Disentangling diffusion: an analysis of municipal tax rate patterns

The geographic diffusion of policy is a well-documented phenomenon, but the mechanisms underlying diffusion are more obscure. This study describes and explains municipal property and sales tax rates. It examines the influence of diffusion in this rate-setting process. Existing literature describes two such mechanisms driving such diffusion: learning and competition, but leaves the question of the relative influence of these mechanisms in significant doubt. An examination of municipal tax rates, financial and demographic data shows that, when setting their own sales and property tax rates, local governments weigh the rates of their neighbors more heavily than other factors. Evidence implies a stronger role for learning and less robust role for tax competition as explanations for municipal tax rate diffusion. Budgetary demands, as well as state-mandated formal rules, also influence local government rate-setting behavior.

DISENTANGLING DIFFUSION: AN ANALYSIS OF MUNICIPAL TAX RATE PATTERNS by Christopher F. Krueger A dissertation submitted to the faculty of The University of Utah in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Political Science The University of Utah May 2016 Copyright © Christopher F. Krueger 2016 All Rights Reserved The Univ e r s i ty of Utah Graduat e School STATEMENT OF DISSERTATION APPROVAL The dissertation of Christopher F. Krueger has been approved by the following supervisory committee members: James J. Gosling Chair August 3, 2015 Date Approved Daniel Levin Member August 3, 2015 Date Approved Matthew J Burbank Member August 3, 2015 Date Approved Brent J Steele Member August 3, 2015 Date Approved Andre H Baksh Member August 3, 2015 Date Approved and by Mark Button Chair/Dean of the Department/College/School of Political Science and by David B. Kieda, Dean of The Graduate School. ABSTRACT The geographic diffusion of policy is a well-documented phenomenon, but the mechanisms underlying diffusion are more obscure. This study describes and explains municipal property and sales tax rates. It examines the influence of diffusion in this rate-setting process. Existing literature describes two such mechanisms driving such diffusion: learning and competition, but leaves the question of the relative influence of these mechanisms in significant doubt. An examination of municipal tax rates, financial and demographic data shows that, when setting their own sales and property tax rates, local governments weigh the rates of their neighbors more heavily than other factors. Evidence implies a stronger role for learning and less robust role for tax competition as explanations for municipal tax rate diffusion. Budgetary demands, as well as state-mandated formal rules, also influence local government rate-setting behavior. TABLE OF CONTENTS ABSTRACT........................................................................................................ iii CHAPTERS 1: INTRODUCTION ............................................................................................. 1 A brief overview of the learning-competition question................................. 2 The goals and findings of this paper...........................................................4 Where this study fits in the literature.........................................................6 2: LITERATURE REVIEW...................................................................................... 8 The municipal revenue landscape............................................................. 8 Lacking parsimony: Same findings, multiple explanations........................ 14 Endogenous theories.............................................................................. 15 Diffusion-through-learning..................................................................... 22 Diffusion-through-competition................................................................. 32 Learning or competition: Recent literature............................................... 47 Summary................................................................................................ 61 3: METHODS.................................................................................................... 64 The nature of spatial analysis.................................................................. 64 Variables............................................................................................... 67 Primary hypotheses............................................................................... 68 Secondary hypotheses.......................................................................... 72 Additional tests...................................................................................... 78 4: RESULTS....................................................................................................... 83 The state tables.................................................................................... 83 A summary of the evidence for diffusion................................................. 87 Competition-learning tests (H1-H3)........................................................ 89 Secondary tests......................................................................................91 Inductive test results............................................................................. 99 Summary............................................................................................. 115 5: CONCLUSION............................................................................................ 116 Diffusion, again......................................................................................116 The competition-learning question..........................................................120 A more complex picture..........................................................................130 Advice to city managers.........................................................................147 Summary............................................................................................. 152 APPENDICES A: STATE CORRELATION TABLES..................................................................... 155 B: RATE HISTOGRAMS................................................................................... 177 C: ADDITIONAL DATA TABLES........................................................................ 199 D: RATE-TO-REVENUE SCATTERPLOTS............................................................. 200 E: VERTICAL CONTROLS SURVEYS.................................................................. 212 F: DATA SET DETAILS..................................................................................... 216 REFERENCES.................................................................................................. 219 v CHAPTER 1 INTRODUCTION This paper explains the mechanism(s) behind the adoption of municipal local option sales tax (LOST) and municipal property tax rates. To a lesser extent, it predicts those rates. This study examines the influence of horizontal interactions among governments in a regional network as those municipalities choose their rates. Significant scholarship has been dedicated to understanding this horizontal behavior. Although endogenous variables (e.g. residents' preferences) affect political entities' policy choices, scholarship identifies the crucial exogenous influence of peers within these horizontal networks. Governments copy their neighbors. Although several theories have been offered by scholars to explain this diffusion, two have emerged as the most salient: diffusion-through-learning and diffusion-through-competition. These somewhat competing/somewhat complementary theories have been aptly compared in several recent works.1 This dissertation furthers the study and differentiation of these two mechanisms. The learning model conceptualizes cities' behavior within peer-to-peer networks as a positive-sum game. One city emulates another's success in an effort to accrue some of the benefits of that success. For example, Dallas, Texas, copies its successful neighbor, Fort Worth, Texas, because Dallas expects to accrue some of the benefits enjoyed by Fort Worth without any necessary detriment to Fort Worth. 1 See Baybeck, Berry and Siegel 2011; Boehmke and Witmer 2004; Burge and Piper 2012; da Silvia Costa and Carvalho 2013; Shipan and Volden 2008; Ting and Carpenter 2008; Volden, Brueckner, and Saavedra 2001. 2 In contrast, the competition model describes cities behaving as though they were participating in a zero-sum game. Governments copy each other in an effort to "steal" benefits away from their neighbors for themselves. In this case, Dallas is motivated to seize benefits from Fort Worth by copying Fort Worth's behavior. Dallas benefits at Fort Worth's expense. This study examines municipal sales and property tax rates to test these competing explanations. It finds robust support for diffusion in general and fairly strong support for the learning mechanism. Some support for the competition is also indicated. A brief overview of the competition-or-learning question At its core, the diffusion-through-learning mechanism ("learning mechanism") conceptualizes cities as agents that copy neighbors' actions in a positive-sum setting. Jack Walker's (1969) early work on the learning mechanism has since been further developed by many scholars, notably by William and Frances Stokes Berry (2007), in more recent scholarship. According to this model, actions by one city do not necessarily take benefits from another. In this model, such policy mimicry may deliver increased benefits to residents, increasing the approval of city leaders by its residents and-perhaps-increase the long-term economic and cultural health of the city (Meseguer 2005; Shipan and Volden 2012; Volden 2005;). According to this model, as governments copy their neighbors' behavior, geographic proximity increases the likelihood neighboring jurisdictions will adopt similar policies. This diffusion-through-learning has been variously identified as "mimicking," "copycat behavior," "yardstick competition," and "policy diffusion" in the political science literature (for additional discussion see Maggetti and Gilardi 2013).2 Furthermore, a jurisdiction will have more influence on the policies of its immediate neighbors than on jurisdictions far away. The policies of Fort Worth, Texas have significant influence on the policies of Arlington (a contiguous neighbor), less influence on Dallas (30 miles), and significantly less influence on El Paso (about 600 miles). The learning model assumes that near neighbors' policies are more visible than distant neighbors' policies. This study, which considers tax rates as its focus, assumes that Fort Worth's tax rates are more visible to the leaders and residents of Dallas than they are to El Paso. Evidence presented by this study strongly supports this assumption. The diffusion-through-competition model ("competition mechanism") grew from a wealth of research from the mid-twentieth century. Charles Tiebout's "A Pure Theory of Local Expenditures" (1956) suggests that governments' behavior can be likened to that of firms, while residents and resident businesses of jurisdictions can be compared to consumers. Like consumers purchasing higher quality goods at lower prices, Tiebout contends residents and businesses "shop around" for places to live and work, offering the best services at the comparably lowest tax rates. Dozens of studies have confirmed this "shopping with your feet" phenomenon (for reviews, see Dowding, John and Biggs 1995; Genschel and Schwarz 2011). Inextricably linked to this competition for residents and businesses is the concept of tax competition (Hendrick, Wu and Jacob 2007; Wildasin 1988; Wilson 1999). If the sales tax rates of Dallas are substantially higher than those in nearby Fort Worth, Fort Worth might see increased sales as residents of Dallas flock to Fort Worth to make those purchases. Given the right set of rates and sales, Fort Worth might paradoxically generate more sales tax revenue than Dallas even though it has Each of these "sub-types" has adherents differentiating their theories from Berry and Berry's (2007). This dissertation will make considerable effort to disentangle some of these terms and their corresponding theories in chapter two. But in short, these four terms are (at the least) in the same general theoretical family, even if they are just shy of synonymous. 3 4 lower tax rates. The primary tests of this dissertation rest on the assumption that if competition is the primary force driving policy diffusion, municipal tax rates will show evidence of such competition. As a result, competing cities will attempt to poach revenue from their neighbors. Because cities compete more intensely with their near neighbors than their distant neighbors, cities should be more likely to poach revenue from their near neighbors than their distant ones. But the evidence generated and considered by this study concludes that revenue poaching is weak; diffusion seems to be primarily driven by the learning mechanism. The goals and findings of this paper The first question addressed by this dissertation is the degree to which cities copy each other across geographic space. Existing literature has examined this question so extensively as to expect this finding as a matter of course. Other, intrinsic explanations offer alternatives to the diffusion-through-learning and diffusion-through-competition models. Behavioral models, tax availability models, and budget constraints are among the most visible of these competitors. Extrinsic constraints imposed by superordinate governments may also play a role. But given the strong evidence from other studies describing the spread of policy throughout a geographic region, these other factors should play a less significant role. Diffusion should play the most significant role in determining tax rates. The evidence gathered in this study confirms this assumption. The second question, contingent on the first, is whether diffusion can be best explained by the learning or competition mechanisms. This paper examines the extent to which cities copy their neighbors' sales tax rates, then compares that to the extent to which cities copy their neighbors' property tax rates. Because the competition model predicts sales tax rates will be more responsive to neighboring rates, stronger evidence of sales tax rate diffusion will imply a more powerful role for 5 the competition mechanism. The evidence gathered by this study shows cities copy both tax rates with near-equal frequency, indicating learning is more likely to be the mechanism driving diffusion. The third question is whether cities act to maximize revenue. If governments act as revenue maximizers, this would have profound implications for the nature of government. The predominant view in political science is that governments are not revenue maximizers (Zodrow and Mieszkowski 1986) or, at a minimum, do not prioritize revenue maximization. The evidence gathered in this study provides strong evidence confirming this predominant view. The fourth question addressed by this study is whether cities, and by extension all governments, act rationally. This question has not been firmly settled in the literature (Dickson 2014; Downs 1957; Ostrom 1999). This essay examines evidence that cities could make revenue choices with little-to-no cost and substantial payoff. Do cities make these rational choices? The relevant evidence gathered and scrutinized by this investigation is less conclusive than others considered in this paper. Nevertheless, it appears as though governments act with little evidence of rational behavior, at least in the short term. The fifth question of this paper is one of the vertical interactions of governments, specifically the effect of state government rules on municipal behavior. Do superordinate rules influence subordinate behavior within peer-to-peer networks? The existing literature on this point is less robust and conclusive (McKinnon and Nechyba 1997; Tannenwald 1991). With less existing scholarship on this subject, especially on state laws and their effect on local policy diffusion, conclusions are more difficult to predict. The evidence generated by this study suggests that state rules do affect the patterns of diffusion throughout networks, albeit weakly. Finally, from a practical point of view, this dissertation will serve as a "political 6 survival guide" for city administrators throughout the United States (and beyond) when they are faced with the prospect of implementing new taxes or raising existing rates. As of the writing of this paper, such information is largely unavailable or has been obfuscated in the current literature (Krane, Ebdon, and Bartle 2004). It will help city administrators and council members determine effective, competitive, and palatable rates. It will provide particular assistance to these leaders as they weigh the complex array of variables-most importantly the behavior of their neighbors-in raising rates or implementing new ones. Further, by determining which formal and informal state rules influence local rates more robustly, this paper offers some guidance to local government lobbyists/leagues as they advise their respective state legislatures to change sales and property tax laws. In short, this study suggests cities should largely conform to their neighbors' rates, and should lobby their state legislatures for more discretion. Where this study fits in the literature At least 800 articles have been published in the last 50 years on the subject of policy diffusion (Graham, Shipan and Volden 2013). Some basic questions, including the fundamental competition-or-learning question investigated by this dissertation, remain unanswered and disputed (Shipan and Volden 2012). This dissertation will join the hundreds of others searching for evidence of diffusion. It also finds evidence of diffusion as the product of learning rather than of competition, but such a brief summary unjustly simplifies the significant evidence gathered and analyzed by this study. There is some evidence of diffusion-through-competition, especially in a long-run scenario, and especially when contextual evidence is considered. Methodologically, this paper offers an unusual tool to inquire into the topic of diffusion. Existing literature is profligate with studies of nominal variables and 7 interstate data. Both the diffusion-through-learning and the diffusion-through-competition literature have made repeated calls for research using continuous3 variables and on levels other than the state-to-state level (e.g., Berry and Baybeck 2005; Genchel and Schwarz 2011; Brueckner 2003). This study does both. In addition, the large size and numerical data offers additional descriptive and explanatory power over the categorical, smaller data sets used in previous studies. Finally, this investigation offers a vehicle for future study of these mechanisms by adding a new4 tool to the relatively new sub-field of GIS-based statistical analysis (Berry and Baybeck 2005), should future scholars wish to emulate and/or refine its methods. This dissertation consists of six chapters: This introduction is followed by Chapter 2: an overview of competing theoretical frameworks. A brief review of the existing municipal revenue landscape will be followed by a survey of alternate theories to explain and describe municipal rate-setting behavior. The third and fourth sections of Chapter 2 examine the primary theories driving tax rates under scrutiny: diffusion-through-competition and diffusion-through-learning. Chapter 3 is an overview of the methodology and data undertaken in this investigation; it also lays out the crucial hypotheses to be tested and lists the data gathered. Chapter 4 will present the results. Chapter 5 will conclude this dissertation, discussing the implications of these findings on the questions posited above and summarizing the findings. 3 Continuous variables are those in which the values to be analyzed can take any value between the minimum and maximum. 4There have been several other studies that have used GIS (e.g., Berry and Baybeck 2005), but not in the specific way this study does. CHAPTER 2 LITERATURE REVIEW The municipal revenue landscape Property taxes are older than the United States. Until the 1930s, property taxes were by far the largest single source of revenue for local governments (Fisher 1996). Declining property values and inability of homeowners to pay during the Great Depression led to a decreased reliance on the property tax. In the 1970s, a wave of reforms further reduced the salience of the property tax. These laws were made by state legislatures, referenda, and/or initiatives. California's Proposition 13 (1978) severely curtailed local government revenue by freezing tax rates and property values. A parallel wave of "Tax Expenditure Limits" passed several other states through popular or legislative means (ACIR 1995). Several studies have described these vertical constraints within the context of property taxes. Such rules have been shown to impact the type and rates of these local taxes, both through their top-down restrictions as well as through their affects on the relationship of subordinate units within the peer-to-peer network (Bartle 2003; Bowler and Donovan 2004; Burge and Rogers 2011; Chicoine and Walzer 1986; Dye and McGuire 1997; Henderson 1994; Johnston, Pagano and Russo 2000). This study finds little evidence that such vertical rules significantly affect property tax diffusion patterns. Local option sales taxes are a comparatively newer phenomenon. New York City was the first city to adopt the Local Option Sales Tax in 1934, in part as a reaction to declining property tax revenue during the Great Depression (ACIR 1989). 9 Many states now allow local governments to set their own sales tax rates, but these authorizations vary significantly from state to state. Some states allow sales taxes to be levied only at the county level (e.g. South Carolina, Ohio, and Nevada). Counties in these states do sometimes have to share revenue with cities within their borders. Some states mandate a set municipal sales tax rate for every city within the state (e.g. New Jersey and Virginia), creating a "LOST" without the "optional" component of the term. This study will be more limited in scope, focusing on the 22 states that allow at least some of their cities at least some discretion in setting general sales tax rates.5 As of 2008, these states were: Alabama, Alaska, Arizona, Arkansas, California, Colorado, Idaho, Illinois, Iowa, Kansas, Louisiana, Minnesota, Missouri, New Mexico, North Dakota, Nebraska, Oklahoma, South Dakota, Tennessee, Texas, Utah, and Washington. Although many states have imposed complex rules regarding the timing, rates, and means through which property is assessed and taxed, states' rules regarding sales taxes are at least as complex. These rules include rate ceilings, rate increase maximums, rate minimums, referendum requirements, revenue sharing, and sunset laws on rates. As with property tax, several studies suggest those limits will almost certainly affect the diffusion pattern of rates throughout these various states (Burge and Rogers 2011; Cornia, Grimshaw, Lewis and Barbour 1999; Luna, Bruce and Hawkins 2007; Nelson and Walker 2010; Zhao 2005). However, the literature on vertical constraints and their effects on LOST rates is not as robust as that covering property tax. Even these studies discuss the role of vertical restrictions as an afterthought-perhaps with the exception of Burge and Rogers 5 Hawaii and Vermont are not included in this study. As of 2008 each of these states had exactly one city with a LOST, while the rest of the cities in the state had none. Inclusion of these states would produce no meaningful data. Finally, states with county-only data are not included. First, because there is enough data with the city-only LOST states to produce meaningful results and second, because in many states the counties are so large (especially Western States like Nevada) as to make meaningful tax competition through cross-border shopping negligible. 10 (2011). This study will narrow this gap in the literature, demonstrating that vertical rules have little influence on diffusion patterns. Revenue Sources A review of financial trends provides a context for understanding the role of property and sales taxes in the greater municipal revenue landscape. Cities have several potential sources of revenue: sales taxes, property taxes, income tax, franchise taxes, user fees and intergovernmental transfers (e.g., federal, state, and local grants and payments). The basic trends for these different revenue lines are illustrated in Figure 2.1, which demonstrates the decline in intergovernmental transfers during recent years. Figure 2.1 shows the corresponding increase in local governments' taxes and fees (fees are included in "other revenue."). These trends are confirmed in articles from Bartle (2003), Bartle, Ebdon and Krane (2003) and Johnston, Pagano and Russo (2000). 100% fll 90% Intergovernmental 3 Transfers c 80% <D i Other Revenue <D 70% tc +J C 60% i Other Tax <D E 50% c Property Tax <D > 40% O u 30% i General Sales Tax ro uA u 20% _l ■ Income Tax 4-1 c 10% <D U 0% <D 1952 1968 1978 1992 2002 2008 Year Figure 2.1: Local government by source, 1952-2008. From the U.S. Census Bureau's Statistical Abstracts of the United States. This figure represents all revenue sources for all states. 11 Looking more closely at the cities under examination in this paper, Figure 2.2 depicts the LOST-only revenue landscape from 2011. This figure includes the Census Bureau's results from the 2011 Survey of Local Government Finances and tallies the revenue amounts by category, but only for the 22 under the consideration of this paper. However, the Survey data in Figure 2.2 indicates a much larger portion of revenue coming from fees. Despite their differences, three trends are evident. First, cities have significantly increased dependence on fee-based revenue. Second, intergovernmental transfers constitute a smaller portion of the budget compared to 1970. Third, sales tax revenue and property tax are both important sources of revenue. ■ Intergovernmental Transfers ■ Other Tax ■ Fees ■ General Sales Tax ■ Property Tax Figure 2.2: LOST-state local government revenue by source. This figure reflects the same data as Figure 2.1, but only for the study year 2011. The data is also limited to the 22 states studied in this dissertation. From Census Bureau's Annual Survey of State and Local Government Finances, 2011. 12 Again, these trends have been confirmed in several other studies, notably Bartle (2003), Bartle, Ebdon and Krane (2003) and Johnston, Pagano and Russo (2000). This paper focuses on tax rates and their corresponding revenues. Figures 2.3 and 2.4 clarify those specific sources of funding. Figure 2.3, like Figure 2.1, illustrates trends over 1952-2008 and again includes data from county and special districts as well as cities. But unlike Figure 2.1, only the tax revenue is included in Figure 2.3. This makes the relationship between the various sources of taxes more clear, specifically local government's dependence on sales tax and property tax revenue, among other sources of taxes. 100% <u 3 C <D ><D tcX (0 H4 -1> o u rouo c<D U <D Q. i Other Tax Property Tax ■ General Sales Tax Income Tax 0% 1952 1968 1978 1992 2002 2008 Year Figure 2.3: Local government tax revenue by source, selected years 1952-2008. This is the same data as in Figure 2.1, with only the tax revenue included. The figure also includes only the data from the 22 study states of this investigation. Compiled from various Statistical Abstracts of the United States. 13 Other Tax ■ General Sales Tax ■ Property Tax Figure 2.4: Municipal government revenue by tax, LOST states only, 2011 is a graphical depiction of the same data as in Figure 2.3, but with only tax revenue included. From U.S. Census Bureau's Annual Survey of State and Local Government Finances, 2011. Figure 2.4 further narrows the scope of the data. Like Figure 2.2, it includes only the tax data from the cities in the 22 states under the microscope of this paper. Other forms of taxation (i.e., franchise tax, excise tax) remain fairly steady throughout the period, while the local government income tax grew quickly at first but then leveled out in the 1970s (see also Bartle 2003). Only the sales tax has grown significantly and fairly steadily to "make up" for a decline in property tax revenue. This chart demonstrates the significant share of municipal revenue commanded by sales and property tax. The importance of these taxes makes this topic worthy of investigation, as city managers make decisions about which taxes to raise, and to what extent to raise them. 14 Lacking parsimony: Same findings, multiple explanations Complex political behavior begets complex explanations. As with any complex behavior, multiple forces are at work in the setting of municipal tax rates. There are three leading endogenous approaches to the study of municipal taxation competing with the "mechanisms of diffusion" theories. These will be surveyed before proceeding to a more detailed review of the competition and learning models. This study finds evidence of many mechanisms at work. First, the behavior-driven model will be described, in which constituent preferences dictate the behavior (in this case, tax types and rates) of officials and therefore of cities. Second, the paper will turn to the tax availability model, which emphasizes the pragmatic nature of taxation; governments tax that which they can most easily and efficiently tax. Third, the budget obligations model will be considered. This theory stresses government's need to set tax rates congruent to its budget demands. A fourth mechanism will be considered at the end of this section. The institutional mechanisms theory highlights the role of formal, vertical intergovernmental mechanisms as the primary force behind government action. This study will devote considerable effort to examining the role of such vertical mechanisms as predictors of tax rates. The second and third sections of this chapter review the existing literature on the diffusion-through-learning model and the diffusion-through-competition model. Recent academic efforts have attempted to disaggregate and evaluate these sticky mechanisms. The terms have been and continue to be disputed and conflated in the literature. This paper will clarify these mechanisms in the face of convoluted, conjoined terminology. The final, fourth section of this chapter reviews the six studies most relevant to the goals and methods of this dissertation. 15 Endogenous theories The Behavioral Model The traditional behavioral model, most prominently articulated in Campbell, Converse, Stokes, and Miller's (1960) "The American Voter," emphasized the role that public opinion plays in shaping public policy. Dozens of other studies since then (e.g., Carmines and Stimson 1989; Dalton 2005 Jacobson 2004; Page and Shapiro 1983) provide compelling evidence that political leaders are poignantly aware of and responsive to public opinion. Studies of local opinion and taxation behavior confirm that at least some of the behavior of political leaders is dictated by opinion. Therefore, public opinion should significantly influence tax-setting behavior. Less unpopular taxes should be substituted for those that are truly despised by the public (Krishna and Slemrod 2003). The general level of opposition to taxation, and towards the public sector in general, should dictate lower overall tax rates. Several studies recognize this contention, not only acknowledging the variables shaping public opinion, but also explicating the mechanisms through which public opinion influences the behavior of leaders (e.g., Ansolabehere and Snyder 2006; Bowler and Donovan 1995). Several other studies implicate those decisions in the context of tax rates (Ashworth and Heyndels 1997; DeHoog, Lowery and Lyons 1990; Henderson 1994; Hendrick, Wu and Jacob 2007; Stine 1998). Bryan Caplan (2001) goes so far as to test the interaction between property tax rates and voting, and concludes the virtual monopoly cities have over services virtually prohibits residents and resident firms from relocating. Such firms' and residents' only recourse is to vote for new city leaders in order to affect local tax rates. But there is substantial evidence to equivocate the impact of a unidirectional "bottom-up" behavioral model. First, John Zaller's (1989) interaction model suggests political leaders' opinions are exchanged with those of the general public; 16 opinions evolve together. This might allow a mayor or city councilwoman to "sell" a type of tax or tax rate to an otherwise strongly opposed electorate, reducing the overall correlation between opinion and taxes. Second, "All politics is local." Eriksen, Wright, and McIver (1989) demonstrated a tremendous regional variation among partisans. To assume that Republican voters in Salt Lake City, Utah will share identical opinions on property taxes as Republican voters in Murray, Utah, let alone Murray, Kentucky, stretches the assumption of ideological consistency too far. Third, the mundane nature of local taxation means that it often escapes the kind of visibility of other contentious issues (e.g. the location of adult novelty stores). Several scholars, notably Popkin (1994), suggest this allows leaders a greater degree of autonomy in making tax rate decisions (see also Krishna and Slemrod 2003). Fourth, a host of opinion-distorting mechanisms play out in the election cycle(s). City leaders disguise tax increases (Krishna and Slemrod 2003), raise less visible taxes, and/or raise taxes in off-election years or even as they are retiring from politics (Berry and Berry 1994; Bordignon, Cerniglia and Revelli 2003). Fifth, city leaders are often judged by factors not associated with policy choices, such as charisma and political connectedness (DeHoog, Lowery, and Lyons 1990). Sixth, interest groups play a disproportionate role in determining tax rates, distorting popular will in the setting of tax rates (Gill and Haurin 2001). These six factors contribute to the detachment of tax policy from public opinion. The Tax Availability Model Political theory also offers an expediency explanation for policymaking. Governments tax what is practical and efficacious to tax. Prior to about 1900, income tax would have been a practical impossibility. For most people, income was neither reported nor paid with enough documentation to enable vigorous, fair income tax collection. To demonstrate, practical enforcement of sales taxes is currently 17 facing a huge challenge through Internet sales (Cornia, Sjoquist, and Walters; 2004; McLure 1999). Another related tax availability issue is volatility. Sales taxes are much more sensitive to economic fluctuations than are property taxes. In an economic downturn, sales taxes are liable to see significant dips (Fisher 2009). In times of prosperity, though, sales tax revenue can be of great use to municipal leaders (Sjoquist, Walker and Wallace 2005). A surge in retail development may entice city leaders or voters to begin raising sales tax rates to capitalize on a previously untapped revenue stream (Dye and McGuire 1991; Henderson 1994). A question of implementation is another important availability factor. Property tax collection requires a substantial number of bureaucrats to process and enforce such taxation. The more complex the tax structure, the more substantial the transaction cost associated with the cost. Smaller cities, or cities with tight budgets, may simply not have the resources to collect the tax in question (Slemrod and Yitzhaki 2002). In contrast, cities and counties with strong property values may take advantage of this high transaction cost and raise taxes surreptitiously (Stine 2005). Some cities may not possess the tax base needed to effectively raise revenue through specific types of taxation (e.g. Blackley and DeBoer 1987). Cities with abysmally low property values, for instance, would find it difficult to raise significant revenue through property tax. Cities with a great deal of public sector property would also struggle to raise significant revenue through property taxes. Cities with few retail stores may find it difficult to raise revenue through sales tax (see Luna, Bruce and Hawkins 2007). These practical constraints will certainly affect the rates and type of taxes, and will also interact with tax competition issues. Finally, the fuzzy nature of policymaking requires a brief discussion of the entanglement between the availability model and other mechanisms at work in the 18 setting of municipal tax rates. Volden, Ting and Carpenter (2008) demonstrate that over time, cities learn which policies work best. Although this might seem like unequivocal diffusion-through-learning, their study discusses the way(s) in which jurisdictions craft policy based on internal successes or failures. Recall from above that leaders execute implementable policies. Seattle might initially adopt a policy a it learns from Tacoma, but over time Seattle learns from its own successes and failures as it modifies the policy to a point more effective and beneficial than Tacoma's was or became. The tax availability model fits this scenario well. Seattle might initially adopt a LOST because it perceives that nearby Tacoma had success with its LOST. But then, as Seattle learns from internal experimentation which revenues are most available, it changes its policy not because of outside pressure but (mostly) because of internal pressure. The sticky nature of these causal mechanisms makes it difficult to disentangle this endogenous learning from external learning. However, this paper's significant data and analysis makes significant progress in measuring and disentangling the two. Budget Obligations Budget obligations may drive cities' type and rate decisions. For instance, cities whose responsibilities include public education will have a significant fiscal obligation perhaps absent in cities in neighboring states, or even cities in the same state. Cities may or may not include fire departments, medical services, sewage, or mosquito abatement within their operational budget (for a survey, see Wallis 2000). Cities undertaking such additional service commitments are under pressure to raise more revenue than their neighbors without such obligations (Inman 1989). Epple and Schipper (1981) consider pensions as specific fiscal demands and find such pensions do affect budgetary decisions. Alm, McKee and Skidmore (1993) find strong evidence fiscal distress is a leading cause for states to adopt lotteries. Bartik 19 (1992) offers a review of the literature investigating the complex relationship between taxes as a means to raise revenue for creating jobs. As above, the obligations may entangle with voter preferences. Line item tax increases may be more likely to pass a voter referendum if such tax increases are dedicated beforehand to a specific spending item (Green 2006). Not all the obligations are as clear; many fall along a continuum. Some of the more subtle revenue demands include different levels of service, such as more fire protection funding in cities with older homes, more police protection in cities with higher crime rates, or fewer schools per capita in cities with large number of private school attendees. These three endogenous approaches are far from exclusive. A city with demanding citizens would generate budget obligations, which would then need to be met by increased taxation. For example, a positive feedback loop in which citizens demanded more libraries could lead to more legacy operating costs and capital debt. But those robust libraries and corresponding services would generate a clientele expecting more of those services. A similar scenario might develop in a city with an existing strong tax base. Citizens might expect their city to act differently because their property tax revenues were robust, and then call on city leaders to tap into those resources more aggressively. These endogenous mechanisms also interact with the exogenous behavior of neighbors. For instance, both of the examples above are deeply linked to peer-to-peer influences. City residents would be more likely to demand more libraries and library services if a neighboring city had better services, and property tax revenue only seems untapped when compared to the revenue those taxes are generating in a neighboring city. Existing literature offers support for these interactions. Zodrow and Mieszkowski (1986) and Allers and Elhorst (2005) discuss the horizontal spillover 20 effects of public spending within interjurisdictional settings. This can create interaction between endogenous and exogenous mechanisms. A popular example in the literature is law-enforcement spending. If Seattle, Washington were to dramatically increase its spending on police, some crime would invariably move to neighboring Tacoma, pressuring Tacoma to increase its spending as well (Lawton, Taylor and Luongo 2005). A city with a high crime rate might not feel pressured to fund police more vigorously if such a city's neighbors had even worse crime. In a related phenomenon, the negative consequences of one activity spill over into a neighboring jurisdiction, imposing costs on its neighbors (Allers and Elhorst 2005; Case, Rosen and Hines 1993; Sjoquist et al 2007). For instance, limited riverboat gambling in Davenport, Iowa, might increase crime in neighboring Bettendorf. Bettendorf collects none of the gambling revenue of Davenport, but does have to deal with the externalities, forcing Bettendorf to at least consider adopting riverboat gambling. These mechanisms lurk within the diffusion-through-learning and the diffusion-through-competition theories. Institutional Constraints A fourth endogenous factor involves vertical intergovernmental relationships. These legal constraints are not endogenous to the local governments the way voter preference, budget obligations, and tax availability are, but neither are these legal constraints dependent on the exogenous peer-to-peer horizontal network, the major focus of this paper. In a broad sense, the constraints are part of the legal framework of a city, and as such are correctly labeled "endogenous." Legal constraint is a powerful variable influencing municipal behavior (Bowman and Kearney 2012; , Zimmerman 2002; Zimmerman 2004) and can specifically affect rates as well. The limits on the rates are usually present, and many times such limits fall within a very narrow band of discretion. Many states set property tax or sales tax rate ceilings 21 and/or floors on their subgovernments, significantly reducing their discretion to raise revenue and often forcing the use of and dependence on other mechanisms for revenue (e.g. Cutler, Elmendorf, and Zeckhauser 1999; Chapman 2003; Dye and McGuire 1997; Johnston, Pagano and Russo 2000; Mullins and Joyce 1996; Pagano and Johnston 2000; Sokolow 1998). Although such states will not be considered in this paper, many forbid any local option sales taxes at all, completely preventing diffusion of such rates in any form (Dye 2008). Vertical restraints affect the rate-setting behavior of cities within this intergovernmental setting, as well as the degree to which competition can even take place (Boyne 1996). States also impose legal mechanisms facilitating or hindering taxation. If cities are required to obtain taxes through citizen referenda, it is more difficult to alter rates. Some states even require supermajorities of citizen referenda to increase taxes beyond a certain level. Several states allow local governments to raise rates, but then those rate hikes can be repealed by ballot initiative (for a review, see Sokolow 1998), and at least one study confirms local voters actively control the rates of their cities through those mechanisms (Biegeleisen and Sjoquist 1988). A few states place tight controls on the rate variations allowed. For instance, Georgia limits its LOST rates to either 1% or 0%. A few states require cities to share a portion of their property tax and/or sales tax revenue with other cities in the state. Both Utah (sales) and Texas (property) are among the states with such requirements. My research addresses these limitations and constraints, and offers some quantification of the effects of these constraints. However, like the ideological issues above, to disentangle state controls from municipal cooperation is difficult. Theory suggests states may be brought in on behalf of cities to set rules (Allers and Elhorst 2005; Bednar, Eskridge, and Weingast 2001; McKinnon and Nechyba 1997) making 22 it more difficult for cities to undercut one another. Will state controls result in higher or lower rates? The limited literature suggests more restrictions will reduce local rates, but the evidence is far from conclusive (Brueckner and Saavedra 2001; Mullins and Joyce 1996; Sjoquist et al 2007). Will state controls result in more or less rate variation? Existing studies certainly indicate state controls do impact local systems (Boschken 1998; Bowman and Kearney 2012; Bruckner and Saavedra 2001; Burge and Piper 2012; Burge and Rogers 2011; Chicoine and Walzer 1986; Mullins and Joyce 1996), but the literature here is less robust than the literature examining the effect of federal restrictions on state behavior. This paper increases the knowledge and understanding of state restrictions on the variation, diffusion, and level of local government tax rates. These findings will have implications for the effects of all superordinate governmental rules on policy diffusion throughout their subordinate units. Diffusion-through-learning There is overwhelming evidence that governments copy each other, especially within peer-to-peer networks whose jurisdictions are in close geographic proximity. Two leading theories have been offered to explain this phenomenon: diffusion-through- learning and diffusion-through-competition. Academics have used both terms in multiple contexts over the past several decades of diffusion scholarship. It is therefore necessary to distinguish their meaning in this study. Learning by Other Names A commonly used word like "learning" comes with many possible connotations. As it is used in this paper, diffusion-through-learning is easily distinguished from an endogenous learning mechanism. One of the most prolific scholars of diffusion, Craig Volden, and his colleagues (Volden, Ting and Carpenter 23 2008) describe such an endogenous "decision theoretic" mechanism. The decision theoretic model describes governments learning from their own policy experiments, adopting policies based on their jurisdiction's internal preferences, successes and failures. This is an important use of the word "learning"; governments do learn from their own experiments. But since the primary goal of this paper is to evaluate the exogenous forces driving diffusion, it will focus almost entirely on the exogenous learning described in their "game-theoretic" model. 6 However, some the findings of this paper do have implications for this endogenous learning model, as addressed below. Further, the concept of diffusion-through-learning has taken many names (see Maggetti and Gilardi (2013) for further discussion of the terminology problems). "Yardstick" is a notable competitor to "learning." The term "yardstick competition" has been applied to diffusion-through-learning. Some scholars differentiate "yardstick" from "learning" but they are close enough (if at all separate) to deserve treatment as synonymous terms throughout this paper.7 This paper will use the term "learning" almost exclusively. The peer-to-peer observe-act-observe-react cycle more closely resembles a classroom setting where students compete against, but also work in cooperation with each other, more than it resembles an athletic competition. 6 Volden, Ting and Carpenter also offer a peer-to-peer diffusion mechanism, labeled the "game-theoretic" diffusion model. Volden, Ting, and Carpenter's use of the "game theoretic" terminology is weak. Game theory by definition requires not only two or more players (Volden, Ting and Carpenter satisfy this part of the definition) but also the idea players are trying to maximize their utility based on the moves of the other player. Moves are necessarily interdependent. To qualify as "game theory," Tacoma's actions must affect Bellingham's and vice-versa. Finally, Volden, Ting, and Carpenter aggregate the two mechanisms of diffusion this paper attempts to disaggregate. Their exogenous "game theoretic" mechanism may take place in a positive or zero-sum environment, the principal means through which the two mechanisms are distinguished in this paper. 7 The term "yardstick" is important to discuss at least briefly here because of the implications an actual yardstick has as a measuring tool. Just as sports teams measure success against competitors, when a government, or set of governments, is perceived to be successful with a policy (e.g., a lottery generates $200m for public education), neighboring governments and constituents will use a figurative "yardstick" to measure the success of their own policies. If such a lottery is perceived to be more successful than the current policy, leaders may adopt policy to "measure up" to their neighboring jurisdictions (see Besley and Case 1995). 24 Positive-sum Games and "Learning" Conformity powerfully affects human behavior. The tendency to copy others' behavior can be so powerful as to lead a person to go against what he/she knows to be true (Asch 1956). Individuals learn by mimicking their peers, especially when they perceive successful individuals have adopted a behavior. In this respect, the diffusion-through-learning mechanism likens the spread of governmental policy to the spread of popular icons or tools. These icons and tools can spread purposefully or inadvertently throughout human populations. Trends in government spread in similar patterns. One jurisdiction may adopt a policy because that policy is successful or is perceived as successful. Adoption may also occur despite the fact that the policy may not necessarily deserve to be adopted, but because the policy is popular (Meseguer 2005; Walker 1969). Similarly, policies may be rejected when a policy seems to fail even when the evidence used to evaluate the degree of success or failure is itself unclear (Volden 2010). Copying often stands in place of formal policy analysis. (Allers 2012; Allers and Elhorst 2005; Berry and Berry 1992; Besley and Case 1995; Ladd 1992; Maggetti and Gilardi 2013). Tacoma, Washington, might perceive neighbor Bellingham's franchise tax as successful. Whether the tax is truly successful or not, Tacoma may rush to adopt it, even without conducting a careful analysis of the applicable similarities and differences with Bellingham. Hoover is a small city in the Birmingham, Alabama metro area. Imagine a scenario in which all the cities in the metro area adopt a particular successful policy, but Hoover adopts the policy last (i.e. the "loser of the race" if it were a sports competition). According to the learning mechanism, Hoover will not necessarily lose anything. Even if Hoover lost the race to adopt the policy in question, Hoover's loss doesn't benefit the other cities in the network. If other cities do benefit from 25 Hoover's reluctance, a diffusion-through-competition mechanism is applicable. Finally, the "learning" terminology is preferable to other terms ("yardstick") because-unlike measuring the success of an athlete in the long jump-the success of a policy is often subjective. Some policies are not easily measured and others defy measurement completely (Allers 2012). Cities and their residents measure municipal success in both tangible and intangible means. San Diego and its neighbors, Chula Vista and Encinitas must balance public spending between golf courses and art museums. A yardstick is too narrow a term to use in judging the quality of golf courses and art museums between the cities. Very real semantic difficulties are embedded in "diffusion-through-learning." Even with these caveats, a consistent use of "diffusion-through-learning" throughout the remainder of this paper facilitates review, discussion, testing and analysis. However, due to the importance and relevance of this discussion to the goals of this paper, these terminology problems will be revisited in the semantics subsection at the beginning of the diffusion-through-competition section. Purposeful Learning and Blind Copying The goals section of this paper included the question of governments as rational actors. Another semantic problem of "learning" is that of rational intent (Shipan and Volden 2008). Can true learning be said to occur if the copying is occurring without any rational intent? Human conformity can be based on an informed decision, or it may be relatively blind. If Mike adopts Tom's behavior because Mike perceives such a behavior benefits Tom, Mike is making a rationally informed decision to conform. This situation is more likely when the effect of the specific behavior has a direct path to a measureable outcome. (e.g., Tom has a comfortable retirement and contributed aggressively to his IRA when he was young.) But if Mike blindly copies Tom's behavior because Tom is successful-or because Tom 26 is a friend-then the decision to conform has been made blindly. This is more likely to occur when the behavior does not have a direct causal path with an outcome or the outcome is not easily measured (e.g., Tom has a comfortable retirement and regularly has his teeth whitened.) In this second case, Mike is simply using a shortcut to save time making decisions, and mimicking a behavior of a person he trusts. This second behavior might not be completely irrational, but it is nevertheless far from purely rational. The blind/informed nature of diffusion-through-learning falls along a similar continuum (Maggetti and Gilardi 2013; Shipan and Volden 2008). In part, this is due to a lack of perfect information, which forces governments to make imperfectly informed decisions. When information is easily available, cities are more likely to adopt a neighbor's policy with informed intent. Encinitas, California, can easily see property values escalating rapidly around neighbor Chula Vista's newly built public golf course. Encinitas sees the Chula Vista golf course pays for itself from user fees. Chula Vista officials report high levels of satisfaction among their residents. Given all these Chula Vista positives, and some undeveloped land in Encinitas available for a reasonable cost, Encinitas will build the course. Other times, information is not as easily visible. The impact of zoning decisions is notoriously difficult to evaluate (Pogodzinski and Sass 1991). For instance, zoning an area high or low-density residential may result in different patterns of growth and development. Chula Vista observes one area of high-density housing in San Diego that attracts high-end, wealthy residents and another that does not. Still, San Diego is perceived as a successful city and has a plentitude of high-density neighborhoods. If Chula Vista has none, it will feel pressure to adopt a similar policy. Chula Vista hasn't acted completely irrationally in this situation, but it is far less rational than the carefully considered golf course situation above. 27 Some decisions must be made even more blindly. At the extreme end of the blind copying example, Chula Vista will copy one or more of Encinitas' behaviors for no other reason than because Encinitas is perceived to be successful, and if Chula Vista wants to be successful it feels compelled to follow suit. The value of services delivered by a municipal arts council may not be completely impossible to measure, but it lies near the end of the unquantifiable end of the spectrum. Encinitas is perceived to be a city with status and wealth. Encinitas has an art council. Chula Vista wishes to emulate Encinitas in all things. Chula Vista creates an art council. Although it is a secondary goal, this paper will distinguish between these two forms of learning and offer evidence as to which of the two appears to be more influential within the learning model. The nature of the massive, numerical data in this study facilitates this analysis. In short, this paper finds more evidence of blind copying than of intentional learning. However, given that this goal is secondary to the overall goal of weighing competition and learning, "learning" as it is used in this paper will usually cover both blind copying and intentional copying.8 Patterns Diffusion often describes adoption patterns similar to a sigmoid (s-curve) mathematical curve or exponential curve (Figures 2.5 and 2.6). Such curves are found often in nature, in the adoption of a beneficial adaptation throughout a species (Boyd 1988). Human behavior is profligate with examples of this sigmoid adoption curve. From technological adoption to popular culture, pioneers will lead the way by experimenting. Adoption by others will be slow at first. The adoption rate then accelerates until the majority owns the technology or has seen the movie (Gladwell 2000). Then, the adoption rate slows as a few holdouts resist change. 8 On the occasions when this paper does distinguish between blind copying and informed copying, it will be clearly stated. Otherwise, "learning" will refer to any behavior along the continuum. Percentage of adoptees 28 Many Few Earlier Time Later Figure 2.5: A sigmoid policy adoption cure (from Pemberton 1936 and Gray 1973). Figure 2.6: An exponential policy adoption-rate curve. According to this model, a positive feedback loop dominates the diffusion pattern until every member in the community has adopted the policy. 29 Existing literature demonstrates that tax policy follows such a curve (e.g., Henrich 2001 and Zhao 2005). Burge and Piper (2012) demonstrate that the spread of LOSTs follows this s-curve. They infer from this pattern that learning has played a significant role in spreading the adoption of such a policy. This study demonstrates strong support for this sigmoid pattern of adoption. The findings in Chapter 4 and indicate broad support for the general form this sigmoid curve takes. Chapter 5 will return to discuss the implications of this phenomenon. Academic study of the spread of innovations throughout a community has been profligate. The study of diffusion-through-learning holds this sigmoid pattern up as a demonstration of that mechanism at work. However, such a pattern does not indicate exclusivity for the learning mechanism. In contrast, adoption sometimes fails to spread according to the S-curve pattern (e.g., Henrich 2001). Such scholarship has suggested several alternative patterns of diffusion (see also Aoki, Lehmann, and Feldman 2011). One alternate pattern in both biological and cultural systems can be described by an exponential curve, as in Figure 2.6. In this case, policy is so successful that no member dare resist adoption. This pattern supposedly indicates a more rational policy-adoption process (Henrich 2001). This curve has also been more closely associated with a competition model (e.g., Brueckner and Saavedra 2001). However, the evidence gathered by this study does not confirm the exponential curve, reducing support for both rational learning and for competition. Variables These two patterns, exponential and (even more ubiquitous) the S-curve, dominate the literature. There are several theoretical mechanisms offered to explain the rate and shape of the adoption curve. Gatignon and Robinson (1985) provide a relatively early, competent review of 30 such mechanisms. Of the many factors identified in their review, several are particularly relevant to the adoption of tax rates. First, the status of the early adopters, for instance, generally speeds adoption by the rest of the community and tends to create an adoption pattern resembling the "S-curve." Thus, a large, wealthy city increasing or changing its tax rates should be more persuasive than a small, poorer city. Second, a clear, unimodal distribution of the group's attitudes into supporting and resisting roles encourages the S-curve pattern. Thus, within a group of communities, a clear division between those favoring property tax and those favoring sales tax as sources of revenue should create the sigmoid pattern. Third, increasing ambiguousness of the results of an adoption generally increases the chance of an S-curve pattern. If the costs and benefits of adopting, raising and lowering rates are clear, the pattern will be more likely to conform to the exponential curve. In a recent review article, Shipan and Volden (2012) offer an update to Jack Walker's (1969) diffusion study. In their review of hundreds of diffusion articles, they develop "seven lessons" from the research that prove helpful not only to future scholars but also to practitioners of public policy. First, echoing Walker's work from 40+ years before, Shipan and Volden say diffusion is not only about geography. If Dallas, Texas, shares economic, cultural, and/or ideological similarities with Melbourne, Australia, Dallas will be more likely to copy the actions of Melbourne. Second, again echoing Walker, jurisdictions really do compete against each other: for tax revenue, against negative spillovers, and even for intangible "prestige." This theoretical competition crosses over into the diffusion-through-competition mechanism and will be revisited in the next section. Third, governments learn from each other. They don't just copy each other haphazardly, they observe their neighbors' policies to see which are successful and then intelligently choose from the 31 menu of policies that seem to work best. Fourth, governments do sometimes copy each other haphazardly if those behaviors are widely spread throughout the system because those behaviors are associated with success. This can take the form of coercion, leading to a "race to the bottom" to be addressed in detail in the next section. Fifth, again reiterating Walker's (and many others') scholarship, endogenous political structures (e.g., Walker's close electoral margins or jurisdictional wealth) play a significant role in how rapidly a policy will diffuse. Sixth, Shipan and Volden echo many of the variables discussed in Gatignon and Robinson (1985). The policies themselves can have a dramatic effect on the speed and penetration of adoption. The more complex, compatible, measureable, and trialable a policy is, the faster it will diffuse. Finally, Shipan and Volden identify the role decentralization plays in diffusion. Without at least some degree of autonomy given to subgovernments to make policy, there would not be any diffusion to speak of; the experimental power of (for instance) federalism will count for naught. In the most up-to-date scholarship exploring the variables affecting and describing the diffusion-through-learning model, Butler and Volden (2013) follow up the Shipan and Volden (2012) work by reporting the results of a survey of city leaders. The survey finds leaders are indeed eager to learn from their neighbors, are more eager to learn from success than failure, are more likely to copy larger cities rather than smaller cities, and are more likely to seek out and learn from ideologically like-minded cities over other factors. This latest effort confirms the earlier work of Grossback, Nicholson-Crotty, and Peterson (2004). Although testing and explaining these variables is only a peripheral goal of this paper, the findings of this paper indirectly inform them, and thus add to the relevant literature. More importantly for the purposes of this dissertation, these adoption curves and associated variables assist in disaggregating the diffusion- 32 through-learning mechanism from the diffusion-through-competition mechanism and will therefore be reconsidered in the methods, results, and discussion. Finally, these variables also add to the overall complexity of the tax rate diffusion milieu. This study concludes that such variables help to the complex manner in which the diffusion-through-learning and diffusion-through-competition mechanisms interact with endogenous factors. Diffusion-through-competition Returning to the primary question posed by this study: How much of a role does learning play in tax adoption and rate setting, especially compared to the competition mechanism? The diffusion-through-learning mechanism is difficult to disaggregate from the learning-through-competition mechanism. The Semantics of "Competition" To meet its goals, it is essential for this paper to clarify the language of these mechanisms. Otherwise, any effort to differentiate between the two with data will be moot. The existing literature on the relevant terminology is robust, if often incongruous. Particularly relevant among all the disjointed literature discussing the terminology in question are studies from Kenyon (1997), Boyne (1996), Meseguer (2005) and Salmon (2013).9 But for all their complex interactions and overlapping semantics, one connotation most effectively distinguishes between the terms. In the learning model, the success of one jurisdiction does not necessarily take resources away from a neighbor, creating a positive-sum game. In contrast, the competition model requires the loss of one jurisdiction in order for another to benefit, although a 9 I have attempted to synthesize these discussions in such a way as to make my key terms more discrete than they might otherwise be treated by these other studies. Consequently, the semantic distinctions in this study do not regurgitate any single study among this group, although my terminology comes closest to Salmon (2013). As noted below, among the studies under close scrutiny, the terminology in this paper is closest to that of Burge and Piper (2012). But Burge and Piper (2012) is an empirical study, unlike the several listed here, and as such its discussion of semantics is less robust. 33 pure zero-sum game is not necessarily required. If Encinitas builds a public golf course, neighbor Chula Vista might follow suit. Within a pure version of the learning model, Encinitas can't lose anything when Chula Vista copies it. Encinitas increases its status by building a beautiful public course. Chula Vista's status increases when it copies that behavior. There is no "finite" amount of status for which the two cities compete against each other. This is a positive-sum game. The competition model portrays things quite differently. According to this mechanism, there is a finite amount of the resource in question. One fairly common example of the competition mechanism is municipal recruitment of industry. Consider a scenario where Huntsville, Alabama and Jackson, Mississippi are among the final choices for a new BMW factory in North America. There is only one factory. Both cities want it desperately. One city will win the factory; the other loses a corresponding amount of revenue, income, status, etc. This is a zero-sum game. A secondary, but important, additional connotation to the term "competition" assumes players act strategically. Within a network, jurisdictions make moves based on what each believes will maximize their utility. Furthermore, every player's actions affect every other player.10 Finally, every actor tries its best to predict the moves of the other player(s) in order to "beat" the other players to a limited supply of resources. To illustrate the strategic nature of these moves and counter-moves, reconsider the industrial recruitment scenario, introducing a third city, Chattanooga, Tennessee into the bid for the plant. To lure the BMW plant to their cities, all three offer tax incentives to the corporation. Huntsville is the largest. It has the deepest 10 Although game theory often insists (Dixit and Skeath 1999) players act strategically, the simplest games in this study do not. If a city lowers its tax rate to poach revenue from its neighbor, but does not consider what the neighbor's countermove might be, it would still be a "competitive move" if not a "strategic" one. 34 pockets. Jackson is the next wealthiest. Jackson could perhaps afford to get into a bidding war with Chattanooga but-game theory predicts-it will not, since Jackson knows Huntsville will beat any deal they offer; it will not even try. But Jackson may be able to offer other incentives (e.g., cheaper land, lower cost of living) that might attract BMW. While each city is predicting and moving against each other, players also try to predict and respond to BMW's moves. Although it is a secondary goal of this paper, this study finds little evidence of such moves and counter-moves, increasing the certainty with which this paper concludes a more powerful (but not unequivocal) role for the learning mechanism. These strategic considerations will create additional complexities in any system through which policy diffuses. As Chattanooga, Jackson and Huntsville compete for the BMW plant, they try to guess what incentives the other players might offer and respond with what they believe BMW will consider a better offer. Such strategic moves require the players to attempt to predict the moves of their rivals. For instance, the calculation, "If I adopt strategy X, my opponent will probably adopt strategy Y, resulting in a net benefit/loss to me," is strategic. Or even, "I expect my opponent will use strategy Y against me; thus I will signal a willingness to adopt X, even if X will cost me, in order to deter my opponent from adopting Y." If Huntsville were fairly certain its competitors were unable/unwilling to increase tax incentive above a certain point, it might choose to signal a willingness to raise incentives above that point. On final implication of the strategic nature of the diffusion-through-competition mechanism is its effect on the role of motives. The diffusion-through-learning mechanism depicts a scenario in which players ignore the effect(s) of their actions on peers within the network and vice-versa. The competition mechanism says that not only do players act to steal benefits from one another, but also that 35 each player perceives the others as agents trying to take their benefits. The learning/competition semantic dichotomy is as dependent on the motives and perceptions of the players as it is on the actual choices. This paper's tests depend on these motives as the tests pry apart the two mechanisms. A Continuum The zero-sum game is not as cleanly detached from the learning mechanism as economists and political scientists would like (Bordignon, Cerniglia, and Revelli 2004). The learning/competition question is not truly binary. However intangible, and however small, Encinitas still loses some status to Chula Vista when Chula Vista builds a beautiful public golf course. Why? There is a finite amount of demand for golf courses in the greater San Diego area. Encinitas loses some of that demand to Chula Vista when the latter builds a top-notch course. Even though this scenario is far less competitive than the zero-sum game depicted in the BMW scenario, Chula Vista still benefits due to Encinitas' loss. Beneath the surface of many learning scenarios lurk competitive and strategic motives. Imagine what at first might seem like a simple diffusion-through-learning scenario: Los Angeles wants to recruit a professional football team to emulate the perceived success and popularity of San Francisco. This might appear superficially like a pure example of learning. But consider: Los Angeles builds a city-funded stadium at a cost of hundreds of millions of dollars. Now the Chicago Bears threaten to move to Los Angeles for a "free" stadium with the potential to generate more revenue. Chicago must build a new stadium or rebuild Soldier Field at taxpayer expense. This touches off an arms race in which city after city must follow suit. Cities run faster and faster to stay in the same place, building bigger and better stadiums, which cost the cities more without attracting any additional revenue or prestige. Acting strategically, the other cities might put political pressure on Los 36 Angeles, discouraging them from pursuing an NFL team. Aspects of competition penetrate further into this superficial "learning" scenario. Even if the NFL expands and creates a new team for Los Angeles, every other member team loses a little revenue. Before the expansion, football fans in the Los Angeles area might travel to San Diego to watch the Chargers, buy jerseys from the 49ers, or pay for a cable TV channel guaranteeing delivery of all the Raiders games throughout the season. All the teams see some loss. Even if the overall revenue generated by the NFL goes up after the expansion, each team might lose $2 million in revenue (for a total $52m loss) while the new LA team might net $60m per year. Nevertheless, acting strategically, other cities might improve their stadiums to prevent Los Angeles from even considering entering into an arms race. Therefore, it may be difficult, or even impossible to recount or even imagine a scenario operating completely within the learning or competition regimens posited here. Nevertheless, this paper asserts these two conflated mechanisms manifest themselves differently enough to warrant investigation, both for theoretical and practical reasons. Is Los Angeles mostly acting strategically to poach an NFL team from Chicago, or is it acting without strategic consideration? Chicago and the NFL need to know to what degree they need to watch Los Angeles' moves. Encinitas city leaders (and political scientists) should know if Chula Vista is building a golf course with the hope of stealing the love of golfers or simply to raise property values within their city. Tax Competition Among scholars, there seems to be little doubt tax competition exists (for a review see Genschel and Schwarz 2011), although the degree of such competition is debated. This dissertation assumes the most salient causal mechanism in setting tax rates is the policies of neighbors, taking the form of competition or learning. This 37 subsection further details the tenants of the competition mechanism and its applicability to basic questions of tax competition and revenue maximization. Tax competition is complex, not simply because there are more variables, but also because there are usually many more than two players. But tax competition can be reduced to a simplified game in which players make moves against one another based on predicted outcomes and predicted opponents' responses. Figure 2.7 depicts the most important variables involved in tax competition. Figure 2.7 demonstrates how each causal mechanism impacts the system in multiple ways. Raising sales tax rates will increase revenue, but will also decrease spending in the home city as residents cross-border shop (Luna 2004; Luna, Bruce and Hawkins 2007; Nelson 2002; Walsh and Jones 1988). Higher sales tax rates Municipal revenue Increasing retail purchasing Increasing property tax revenue^ T Increasing number of retailers t + Higher property tax rates Figure 2.7: Simplified municipal revenue determinants. This figure depicts the interacting forces at work as cities try to maximize their revenues. The interaction of rates is more complex than a simple positive feedback loop. 38 Higher sales tax rates will also reduce the likelihood of new retailers coming into the city, since such retailers believe their sales will be higher in cities with lower rates (Edmiston and Turnbull 2003; Fisher 1980; Lewis and Barbour 1999; Torralba 2004). Higher property taxes affect residential behavior as well. Residents move into communities to avoid paying higher taxes while benefitting from the amenities of a neighboring higher-rate city (Wildasin 1989). Evidence of tax competition abounds. Cornia, Grimshaw, Nelson, and Walters (2010), Luna (2004), Zhao (2005), and Devereux, Lockwood, Redoano (2007) all found strong evidence of such behavior in their study on competitive sales taxes. Scholarship also indicates property tax competition is widespread (Brueckner and Saavedra 2001; Heyndels and Vuchelen 1998; Wu and Hendrick 2009). However, the general conclusion of these works also indicates that property taxes are less mobile than sales taxes because transaction costs are generally lower in cross-border shopping than in relocation of a firm or resident (Caplan 2001; Chapman 2003; Coates 1993; Hendrick, Wu and Jacob 2007; Krmenec 1991; Ladd 1992; Powell 2004; Wilson 1996;). Even so, the literature on this question is not unanimous (e.g., Goodspeed 1998; Lewis and Barbour 1999; Lyytikainen 2012), mostly owing to the significant confounding variables and feedback loops inherent in the property-sales tax feedback loop (see Figure 2.7). City leaders are familiar with these mechanisms and their consequences (Bartle 2003; Cornia, Grimshaw, Nelson, and Walters 2010; Luna 2004; Luna, Bruce and Hawkins 2007; Sjoquist, Smith, Walker, and Wallace 2007). As such, city leaders watch the moves of their neighbors, and are therefore able to calculate the impacts of their neighbors' moves on their city and its revenue. Figure 2.7 illuminates these mechanisms; higher sales tax rate decreases retail purchasing by driving up cross-border shopping in neighboring cities. Higher property tax rates 39 may drive retailers away, encouraging a retailer to relocate from their home city to a neighboring one. Such relocation would be likely only if the property tax savings outweighed a potential decline in sales. Perhaps the neighboring city with lower tax rates did not have competitive infrastructure. Perhaps the home city is known for its concentration of high-end retail. But perhaps realtors see the neighboring city as a rising market, one that will quickly gentrify. A retailer, firm or resident might risk moving into such a neighborhood. Cities also compete in myriad ways besides taxes-as evidenced (saliently) in the BMW and (less so) golf course examples. Imagine a revised Figure 2.7 that included a fairly comprehensive overview of these factors. A figure complex to the point of confusion, even uselessness, would result. This wider competition has been considered in the literature (Baldwin, Forslid, and Martin 2005; Kenyon and Kincaid 1991) but because of the sheer complexity is usually handled through qualitative study. Even the broad scope of this study focuses on narrow cross sections of public policy. The wider competition will nonetheless be revisited later in this paper. A Two-Player Game Probably the best-known scenario in game theory, the prisoner's dilemma, posits a situation in which two players must decide to cooperate with or defect against the other player. Traditionally, this is portrayed as two suspected criminals (A & B) who would collectively benefit if they cooperated with one another and refused to testify against the accomplice. The worst outcome for A is when he refuses to testify but B defects, landing A with a severe penalty and B with a minor slap on the wrist. This scenario is entirely applicable to the tax competition model, both in terms of property and sales taxes. With respect to LOSTs, the mechanisms of such a 40 scenario are illustrated in Table 2.1, but the same matrix could be applicable to LPTRs. To simplify, imagine this tax competition free from many of the confounding variables that will be introduced later. In this oversimplified scenario, there are two neighboring cities, both relatively isolated from others. Further, imagine each can set its LOST rate either high (e.g., 8%) or low (e.g., 3%). La Verkin and Hurricane, Utah, will serve in the following explanation, even if in reality their situation is not this simple. Turning to Table 2.1, if La Verkin sets a high LOST rate, Hurricane will seize this opportunity to set its rate low. In this scenario, Hurricane defects against La Verkin, and poaches retail activity by attracting cross-border shopping and increasing the number of retailers locating in Hurricane. As a result, La Verkin nets a weak $1m in revenue and Hurricane takes in a much-higher $4m. These figures (especially the differences between and within each cell) almost certainly exaggerate a real situation. In other words, adopting lower tax rate leads to increased revenue for your city, but only if you have neighbors with high rates to steal shoppers from and only if your city has a high enough rate to capitalize on such cross-border shopping. Since both La Verkin and Hurricane should know this, game theory predicts (Nash 1951; Oates 1972) both cities will set low rates so that neither can steal shoppers or retailers from the other, and each will wind up with a modest, but not terrible, $2m. Table 2.1: A prisoners' dilemma payoff matrix for intermunicipal tax competition. Adapted from Dixit and Skeath (1999). The rates for each city are set outside the matrix, while the revenue generated by each of the four scenarios is listed in each of the four interior cells. La Verkin Low sales tax rates High sales tax rates Low sales tax rates $2m, $2m $4m, $1m High sales tax rates $1m, $4m $3m, $3m 41 There is substantial evidence-more theoretical than empirical-to show governments do race to the bottom in this way (Asplund, Friberg, and Wilander 2007; Genschel 2002; Oates 1972; Oates and Schwab 1988; Wilson 1999; Zodrow and Mieszkowski 1986). However, there is also substantial evidence demonstrating governments do not engage in a race to the bottom (e.g. Basinger and Hallerberg 2004; Baskaran and Lopes da Fonesca 2013; Chirinko and Wilson 2009; Genchel and Schawrz 2011; Mendoza and Tesar 2005). If players cooperate, they can potentially reach the best scenario, the Pareto optimum, which occurs when Hurricane cooperates with La Verkin. In this case, both cities set their rates relatively high (the high, high outcome). Hurricane and La Verkin must resist temptation to lower rates in this scenario, since this would instigate the race to the bottom, temporarily netting the defecting city more revenue and more political popularity for its leaders, but then inviting retaliatory lowering of rates from their neighbor. Instead, if both cities keep rates high, they both benefit from higher revenue. This is the best mutual scenario; collectively, La Verkin and Hurricane net $6m in sales taxes. This investigation elucidates these patterns of behavior. More Players, Continuous Rates, Equilibria The simplified scenario above must now be made more complex to illustrate a truer rendition of the mechanisms involved. First, rates are usually fluid. Most states with LOSTs and LPTRs allow their cities to change rates in smaller increments. Second, most cities have more than one neighbor. Figure 2.8 allows for more continuous rate changes, depicting a continuous response scenario to a fixed metro-area rate. This fixed metro rate could come from a single neighboring city, or from a dozen neighbors each with identical rates. Although this figure also oversimplifies matters, it illustrates the prisoner's dilemma in a more realistic environment. 42 High (U D C (U > $ ^ Medium (U > o o X Low Figure 2.8: Hoover's revenue-per-rate versus a constant metro-area tax rate. Adapted from Morrow (1994). Hoover's best response to a fixed metro-area rate falls along the curve. y is the rate at which revenue is maximized. Rate 9 might produce in slightly higher rates than rate y, but with lower overall revenue. Suppose Hoover, Alabama is considering whether to raise or lower its rates. Hoover surveys the tax environment surrounding it and notes the average LOST rate of its neighboring cities is relatively high, perhaps 4%. If the metro rate is above the Nash equilibrium, Hoover's best response is to set rates a little lower than the average metro rate. This will increase sales and retail development. In turn, Hoover will see a significant net increase in revenue, even though its rates are lower. If the average metro rate changed, Hoover's best response curve would change with it, and Hoover would be forced to change its rate accordingly. And although Figure 2.8 is devoted to sales tax rates and revenue, the same figure could be used to illustrate the property tax rates and revenues. Residents will move into Hoover's maximum retail sales volume, but not maximum revenue Hoover's total revenue varies along this curve as its rates vary Hoover's maximum revenue given a constant metro rate (MR) Y ir , Metro rate held "constant here Hoover's Tax Rate 43 areas with lower LPTRs to avoid higher taxes (Dowding, John, and Biggs 1996). And, given a subsequent boost in property values, Hoover would see a net gain in revenue even though its rates were lower. Given a fixed metro rate, Hoover's best response is again at point y. But Figure 2.8 doesn't allow Hoover's neighbors to react to Hoover's changes. Adding another layer of complexity, Figure 2.9 is a depiction of how two cities might interact along a dual continuum of rates rather than the dichotomy in Table 2.1. This two-city interactive response curve shows only the "best responses" of each city's tax rates to its single neighbor's tax rates. -cMu ru cL X |TU High CD tn c o Ci tn CD Medium -M tn cu CO .tn cu > ru Low w d g C3 Hurricane's Best Response Tax Rate Figure 2.9: A simple, two-jurisdiction best response graph. Here, two players can adjust their rates to one another along continuous rates, rather than one city adjusting its rates to a fixed competitive rate as in Figure 2.4. (Adapted from Dixit & Skeath 1999). 44 La Verkin and Hurricane, Utah, are adjoining communities somewhat isolated from other cities. With only two players, Hurricane and La Verkin only need to consider the moves of a single neighbor as they set their rates. Each can adjust its rates along a continuum. Hurricane's best responses fall along the red curve, and La Verkin's fall along the blue curve. As La Verkin raises its rates above the Nash point ^, so will Hurricane, though not by quite as much. To illustrate, if La Verkin sets a high rate, Hurricane will undercut that rate by just a bit, at point ^. Hurricane's slightly lower taxes-per-sale will be offset by more sales. If La Verkin sets its rates far below the Nash equilibrium, Hurricane will set its rates slightly higher (at point y), with slightly lower sales than La Verkin, but its slightly higher tax rate will generate more overall revenue per capita. Both cities end up losing revenue in this second scenario. As in the simple model, Nash (1951) predicts La Verkin and Hurricane will move and counter-move against each other as each tries to maximize its own revenue given its opponents' moves. As each seeks to maximize revenue, theory predicts both cities should move towards Nash equilibrium, point ^ in Figure 2.9 (Oates 1972). But, as in the simple model, this race to the bottom, more aptly called the "race to the Nash equilibrium," is not the best-case scenario for both cities. Much as in the prisoner's dilemma, both cities could do better if they cooperate and reach point p. However, both have a strong incentive to defect (cheat) as well, undercutting their neighbor's rate to poach revenue through cross-border shopping and attracting more retail firms. Thus, cities face a collective action problem as they try to set rates above the Nash Equilibrium (Friedman 1971; Ostrom 1998). If cities cooperate, and don't undercut each other, they can reach a "win, win" scenario as illustrated in Table 2.1, where both La Verkin and Hurricane set high 45 rates. This point is illustrated in Figure 2.9 by point p. But to reach this point, cities must not defect. Both cities must resist the urge to lower their rates, since doing so will reduce their collective benefits and instigate a race towards the Nash Equilibrium, as discussed in the previous two subsections. Most municipalities in the United States reside in areas with multiple jurisdictions, but adding more players to Figure 2.9 would make the figure unreadable. Figure 2.10 solves that problem by clustering multiple players into cooperators and defectors. This is still an oversimplification, because it lumps players into defectors (derisively, "cheaters") or cooperators, when in fact cities fall along a continuum, as in Figure 2.9. U JZu fC V VD CV > a u !_ Va 75 4-1 ,o Less More Level of rate-cooperation Figure 2.10: An n-player, defector-cooperator incentive model. This figure puts all the cooperators along the "cooperators" line and all the cheaters along the "defectors" curve. Note as more cities in the metro area cooperate, the overall benefit for all the cooperators goes up, but so does the incentive to cheat (t). The best scenario for any one city would be to be the only cheater, point 0. (Adapted from Ostrom 1998). The Nash equilibrium in this figure is at point v. 46 In Figure 2.10, the total per capita sales tax revenue generated is represented on the y-axis, and the amount of cooperation on the x-axis. The two lines, "cooperators" and "defectors", represent the revenue of the cooperating and noncooperating players, respectively. The black line is the break-even point; any revenue above the line beats expected revenue generated from the Nash Equilibrium. e is the point reached by all players if everyone cooperates. 0 is the point reached by the one player who does not cooperate in a large-N setting. t represents the incentive for any one player to defect. In this example, the payoffs from defectors remain relatively constant. The number of cooperating players is the only factor affecting the payoffs of the defectors(s) and cooperator(s). Existing empirical and experimental literature demonstrates cooperation is harder to maintain with more players (Isaac and Walker 1988; Kremenec 1991, Lewis and Barbour 1999). Thus, more cities in an area should create more tax competition. Nevertheless, it is in the collective best interest of all cities for everyone to cooperate. Not only is more revenue generated, but cooperation reduces political costs as well. There are several theoretical and empirically tested mechanisms encouraging cooperation (Baskaran and Lopes da Fonesca 2013; Heckathorn and Maser 1990; Ostrom 1998; Palfrey and Rosenthal 1994). Some of these mechanisms are formal (e.g., a built-in system of "punishment" for defectors) and others informal (e.g., a long history of working together). This theoretical discussion will aid in answering the primary goal of this paper: Does policy diffuse through learning or competition? Examining tax rate distribution patterns and rate-to-revenue correlations elucidates the degree to which cities cooperate and compete. This investigation finds that cities are neither cooperating nor racing to the bottom, at least not in any directly measureable sense, 47 once again indicating a stronger role for learning than for cooperation. Learning or competition: Recent literature The following six recent works are particularly relevant to this dissertation. These studies, among hundreds of others, are the ripest and the most germane to the goals and methods of this paper. They have not only provided considerable theoretical and empirical background for this dissertation, but have also provided considerable guidance on the methodology and semantics of the subject. This study continues their work by examining the municipal property and sales tax rate-setting behavior. Shipan and Volden (2008) Shipan and Volden's (2008) paper on the diffusion of smoking policies is the single most relevant scholarly work to this dissertation.11 In some ways, other papers under close scrutiny more closely resemble the goals of this paper. But Shipan and Volden make the most concerted effort to disaggregate the diffusion-through- learning and the diffusion-through-competition mechanisms. They find evidence of competition and learning, but struggle to differentiate the degree to which cities learn or compete as policy diffuses throughout a region. Shipan and Volden posit four mechanisms of diffusion, all of which have been discussed in the preceding pages. The first and second are easily recognizable, they are nearly identical to the diffusion-through-learning and diffusion-through-competition models. Their third, the imitation hypothesis, is similar to blind copying as described in the "purposeful learning and blind copying" subsection of this chapter. Shipan and Volden recognize the difficulties in distinguishing between the 11 The Shipan and Volden (2008) paper was extremely helpful in rescuing this dissertation, which began as: "Do cities compete on tax rates or do they cooperate?" After preliminary data made it plainly clear they do not cooperate, I investigated several other avenues of inquiry. Shipan and Volden (2008) was particularly helpful in setting me in this new direction. 48 two forms of copying with empirical tests. Even though it is not the primary goal of this paper, the data and analysis presented in this paper will also pry apart some of the differences between these blind copying and purposeful learning mechanisms. Shipan and Volden's fourth mechanism, the coercion hypothesis, is substantially related to the goal of identification and analysis of the vertical intergovernmental restrictions limiting subordinate government's rate-setting discretion. If the California state government restricted municipalities' discretion in buying, building, and maintaining golf courses, such an action would significantly affect the ways in which golf course diffuse through the greater San Diego metropolitan area. As indicated earlier, this paper finds little evidence of state policy affecting patterns of tax rate diffusion. As for methodology, Shipan and Volden assemble tests for each of their hypotheses. For instance, they conduct a "competition" test. For every city without an antismoking policy, they find all the cities within 10 miles that also haven't adopted an antismoking policy, then they weight this number by population. If smaller towns do not adopt smoking restrictions in their restaurants because their larger neighbors have not either, Shipan and Volden suggest this should be an indication the smaller city is afraid to adopt for fear of losing restaurant business to the larger city. Shipan and Volden's primary goal is to disentangle and weigh the strength of these various diffusion methods, mirroring the primary goals of this paper. However, their methods have some weaknesses. Even though their tests are essentially valid, there are too many lurking variables to make all their tests truly convincing. For instance, in the case of the competition test above (their most important test) there is significant overlap with the principles behind other mechanisms, like imitation and learning. Simply because a small city waits for a large neighbor to act does not 49 necessarily mean the small city fears competition. It may simply be emulating the policies of its larger neighbor. In addition, because their tests do not offer an "either-or" decision rule between the diffusion-through-competition and diffusion-through- learning models, the fact that stronger correlations exist for their learning tests does not solidly undermine their competition model. Shipan and Volden admit their tests are muddled and call for further tests using different methods. Their methodological shortcomings will be revisited in the methods section of this paper. Baybeck, Berry and Sigel (2011) In 2011, Brady Baybeck, William Berry, and David Siegel (BBS) published a work highly relevant to this dissertation. Their article is mostly an exploration of the diffusion-through-competition mechanisms, with very little effort dedicated to differentiating between learning and competition. For this reason, the BBS work is not quite as relevant to this dissertation as the Shipan and Volden paper. In their article, BBS describe and explain the mechanisms behind revenue competition through a spatial analysis of lottery adoptions. While BBS make it clear they respect the methods and goals of Shipan and Volden (2008), they claim the diffusion-through- competition model is woefully under-researched. As such, they make a concerted effort to disaggregate and measure the mechanisms behind such competition: defensive competition (adopting a policy to stop revenue flowing to another state), offensive competition (adopting a policy to poach revenue from nonadopters) and anticipatory competition (adopting a policy to pre-empt a competing state from poach revenue from you). Almost as an afterthought-since they do not include the mechanism in their model-the BBS study offers only minimal attempts to measure the power of the learning component of lottery diffusion. Although there are several reasons to bring BBS under close examination, 50 methodology is the single most important of these. Of the six studies under close scrutiny, BBS' methodology is the most relevant to the core hypotheses and tests of this dissertation. Their core tests are also the most relevant to my research. BBS use GIS to give each nonadopting state a score on three measures, each testing one of their three forms of diffusion-through-competition. The tests hinge on residents cross-border shopping for lottery tickets. Such cross-border shopping is the key instrument in this study's attempt to differentiate learning and competition. BBS' methods are directly relevant to the goals and methods of this study, whose core hypotheses assumes property tax rates are less dependent on neighbors' property tax rates than sales tax rates are dependent on neighbors' sales tax rates, because sales taxes are more mobile than property taxes. Like BBS, this paper asserts that jurisdictions (is this case, cities) act defensively to try and prevent cross-border shopping out of their jurisdiction and to encourage cross-border shopping into their jurisdiction. This study's inferential data tests and the mechanisms they imply are quite similar to those of BBS. BBS' data and analysis demonstrate strong support for diffusion-through-competition. However, BBS go even further, claiming inferential support for governments acting strategically, as in the case of the competition for the BMW plant. In their study, BBS find states do not adopt the lottery because such states do not want to trigger reciprocal adoptions from neighbors who have not yet adopted. BBS contest this demonstrates governments are behaving strategically.12 Their argument has merit, but is far from convincing. The same result may arise from offensive competition from other states.13 The tests in this paper are less 12 Will Chula Vista copy Encinitas' golf course? Maybe not, say BBS, since that will trigger Encinitas to build even more courses, diminishing the overall returns both cities make on the courses. I would be more convinced of the validity of BBS' tests if the cost of lottery adoption (say, a massive investment in infrastructure) was high, but the lottery adoption "price" is relatively low. 13 This is complicated. If neither Illinois nor Wisconsin have lotteries, but Iowa does, then BBS 51 focused on distinguishing the role strategic moves play in rate diffusion, but implications from the results of this paper indicate (weakly) that cities do not act strategically. Finally, BBS discuss the long-run economic consequences of such lottery competition and by extension any kind of revenue-based competition. This significantly informs the race-to-the-bottom scenario. Adopting policies earning economic rent in the short term (poaching revenue from neighbors via the lottery) should eventually lead to all states adopting the lottery. The result: Utah is one of only five states in the lower 48 without a lottery. Now, very few states get a free ride (Idaho on Utah, Louisiana on Alabama, etc.). The implication for this dissertation is all cities should push towards the Nash point. But they don't, further dispelling the competition mechanism as the most powerful force driving tax rate diffusion. Burge and Piper (2012) Burge and Piper (2012) offer another highly relevant work on the "sources of diffusion" question. They examine Oklahoma municipal and county LOST adoptions and rate changes. Like the other articles under close scrutiny in this paper, their study attempts to disaggregate the causes of policy diffusion. Furthermore, Burge and Piper examine several endogenous factors as well as the diffusion mechanisms, devoting a greater percentage of their effort to that cause than this paper does. But of particular relevance to this investigation, their research explores the mechanisms behind LOST diffusion as well as the interaction between property and sales tax rates and revenue. For instance, they offer a lengthy discussion of the revenue trade-offs say Illinois will not adopt because they do not want to start a lottery war with Wisconsin. But it could show up as a negative correlation because Wisconsin acts first to poach Chicago's lottery ticket-buying consumers. This shows up as a negative correlation for Illinois because it reduces the comparative speed with which Wisconsin adopts the lottery. 52 between raising and lowering tax rates, tax exporting via cross-border shopping, and voter attitudes towards such rates. An additionally relevant component of their research pays considerable attention to vertical relationships as well. However, Burge and Piper pay only superficial attention to distinguishing between diffusion-through- learning and diffusion-through-competition. They have no direct tests to disaggregate the two. Like Baybeck, Berry and Siegel (2011), Burge and Piper use a modified Event History Analysis to examine LOST rate changes over time. Their learning14 tests rely on a very similar strategy to the tests used by BBS and others. The adoption of LOSTs by the neighbors of Tulsa significantly increases the chances of Tulsa adopting a LOST itself. As expected, they find a robust relationship. However, like BBS' tests, this simple test alone does not constitute a means to disaggregate between the diffusion-through-competition and diffusion-through-learning. The adoption of a policy simply because a neighbor did may result from a competition over revenue. But it may also simply be a measure of a neighbor wanting to copy another's success, or even perceived success. Burge and Piper do not dispute the weakness of this test to differentiate the two mechanisms. Burge and Piper do attempt to pry apart the learning and competition mechanisms with two other tests. One of these tests demonstrates that cities with large retail activity are more likely to adopt LOSTS. Burge and Piper assert that such behavior indicates tax competition. Tulsa, a city with a large existing retail center, may attempt to tax export to neighboring cities by taxing other cities' residents who 14 Burge and Piper (2012) use the term "yardstick" to describe their learning model. As above, there are multiple uses of the term in the literature. In fact, they lump yardstick competition/conformity/learning together but use "yardstick" as an umbrella term. However, at times, they discuss "policy diffusion" as a separate entity from the "yardstick" mechanism (see Besley and Case 1995; Bordignon, Cerniglia, and Revelli 2004). Burge and Piper also use the economic term "spillover" more liberally than is warranted. For more information, see the discussion of semantics in the "learning by other names" and "semantics of competition" subsections of this paper as well as the sources listed in those subsections. 53 cross-border shop in the city with the large retail base. This is competition in the sense that Tulsa is poaching potential revenue from other cities in the metro area. Although some tax competition is indicated by this data, because such action could also be a measure of the endogenous "tax availability" mechanism, this first test cannot fit behaviors discretely into the competition15 mechanism as defined by this paper. The second Burger and Piper competition test asks whether cities are raising rates in response to neighbors' raising and lowering rates. Their data analysis indicates cities are not raising rates to mirror each other. Coupled with their other endogenous and exogenous tests, Burge and Piper find evidence that diffusion-through- learning, vertical intergovernmental interaction, and endogenous factors act as agents to spread the policy. Such learning may include both intentional learning and the blind copying behavior described earlier. In short, Burge and Piper conclude there is weak support for the diffusion-through- competition mechanism. They offer little conclusive data and analysis to differentiate the mechanisms of diffusion as described in this dissertation. Like BBS and Shipan and Volden, they call for more study and more effort to accomplish this. Boehmke and Witmer 2004 Generally, this "in-depth" section has focused on articles published with in the last five years, but the Boehmke and Witmer (2004) study is one of two earlier articles relevant enough to include in this section. It is difficult to imagine a more appropriate title to this dissertation than, "Disentangling Diffusion: The Effects of 15 Burge and Piper do not discuss the revenue curve (see Figures 2.8 and 2.9). Instead, they simply discuss the tax-exporting nature of tax competition. This is overly simplistic. The model proposed in this study assumes cities want to maximize revenue while minimizing cost. If a higher tax rate leads to more revenue, ceteris paribus, then the city will raise its rates, at least up to the Nash point, where theoretically its revenue will begin to be poached by neighboring cities. This leads-again-into a semantics problem as the diffusion-through-competition model visualized in this paper includes some of what Burge and Piper include in their definition of "yardstick." 54 Social Learning and Economic Competition on State Policy Innovation and Expansion." In this study, they devise several tests to do just that. More than any other article under close scrutiny, Boehmke and Witmer choose and use terminology most closely mirroring the language of this paper. Diffusion-through-competition (which they simply call "economic competition") refers to the need for a given state, G, to adopt a policy to poach revenue from neighbors or prevent poaching from G. Diffusion-through-learning (which they call "social learning diffusion") refers to the comparative value judgments states make as they copy each other in the hopes of adopting good policy. In this case, "Are the benefits of Indian Casinos (potential revenue) worth the disadvantages (negative spillovers)?" By watching the success of other neighboring states that have adopted, state G might decide that the trade-offs are worth adopting the policy. Boehmke and Witmer also make a strong case for the need to further clarify and explain the need to differentiate between diffusion-through-learning and diffusion-through-competition, both for practical reasons and for furthering the discipline. They correctly claim that the literature prior to 2004 lacked a robust treatment of the diffusion-through-competition mechanism and instead focused almost solely on the diffusion-through-learning mechanism. Boehmke and Witmer's subject matter is also relevant to this study: the adoption of Indian gaming compacts throughout the United States. Accelerating since 1988, states have allowed the expansion of Indian-owned casinos. Boehmke and Witmer measure the number of compacts and the degree of gaming offered by the casinos through an Event History Analysis. This allows examination of adoption of policies that potentially diffuse via competition and/or learning. Boehmke and Witmer's tests are far from decisive. They conduct two tests on the gaming data. Their first test is of innovation, when a state adopts Indian gaming 55 for the first time, and to what degree that innovation is based on the behavior of that state's neighbors. Their second test is of expansion, in which they measure the growth of existing Indian gaming within a state and measure to what degree that expansion is related to neighbors' Indian gaming. They assume that the expansion of Indian gaming is a measure of "competition," but not of "learning." That is, states expand (increase the number and variety of) Indian gaming ventures in an effort to pre-empt other neighboring states from stealing revenue only due to economic competition pressure. But they claim the innovation (the initial adoption of Indian gaming) is due to both "learning" and "competition." Boehmke and Witmer's assumptions have merit, but are flawed enough to significantly undermine their conclusions. Learning can influence both innovation and expansion. Returning to the golf course example, Chula Vista considers copying Encinitas' blue ribbon course. In this innovation stage, Chula Vista simultaneously learns from and competes against Encinitas. They try to copy the perceived success of Encinitas while competing against it (for the "finite" interest in golf in the greater San Diego area) at the same time. Then, in Boehmke and Witmer's expansion stage, the competition mechanism becomes the sole driver as Chula Vista adds features to their existing course. Competition seems like a more plausible causal mechanism behind expansion, but consider expansion applied to a "learning" positive-sum environment. It is equally possible that Chula Vista is simply continuing to "learn" from Encinitas as it adds features to its blue ribbon course. Even in the BMW example, Jackson might perceive Huntsville benefits from setting up an industrial park, then copy that action with the belief that Jackson will benefit even if it doesn't win the BMW factory. 56 Brueckner and Saavedra (2001) Jan Brueckner and Luz Saavedra (2001) offer the earliest study under close scrutiny in the competition-versus-learning literature. Despite its age, this germane study, "Do Local Governments Engage in Strategic Tax Competition," was one of the first to use a spatial analysis correlation to test property tax rates of the municipalities in the Boston metropolitan area. Like this paper, theirs is an effort to describe and explain the landscape of local government rate setting. They find strong evidence that tax competition motivates rate-setting behavior. Brueckner and Saavedra deserve an in-depth look for several reasons. First, this paper was not the first to use spatial correlations, but of the early works in the field, their methodology closely resembles that of this paper. In particular, they compare the behavior of multiple types of rates under multiple conditions, drawing conclusions from the differences in the strengths of these correlations. The core data and tests of this paper are related to and derived from spatial correlations. Second, they offer a robust discussion of voter preferences vis-a-vis strategic rate setting. Brueckner and Saavedra suggest that some residents will tolerate-even welcome-higher tax rates to meet those preferences. This significant treatment of preferences is a worthy discussion of the behavioral model. Third, their study includes a serious discussion on the interaction between property values, tax rates, and behavioral preferences. Firms and residents shopping for property will, ceteris paribus, discount the land located in a higher-rate location (see Figure 2.7). But public goods offered by higher-rate jurisdictions might offset such losses caused by the higher economic rents of such jurisdictions. Property values are therefore based on a complex chain of interacting variables. Fourth, Brueckner and Saavedra offer a strong discussion of the role of vertical rate restrictions. Akin to Proposition 13 in California, citizens of Massachusetts passed Proposition 21/2 in 1980, severely limiting 57 cities and towns' ability to raise revenue through the local property tax. Brueckner and Saavedra examine the behavior of cities and towns after its passage, and find that the level of tax competition declined. More than any other study under close scrutiny, Brueckner and Saavedra produce empirical evidence that state law affects tax competition, illustrating the need to consider the impact of state restrictions as a secondary goal of this paper. Brueckner and Saavedra's paper differs most significantly from this investigation in one way. They lack serious interest in and discussion of the learning model (which they call "mimicking"). Their paper does consider factors other than tax competition (e.g. the behavioral model as above), but omits nearly all discussion of the core question of this study. They conclude that competition drives copying behavior, but do not satisfactorily address the possibility that learning could drive diffusion. This is a problem, because as was shown in other studies (Bordignon, Cerniglia, and Revelli 2004; Lyytikainen 2012), a simple spatial correlation cannot, by itself, eliminate the possibility of diffusion-through-learning as the primary mechanism motivating rate-setting behavior. However, despite their lack of discussion of the diffusion-through-learning mechanism they do include several tests that support the competition model more thoroughly than the learning model. For instance, they show that business tax rates have a stronger spatial correlation than residential property tax rates. They are correct in their claim that this difference supports the competition model. It is reasonable to conclude that firms are more willing to move locations to capture better tax rates. Residents are more likely to stay in an area for reasons other than tax rates. But even if their premise is basically sound, it too easily dismisses the learning factors at work. Businesses choose locations based on factors other than 58 tax rates and residents do choose homes based on tax rates, undermining the significance of this test as a decisive one. A second test shows that after the passage of Proposition 21/2, the spatial correlations of residential tax rates declined to almost zero. Proposition 2 / imposed restrictive rate limits on Boston-area municipalities. This led cities to reduce their rates between 1980 and 1990 to the effective rate ceiling, a ceiling far below the Nash equilibrium. Virtually all cities in the area were leveled to this ceiling, eliminating any variation and thus competition. This test implies support for the competition model. If learning were driving rate-setting behavior, a ceiling cap on rates might have resulted in a new s-curve (see Figure 2.5) rather than cities' rates clustering at the upper limit allowed by law. In comparison with this admittedly strong study, this study more specifically addresses the competition-or-learning question and offers more robust tests. Da Silva Costa and Carvalho (2013) This section of in-depth reviews concludes with a paper from Jose da Silva Costa and Armindo Carvalho (2013), "Yardstick Competition among Portuguese Municipalities: The Case of Urban Property Tax." As with the other papers in this section, da Silvia Costa and Carvalho wade into the sticky discussion of diffusion-through- learning and diffusion-through-competition by attempting to disentangle tax competition from other mechanisms affecting rate-setting behavior. In particular, da Silvia Costa and Carvalho discuss "yardstick competition" at some length and contrast it with what they call "strategic competition." Voters use a metaphorical "yardstick" to measure the performance of their city leaders, but they primarily make those measurements in comparison to other cities, in this case residents compare their property tax rates to neighboring cities' property tax rates. If residents perceive that their rates do not compare favorably to their neighbors' rates, they are more likely to vote their leaders out of office. While this might sound like 59 competition, recall the fundamental principle differentiating diffusion-through-learning and diffusion-through-competition in the semantics sections: is the city setting rates to be successful (including looking good to its residents), or because it wants to poach revenue at their neighbors' expense? Recognizing that the two mechanisms are never completely inseparable, the former is the diffusion-through-learning mechanism and the latter is diffusion-through-competition. Even though da Costa Silva and Carvalho use the term "competition," the mechanism they describe is, at its core, nonetheless a learning mechanism. Only if the adoption of a rate by city X would lead to a revenue increase, at city Y's expense, would the term "competition"16 be applicable. Therefore, as da Costa Silva and Carvalho distinguish between what they call strategic tax competition and yardstick competition, they are really offering a mechanism to differentiate between diffusion-through-learning and diffusion-through-competition. Methodologically, their paper offers an interesting means of investigation. Like many of the papers under close scrutiny in this paper, da Costa Silva and Carvalho use a panel study and track tax changes over time. And, like Brueckner and Saavedra (2001), da Silva Costa and Carvalho posit that less rate homogeneity (among certain cities) indicates more competition and less learning. More creatively, da Costa Silva and Carvalho use ideology and partisanship in testing the learning/competition mechanisms. Like nearly all the works considered by this paper, da Costa Silva and Carvalho discuss political preferences as an endogenous variable. For instance, they discuss the likelihood that cities with strong financial resources (e.g., an industrial center with high-paying jobs) will be less 16 Like many of the papers under close scrutiny in this paper, and many others mentioned more briefly, this scenario skirts the gray line between the two mechanisms. Da Costa Silva and Carvalho discuss the possibility that voters might be driven so far as to move to the neighboring (lower tax rate) city if voting the tax-raising city council members out of office alone does not result in enough satisfaction This, in turn, may lead to an increase in property values for the lower-rate neighbor (see Figure 2.7). In the long run, then, the learning mechanism might transition into a competition mechanism. 60 willing to move towards a Nash equilibrium. They also note the role that information cost plays in determining people's willingness to "shop with their feet" and/or "vote with their feet" by moving out of a particular jurisdiction. Furthermore, political affiliation and city council composition play a significant role in setting tax rates. A left-leaning city council is more willing, for instance, to raise taxes to pay for social programs. This is not a significant departure from previous studies. But using the ideological composition of the city councils is actually at the core of their assumptions and tests to distinguish between the learning and competition mechanisms. Their inventive test and corresponding hypothesis is this: narrowly-divided city councils have less discretion in setting tax rates and are more dependent on the tax rates of their neighbors. To demonstrate, the Dallas city council is narrowly divided between Democrats and Republicans. Its voters regularly flip control of the city from Democrats to Republicans and vice versa. A councilman in Dallas would have to worry a lot more about his rate-setting votes than his counterpart in neighboring Fort Worth, whose city council is solidly Republican. The Fort Worth councilman knows that his votes are relatively safe. Given that premise, da Costa Silva and Carvalho posit that if the tax rates of city with narrower partisan divisions among its leadership (closer to parity) move over time toward the regional average, then city leaders' behavior is being shaped by yardstick competition (the diffusion-through- learning mechanism). In other words, the Dallas city councilman, fearing for his job, moves towards Fort Worth's rates so that Dallas voters do not judge Dallas taxes "unreasonable" and vote the precarious city councilmen and women out of office. "Safe" Fort Worth city leaders, in contrast, face less danger from their voters. They have relative impunity to set rates as they wish, and they wish to set rates low enough to poach revenue from their neighbors. Thus, if large-majority city 61 councils move towards the regional average, they are motivated by strategic competition (the diffusion-through-competition mechanism). If these assumptions were acceptable, their conclusion of a more robust role for the learning mechanism would be valid. But their assumptions assume too much. Such questions of the strength of council majorities must play but a minor role in the diffusion of rates. To demonstrate, Dallas residents will not punish city leaders if their tax rates are low-to-average for the region. Dallas residents will only punish leaders if Dallas rates are noticeably above the rates of the surrounding municipalities, as evidenced by the several studies demonstrating the frequency of city leaders seeking political cover in their neighbors' tax increases (e.g., Berry and Berry 1992). They also may be competing for businesses by lowering their rates to the regional average, and thus be engaging in regional competition for firms. Fort Worth leaders may also be interested in appealing to voters as well. Therefore, they may be converging on regional averages due to Yardstick Competition. Their test has some merit. Dallas leaders will be more likely to be motivated by the learning mechanism, and the Fort Worth council will be more likely to be motivated by tax competition, but these mechanisms can hardly be called exclusive. Summary These in-depth reviews could go on much longer, but they become repetitious very quickly. There are many more studies that explore tax competition (some via sales taxes, some through property taxes) without examining the role competition plays in diffusion. These are mostly studies that reflect the economic cost and rent-earning potential of poaching cross-border shoppers or residents/businesses that move into a community to take advantage of/despite that community's tax rate. A cavalcade of studies examines the role of either learning or competition alone, and most of these are at the state level (for a very general review see Shipan 62 and Volden 2012). After reviewing hundreds of tangentially related studies, and more than thirty that were directly related to the diffusion question, I find these six studies to be the most relevant. Yet even among these, several issues are unresolved. Table 2.2 summarizes these six papers. But, as helpful as these relevant studies are, each fails to fulfill all the goals of this paper. What is missing is a clear test to distinguish the two mechanisms. Several studies cited in this literature review-even some beyond these six-have a learning-or-competition test. As discussed above, there are significant inherent weaknesses even among the three studies with either-or tests (Boemke and Witmer 2004; Bruckner and Saavedra 2001; and da Costa Silva and Carvalho 2013). Table 2.2: Summary of the six core studies Study Central goal to differentiate learning from competition Includes either-or learning-competition test Comparison of multiple taxing regimens Focus on local jurisdictions Examination of vertical controls Includes spatial (GIS) analysis Boehmke and Witmer (2004) Yes Yes No No No Yes Shipan and Volden (2008) Yes No No Yes Yes Yes Brueckner and Saavedra (2001) No Yes Yes Yes Yes Yes Burge and Piper (2012) Yes No Yes Yes Yes Yes da Costa Silva and Carvalho (2013) Yes Yes No Yes No Yes Baybeck, Barry and Siegel (2011) No No No No No Yes This paper Yes Yes Yes Yes Yes Yes 63 To be fair, calling the tests offered by those three "either-or" is an overstatement. Even this paper's stronger test to disaggregate the two is hardly perfect. Despite weaknesses in the validity of my tests, the study remains a worthwhile endeavor. Several studies, including a few meta-studies (e.g., Goodspeed 1998; Maggetti and Gilardi 2013; Shipan and Volden 2012, Lyytikainen 2012), recognize the lack of definitive studies in this area of scholarship, and call for new tools to measure the mechanisms of tax rate diffusion and their effects. This investigation uses many of the methodological tools and theoretical tenants of previous works, especially of the six detailed above. Nonetheless, the methods used in this study have never been used in a published paper before. The core either-or test, to be discussed below, shows particular originality. These tests are valid and reliable. Finally, none of these six studies offers a mechanism to test the effect of vertical governmental constraints on these diffusion mechanisms. This study will address all these weaknesses in the existing literature. CHAPTER 3 METHODS The primary goal of this paper is to disentangle, describe, and explain the influence of the diffusion-through-learning and diffusion