect experienced after physical activity does not. The predictive effects of positive affect felt during activity engagement hold even among initially sedentary adults at 6- and 12month follow-up” (Van Cappellen et al., 2018, 78). In general, “positive affect experienced during a positive health behavior creates nonconscious [motivations] for cues associated with that health behavior, [motivations] that in turn support subsequent and repeated decisions to engage in that behavior” (emphasis added) (Van Cappellen et al., 2018, 81). Van Cappellen et al. note that “people vary--from one another and over time--in 287 the extent to which they derive pleasant affect from positive health behaviors” though research has identified multiple “limiting favors, which reduce the extent to which one can derive pleasant affect from health behaviors” ranging from lack of social support to physical limitations to psychological factors (Van Cappellen et al., 2018, 83). Based on this logic, it is far less likely that individuals would wish to engage in COVID-19 protective behaviors over a long period of time. Not only are the intended results or protective action unnoticeable in that they prevent a negative outcome from occurring, there are also unlikely to be sustained external factors that generate the necessary positive affect while engaging in the behavior. Those potential factors that may generate, and in fact amplify, a positive affect are called vantage resources by the Upward Spiral Theory: “whether biological, cognitive, psychological or social...vantage resources amplif[y]...the positive affect gained from positive health behaviors and...these vantage resources are in turn further augmented by the experience of positive affect. These reciprocal effects also foster the dynamics of upward spirals” (Van Cappellen et al., 2018, 83). The discussion of information processing and the formation of behavioral intentions are proposed here as a set of guidelines to encourage initial action in members of the public--that is, a set of guidelines that are meant to enable public health officials and others in positions of authority better communicate about the sound, evidence-based practices meant to protect the health of individuals and populations, particularly during a pandemic. The addition of these important findings on habitual, subconscious motivations underlying behaviors is meant to close the intention-behavior gap to encourage the maintenance of those behaviors. 288 FINAL THOUGHTS As Schillinger et al. note, “There is little question that social media has the potential to facilitate or undermine public health efforts” (Schillinger et al., 2020, 1393). Indeed, social media’s influence in public health is vast, with “three quarters of US adults us[ing] social media; [and] of these, three quarters engage at least once daily and nearly 50% report [that] information found via social media affects the way they deal with their health” as of 2020 (Schillinger et al., 2020, 1393). Meanwhile in China, more than 740 million individuals (> 50% of the population) have social media accounts with which they daily engage, and more than 70% of WeChat’s (a Chinese messaging, social media, and mobile payment app) 570 million users report it to be their primary sources of health education” (emphasis added) (Schillinger et al., 2020, 1393). Additionally, “social media has increased the capacity of communication to influence public health” beyond the mere numbers of users: to an unprecedented degree, the popularity and technical sophistication of social media platforms have translated into health discourse becoming more ubiquitous; content becoming more creative, innovative, and engaging; production becoming both more democratized and more market sponsored; communications becoming massively scalable and rapidly spreadable by influencers and autonomous bots; artificial intelligence enabling high-volume tailoring and targeting of communications; and governments, regulatory agencies, corporate and sponsoring entities, and social media platforms themselves having the capacity to control the flow of communications (Schillinger et al., 2020, 1396). Despite this influence, “scientists’ and public health practitioners’ abilities to make sense of the myriad ways that social media can influence public health have lagged” and “no widely-accepted model exists for examining the roles that social media can play with respect to population health” (Schillinger et al., 2020, 1396, 1393). This paper echoes Schillinger et al.’s call for additional research in this vital, yet emerging, 289 communication channel. In public health, “mass communications--whether generated by the private or public sector--influence population health by shaping discourse about exposure, risk, and disease, [in turn] influencing the adoption (or nonadoption) of healthpromoting social policies, linking people to health services, and providing education and motivation that influence behaviors” (Schillinger et al., 2020, 1393). This paper attempted to outline the complexity and subjectivity of this process based on widelyaccepted psychological theories. This paper also focused on the unique conditions of a long-term, global pandemic of a novel infectious disease, as perhaps no other emergency can be of greater importance or urgency in a public health context. However, the framework presented here is informed by theory and exploratory analysis. Future research is needed to take the theories and frameworks for better public health communication presented here and test them with proper experimental controls. Additional research is also needed to integrate the knowledge presented here into public health social media and broader communication strategies--both the introduction to information processing and the conscious formation of behavioral intentions that are vital to initial behavior adoption, as well as the subconscious, habitual behaviors that enable behavior to continue for long periods of time, as is needed during a pandemic. There is a great need to understand exactly how social media functions in a public health context and the best ways to use it to promote healthy outcomes in a variety of contexts and enabling public health officials to make more informed decisions regarding their communication strategies to achieve these goals can only be a benefit, both for those officials working to promote health but also to those societies and communities they serve. 290 REFERENCES Afshar, Z., Ebrahimpour, S., Javanian, M., Koppolu, V., Vasigala, V., & Hasanpour, A. (2020, Sept.-Oct.). Coronavirus disease 2019 (COVID-19), MERS and SARS: Similarity and difference. Journal of Acute Disease, 9(5), 194-199. 10.4103/2221-6189.291283 Agrawal, A., Gindodiya, A., Deo, K., Kashikar, S., Fulzele, P., & Khatib, N. (2020). A Comparative Analysis of the Spanish Flu 1918 and COVID-19 Pandemics. The Open Public Health Journal, 14, 128-134. 10.2174/1874944502114010128 Ahmad, T., & Khan, M. (2020, Feb 27). COVID-19: Zoonotic aspects. Travel Medicine and Infectious Disease, 36, 1-2. https://doi.org/10.1016/j.tmaid.2020.101607 Ajzen, I., Albarracín, D., & Hornik, R. (Eds.). (2007). Prediction and Change of Health Behavior: Applying the Reasoned Action Approach. Lawrence Erlbaum Associates. Alberts, B., Johnson, A., & Lewis, J. (2002). Molecular Biology of the Cell, (4th ed.). Garland Science. https://www.ncbi.nlm.nih.gov/books/NBK21070/ Asimakopoulou, K., Hoorens, V., Speed, E., Coulson, N. S., Antoniszczak, D., Collyer, F., Deschrijver, E., Dubbin, L., Faulks, D., Forsyth, R., Goltsi, V., Harsløf, I., Larsen, K., Manaras, I., Olczak-Kowalczyk, D., Willis, K., Zenou, T., & Scrambler, S. (2020). Comparative optimism about infection and recovery from COVID-19; Implications for adherence with lockdown advice. Health Expectations, 23, 1502-1511. 10.1111/hex.13134 Associated Press. (2021, Jan 19). 'We know this is real': New clinics aid virus 'long-haulers'. Modern Healthcare. Retrieved Mar 21, 2021, from https://www.modernhealthcare.com/providers/we-know-this-real-new-clinics-aid-viruslong-haulers 291 Ayers, S., Baum, A., & McManus, C. (Eds.). (2007). Cambridge Handbook of Psychology, Health, and Medicine (2nd ed.). Cambridge University Press. https://searchcredoreferencecom.ezproxy.lib.utah.edu/content/title/cupphm?tab=entry_view&heading=the_health_bel ief_model1&sequence=0.pdf Azer, J., Blasco-Arcas, L., & Harrigan, P. (2021). #COVID-19: Forms and drivers of social media users' engagement behavior toward a global crisis. Journal of Business Research, 135, 99-111. https://doi.org/10.1016/j.jbusres.2021.06.030 Bar, M. (2007, June 4). The proactive brain: using analogies and associations to generate predictions. TRENDS in Cognitive Sciences, 11(7), 280-289. 10.1016/j.tics.2007.05.005A Baumeister, R. F., & Vohs, K. D. (2007). Encyclopedia of Social Psychology. SAGE Publications, Inc. Beer, T. (2020, Sept 7). ‘Reckless’ And ‘Reprehensible’ Frat Party Linked To Coronavirus Outbreak At UNH: Here Are The Latest College Coronavirus Updates. Forbes. https://www.forbes.com/sites/tommybeer/2020/09/07/reckless-and-reprehensible-fratparty-linked-to-coronavirus-outbreak-at-unh-here-are-the-latest-college-coronavirusupdates/?sh=7076713bb3a0 Berkman, N. D., Davis, T. C., & McCormack, L. (2010, September 15). Health Literacy: What Is It? Journal of Health Communication, 15(2), 9-19. https://doi.org/10.1080/10810730.2010.499985 Block, L. G., & Keller, P. A. (1995, May). When to Accentuate the Negative: The Effects of Perceived Efficacy and Message Framing on Intentions to Perform a Health-Related Behavior. Journal of Marketing Research, 32, 192-203. 292 Bogoch, I., & Detsky, A. (2020, May 16). The Term 'Asymptomatic' Is More Complicated Than It Seems When It Comes To Coronavirus Infections. Forbes. https://www.forbes.com/sites/coronavirusfrontlines/2020/05/16/the-term-asymptomaticis-more-complicated-than-it-seems-when-it-comes-to-coronavirusinfections/?sh=333da82161d9 Bond, L., & Nolan, T. (2011). Making sense of perceptions of risk of diseases and vaccinations: a qualitative study combining models of health beliefs, decision-making, and risk perception. BMC Public Health, 11. https://doi.org/10.1186/1471-2458-11-943 Brewer, N. T., Chapman, G. B., Gibbons, F. X., Gerrard, M., McCaul, K. D., & Weinstein, N. D. (2007). Meta-analysis of the Relationship Between Risk Perception and Health Behavior: The Example of Vaccination. Health Psychology, 26(2), 136-145. https://doi.org/10.1037/0278-6133.26.2.136 Broniatowski, D. A., Jamison, A. M., Qi, S., AlKulaib, L., Chen, T., Benton, A., Quinn, S. C., & Dredze, M. (2018, Oct.). Weaponized Health Communication: Twitter Bots and Russian Trolls Amplify the Vaccine Debate. AJPH, 108(10), 1378-1385. 10.2105/AJPH.2018.304567 Brooks, Z. C., & Das, S. (2020, Nov). COVID-19 Testing: Impact of Prevalence, Sensitivity, and Specificity on Patient Risk and Cost. American Journal of Clinical Pathology, 154, 5755584. 10.1093/AJCP/AQAA141 Brown, V. J. (2014, October). Risk Perception: It's Personal. Environmental Health Perspectives, 122(10), A276-A279. http://dx.doi.org/10.1289/ehp.122-A276 Brüssow, H. (2020). Immunology of COVID-19. Environmental Microbiology, 22(12), 48954908. 10.1111/1462-2920.15302 293 Buarki, H., & Alkhateeb, B. (2017). Use of hashtags to retrieve information on the web. The Electronic Library, 36(2), 286-304. 10.1108/EL-01-2017-0011 Buckton, C. H., Lean, M. E. J., & Combet, E. (2015). 'Language is the source of misunderstandings'--impact of terminology on public perceptions of health promotion messages. BMC Public Health, 15(579), 1-13. 10..1186/s12889-015-1884-1 Buder, J., Rabl, L., Feiks, M., Badermann, M., & Zurstiege, G. (2020, Dec 11). Does negatively toned language use on social media lead to attitude polarization. Computers in Human Behavior, 116, 1-8. https://doi.org/10.1016/j.chb.2020.106663 Burke, L. (2020, Aug 24). COVID-19 Roundup: Clusters, Parties, and Punishments. Inside Higher Ed. https://www.insidehighered.com/news/2020/08/24/covid-19-roundup-classesdelayed-and-students-sent-home-administrators-crack-down Callard, F., & Perego, E. (2020, October 7). How and why patients made Long Covid. Social Science & Medicine, 268, 1-5. https://doi.org/10.1016/j.socscimed.2020.113426 Celentano, D. D., & Szklo, M. (2019). Gordis Epidemiology (Sixth ed.). Elsevier. Center for Disease Control. (2020, Dec 13). Older Adults and COVID-19. CDC.gov. Retrieved Feb 14, 2021, from https://www.cdc.gov/coronavirus/2019-ncov/need-extraprecautions/older-adults.html#print Centers for Disease Control and Prevention. (2012, May 18). Section 9: Natural History and Spectrum of Disease. CDC.gov. Retrieved Feb 21, 2021, from https://www.cdc.gov/csels/dsepd/ss1978/lesson1/section9.html#:~:text=Natural%20histor y%20of%20disease%20refers,with%20AIDS%20and%20usually%20death. Centers for Disease Control and Prevention. (2020, Apr 22). How to Wear a Cloth Face Covering. CDC.gov. Retrieved June 13, 2021, from 294 https://web.archive.org/web/20200422215517/https://www.cdc.gov/coronavirus/2019ncov/prevent-getting-sick/diy-cloth-face-coverings.html Centers for Disease Control and Prevention. (2021). COVID Data Tracker. Centers for Disease Control and Prevention. Retrieved May 14, 2021, from https://covid.cdc.gov/covid-datatracker/#datatracker-home Chandra, A., & Chandra, S. B. (2020). A Comparative Analysis on SARS, MERS, and COVID19. Journal of Contemporary Medicine, 10(3), 464-470. 10.16899/jcm.769787 Charry, K., & Tessitore, T. (2021). I tweet, they follow, you eat: Number of followers as nudge on social media to eat more healthily. Social Science & Medicine, 269, 1-9. https://doi.org/10.1016/j.socscimed.2020.113595 Chen, Q., Min, C., Zhang, W., Wang, G., Ma, Z., & Evans, R. (2020, April 12). Unpacking the black box: How to promote citizen engagement through government social media during the COVID-19 crisis. Computers in Human Behavior, 110. https://doi.org/10.1016/j.chb.2020.106380 Cohen, M. S., & Corey, L. (2020, May 8). Combination prevention for COVID-19. Science, 368(6491), 551. 10.1126/science.abc5798 Craig, T. Y., & Blankenship, K. L. (2011). Language and Persuasion: Linguistic Extremity Influences Message Processing and Behavioral Intentions. Journal of Language and Social Psychology, 30(3), 290-310. 10.1177/0261927X11407167 Cranley, E. (2020, Feb 15). The US will begin testing patients with flu-like symptoms for coronavirus in an expanded effort to contain the outbreak. Business Insider. Retrieved Feb 6, 2021, from https://www.businessinsider.com/coronavirus-test-patients-flu-likesymptoms-2020-2 295 Dolinski, D., Dolinska, B., Zmaczynska-Witek, B., Banach, M., & Kulesza, W. (2020, May 13). Unrealistic Optimism in the Time of Coronavirus Pandemic: May It Help To Kill, If So-Whom: Disease or the Person? Journal of Clinical Medicine, 9. 10.3390/jcm9051464 Domingo, P., Mur, I., Pomar, V., Corominas, H., Casademont, J., & de Benito, N. (2020, July 29). The four horsemen of a viral Apocalypse: The pathogenesis of SARS-CoV-2 infection (COVID-19). EBioMedicine, 58, 1-12. https://doi.org/10.1016/j.ebiom.2020.102887 Duncombe, C. (2019). The Politics of Twitter: Emotions and the Power of Social Media. International Political Sociology, 13, 409-429. 10.1093/ips/olz013 Edwards, E. (2021, Mar 1). Inside 'post-Covid' clinics: How specialized centers are trying to treat long-haulers. NBC News. Retrieved Mar 21, 2021, from https://www.nbcnews.com/health/health-news/inside-post-covid-clinics-how-specializedcenters-are-trying-treat-n1258879 Endsley, M. R. (2018, December). Combating Information Attacks in the Age of the Internet: New Challenges for Cognitive Engineering. Human Factors: The Journal of the Human Factors and Ergonomics Society, 60(8), 1081-1094. 10.1177/0018720818807357 Erz, A., Marder, B., & Osadchaya, E. (2018). Hashtags: Motivational drivers, their use, and differences between influencers and followers. Computers in Human Behavior, 89, 4860. https://doi.org/10.1016/j.chb.2018.07.030 Eyre, D. P., & Littleton, J. R. (2012). Shaping the zeitgeist: Influencing social processes as the center of gravity for strategic communications in the twenty-first century. Public Relations Review, 38, 179-187. 10.1016/j.pubrev.2012.01.006 Fernández-de-las-Peñas, C., Palacios-Ceña, D., Gómez-Mayordomo, V., Cuadrado, M. L., & 296 Florencio, L. L. (2021, March 5). Defining Post-COVID Symptoms (Post-Acute COVID, Long COVID, Persistent Post-COVID): An Integrative Classification. International Journal of Environmental Research and Public Health, 18, 1-9. https://doi.org/10.3390/ijerph18052621 Fitzsimons, T. (2020, Nov 2). Thousands in Utah attend Halloween rave to protest coronavirus rules, sheriff says. NBC News. https://www.nbcnews.com/news/us-news/thousands-utahattended-halloween-rave-protest-coronavirus-rules-sheriff-says-n1245777 Flaxman, S., Goel, S., & Rao, J. M. (2016). Filter Bubbles, Echo Chambers, and Online News Consumption. Public Opinion Quarterly, 80(Special Issue), 298-320. 10.1093/poq/nfw006 Fung, I. C.H., Jackson, A. M., Mullican, L. A., Blankenship, E. B., Goff, M. E., Guinn, A. J., Saroha, N., & Tse, Z. T. H. (2018). Contents, Followers, and Retweets of the Centers for Disease Control and Prevention's Office of Advanced Molecular Detection (@CDC_AMD) Twitter Profile: Cross-Sectional Study. JMIR Public Health and Surveillance, 4(2). 10.2196/publichealth.8737 Gandhi, M., Yokoe, D. S., & Havlir, D. V. (2020, May 28). Asymptomatic Transmission, the Achilles' Heel of Current Strategies to Control Covid-19. The New England Journal of Medicine, 382(22), 2158-2159. George, N., Britto, D. R., Krishnan, B., dass, L. M., Prasant, H. A., & Aravindhan, V. (2018). Assessment of hashtag (#) campaigns aimed at health awareness in social media. Journal of Education and Health Promotion, 7(114). 10.4103/jehp/jehp_37_18 Godoy, M. (2020, Mar 17). How COVID-19 Kills: The New Coronavirus Disease Can Take A Deadly Turn. WBUR News. https://www.wbur.org/npr/805289669/how-covid-19-kills- 297 the-new-coronavirus-disease-can-take-a-deadly-turn Goodnough, A., & Sheikh, K. (2020, March 31). C.D.C. Weighs Advising Everyone to Wear a Mask. The New York Times. https://www.nytimes.com/2020/03/31/health/cdc-maskscoronavirus.html Griskevicius, V., Cialdini, R. B., & Goldstein, N. J. (2008). Social Norms: An Underestimated and Underemployed Lever for Managing Climate Change. International Journal for Sustainability Communication, 3, 5-13. www.ijsc-online.org Guarner, J. (2020, April). Three Emerging Coronaviruses in Two Decades. American Journal of Clinical Pathology, 153, 420-421. 10.1093/ajcp/aqaa029 Haim, M., Kümpel, A. S., & Brosius, H.-B. (2018). Popularity cues in online media: A review of conceptualizations, operationalizations, and general effects. Studies in Communication, 7(2), 186-207. https://doi.org/10.5771/2192-4007-2008-2-58 Haman, M. (2021). Twitter Followers of Canadian Political and Health Authorities during the COVID-19 Pandemic: What Are Their Activity and Interests. Canadian Journal of Political Science, 54, 134-149. 10.1017/S0008423921000020 Harris, J. M., Ciorciari, J., & Gountas, J. (2018, Jan 31). Public health social media communications and consumer neuroscience. Cogent Psychology, 5, 1-13. https://doi.org/10.1080/23311908.2018.1434058 Hayes, R. A., Carr, C. T., & Wohn, D. Y. (2016). One Click, Many Meanings: Interpreting Paralinguistic Digital Affordances in Social Media. Journal of Broadcasting and Electronic Media, 60(1), 171-187. 10.1080/08838151.2015.1127248 Heldman, A. B., Schindelar, J., & Weaver III, J. B. (2013). Social Media Engagement and Public Health Communication: Implications for Public Health Organizations Being Truly 298 "Social". Public Health Reviews, 35(1), 1-18. https://link.springer.com/content/pdf/10.1007/BF03391698.pdf Hernández, J. C., & Gorman, J. (2021, Feb 12). On W.H.O. Trip, China Refused to Hand Over Important Data. The New York Times. https://www.nytimes.com/2021/02/12/world/asia/china-world-health-organizationcoronavirus.html Hjelmgaard, K. (2021, Feb 9). WHO will end research into 'extremely unlikely' theory that COVID-19 originated in Wuhan lab. USA Today. Retrieved Feb 13, 2021, from https://www.usatoday.com/story/news/world/2021/02/09/covid-origin-world-healthorganization-end-research-wuhan-lab-theory/4446839001/ Hong, S., & Cameron, G. T. (2018). Will comments change your opinion? The persuasion effects of online comments and heuristic cues in crisis communication. Journal of Contingencies and Crisis Management, 26, 173-182. 10.1111/1468-5973.12215 Houts, P. S., Doak, C. C., Doak, L. G., & Loscalzo, M. J. (2006). The role of pictures in improving health communication: A review of research on attention, comprehension, recall, and adherence. Patient Education and Counseling, 61, 173-190. 10.1016/j.pec.2005.05.004 Hu, L., Min, Q., Han, S., & Liu, Z. (2020, June 13). Understanding followers' stickiness to digital influencers: The effect of psychological responses. International Journal of Information Management, 54, 1-14. https://doi.org.10.1016/j.ijinfomgt.2020.102169 Huang, I., & Pranata, R. (2020, May 24). Lymphopenia in severe coronavirus disease-2019 (COVID-19): systematic review and meta-analysis. Journal of Intensive Care, 8(36), 110. https://doi.org/10.1186/s40560-020-00453-4 299 InformedHealth.org. (2020, Jul 30). The innate and adaptive immune systems. Institute for Quality and Efficiency in Health Care. Retrieved Feb 25, 2021, from https://www.ncbi.nlm.nih.gov/books/NBK279396/ Jackson, S. J., & Welles, B. F. (2015). Hijacking #myNYPD: Social Media Dissent and Networked Counterpublics. Journal of Communication, 65, 932-952. 10.1111/jcom.12185 Jalali, N. Y., & Papatla, P. (2019). Composing tweets to increase retweets. International Journal of Research in Marketing, 36, 647-668. https://doi.org/10.1016/j.ijresmar.2019.05.001 Javelle, E., & Raoult, D. (2021, Apr 1). COVID-19 pandemic more than a century after the Spanish flu. The Lancet: Infectious Diseases, 21(4), e78. https://doi.org/10.1016/S14733099(20)30650-2 Jiang, J., Ren, X., & Ferrara, E. (2021). Social media polarization and echo chambers: A case study of COVID-19. https://arxiv.org/pdf/2103.10979.pdf Jingnan, H., Aubrey, A., & Wroth, C. (2020, Mar 31). Should we all be wearing masks in public? Health experts revisit the question. NPR. Retrieved Mar 27, 2021, from https://www.npr.org/sections/health-shots/2020/03/31/824560471/should-we-all-bewearing-masks-in-public-health-experts-revisit-the-question Johansson, M. A., Quandelacy, T. M., Kada, S., Prasad, P. V., Steeele, M., Brooks, J. T., Slayton, R. B., Biggerstaff, M., & Butler, J. C. (2021, January 7). SARS-CoV-2 Transmission From People Without COVID-19 Symptoms. JAMA Network Open, 1-8. 10.1001/jamanetworkopen.2020.35057 Johnston, C., & Davis, W. E. (2019). Motivating exercise through social media: Is a picture always worth a thousand words? Psychology of Sport & Exercise, 41, 119-126. 300 https://doi.org/10.1016/j.psychsport.2018.12.012 Jordan, M. (2020, Jan 30). Inside the California Military Base a Coronavirus Evacuee Tried to Flee. The New York Times. https://www.nytimes.com/2020/01/30/us/coronavirusamericans-evacuate.html Kahan, D. M., Braman, D., Cohen, G. L., Gastil, J., & Slovic, P. (2010, December). Who Fears the HPV Vaccine, Who Doesn't, and Why? An Experimental Study of the Mechanisms of Cultural Cognition. Law and Human Behavior, 34(6), 501-516. https://www.jstor.org/stable/40961856 Karmarkar, U. R., & Tormala, Z. L. (2010, April). Believe Me, I Have No Idea What I'm Talking About: The Effects of Source Certainty on Consumer Involvement and Persuasion. Journal of Consumer Research, 36(6), 1033-1049. https://www.jstor.org/stable/10.1086/648381 Kass-Hout, T. A., & Alhinnawi, H. (2013). Social media in public health. British Medical Bulletin, 108, 5-24. 10.1093/bmb/ldt028 Kaye, L. K. (2021). Exploring the "socialness" of social media. Computers in Human Behavior Reports, 3. https://doi.org/10.1016/j.chbr.2021.100083 Kim, E., Sung, Y., & Kang, H. (2014). Brand followers' retweeting behavior on Twitter: How brand relationships influence brand electronic word-of-mouth. Computers in Human Behavior, 37, 18-25. http://dx.doi.org/10.1016/j.chb.2014.04.020 Knapp, L., & Baum, N. (2015, Sept/Oct). Hashtags and How to Use Them on Social Media. The Journal of Medical Practice Management, 31(2), 131-133. Lahuerta-Otero, E., Cordero-Gutiérrez, R., & De la Prieta-Pintado, F. (2018). Retweet or like? That is the question. Online Information Review, 42(5), 562-578. 10.1108/OIR-04-2017- 301 0135 Larremore, D. B., Wilder, B., Lester, E., Shehata, S., Burke, J. M., Hay, J. A., Tambe, M., Mina, M. J., & Parker, R. (2021, Jan 1). Test sensitivity is secondary to frequency and turnaround time for COVID-19 surveillance. Science Advances, 7(1), 1-10. 10.1126/sciadv.abd5393 La Torre, F., Leonardi, L., Giardino, G., Volpi, S., Federici, S., Soresina, A., Cancrini, C., Lougaris, V., Castagnoli, R., Corrente, S., & Cardinale, F. (2020). Immunological basis of virus-host interaction in COVID-19. Pediatric Allergy Immunology, 31(Suupl. 26), 7578. 10.1111.pai.13363 Lee, B., Liu, J., Choung, H., & McLeod, D. (2019). Framing Risk with Numbers: The Framing Effects of Risk Assertions and Number Formats on Emotions and Risk Perceptions. Mass Communication and Society, 22, 344-364. https://doi.org/10.1080/15205436.2018.1536790 Lee, J. (2014). Are Some People Less Influenced by Others' Opinions? The Role of Internal Political Self-Efficacy and Need for Cognition in Impression Formation on Social Networking Sites. Cyberpsychology, Behavior, and Social Networking, 17(9), 571-577. 10.1089/cyber.2013.0713 Lee, M., Kim, H., & Kim, O. (2015). Why do people retweet a tweet?: Altruistic, Egotistic, and reciprocity motivations for retweeting. Psychologica, 58, 189-201. Lewis, D. (2020, July 8). Mounting evidence suggests coronavirus is airborne--nut health advice has not caught up. Nature. https://www.nature.com/articles/d41586-020-02058-1 Lin, Y., & Lee-Won, R. J. (2017). When retweets persuade: The persuasive effects of dialogic retweeting and the role of social presence in organizations' Twitter-based communication. 302 Telematics and Informatics, 34, 422-433. http://dx.doi.org/10.1016/j.tele.2016.09.003 Liu, K., Chen, Y., Lin, R., & Han, K. (2020, Mar 27). Clinical features of COVID-19 in elderly patients: A comparison with young and middle-aged patients. Journal of Infection, 80, e14-e18. https://doi.org/10.1016/j.jinf.2020.03.005 Lovejoy, K., Waters, R. D., & Saxton, G. D. (2012). Engaging stakeholders through Twitter: How nonprofit organizations are getting more out of 140 characters or less. Public Relations Review, 38, 313-318. 10.1016/j.pubrev.2012.01.005 Manoylov, M. (2020, Nov 6). What are cytokines? LiveScience. Retrieved Feb 27, 2021, from https://www.livescience.com/what-are-cytokines.html Manzanares-Meza, L. D., & Medina-Contreras, O. (2020). SARS-CoV-2 and influenza: a comparative overview and treatment implications. SARS-CoV-2 e influenza: revisión comparativa e implicaciones del tratamiento. Boletin medico del Hospital Infantil de Mexico, 77(5), 262–273. https://doi.org/10.24875/BMHIM.20000183 Marin, L. (2020). Three contextual dimensions of information on social media: lessons learned in the COVID-19 infodemic. Ethics and Information Technology. https://doi.org/10.1007/s10676-020-09550-2 Marshall, M. (2020, Sept 17). How COVID-19 Can Damage the Brain. Nature, 585, 342-343. Mascarenhas, L. (2021, Jan 28). 'Long Covid' still puzzles doctors but treatment is possible. CNN Health. Retrieved Mar 21, 2021, from https://www.cnn.com/2021/01/28/health/understand-treat-long-covid/index.html McEnvoy, J. (2020, Aug 13). College Shuts Down 20 Parties As Officials Say Reopening Is 'Recipe For Disaster'. Forbes. https://www.forbes.com/sites/jemimamcevoy/2020/08/13/college-shuts-down-20-parties- 303 as-officials-say-reopening-is-recipe-for-disaster/?sh=55c2c3f06de6 Meppelink, C. S., Smit, E. G., Fransen, M. L., & Diviani, N. (2019, Mar 21). "I was Right about Vaccination": Confirmation Bias and Health Literacy in Online Health Information Seeking. Journal of Health Communication, 24(2), 129-140. https://doi.org/10.1080.10810730.2019.1583791 Merriam-Webster. (n.d.). Merriam-Webster Dictionary. Merriam-Webster. https://www.merriam-webster.com/medical/lymphopenia Metzger, M. J., & Flanagin, A. J. (2013). Credibility and trust of information in online environments: The use of cognitive heuristics. Journal of Pragmatics, 59, 210-220. http://dx.doi.org/10.1016/j.pragma.2013.07.012 Miller, C. H., Lane, L. T., Deatrick, L. M., Young, A. M., & Potts, K. A. (2007). Psychological Reactance and Promotional Health Messages: The Effects of Controlling Language, Lexical Concreteness, and the Restoration of Freedom. Health Communication Research, 33, 219-240. 10.1111/j.1468-2958.2007.00297.x Mina, M. J., & Andersen, K. G. (2021, Jan 8). COVID-19 testing: One size does not fit all. Science, 371(6525), 126-127. http://science.sciencemag.org/ Moghadas, S. M., Fitzpatrick, M. C., Sah, P., Pandey, A., Shoukat, A., Singer, B. H., & Galvani, A. P. (2020, July 28). The implications of silent transmission for the control of COVID19 outbreaks. PNAS, 117(30), 17513-17515. https://doi.org/10.1073/pnas.2008373117 Mokhtari, T., Hassani, F., Ghaffari, N., Ebrahimi, B., Yarahmadi, A., & Hassanzadeh, G. (2020, October 4). COVID-19 and multiorgan failure: A narrative review on potential mechanisms. Journal of Molecular Histology, 51, 613-628. https://doi.org/10.1007/s10735-020-09915-3 304 Moyer, J. (2020, Aug 20). Penn State classes haven’t yet started. But large parties already have, raising community concern. Centre Daily. https://www.centredaily.com/news/local/education/penn-state/article245105540.html National Human Genome Research Institute. (n.d.). Lymphocyte. genome.gov. Retrieved Feb 21, 2021, from https://www.genome.gov/geneticsglossary/Lymphocyte#:~:text=Lymphocytes%20are%20cells%20that%20circulate,destro y%20invading%20viruses%20or%20bacteria. Nature Publishing Group. (2020, Dec). Meeting the challenge of long COVID. Nature, 26(12), 1803. www.nature.com/naturemedicine Newberry, C. (2021, Feb 3). 36 Twitter Stats All Marketers Need to Know in 2021. Hootsuite. Retrieved May 1, 2021, from https://blog.hootsuite.com/twitter-statistics/ New York Post. (2020, Aug 17). Georgia college kids prepare for coronavirus semester with wild party. Fox News. https://www.foxnews.com/us/georgia-college-kids-preparecoronavirus-semester-wild-party Nguyen, A., & Catalan-Matamoros, D. (2020, June 26). Digital Mis/Disinformation and Public Engagement with Health Science Controversies: Fresh Perspectives from Covid-19. Media and Communications, 8(2), 323-328. 10.17645/mac.v8i2.3352 Nichols, C. M., Bristow, N., Ewing, E. T., Gabriel, J. M., Montoya, B. C., & Outka, E. (2020). Reconsidering the 1918-1919 Influenza Pandemic in the Age of COVID-19. The Journal of the Gilded Age and Progressive Era, 19, 642-672. 10.1017/S1537781420000377 Noor, F. M., & Islam, M. M. (2020, Sept 12). Prevalence and Associated Risk Factors of Mortality Among COVID-19 Patients: A Meta-Analysis. Journal of Community Health, 45, 1270-1282. https://doi.org/10.1007/s10900-020-00920 305 Peeples, L. (2020, Oct 6). Face masks: what the data say. Nature. https://www.nature.com/articles/d41586-020-02801-8 Pennycook, G., McPhetres, J., Zhang, Y., Lu, J. G., & Rand, D. G. (2020). Fighting COVID-19 Misinformation on Social Media: Experimental Evidence for a Scalable Accuracy-Nudge Intervention. Psychological Science, 31(7), 770-780. 10.1177/0956797620939054 Petty, R. E., & Cacioppo, J. T. (1986). The Elaboration Likelihood Model of Persuasion. Advances in Experimental Social Psychology, 19, 123-205. https://www.researchgate.net/publication/270271600_The_Elaboration_Likelihood_Mod el_of_Persuasion Petty, R. E., & Cacioppo, J. T. (1996). Attitudes and Persuasion: Classic and Contemporary Approaches. Westview Press. Pfeiler, S., Massberg, S., & Engelmann, B. (2014, May 1). Biological basis and pathological relevance of microvascular thrombosis. Thrombosis Research, 133(Supplement 1), S35S37. https://doi.org/10.1016/j.thromres.2014.03.016 Platoff, E., Mulcahy, S., & Novak, A. (2020, Sept 22). As Texas college town emerge as coronavirus hot spots, universities try to keep students from infecting locals. The Texas Tribune. https://www.texastribune.org/2020/09/22/texas-coronavirus-universities/ Pollock, A. M., & Lancaster, J. (2020, Dec 21). Asymptomatic transmission of covid-19. BMJ, 371. https://doi.org/10.1136/bmj.m4851 Porter, S. C., Rheinschmidt-Same, M., & Richeson, J. A. (2015). Interring Identity From Language: Linguistic Intergroup Bias Informs Social Categorization. Psychological Science, 27(1), 94-102. 10.1177/09567976156122202 Pradhan, D., Biswasroy, P., Naik, P. K., Ghosh, G., & Rath, G. (2020, Apr). A Review of 306 Current Interventions for COVID-19 Prevention. Archives of Medical Research, 51, 363374. https://doi.org/10.1016/j.arcmed.2020.04.20 Qiu, L., Tang, Q., & Whinston, A. B. (2015). Two Formulas for Success in Social Media: Learning and Networking Effects. Journal of Management Information Systems, 32(4), 78-108. 10.1080/07421222.2015.1138368 Quintana, C. (2020, Aug 17). UNC-Chapel Hill fall semester going online amid COVID-19 outbreaks, one week into classes. USA Today. https://www.usatoday.com/story/news/education/2020/08/17/unc-chapel-hill-covid-casesfall-semester-online-classes/3383874001/ Rand, B., & Yamada, H. (2021, Mar 25). COVID-19 survivors report eye pain, adding to list of potential COVID-19 side effects. ABC News. https://abcnews.go.com/Health/covid-19survivors-report-eye-pain-adding-list/story?id=76664513 Rauschnabel, P. A., Sheldon, P., & Herzfeldt, E. (2019). What motivates users to hashtag on social media? Psychol. Mark., 36, 473-488. 10.1002/mar.21191 Reynolds, K. J., Subašić, E., & Tindall, K. (2015). The Problem of Behaviour Change: From Social Norms to an Ingroup Focus. Social and Personality Psychology Compass, 9(1), 45-56. 10.1111/spc3.12155 Richards, A. S., & Banas, J. A. (2015). Inoculating Against Reactance to Persuasive Health Messages. Health Communication, 30, 451-460. 10.1080/10410236.2013.867005 Rim, H., & Song, D. (2016). "How Negative Becomes Less Negative": Understanding the Effects of Comment Valence and Response Sidedness in Social Media. Journal of Communication, 66, 475-495. 10.1111.jcom.12205 Rittenberg, A. (2020, July 29). Report: Rutgers football COVID-19 outbreak linked to on- 307 campus party. ESPN. https://www.espn.com/collegefootball/story/_/id/29559800/rutgers-football-covid-19-outbreak-linked-campus-party Rook, L. (2013). Mental models: a robust definition. The Learning Organization, 20(1), 38-47. 10.1108/09696471211288519 Rudd, R. E. (2007, Sep/Oct). Health Literacy Skills of U.S. Adults. American Journal of Health Behavior, 31, 8-18. Rudd, R. E. (2010, Dec. 9). Improving Americans' Health Literacy. The New England Journal of Medicine, 363(24), 2283-2285. Rypdal, K., Bianchi, F. M., & Rypdal, M. (2020, Dec 21). Intervention Fatigue is the Primary Cause of Strong Secondary Waves in the COVID-19 Pandemic. International Journal of Environmental Research and Public Health, 17, 1-17. 10.3390/ijerph17249592 Salter, S. (2020, Sept 24). Reinventing Cloth Masks in the Face of Pandemics. Risk Analysis, 114. https://doi.org/10.1111/risa.13602 Schillinger, D., Chittamuru, D., & Ramirez, A. S. (2020, September). From "Infodemics" to Health Promotion: A Novel Framework for the Role of Social Media in Public Health. APJH Methods, 110(9), 1393-1396. 10.2105/APJH.2020.305746 Schumaker, E. (2020, Feb 4). Here are the coronavirus symptoms to watch out for amid outbreak. ABC News. Scribner, H. (2021, Feb 10). COVID-19 likely came from an 'intermediary host species,' WHO says. Deseret News. Shah, P. (2020, Jun 27). Sentiment Analysis using TextBlob. Toward Data Science. Retrieved May 24, 2021, from https://towardsdatascience.com/my-absolute-go-to-for-sentimentanalysis-textblob- 308 3ac3a11d524#:~:text=TextBlob%20is%20a%20python%20library%20for%20Natural%2 0Language%20Processing%20(NLP).&text=Polarity%20lies%20between%20%5B%2D 1,help%20with%20fine%2Dgrained Shahi, G. K., Dirkson, A., & Majchrzak, T. A. (2020, Aug 24). An Exploratory Study of COVID-19 Misinformation on Twitter. https://arxiv.org/abs/2005.05710 Shepperd, J. A., Waters, E. A., Weinstein, N. D., & Klein, W. M. P. (2015). A Primer on Unrealistic Optimism. Current Discussions in Psychological Science, 24(3), 232-237. 10.1177/0963721414568341 Shi, J., Lai, K. K., Hu, P., & Chen, G. (2017). Understanding and predicting individual retweeting behavior: Receiver perspectives. Applied Soft Computing, 60, 844-857. http://dx.doi.org/10.1016/j.asoc.2017.08.044 Smith, C. T., De Houwer, J., & Nosek, B. A. (2012). Consider the Source: Persuasion of Implicit Evaluations Is Moderated by Source Credibility. Personality and Social Psychology Bulletin, 39(2), 193-205. 10.1177/0146167212472374 Snape, M. D., & Viner, R. M. (2020, Oct 16). COVID-19 in children and young people. Science, 370(6514), 286-287. Snyder, L. B. (2007, Mar-Apr). Health Communication Campaigns and Their Impact on Behavior. Journal of Nutrition Education and Behavior, 39(2 Suppl), S32-40. 10.1016/j.jneb.2006.09.004 Song, J.-W., Zhang, C., Fan, X., Meng, F.-P., Xia, P., Cao, W.-J., Yang, T., Dai, X.-P., Wang, S.-Y., Xu, R.-N., Jiang, T.-J., Li, W.-G., Zhang, D.-W., Zhao, P., Shi, M., Agrati, C., Ippolito, G., Maeurer, M., Zumla, A., … Zhang, J.-Y. (2020, July 8). Immunological and inflammatory profiles in mild and severe cases of COVID-19. Nature Communications, 309 11, 1-10. https://doi.org/10.1038/s41467-020-17240-2 Southwell, B. G., Kelly, B. J., Bann, C. M., Squiers, L. B., Ray, S. E., & McCormack, L. A. (2020). Mental Models for Infectious Diseases and Public Understanding of COVID-19 Prevention. Health Communication, 35(14), 1707-1710. https://doi.org/10.1080/10410236.2020.1837462 Stieg, C. (2020, Feb 6). Cruises quarantined, flights stopped: Why risk of Americans getting coronavirus at airports, cruises, theme parks is still low. CNBC. https://www.cnbc.com/2020/01/31/travel-safety-tips-for-coronavirus-cruises-themeparks-airports.html Subbarao, K., & Mahanty, S. (2020, June 16). Respiratory Virus Infections: Understanding COVID-19. Immunity, 52, 905-909. https://doi.org/10.1016/j.immuni.2020.05.004 Sutton, J. (2018, October 1). Health Communication Trolls and Bots Versus Public Health Agencies’ Trusted Voices. American Journal of Public Health, 108(10), 1281-1282. https://doi.org/10.2105/AJPH.2018.304661 Szkaradkiewicz-Karpińska, A. K., & Szkaradkiewicz, A. (2021). Towards a more effective strategy for COVID-19 prevention (Review). Experimental and Therapeutic Medicine, 21(33). 10.3892/etm.2020.9465 Tang, J. W., Tambyah, P. A., & Hui, D. S. C. (2020, January). Emergence of a novel coronavirus causing respiratory illness from Wuhan, China. Journal of Infection, 80, 350-352. https://doi.org/10.1016/j.jinf.2020.01.014 Thackeray, R., Neiger, B. L., Smith, A. K., & Van Wagenen, S. B. (2012). Adoption and use of social media among public health departments. BMC Public Health, 12, 1-6. http://www.biomedcentral.com/1471-2458/12/242 310 Thornton, C. (2021, Feb 9). WHO says an "intermediary host species" is most likely how Covid19 was introduced to humans. CNN World. Retrieved Feb 13, 2021, from https://www.cnn.com/world/live-news/coronavirus-pandemic-vaccine-updates-02-1021/h_5491b041b87a80d6cfb16f6d12c39d8c Tong, K. K., Chen, J. H., Yu, E. W., & Wu, A. M.S. (2020). Adherence to COVID-19 Precautionary Measures: Applying the Health Belief Model and Generalised Social Beliefs to a Probability Community Sample. Applied Psychology: Health and WellBeing, 12(4), 1205-1223. 10.1111/aphw.12230 Tormala, Z. L., Briñol, P., & Petty, R. E. (2006). When credibility attacks: The reverse impact of source credibility on persuasion. Journal of Experimental Psychology, 42, 684-691. 10.1016/j.jesp.2005.10.005 United Nations Department of Global Communications. (2020, March 31). UN tackles 'infodemic' of misinformation and cybercrime in COVID-19 crisis. UN.org. Retrieved Jan 9, 2021, from https://www.un.org/en/un-coronavirus-communications-team/un-tackling%E2%80%98infodemic%E2%80%99-misinformation-and-cybercrime-covid-19 United Nations Environment Programme. (2020). Preventing the Next Pandemic: Zoonotic diseases and how to break the chain of transmission [Report]. In Frontiers Report Series. United Nations. Van Cappellen, P., Rice, E. L., Catalino, L. I., & Fredrickson, B. L. (2018). Positive affective process underlie positive health behavior change. Psychology & Health, 33(1), 77-97. https://doi.org/10.1080/08870446.2017.1320798 Van der Meer, T. G. L. A., Hameleers, M., & Kroon, A. C. (2020). Crafting Our Own Biased Media Diets: The Effects of Confirmation, Source, and Negativity Bias on Selective 311 Attendance to Online News. Mass Communication and Society, 23(6), 937-967. https://doi.org/10.1080/15205436.2020.1782432 Waldrop, T. (2021, Feb 22). Clinics are spring up around the country for what some call a potential second pandemic: Long Covid. CNN Health. Retrieved Mar 21, 2021, from https://www.cnn.com/2021/02/22/health/long-covid-clinics/index.html Wallsten, K., & Tarsi, M. (2015). Persuasion from Below? An experimental assessment of the impact of anonymous comments sections. Journalism Practice, 10(8), 1019-1040. http://dx.doi.org/10.1080/1512786.2015.1102607 Wang, R., Zhang, X., Irwin, D. M., & Shen, Y. (2020, January). Emergence of SARS-like coronavirus poses new challenge in China. Journal of Infection, 80, 354-356. https://doi.org/10.1016/j.jinf.2020.01.017 Watson, J., Whiting, P. F., & Brush, J. E. (2020, May 12). Interpreting a covid-19 test result. BMJ, 369, 1-7. 10.1136/bmj.m1808 Weise, E., & Weintraub, K. (2021, Jan 16). Where did COVID-19 come from? USA Today. World Health Organization. (2020, Jan 5). Pneumonia of unknown cause -- China. World Health Organization. Retrieved Feb 13, 2021, from https://www.who.int/csr/don/05-january2020-pneumonia-of-unkown-cause-china/en/ World Health Organization. (2021). WHO Coronavirus (COVID-19) Dashboard. World Health Organization. Retrieved May 14, 2021, from https://covid19.who.int/ Yeager, D. S., Dahl, R. E., & Dweck, C. S. (2018). Why Interventions to Influence Adolescent Behavior Often Fail but Could Succeed. Perspectives on Psychological Science, 13(1), 101-122. 10.1177/17456991917722620 Yong, E. (2020, Dec 14). How Science Beat the Virus. The Atlantic. 312 https://www.theatlantic.com/magazine/archive/2021/01/science-covid-19-manhattanproject/617262 Zheng, Z., Peng, F., Xu, B., Zhao, J., Liu, H., Peng, J., Li, Q., Jiang, C., Zhou, Y., Liu, S., Ye, C., Zhang, P., Xing, Y., Guo, H., & Tang, W. (2020, April 23). Risk factors of critical & mortal COVID-19 cases: A systematic literature review and meta-analysis. Journal of Infection, 81, e16-e25. https://doi.org/10.1016/j.jinf.2020.04.021 Zhu, N., Zhang, D., Wang, W., Li, Z., Yang, B., Song, J., Zhao, Z., Huang, B., Shi, W., Lu, R., Niu, P., Zhan, F., Ma, Z., Wang, D., Xu, W., Wu, G., Gao, G. F., & Tan, W. (2020, February 20). A Novel Coronavirus from Patients with Pneumonia in China, 2019. The New England Journal of Medicine, 382(8), 727-733. Ziegler, R., Diehl, M., & Ruther, A. (2002, Apr). Multiple Source Characteristics and Persuasion: Source Inconsistency as a Determinant of Message Scrutiny. PSPB, 28(4), 496-508. 313 Appendix A CDC Instructions for cloth face masks as of April 22, 2020 314 315 316 Appendix B Analysis of Utah Coronavirus Info Twitter Account Likes & Retweets 317 318 Appendix C Use of COVID-19 pandemic-related terms for each account 319 320 321 Name of Candidate: Rachel Nelson Date of Submission: February 28, 2022