Why do some firms pay more than the market wage rate? A case study with longitudinal data

This research studies the three main economic theories of wage as they apply to a specific case study using longitudinal panel data. The firm in the case study pays employees above market wage which presents an interesting opportunity to scrutinize the validity of each wage theory. The economic theories associated with the neoclassical school of thought, the institutional school of thought and power relationships are each tested for a fit with the case study data. Relative to the neoclassical theory, the efficiency wage theory and agency theory are tested as these are the elements of neoclassical thought which support wages above market rates. Two tests are used to evaluate the hypotheses. The first is a formal econometric test of the shirking model of the efficiency wage theory using production and pay data at the factory level. This is made possible because the four factories in the case study have essentially the same cultural and institutional environment, along with shared production technology and similar product output. As such, control is provided for elements other than wage differences between the factories. The second test is a survey of plant management. The survey provides a view of both what plant management was trying to accomplish with wage policy, and also how effective they felt the pay practices were. In addition to the survey, the culture within the firm is analyzed logically for its connection to pay practices. This research concludes that for the four factories studied there is no support for the shirking model of the efficiency wage theory or agency theory. As a result the standard neoclassical wage theory cannot explain the above market wage and is called into question, at least as a generalized model. In addition, there is substantial support provided for the institutional theories around wage as well as evidence of the power of the threat of union organizing. The management survey, the logical analysis of firm culture, and the comparisons to the organizational behavioral research all support the findings that institutional factors heavily influence wages in the sample firm.

WHY DO SOME FIRMS PAY MORE THAN THE MARKET WAGE RATE? A CASE STUDY WITH LONGITUDINAL DATA by Richard Parsons 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 Economics University of Utah August 2011 Copyright © Richard Parsons 2011 All Rights Reserved The Uni v e r s i t y o f Ut ah Gr adua t e S choo l STATEMENT OF DISSERTATION APPROVAL The dissertation of Richard Parsons has been approved by the following supervisory committee members: David Kiefer , Chair 3/15/2011 Date Approved Thomas Maloney , Member 3/28/2011 Date Approved Norman Waitzman , Member 3/28/2011 Date Approved Peter Philips , Member 4/12/2011 Date Approved Vicki Whiting , Member 4/4/2011 Date Approved and by Peter Philips , Chair of the Department of Economics and by Charles A. Wight, Dean of The Graduate School. ABSTRACT This research studies the three main economic theories of wage as they apply to a specific case study using longitudinal panel data. The firm in the case study pays employees above market wage which presents an interesting opportunity to scrutinize the validity of each wage theory. The economic theories associated with the neoclassical school of thought, the institutional school of thought and power relationships are each tested for a fit with the case study data. Relative to the neoclassical theory, the efficiency wage theory and agency theory are tested as these are the elements of neoclassical thought which support wages above market rates. Two tests are used to evaluate the hypotheses. The first is a formal econometric test of the shirking model of the efficiency wage theory using production and pay data at the factory level. This is made possible because the four factories in the case study have essentially the same cultural and institutional environment, along with shared production technology and similar product output. As such, control is provided for elements other than wage differences between the factories. The second test is a survey of plant management. The survey provides a view of both what plant management was trying to accomplish with wage policy, and also how effective they felt the pay practices were. In addition to the survey, the culture within the firm is analyzed logically for its connection to pay practices. iv This research concludes that for the four factories studied there is no support for the shirking model of the efficiency wage theory or agency theory. As a result the standard neoclassical wage theory cannot explain the above market wage and is called into question, at least as a generalized model. In addition, there is substantial support provided for the institutional theories around wage as well as evidence of the power of the threat of union organizing. The management survey, the logical analysis of firm culture, and the comparisons to the organizational behavioral research all support the findings that institutional factors heavily influence wages in the sample firm. TABLE OF CONTENTS ABSTRACT ................................................................................................................................................. iii LIST OF FIGURES .................................................................................................................................... vii 1 INTRODUCTION ........................................................................................................... 1 1.1 Wage Theory ..................................................................................................... 4 1.2 Setting Up the Test ......................................................................................... 9 2 LITERATURE REVIEW .............................................................................................. 12 2.1 Efficiency Wage Theory Literature Review ......................................... 12 2.2 Institutional Wage Theory Literature Review .................................... 19 2.3 Power Relationship Wage Theory Literature Review ..................... 26 3 TESTING EFFICIENCY WAGE THEORY .............................................................. 32 3.1 Overview of Efficiency Wage Test Methodology ............................... 32 3.2 Plants and Data ............................................................................................... 35 3.3 Choosing a Production Function .............................................................. 47 3.4 Calculating the Wage Premium ................................................................ 62 3.5 Testing the Wage Premiums ...................................................................... 73 4 TESTING INSTITUTIONALISM AND POWER RELATIONSHIPS ................ 84 4.1 Mars Culture .................................................................................................... 84 4.2 Survey ................................................................................................................. 87 5 MOTIVATIONAL THEORY AND ITS RELATIONSHIP TO ECONOMIC THEORIES OF WAGE ......................................................................... 98 5.1 Background ...................................................................................................... 99 5.2 Neoclassical Wage Theory ......................................................................... 102 5.3 Institutional Wage Theory ........................................................................ 105 5.4 The Relationship Between Motivational Theory and Economic Wage Theories ........................................................................... 107 vi 6 CONCLUSIONS ............................................................................................................ 111 6.1 Conclusion I ..................................................................................................... 111 6.2 Conclusion II ................................................................................................... 113 6.3 Conclusion III .................................................................................................. 114 6.4 Conclusion IV .................................................................................................. 115 6.5 Implications .................................................................................................... 116 6.6 Additional Research ..................................................................................... 117 APPENDIX A. PRODUCTION DATA ................................................................................................. 120 B. PAYROLL DATA .......................................................................................................... 125 C. ALTERNATIVE TEST WITH TEN YEARS OF DATA ....................................... 127 D. CHOOSING A PRODUCTION FUNCTION ........................................................... 135 E. A WIDE RANGE FOR EFFICIENCY WAGE? ....................................................... 139 REFERENCES ......................................................................................................................................... 143 LIST OF FIGURES Figure Page 1.1 Model of Wage Theories .......................................................................................... 10 2.1 Wage and Effort Relationship ................................................................................ 12 2.2 Effort and Output Relationship ............................................................................. 12 3.1 Average Pay by Plant ................................................................................................ 42 3.2 Output and SOC ........................................................................................................... 44 3.3 All Lines-Output and Headcount .......................................................................... 45 3.4 Bar Lines-Output and Headcount ........................................................................ 45 3.5 Shell Coating - Output and Headcount ............................................................... 46 3.8 Resulting Cobb-Douglas Functional Form ........................................................ 63 3.9 Resulting Constant Elasticity of Substitution Functional Form ............... 64 3.10 Resulting Translog Production Functional Form .......................................... 65 3.11 Resulting Additive Functional Form ................................................................... 66 3.12 Premium Pay ................................................................................................................ 72 4.1 Survey Results and Confidence Intervals .......................................................... 96 4.2 Survey Results Percentiles ..................................................................................... 97 5.1 Relationship Between Motivational and Economic Theories .................. 108 E.1 Wage to Effort ............................................................................................................. 140 E.2 Effort to Output .......................................................................................................... 140 E.3 Wages to Profit ........................................................................................................... 141 E.4 A Flat Wage Profit Relationship ........................................................................... 142 1. INTRODUCTION Wages are a keystone of economic theory; thus understanding the wage mechanism is critical for both economic and business policy. Wages are important for both the economy and for the individual. For the economy, wages provide a key input helping determine the level of employment, a division of income between capital and labor, a source of aggregate demand and an element of public and company policy. For the individual, wages provide a means of sustenance, a social recognition of an individual's contribution, and serve as a building block of an individual's sense of self worth. This research will review the major theories of economic wage applied to a specific set of longitudinal data. These data are for four substantially similar production plants owned by Mars, Incorporated over a 10-year period. This specific situation is of interest because at all four sites wages are paid above the local market wage rate leading to the question: Why do some firms pay more than the market wage rate? Three connected hypotheses will be tested: Null Hypothesis One: The neoclassical wage theory accounts for the wage policy at these four plants for the study period. Null Hypothesis Two: The institutional wage theories account for the wage policy at these four plants for the study period. Null Hypothesis Three: The power relationship wage theories account for the wage policy at these four plants for the study period. 2 The alternative hypothesis in each case is that the given wage theory does not account for the wage policy at the four plants for the study period. A widespread dispersion of wages paid, both vertical and horizontal, is well documented throughout industrial countries (Guttschalk & Smeeding, 1997). For example, in July of 2008 management professionals in the United States earned an average wage of $43.60 per hour, while fast food counter workers earned an average of $9.04 (Bureau of Labor Statistics, May 2008). Actors and athletes can earn millions while others make a minimum wage, sometimes even less. This vertical wage variability is a reality in modern societies. Suggested reasons for the variability include different values of human capital, different training and skills, and of course a commonly heard explanation, supply and demand. Neoclassical wage theory suggests that workers doing the same job, even in different companies, would be paid the same wage. In this research the variability in wage in the same job will be called horizontal wage variance. The forces of competition should press wages and production processes into equivalent efficiency. Equilibrium should be achieved as wage adjustments cause employees to move between companies until wages for the same jobs are in balance. While the vertical wage difference between the management professional and the fast food counter worker is reasonable to many people, it might be surprising to them to find that counter to neoclassical wage theory, the same job in the same city is often paid very different wages. 3 As one example, Table 1.1 shows that in Chicago for May 2008 procurement clerks earned the following wages by percentile (Bureau of Labor Statistics, May 2008). The top 10% of these workers earn over two times the amount those paid in the bottom 10% earn. This research will study production plant workers. An example for production plant workers from the same wage survey is shown in Table 1.2 for Chicago team assemblers. Again the top 10% of these workers earn over two times the amount paid to those in the bottom 10%. These examples from Chicago are repeated ubiquitously (Guttschalk & Smeeding, 1997). Further analysis will reveal that this wage variability is not accounted for by training, or by marginal productivity, or by any of the standard neoclassical supply and demand determinates in the labor market (Uusitab, 2002; Lal, 1979). Table 1.1 Chicago Procurement Clerk Wages 2008 Bottom 10% Bottom 25% Median Top 25% Top 10% 11.54 14.12 17.31 20.68 24.51 Table 1.2 Chicago Production Plant Workers 2008 Bottom 10% Bottom 25% Median Top 25% Top 10% 7.74 8.32 10.69 14.26 18.40 4 The question that arises from the empirical data is "Why do some firms pay their employees more than the market wage?" Several theories have been offered as possible explanations of this phenomenon. This dissertation will employ longitudinal data to test these major wage theories through the use of microeconomic data and management surveys for a single firm operating in multiple geographies. 1.1 Wage Theory Modern economic wage theories can be found in three general schools of economic thought: neoclassical, institutional, and power relationships. Each of these theories will be briefly described in this introduction and then further developed in Chapter 2. 1.1.1 Neoclassical Theory of Wage The neoclassical wage theory is derived from a concept of supply of, and demand for labor. Supply of labor is based on an individual's utility function in which consumption and leisure are traded off to maximize utility. "The properties of the supply of individual labor result from the combination of substitution effect and income effect. The combination of these effects seemingly leads to a non-monotonic relation between wages and the individual supply of labor" (Cahuc & Zylberberg, 2004, p. 9). The demand for labor is seen in a wider context. Labor is one of the factors of production. A firm chooses the combination of inputs, including labor, that minimizes its cost of production for the profit maximizing output quantity. Given supply of labor and demand for labor, wages are then determined as 5 the price that brings these two forces into equilibrium. The key results from the neoclassical theory of wage determination include, for a competitive environment:  Homogeneous workers and jobs will receive equal wages.  Workers who are unemployed choose to be unemployed as they maximize their utility trading off between work and leisure. These voluntarily unemployed workers prefer leisure to the disagreeability of work at the available wage.  Workers would move quickly and easily between competing jobs when small differences in wage rates and job characteristics appear. The reality of labor markets undermines these expected results. The dispersion of wages was demonstrated in Table 1.1 and Table 1.2 and researched by Guttschalk (1997). Also, involuntary unemployment is a reality across the globe and has been studied in detail since the great depression. And finally workers do not move quickly between jobs over small wage differences (Moscarini & Thomsson, 2007). These empirical realities have spawned a variety of adjustments to neoclassical wage theory. According to Krueger and Summers (1988) there are two types of answers to the question of why some firms pay more for the same job. First, firms may not be profit maximizers, but rather act according to the manager or owners' alternative agenda. This is called an agency theory mechanism. Agency theory is often expressed as a manager maximizing their own personal welfare rather than that of the firm and its owners. The second type of answer is based on the idea that firms may, for different reasons, find it 6 unprofitable to reduce wages to market level. These are efficiency wage considerations. Both of these answers however are focused within the rational world of homo economicus and orthodox neoclassical theory. Because these theories form an important part of the empirical work of this research they will be discussed in more detail in Chapter 2. 1.1.2 Institutional and Social Determination Wage Theory Other economists have developed theories basing wage determination as a function of social institutions and cultural/psychological determinates. These ideas began with Veblen's anthropological economic work. Wages is a fact incident to the relation of employer and employed. It is, in the sense fixed by colloquial use.… The laborer, from the point of view of consumption of products, is no longer "laborer": he is a member of society simply, and his share of the product of industry is the share of an individual member of society. (Veblen, 2002, p. 20) Veblen tells us that wages are set by social and historical standards and with views of equity and reciprocity rather than supply and demand. A key article by Akerlof (1982) brought the idea of fair wage and fair effort back from anthropology to economics. According to Akerlof the employer pays a wage higher than the going rate and the employee in return exceeds the market work standards. These are seen as a sequence of gifts and counter gifts as would also be seen in primitive social organizations. Cahuc and Zylberberg (2004) provide a relationship showing that social norms influence productivity and effort at equilibrium. While there are few universal behaviors that can be expected from this theory, as social norms will vary by group and over time, the identified institutions and social norms for a specific situation can be thought to drive wage relationships 7 in a logical way and the specific institutions such as unionism and bureaucratic hiring rules can be studied for impact. The impact of institutions and social behaviors have also been studied and documented in work with game theory and primitive cultures. 1.1.3 Power Relationships Wage Theory A final theory of wage setting is centered around the power relationship between worker and employer. In certain forms this theory can be traced to the "Father of Economics," Adam Smith. In speaking of labor and wages and the dispute between employer and employee, he says, "It is not, however, difficult to foresee which of the two parties must, upon all ordinary occasions, have the advantage in the dispute, and force the other into compliance with their terms" (Smith, 1981 (1776)). The words "force" and "dispute" are not synonymous with the concepts of an open free market. Perhaps best known in this arena is the significant work that has developed around Marx's theory of social struggle and surplus value. Marx (1844) states "wages are determined through the antagonistic struggle between capitalist and worker" (para. 1). Marx notes that the wage actually paid a given worker may be temporarily above or below the normal wage. Supply and demand forces cause oscillations above and below a "certain mean." The mean is the "natural price . . . determined independently of demand and supply" (Marx, 1891, para 5). While Marx's theory of wages started with the view that capitalism would drive wages to a subsistence level, he asserted that the average wage is "that quantum of the means of nourishment which is absolutely requisite to keep the 8 laborer in bare existence as a laborer," (Marx, 1848, para 22). Subsequently he began to add additional items of historical or cultural need to the definition of subsistence. In 1865, during his address to the General Council of the First International, Marx presented his fully developed theory of wages. There is still a minimum limit on wages that is basically a physiological minimum but there is no maximum. The determination of where a given wage will lie … is based on the respective powers of the combatants. This seems very similar to the class struggle Marx details in the Communist Manifesto. (Bigelow, 1999, Para. 5) In Marx's world the power struggle between the capitalist and the worker over the produced surplus value is reflected through the wage. Another area of research that fits into the power struggle group of wage theories is the work around union bargaining. A large amount of empirical and theoretical work has been developed around union negotiation. Despite the decline in union membership, union bargaining is both historically and currently very important to the world of wage setting. While union density has decreased, collective bargaining coverage is still quite high because in most European countries collective agreements prohibit differentiating between union and nonunion employees. According to the OEDC (1997) union density for member countries is 44% but collective bargaining is 68% (Cahuc & Zylberberg, 2004). Bargaining theory can be traced to Edgeworth (1881) where parties divide goods in a pareto optimal way, and then continues with Nash (1951) and Stahl (1973) and Rubinstein (1982) where game theory was first explored. If the rules of the game (utility preferences of the players and alternative options) can be described, a best case equilibrium can be modeled. Bargaining theory has resulted 9 in many modern economic models. Models around bargaining have been created and incorporated into economic theory. They include the monopoly union model, markup and union power model and insider/outsider model. These models can result in efficient outcomes under certain assumptions (Cahuc & Zylberberg, 2004). Empirical studies suggest that unions do in fact increase wages (Cahuc & Zylberberg, 2004). Figure 1.1 shows a graphic outline of the three wage theories discussed and the areas where these intersect. Each of these theories will be discussed in detail in Chapter 2. 1.2 Setting Up the Test The test in this research will employ both case study and econometric approaches to determine which of the theories being tested fit with the data. The theories tested include the neoclassical (efficiency wage, agency), institutional determination and power relationships. Efficiency wage will be tested directly using statistical regression analysis to determine if productivity is related to wage premium. The other theories will be tested through the use of a management survey and aligning each theory's stylized facts with the empirical evidence based upon the test firm's specific culture, history and institutional arrangements. 10 Figure 1.1 Model of Wage Theories Contributions of this research will include a direct productivity test of the efficiency wage proposition. While the literature is rich in terms of theoretical discussions of efficiency wage and its implications, the empirical work published to date is either at a macro level, with non-US data, with verifiable results or with proxies for productivity such as the level of supervision. Examples of these are addressed in Chapter 2. Production line data sufficient to model total factor productivity by factor (resulting in labor productivity) are rare and this researcher has not found a case where it has been used in empirical work associated with efficiency wage theory. In addition to the econometric work relating productivity to 11 wage, a management survey will be used to test all the wage theories. The survey will provide insight into the motives and results of management practice relating to agency, institutional and power relationship theory, and also will provide backup and support to the econometrics regarding the efficiency wage test. The management personnel that set wage policy at each plant will be the population for this survey. These individuals were responsible for hiring, firing and achieving output objectives. They studied the information available on wages for their site and made the recommendations on wage rates. They also lived with the results of the implemented wage policy. This research will also provide new information and insights associated with the question of "why some firms pay more than the market wage rate," as it explores management policy and practices regarding the various theories on wage determination. In addition, this research will align the economic wage theory with the latest in terms of management theory on motivation, finding additional empirical support for some economic theories and discrediting others. 2. LITERATURE REVIEW 2.1 Efficiency Wage Theory Literature Review Efficiency wage theory is founded on the idea that it can be profit maximizing for a firm to pay workers a higher wage than the going market rate. This theory means that a firm's production costs are actually reduced by paying a higher wage. In its most common model efficiency wage theory suggests that all other things being equal, effort is a function of wage, and that output is a function of effort. The generally assumed shape for the effort curve is an S shaped curve (Stiglitz, 1976). At the low end of wage, little effort is expended. As wages increase E' >0 and effort increases rapidly in response to premium wages. However, a point is reached where E''<0. Here either the income effect of the high wage takes over, or the ability of the worker to increase effort diminishes. Eventually effort is either flat or declines. The Output resulting from the effort is assumed to have a monotonic Effort Output Wage Effort(wage) Figure 2.1 Wage and Effort Relationship Figure 2.2 Effort and Output Relationship 13 functional relationship and diminishing returns with respect to wage as long as additional effort results in some increased output > 0. 2.1.1 History Efficiency wage theory came out of early work on economic development in less developed countries where the links between wages, nutrition and health were hypothesized. Bliss and Stern (1978) and Leibenstein (1957) present empirical evidence that increases in the nutrition of workers can sufficiently increase the productivity of those workers to outweigh the cost to the employer of the higher payment to workers (Weiss, 1990). Based on this "wage to health hypothesis" and related microfoundations, efficiency wage theory emerged as a microeconomic explanation of real world phenomena that was not appropriately explained with existing theory. Since the time of the great depression and the General Theory, macroeconomics has been focused on the causes and reasons for involuntary unemployment and "sticky" wages. But why are wages sticky? Solow in 1979 described a mathematical model where the optimized wage was required to satisfy a condition such that the elasticity of effort with respect to the wage is unity. This mathematical model is called the Solow condition and the optimized wage was called the efficiency wage (Solow, 1979). In addition to sticky wages, further labor market phenomena which needed explanation included both the dual labor market and wage distribution variability for identical workers. With simple extensions, the efficiency wage models could be made to explain these stylized facts as well. The dual labor market 14 can be explained by finding some sectors where the efficiency wage model applies resulting in higher wages and involuntary unemployment, and some sectors where it does not apply allowing for the standard neoclassical results. Wage distribution can also be explained by having firms with different effort productivity relationships. Also of interest during this research certain conditions were identified which allow for a high road - low road result to be found within the same effort productivity relationship (see Appendix E for this insight and a possible future study). 2.1.2 Kinds of Models A proliferation of models occurred in the early 1980s with publications by Stigliz and Shapiro (1984), Salop (1979), Malcomson (1981) and Akerlof (1982) among others. Under the heading of efficiency wage theory, four model groupings have been created in this research to present the literature used to explain why some firms may pay more than the minimum pay needed to attract employees. 1. They are paid the premium wage in order to minimize turnover costs. If firms must bear part of the costs of turnover (search and training), and if turnover is a decreasing function of the wages firms pay, there may be an incentive to raise wages in order to minimize turnover cost (Salop, 1979). 2. Increasing wages raises workers' effort level (shirking model). Workers who are paid only the equivalent wages of jobs on the outside have little incentive to perform well, since losing their jobs would not be costly. By raising wages, firms may make the cost of job loss larger and thereby encourage good performance (Stiglitz & Shapiro, 1984). 15 3. Workers' feelings of loyalty to their firm increase with the extent to which the firm shares its profits with them. These feelings of loyalty may have a direct effect on productivity (Akerlof, 1982). 4. Firms which pay higher wages will find that they attract a higher quality pool of applicants. If worker quality is not directly observable, this will be desirable (Malcomson, 1981). The theorized stylized facts for the efficiency wage model would generally include the following: 1. Pay lower the first year - future increases are an incentive to avoid shirking. 2. Constantly pay above the market rate in each year. 3. Achieve the labor effort necessary to create above market labor productivity. 4. Lower amounts of supervision can be utilized. 5. Productive surplus is generated which can be shared between higher wages and profits. 2.1.3 The Solow Condition Derivation of the Solow Condition: Under efficiency wage theory a firm's output is given by A firm's costs is given by A firm's profit is given by 16 This will be optimized by taking the first order conditions for w and L and setting them to zero : solving for w: This final equation is called the Solow condition and tells us that with unconstrained labor a firm will pay a wage where the elasticity of the wage with respect to effort is equal to 1. The Solow condition tells us that the efficiency wage is not related to the reservation wage. If the relationship between wage and effort were known, it could be plugged into this equation to determine the profit optimizing wage. 2.1.4 Criticisms Some of the efficiency wage models suffer from a possible theoretical difficulty. More complicated labor contracts can in fact restore pareto optimum equilibrium and return the labor market back to neoclassical standard results without paying above market wages. For example, if all firms required an 17 employment fee, it would provide a substitute nonshirking incentive to the threat of unemployment, and also lower labor costs to equilibrium levels. A number of these types of more complicated employment contracts have been proposed including bonding, fines and seniority wages. However, it is possible that more complicated contracts are not practical or efficient, given the difficulties in communication, interpretation and the resulting misunderstandings (Akerlof & Yellen, 1986). Also, these kinds of challenges to the efficiency wage model result in a moral hazard situation for the firm. The firm would have incentive to declare workers as shirkers, and collect the bonds, fees or fire senior higher paid workers. This, of course, may result in workers being unwilling to accept these more complicated contracts. 2.1.5 Empirical Testing Empirical testing of nutrition and output has been conducted in developing countries where below sustenance nutritional levels have been shown to impact output. As previously mentioned Bliss and Stern (1978) and Leibenstein (1957) present empirical evidence that increases in the nutrition of workers can sufficiently increase the productivity of those workers to outweigh the cost to the employer of the higher payment to workers. Another key work is that of Raff and Summers (1987) relative to Henry Ford's wage. Raff and Summers conduct a case study on Henry Ford's introduction of the "five dollar day" in 1914. Their conclusion is that the Ford experience supports efficiency wage interpretations. "Ford's decision to increase wages so dramatically (doubling for most workers) is most plausibly portrayed as the consequence of efficiency wage considerations, with the structure being consistent, 18 evidence of substantial queues for Ford jobs, and significant increases in productivity and profits at Ford. Concerns such as high turnover and poor worker morale appear to have played a significant role in the five-dollar decision. Ford's new wage put him in the position of rationing jobs, and increased wages did yield substantial productivity benefits and profits" (Raff & Summers, 1987, p. S83). However, this is not a true data test as the study was only an interpretive study with Raff and Summers reading the history and providing their view of its meaning. In developed countries testing the connection between wage and productivity has proven more difficult. Effort is not directly observable, especially in team efforts and coordinated production processes. How to measure effort has been the missing link in previous studies. Unable to directly observe labor effort in common modern team oriented production processes, previous studies have relied on indirect measures like the level of supervision (Abowd, Kramarz, & Margolis, 1999) or on management surveys (Campbell & Kamlani, 1997). This is clearly a weak approach because the level of supervision could be influenced by a large number of causal factors other than labor effort. Included in this group of factors could be cost cutting, supervisory capability and effort, and nonremuneration related loyalty programs. Comparable plant level noncurrency data for measuring labor productivity is hard to come by. Most firms track and maintain historical information only for required accounting systems while productivity data requires collection from multiple tracking systems (hours worked, inputs, outputs etc). Sometimes these data are created in short-term operational systems, but rarely are 19 they summarized and saved in historical systems comparable across plants. That is why the data for this particular study are unique and valuable. 2.2 Institutional Wage Theory Literature Review Lloyd Ulman at the University of California, Berkeley proposed that the development of institutional thought could be described by James Tobin's famous statement about market forces i.e., "market forces do not count; market forces do too count; market forces count for everything" (Kerr, 1977, p. vii). The original modern institutionalist, led by J.R. Commons, focused on historical processes and organizational practices and as a result stressed how market forces were overcome by institutional structure; i.e., market forces do not count. Newer institutionalists allow for market forces and then add institutional rules and boundaries to create a realistic explanation of the observed world behavior, i.e., market forces do too count. The traditional neoclassicalist would contend that market forces count for everything. In discussing the views of the modern institutionalist some of the ideas and explanations that will be covered include:  The role of industry structure and the impact of single firm skills  The Balkanization of labor markets  The two-sector labor market with good jobs and sponge jobs.  The concept of sufficient job satisfaction versus maximization of opportunities.  Managers' obligation to employees. 20 2.2.1 The Role of Industry and Firm Structure Institutionalist have attempted to find the industry characteristics that lead to a firm paying high wages. Krueger and Summers (1988) show that interindustry wage dispersion is large and consistent across time and countries. Their evidence suggests a pattern that some firms pay high wages across and within all occupational groups and some firms have low wages across and within all groups (Leonard K. L., 1987). In fact, they are led to conclude that "The competitive model cannot without substantial modification provide a plausible explanation of inter-industry wage variations" (p. 18). This in fact is a specific answer expressed to one of the key questions raised in the introduction to this research. Krueger and Summers (1988) argue that there are two characteristics which determine whether a firm pays high or low wages. They are, first, whether or not a firm can profitably pay high wages and, second, the level of industry union density. The ability of a firm to pay high wages can be assumed to relate to the intensity of competition it faces. Clearly monopolies, oligopolies and monopolistic competitive firms have more ability to pay high wages than do firms in a truly competitive market. Similarly, firms with a low percentage of costs tied up in labor have the ability to pay high wages. However, this correlation they demonstrate should not be taken beyond what it is -- simply a correlation. If the forces that lead to high pay were applied equally across all firms, only those firms with the ability to afford high pay would respond to these forces leaving the observed result the same. Interestingly, Dickens and Katz find it odd that profits are positively correlated with high wages. Their mind set is back in old thought 21 processes where high wages are a tradeoff to high profits (Leonard K. L., 1987). While empirically valid, Kruger and Summers' correlation is not causation. The second characteristic Kruger and Summers point out is the importance of industry union density. Dickens and Katz (1987) show that union density is correlated with high wages for both union and nonunion employees. However, this applies only to large firms and not smaller firms, even within the same industry and labor pool (Leonard K. L., 1987). Apparently small firms are either not responsive to the union pressure or unions have no interest in organizing the small firms. However, once again, this is not causation, only correlation. Unions have tended to concentrate their organization in industries that were known to have the ability to pay high wages and in fact already paid high wages prior to union organization (Kwoka, 1983). Arthur Alexander is a little more specific in his studies about what industry characteristics lead to high wages. He starts with capital intensity. Capital intensity portends a commitment to the long term and perhaps much industry specific human capital. He also includes product market concentration, size of the firm and unionism as referenced above. Not only do industries have specific characteristics but certain firms do as well. These could include production skills and techniques as well as behavioral and bureaucratic expectations. The reasons that this human capital would be firm specific could include unusual or unique production capabilities and or unusual or unique cultural environments. This would be amplified in a privately run company where owners were personally involved such as in the case study for this research. When developed skills may only be used within an individual firm it creates an important worker-firm 22 attachment. In this case it is not efficient for either the firm or the worker to shift employment. Since skills already acquired have sunk costs, turnover will not be called for even in the case of temporarily reduced demand (Kerr, 1977). This discussion around a worker-firm attachment is an institutional version of efficiency wage. It does not make economic sense for turnover to take place and therefore must be avoided. The difference is that the situation derives from the institutional framework rather than self maximizing behavior. 2.2.2 Balkanization The term "Balkanization of the labor markets" was introduced by Clark Kerr in his famous essay by the same name. There is no single labor market and the fragmentation of the labor market is evident across industries and geographies. There is a natural structure logically within labor markets because of differences in skills, geography, preferences and information. These natural frictions are supplemented by the introduction of additional institutional rules which totally fracture the market. Kerr (1977) states, "Institutional rules in the labor market, as we have seen, establish more boundaries between labor markets and make them more specific and harder to cross. They define the points of competition, the groups which may compete, and the grounds on which they compete" (p. 37). Rules can be formal or informal, issued by employers, groups of employees (including unions) and government. Formal rules can be anything from laws against discrimination to union contracts concerning providing employees. Informal rules can include demands from the employer such as no moving costs will be paid, to 5 years of experience will be required. Included in both the formal and nonformal rules will be the 23 far reaching dictate of honoring seniority. Seniority becomes especially important because it ties employees to a certain firm or union. It will provide promotion and security and cannot be transferred with employees when they move locations. These seniority rules, in particular, undermine transfers between jobs. And without free and easy transfers between jobs the idea of the market mechanism breaks down. The additional impact of these institutional rules and balkanization of the labor market is that there are ins and outs. Those making the rules take care of their own. They have created "sovereignty over a job territory." Fraternity triumphs over liberty as "no trespassing" signs are posted in many job markets (Kerr, 1977, p. 25). This leads directly to a discussion of the two-sector labor market. 2.2.3 The Two-Sector Labor Market The dual labor market theory divides the labor market into two sectors, the primary labor market and the secondary labor market. The primary labor market has high pay, stable jobs and promotion opportunities. It also provides a high return to education and years of service (Dickens, 1987). The secondary market provides low pay, low job stability and no promotional opportunities. The theory further argues that the primary market is heavily influenced by institutional factors while the secondary market is primarily driven by the free market. The argument has been made that the primary market arose out of the movement to oligopolies at the end of the 19th and the beginning of the 20th century. As these large oligopolies arose, they turned their attention from short-term competition to long term market dominance. This coupled with a large capital investment sparked the interest in employee stability and development. Personnel departments were created, benefits 24 offered for seniority and job ladders created, and firm and industry specific skills were developed (Reich, 1973). The secondary market controlled by market forces is often seen as relying on immigrant labor and product demands that are more seasonal, cyclical and do not require the level of capital investment that occurs in the primary market. 2.2.4 Sufficient Job Satisfaction In weighting the merits of neoclassical versus institutional theories against the structure of the labor market a key question is does the labor market operate with free competitive exchange or are their sets of rules, constraints and norms that dominate. This section will explain the institutional constrains around mobility and flexibility as opposed to a free flowing open market. The vast majority of workers are not actively looking for a job. Some of them may express interest when approached but even then many will express satisfaction with their current job. "They are attached to their jobs or their areas as in a marriage contract" (Kerr, 1977, p. 5). The job may not be the best job in the world, but it is theirs! The personal rewards to work are much more than wage. And sufficient job satisfaction is not just about avoidance of the real unpleasantness of finding a new job match. The personal readjustments are real and difficult in job transfers but a much stronger issue is the alternative reward elements. Workers feel loyalty and attachment to their jobs. They take pride in the progress that has been made in their firm and feel an ownership and sense of community with the other employees in terms of moving forward. The job, with its good and bad, often becomes a part of the employees' identity and sense of self-worth. Some employees have been known to return to the factory cafeteria for lunch each day after their retirement because of the emotional and 25 social attachment to their employment. The primary studies relating to this behavior are related to management studies around employment satisfaction and rewards. The behavioral management studies will be addressed in Chapter 5. 2.2.5 Manager's Obligation to Employees Managers often view employees as their "team." Lee Iacocca's (1986) assertion that "the chairman [of a publicly held company] is morally accountable to his employees…" (p.284) is a far cry from Friedman's argument that the manager's only role is to provide profits to the shareholders (Friedman, 1970). Much literature has been developed on reciprocal exchange and gift exchange. Managers and employees obtain a psychological utility by working with and supporting each other (Kranton, 1996). The idea of a manager's moral obligation to employees is a social cultural norm that can be developed within societies and institutions. This idea of "fairness" is supplemented with the idea of social connection. Not only is there a moral obligation but there is also a social obligation to certain employees. This connection has even been shown empirically. "We find that when managers are paid fixed wages, they favor workers to whom they are socially connected irrespective of the worker's ability, but when they are paid performance bonuses, they target their effort toward high ability workers irrespective of whether they are socially connected to them or not" (Bandiera, 2009). The works cited above would show the following institutional factors as important in wage determination:  Profitability or intensity of competition/oligopoly  Union density 26  Size of firm  Capital intensity of firm  Industry or firm specific knowledge  Bureaucratic hiring rules relative to experience, personality traits and seniority  Sufficient job satisfaction and cultural "marriage contracts"  Closeness of managers to workers - social and joint output responsibility 2.3 Power Relationship Wage Theory Literature Review A final theory of wage setting is centered on the power relationship between worker and employer. In certain forms this theory can be traced all the way back to the father of economics Adam Smith. In speaking of labor and wages he says, "It is not, however, difficult to foresee which of the two parties must, upon all ordinary occasions, have the advantage in the dispute, and force the other into compliance with their terms" (Smith, 1981 (1776), p. 27). Adam Smith's use of the words "dispute" and "force" are clearly far removed from the neoclassical view of dispassionate optimizing of personal utilities on both parties part. Smith is clearly recognizing a strong social undercurrent present in economic relationships which not only supplements but can overpower the rational optimizing economic man. Perhaps best known in this arena is the significant work that has developed around Marx's theory of social struggle and surplus value. 27 2.3.1 Marxism Marx notes that the wage actually paid a given worker may be temporarily above or below the normal wage. Supply and demand forces cause oscillations above and below a "certain mean." The mean is the "natural price . . . determined independently of demand and supply." Marx also provides an answer to the question, What determines the height of this mean? What forces establish the "natural" or the equilibrium wage toward which actual market wages are incessantly drawn despite their temporary departure (ScarLett, 2008, para. 5). While Marx's theory of wages started with the view that capitalism would drive wages to a subsistence level, he asserted that the average wage is "that quantum of the means of nourishment which is absolutely requisite to keep the laborer in bare existence as a laborer" Communist Manifesto (1848, p.1). Subsequently he then began to add additional items of historical or cultural need to the definition of subsistence. In 1865, during his address to the General Council of the First International, Marx presented his fully developed theory of wages. There is still a minimum limit on wages that is basically a physiological minimum but there is no maximum. The determination of where a given wage will lie … is based on the respective powers of the combatants. This seems very similar to the class struggle Marx details in the Communist Manifesto. (Bigelow, 2008, para. 5) In Marx's world the power struggle between the capitalist and the worker over the produced surplus value is reflected through the wage. This view of the power relationships would imply that wages going up and down would be a result of shifting power. Power could shift in many ways including the threat of unionization or strike, a significant shock to demand or supply of 28 workers, or legal requirements impacting management approaches to managing workers. There is a new and different strain of Marxist understanding developing around the labor-capital relationship. This new view argues against two of Marx's key assumptions that end up equating wage labor to slavery and the relationship of worker and employer to unavoidable revolution. This new strain is explained by Laycock (1999) as a view that labor power is not bought and sold as a commodity and that its use in creating surplus value is not a mystery to the laborer. Laycock explains that the transaction of providing wage labor to an employer is one of rental or lease rather than one of buying and selling and that it does not "masquerade as a commodity-exchange (per Marx)" but is fully understood by both employer and employee. The result of Laycock's argument is that exploitation and expropriation of surplus value is not tied to Marx's labor theory of value, but instead is related to the unequal distribution of capital and land, or the moral injustice of unequal property distribution. This results in a power conflict that is less about the work relationship and wages than it is about the class structure of society itself. According to this view there would be less work conflict over wages as employers and employees share many goals at the place of employment. Instead there would be more social conflict outside of work. The impact on power relationships regarding wages would be somewhat lessened. Empirical studies have shown that power relationships can make a difference in wages. "Strong unions, centralized wage bargaining, a large public sector and left 29 governments have muted and sometimes overcome inegalitarian tendencies." (Pontusson, Rueda, & Way, 2002, p. 307) 2.3.2 Unionism While many of the Marxist discussions are quite theoretical, the academic work relating to unions is very much applied. Here bargaining theory, game theory and negotiation theory are all about achieving results. Despite the current decline in union membership, union bargaining is both historically and currently very important to the world of wage setting. While union density has decreased, collective bargaining coverage is still quite high because in most European countries collective agreements prohibit differentiating between union and nonunion employees. According to the OEDC (1997) union density for member countries is 44% but collective bargaining is 68% (Cahuc & Zylberberg, 2004). Union bargaining theory has developed significantly over the last 2 decades and includes new trends in institutions and the environment which are significant. Centralized wage bargaining which has been found to be the most effective in eliminated wage dispersions is becoming less and less relied upon in the current more complicated and global business environment. Movement toward a more flexible wage settlement is being used by powerful employee groups and unions alike (Pontusson, Rueda, & Way, 2002). Other reasons for the significant change in the approach to union negotiations are the significant decline of union coverage in the private sector, and also the considerable changes in management approaches. Clawson and Clawson (1999) describe the new environment for labor union negotiations and some of its 30 impacts. Perhaps the biggest management change is the move to participatory management. This would include quality circles, team processes and organizations. These management approaches solicit worker input and joint ownership of goals and results, which clearly work against animus between employee and employer thereby creating a different negotiating environment for unions. The widespread use of these approaches has encroached upon anticompany union rules prompting some change to existing regulations. The other main change that Clawson and Clawson identify is the alliances that unions are forming with other grass roots organizations, from political action committees to community organizing groups. The main result of the new environment is an attempt by unions to move from a servicing model back to a militancy model. As a servicing model, distinctions between labor unions and company unions can be blurred, but as a militancy model labor unions can stand independent of company management initiatives (Clawson & Clawson, 1999). These large environmental changes imply that the union negotiation process is also in flux and possibly blur the union impact on wages with other changes. Bargaining theory perhaps goes all the way back to Edgeworth (1881) where parties divide goods in a pareto optimal way, and then continues with Nash (1950) and Stahl (1972) and Rubinstein (1982) where game theory is developed. If the rules of the game (utility preferences of the players and alternative options) can be described, a best case equilibrium can be modeled. Models around bargaining have been created and incorporated into economic theory. They include the monopoly union model, markup and union 31 power model. These models can result in efficient outcomes under certain assumptions (Cahuc & Zylberberg, 2004). Empirical studies suggest that unions do in fact increase wages (Cahuc & Zylberberg, 2004). The three groupings of economic wage theories each describe a different structure and function to the labor market, and as a result end up answering the question, "Why do some firms pay more than the market wage rate?" differently. Because the neoclassical theory predicts no differences in wages, it requires the efficiency wage theory or agency theory to explain how differences in wages can be justified. These appendages remain neoclassical because they rely on the rational man maximizing utility. The institutional theories can explain the difference in wages many ways by referencing the structured rules of the market and the cultural and social objectives that participants honor. The power relationship theories explain an above market wage by recognizing some level of negotiating power that specific workers have achieved. 3. TESTING EFFICIENCY WAGE THEORY 3.1 Overview of Efficiency Wage Test Methodology Chapter 2 identifies several versions of the efficiency wage model. The most widely used and discussed is the shirking model which relates wage to effort (Stiglitz & Shapiro, 1984). This is the model that will be tested econometrically in this chapter. The other models (the turnover model (Salop, 1979) and the quality model (Malcomson, 1981)) related heavily to corporate culture and will be tested in the management survey outlined in Chapter 4. The element of efficiency wage theory that has always eluded empirical work is the concept of effort. Employee effort can be readily seen in piece rate or task oriented jobs. However, in most of our modern work environments effort is unobservable as large teams work together in interconnected roles to produce common output. This has been the missing link in previous efficiency wage studies. Unable to directly observe labor effort in common modern team oriented production processes, previous studies have relied on indirect measure like the level of supervision (Abowd, Kramarz, & Margolis, 1999) or management surveys (Akerlof , 1982). This is clearly a weak approach because the level of supervision could be influenced by a large number of causal factors other than labor effort. 33 Included in this group of factors could be cost cutting, supervisory capability and effort and nonremuneration related loyalty programs. This study will avoid the issue of directly measuring effort by looking instead at output. If Output = f(e) and Effort = E(w) then Output = f(E(w)). It is assumed that throughout the relevant range that the first derivative of both f(e) and E(w) is positive, giving output a monotonic and positive relationship with wage. As a result wage premium's impact on output can be measured directly if all other variables can be controlled. In this case control will be afforded because the production lines used are substantially similar. They are owned and operated by the same company, with the same manufacturing philosophy and policy, with the same technology and making essentially the same products. This will allow us to compare communal labor effort from line to line based on output. The only assumption here is that the level of effort is necessary and sufficient to raise output if the production processes are substantially similar. Because of the direct production information and pay for similar lines across multiple geographies, a production function can be built and a tested for the impact of wage premiums on productivity. It should be noted that these lines are continuous flow process lines and are not assembly lines. Assembly lines can manage effort from employees by managing the speed of the lines. Continuous flow process lines do not require employees to manually interface with the lines to complete processes. Their speeds are contingent upon the engineering of the line rather than the effort of the associated employees. The associates in a continuous process provide equipment monitoring and 34 adjustment; they coordinate the timing of supplies and provide quality checks. Of course in times of equipment failure they provide whatever manual support they can. With extra effort they can improve uptime, reduce waste, reduce start up times and provide a range of improvement suggestions. The steps in this test will involve first choosing an appropriate production function and then testing if wage premiums play a role in the chosen production functions. A production function should be chosen first to ensure that the processes of the production line are appropriately represented in real terms (no dollars). This will assure that the production function, which may or may not be a good fit with the wage premiums included, is an appropriate model. The specific statistical approaches and data use will be discussed in detail as necessary. The other issue of concern goes back to the wage premium of the individual versus the group. At its core, efficiency wage premium is about an economic individual responding to personal incentive. Can the output of a group and the wage premium of the group be measured in the same way? If the wage premiums of each individual within the group go up and down together, then the incented response of the group should be directionally the same as the response of the individual even if not identical in magnitude. As will be explained in the data sections on Mars payroll data, and the BLS base pay data, the data are not available for a true individual comparison, and of course the ability to measure output (a necessary part of this test) is impossible at an individual level given the group dynamics. An overall sense of the group versus individual can be obtained from a comparison of the payroll premium by category of worker across years and across 35 plants. This comparison is shown in Table 3.1. The categories of High, Medium and Low correspond to the pay zones of employees as explained in the payroll data section. These results show a high level of consistency in movement - all premiums increase between 1993 and 2000. They also show horizontal consistency with 75% of the year to year comparisons having the same rank or position of premium within location. Because the data set is small the standard deviation is large. A statistical comparison would show all the data points within the same 95% confidence interval, however this is not particularly meaningful as the interval would also include negative wage premiums. The consistency of the data within and between the groupings gives support to the idea that group wage premiums can be used along with group output to measure the production input of the wage premiums. 3.2 Plants and Data 3.2.1 Manufacturing Plants The manufacturing plants used in this study have been selected from four cities, Albany Georgia, Cleveland Tennessee, Chicago Illinois and Waco Texas. Cleveland Tennessee is near Chattanooga and will be referred to as Chattanooga in the rest of this study. These cities represent a diverse set of locations. None are Table 3.1 Wage Premium by Pay Zone WACO ALBANY CHICAGO CHATTANOOGA 1993 2000 1993 2000 1993 2000 1993 2000 HIGH 23% 47% 10% 42% 7% 15% 28% 36% MEDIUM 51% 61% 52% 59% 24% 55% 36% 81% LOW 26% 37% 49% 87% 13% 69% 20% 23% 36 especially high income areas (normal for the location of manufacturing plants). But Chicago and Chattanooga enjoy significantly higher income and are larger cities than either Albany or Waco. Chicago is heavily unionized with a significant manufacturing base. The other locations do not have a strong union presence. Each area has a sufficient manufacturing base and civilian labor force to provide a competitive labor environment. While smaller, Albany has a major Coors brewery and Procter and Gamble, and Waco has Raytheon, Cargill and Pilgrim's Pride among other manufacturers. The location demographics are shown in Table 3.2. The plants are all well established and range in age from over 60 years old to over 30 years old. A number of the plant production workers at each site have been employed for 20 years or more. During the 1990s each plant was operated by a senior management team composed of a plant manager, an engineering director, an operations manager and a finance and human resource manager. The Albany Table 3.2 Location Demographics ALBANY CHATTANOOGA CHICAGO WACO GEORGIA TENNESSEE ILLIOIS TEXAS YEAR SOURCE City Limits Population 72,092 153,487 2,661,688 108,689 2006 Fact Sheet Metro Area Labor Force 54,000 235,039 4,137,250 101,874 2000 EEO Data Tool Mfg Firms with >20 Employees 70 370 2,185 153 2007 Economic Census Median Family Income 38,772 50,282 53,381 39,791 2008 Fact Sheet Median FI Ratio to US average 61% 80% 84% 63% 2008 Fact Sheet Labor Force College Graduate 22% 24% 30% 19% 2006 Fact Sheet All data from the U.S. Census Bureau Web Site www.census.gov 37 plant was built originally to process peanuts for use in other plants, but also now has lines that produce filled pretzels and sugar candy. Chattanooga is a very large plant that makes a variety of M&Ms candies and Twix bars. Chicago makes a variety of filled bars including Snickers, 3Musketeers and Milky Way Bars. Waco produces Snickers bars and sugar candy under the brand names Starburst and Skittles. Each plant also has a chocolate plant that turns chocolate liquor, cocoa butter, sugar and milk into finished chocolate. The basic production technologies have been in place for many years at each plant. However new technologies are added into the lines to improve quality or efficiency every year. Most of these base technologies can be purchased off the shelf in ready assembled lines. However the off the shelf lines would be much smaller and would not operate to the level of specifications that the Mars lines have been built up to over the years. The technologies can be grouped as bar lines, chocolate plants, sugar lines and shell coated products. Bar lines produce such products as Snickers, Twix and Milky Way in a variety of sizes and packages. The lines are long continuous processes. Nougat, caramel and other ingredients are continuously cooked and formed into a slab on a moving conveyor belt where the product is cooled, sliced, chopped to size and then enrobed in chocolate, decorated and cooled. Given the need to heat and cool the product, the lines can be physically quite long so that the temperature changes can be achieved without slowing the line down. Wrapping is done in high speed single lanes and then the product is bagged or placed in a carton and then put in a case and on a pallet. 38 The role of employees in working on these lines is to observe the operation of the equipment, to adjust speeds and settings and to address emergencies. Additional employees provide the continuously needed supplies and maintain and repair the equipment. From time to time mechanical problems arise which require manual intervention until the problem can be fixed. Chocolate plants are highly automated and require primarily servicing of the equipment and supplies. The sugar lines have very automated processes but require significant manpower in forming and packaging due to the wide variety of packaging and the very high speeds at which the small pieces are processed. Coated or shelled products include M&Ms and Skittles. These products are originally formed as a sheet of nibs or lentil shaped centers. These nibs are turned in a coating pan batch process while multiple layers of sugar shell are applied. The coating process is very technically sensitive to temperature and humidity and requires close monitoring by employees. Additional capital is required to process multiple colors/flavors until final blending. The high speed packaging machinery also requires a significant number of employees to monitor and maintain during operation. The process lines all exhibit certain characteristics. They operate with a variable relation between capital and labor. For all the lines labor can change as shifts change, for example, from three staffed shifts to four while capital remains constant. Also problems or improvements can be addressed with only labor or capital rather than always requiring equal amounts of both. The capital and labor are partially substitutable and partially complementary. The line cannot run 39 without employees and employees cannot produce the product without capital. However, additional capital can replace employees, or additional employees could be used for some operations instead of automating. Because the lines are very large and expensive, additional capacity is regularly achieved by eliminating bottle necks with capital or labor instead of building new lines. New lines are added only irregularly. The data for this study include 23 lines, and during the 10-year period three lines are retired and three lines are started. Because the lines are well established and operate in a consistent range of capacity, over the period of the study, a single return to scale is appropriate regarding labor and capital, rather than a life cycle concept of multiple returns to scale. Labor and capital are added to the lines in small increments of a few percentage points difference per year rather than fundamental changes in operations. The large changes are achieved by adding or subtracting shifts of operation which is near a constant return to scale relative to labor but requires no additional capital. The data used in the efficiency wage test come from two sources, first payroll files from the plants and second centralized production reporting, gathered from the plants. 3.2.2 Payroll Data During the study period Mars placed all nonexempt jobs within a pay zone. While five zones exist, the vast majority of employees operate from within three zones. These three zones are subsequently called high, medium and low. Each zone 40 has several steps of pay (approximately four). Employees (called associates by Mars) are paid at one of the steps within the zone for the job they have. Over time an associate can move up the steps within their zone as long as they are performing well. Once an associate reaches the top step in the zone they will receive no more pay increases unless they are promoted to a different job at a higher zone, or the company changes the pay scale within the zone. Since turnover is low, over time most associates working within the same job end up being paid at the top of the zone and are paid the same amount. It is common to see the same pay for multiple associates at each site. Payroll is made weekly and the payroll process is conducted at each individual plant. The data available for this study are associate number and hourly pay for all nonexempt employees from each plant's payroll. This does not include benefits, vacation or bonuses. But this base amount is the correct amount to compare to the Bureau of Labor statistics hourly pay surveys because the government data also do not include these benefits. There is one small wrinkle in the Mars payroll data. The plants all pay their associates a 10% bonus if they are ontime for work. This is paid out with the weekly pay over 95% of the time and is thought of by the employees as part of their pay. While the raw numbers will remain unchanged, in calculating pay premium versus average local pay the on-time pay bonus will be included. A summary of pay data by site is shown in Tables 3.3 and 3.4 and in Figure 3.1. Average pay is the average hourly rate of all nonexempt plant associates. High and low pay represents the extreme hourly wage in the data. 41 Table 3.3 Mars Payroll Data Table 3.4 Percent Change in Average Wage CHICAGO Headcount Wage Headcount Wage Headcount Wage Headcount Wage Headcount Wage Average 462 13.69 489 14.74 386 18.87 418 18.88 448 20.33 High Wage 19.17 20.22 23.93 24.29 26.11 Low Wage 9.16 9.86 13.22 12.94 12.44 CLEVELAND Average 517 12.04 522 12.95 657 15.27 635 15.87 666 17.09 High Wage 15.74 16.43 19.48 19.78 21.26 Low Wage 6.36 6.36 8 11.47 10.95 WACO Average 582 12.02 635 12.33 709 14.43 755 14.2 666 14.16 High Wage 15.99 16.72 20.48 21.83 24.64 Low Wage 7.27 7.27 6.27 6.77 7.27 ALBANY Average 291 12.93 296 13.83 306 16.63 319 16.72 281 16.96 High Wage 16.45 17.63 20.87 21.18 22.77 Low Wage 8.99 9.63 9.24 12.77 10.08 1993 1994 2000 2001 2005 1994/1993 2000/1993 2001/2000 2001/1993 2005/2001 Chicago 7.7% 4.7% 0.1% 4.1% 1.9% Cleveland 7.6% 3.5% 3.9% 3.5% 1.9% Waco 2.6% 2.6% -1.6% 2.1% -0.1% Albany 7.0% 3.7% 0.5% 3.3% 0.4% 42 Note: estimates between the data points are straight-lined. Figure 3.1 Average Pay by Plant 3.2.3 Production Data One of the company's owners was a trained engineer. He also believed very strongly that lasting competitive advantage could only be obtained through manufacturing excellence and technology. As a result he required every plant to submit engineering plant performance along with financial data. The unique data are available for approximately a 10 year period in the company's centralized historical repository. Mars calls this information the A30 report. The report was filled out by the financial employees at each site with input from the site engineers. Things like the headcount per line and capital spending by line would be obtained 43 from the financial records, while things like standard operating capacity would be provided by the engineers. The key pieces of data used in this study end up being: Output (Q) is measured in metric tonnes. A metric tonne is approximately 2204 lbs. While a tonne of Snickers bars may not have the same dollar value as a tonne of M&Ms it would be a consistent output variable within types of lines and even across lines relative to the inputs. The metric used is yearly total output moved into finished goods (this excludes any waste or rejected product). Headcount (L) is reported in equivalent full time people and is tracked through the payroll system. The number used is the average of weekly headcount for the year. Standard operating capacity (K) is also measured in metric tonnes. A consistent set of rules is used by all plant industrial engineers to calculate the standard operating capacity. Standard operating capacity is used as a proxy for capital because of several reasons. The first is that dollar value of capital is recorded in the accounting records at cost and so equipment that is 30 years old would not have a comparable value to recently added equipment. The second is that much of ongoing capital spending is related to issues of safety and quality which do not affect output quantities. Standard operating capacity shows the impact on output quantity of capital used on the line without regard to labor. The following chart, Figure 3.2, shows the data points relating output tonnes to standard operating capacity for all lines. This chart demonstrates a consistent overall relationship between standard operating capacity and output tonnes. 44 Figure 3.2 Output and SOC The next chart, Figure 3.3, shows the data points relating output tonnes to headcount for all lines. Because there is a very limited relationship R squared of 7% it is apparent that the production lines have different headcount relationships. The next two graphs, Figures 3.4 and 3.5, show the same data limited to Bar Lines and Shell Coating lines which gives a much improved relationship (R squared 28% and 83%). These figures demonstrate that the different technology groups have different headcount relationships. In order to capture this difference the fixed effects technique using lines as a grouping proves beneficial in the econometric production modeling. The following chart, Table 3.5, shows the data that are available for use in the study, by line by year and a range for each of the data points. This demonstrates the breadth of the data as well as the variance sufficent for regression analysis. 45 Figure 3.3 All Lines- Output and Headcount Figure 3.4 Bar Lines - Output and Headcount 46 Figure 3.5 Shell Coating - Output and Headcount Table 3.5 Analysis of Production Data A30 PRODUCTION DATA Plant Line DATA Volume Assets Labor K/L Y/L Albany Baked Filled Ln 1 93-2002 122% 74% 33% 34% 106% Albany Baked Filled Ln 2 93-95 22% 24% 13% 9% 29% Albany Peanut Process 83-2002 21% 55% 49% 99% 67% Chicago Filled Bar Ln 1 Fs/S 93-2002 82% 115% 132% 39% 54% Chicago Filled Bar Ln 2 Sng 93-2002 81% 122% 73% 49% 32% Chicago Filled Bar Ln 3 Snck 93-2002 143% 84% 237% 266% 136% Chicago Chocolate Plant 93-2002 18% 98% 61% 102% 44% Cleveland Flex Line 93-2002 114% 63% 89% 31% 180% Cleveland International 93-2002 231% 20% 700% 684% 224% Cleveland M&M Peanut 93-2002 18% 29% 18% 30% 35% Cleveland M&M Plain 93-2002 77% 60% 48% 37% 40% Cleveland Twix I 93-2002 594% 22% 119% 118% 416% Cleveland Twix II 95-2002 500% 55% 120 71% 173% Cleveland chocolate Plant 93-2002 48% 37% 61% 69% 63% WACO Line 1 95-2002 122% 63% 57% 123% 134% WACO Line 2 Snickers 93-2002 30% 133% 70% 55% 53% WACO Starburst 93-2002 80% 87% 55% 113% 101% WACO Skittles 93-2002 64% 32% 35% 60% 81% WACO Chocolate plant 93-2002 24% 90% 59% 95% 87% Source Mars A30 production reports 1993-2002 RANGE OF DATA (Max-Min)/Min 47 3.3 Choosing a Production Function The purpose in this study of choosing a production function is to allow a drilled down look into the role labor plays in producing output as a basis for testing the efficiency wage theory. Efficiency wage theory suggests that all other things being equal, effort is a function of wage, and that output is a function of effort. The Output shape resulting from the effort is assumed to have a similar shape to a traditional production function, i.e., additional effort results in some increased output > 0, with diminishing returns. A production function must be selected that provides a reasonable explanation of labor's role in the output process so that the theories around wage, effort and output can be tested. Given a production function, the role of labor can be defined mathematically with a marginal product of labor, a technical rate of substitution and the elasticity of substitution. The selection of the function will be accomplished using a two-step process. Step 1 : Pick a functional form that logically matches the design characteristics of the production lines being tested. Step 2: Run regressions of the various acceptable models to see which ones provide the best empirical fits. The judgment of functional fit is usually accomplished by measuring residual errors as either root mean squared errors or mean absolute percentage errors (Wilson, 2009). In addition the coefficients will be measured for significance and the overall formula for reasonability. 48 3.3.1 Step One - Logical Match The various functional forms for production functions have different characteristics built into the mathematics. These would include the inputs coming in fixed or variable proportion and being either complements or substitutes. The functional form would also include the returns to scale being limited to a single form or having the ability to change from diminishing returns to increasing returns within the same function. The features of difference would also include how technology and scale are applied to the process. For a more complete discussion of the features of various production functions refer to Appendix D. Please also refer to the description of the Mars production lines and the data in the previous section. The features of the Mars production lines being studied would clearly include: a. having variable proportions of inputs possible and b. the inputs can act as either substitutes or complements and c. a single return to scale (approximately constant) would be used through the entire production function. Given the above we can eliminate Leontief and Additive1 production forms from consideration. We are left with Cobb-Douglas, Constant Elasticity and Translog formats, which can fit all of our logical constructs. These formats will be tested. 1 See section 3.3.2.4 for a revision of this conclusion. 49 3.3.2 Step Two - Empirical Fit The data are all in physical measures with no dollar values. Production tonnes are used for volume, standard operating capacity is used as a proxy for capital2 and equivalent headcount is used to measure labor. The data are edited for obvious errors in the standard operating capacity (see data description section for detail). No production lines have been eliminated for the overall fit even though some of them have high estimated errors. There are several options which can assist in fitting the functional forms. Simple linear regression can be used, or panel data techniques can be used. Because the data are structured by years for each line and plant, the data fit the classical panel data format. Classical linear regression and nonlinear techniques will give a slope and intercept that fits the data as a whole. However there may be different relationships within an individual plant or line (panel data structure) that are "averaged" with this classical approach. This difference has been previously referenced in the data section of this chapter. The different effects that lines or plants might have on the relationships can be teased out using panel data techniques called fixed or random effects. It can easily be confirmed that the relationships within a plant or line are significantly different from the overall data relationship with the following example. Table 3.6 shows the correlation coefficient between volume and headcount for the overall data and for two specific lines. 2 If dollars were used to measure capital, then there would be issues with age of the assets and accounting book value as well as the fact that some assets are put in place for safety or quality measures which do not impact o utput tonnes. 50 Table 3.6 Correlation Between Volume and Headcount Data Set OLS coefficient t statistic Overall .0788 3.98 Chicago Line 1 .1095 8.15 Waco Chocolate -.247 2.35 Table 3.6 shows significant, (t statistic) but very different, relationships. These differences should be captured in the econometric fits of the production function. The fixed effect model for the Cobb-Douglas form is as follows: Ln (volume)it = Ln(technology) + αLn(Capital) it + βLn(Labor) it +C i + ε it Where i represents the fixed effect group and t represents time. Here Ci represents the impact of the specific group. This essentially allows for a different intercept (technology) for each group. A further test can be conducted to determine if the dummy variables add significance to the resulting equation. Random effects are based on a different assumption. This technique assumes that the data are a sample from a normally distributed population. As such the error term could be different within each grouping. The most likely error terms by group are calculated. The random effect model is as follows: Ln (output)it = Ln(technology) + αLn(Capital) it + βLn(Labor) it +u i + ε it where ui is normally distributed. An equivalent interpretation is that the fixed Ci's are distributed along a normal distribution. In deciding whether random or fixed effects is a better way to capture the group impact from the data, Greene (2010) identifies some key differences as:  Random Effects 51 o Small number of parameters o Efficient estimation o Restrictive assumption for independence of errors.  Fixed Effects o Robust - generally consistent o Large number of parameters (can handle without losing degrees of freedom) The heart of the problem in deciding between fixed and random effects is whether the explanatory variables are correlated with the error terms - endogeneity. If there is any correlation then fixed effects are preferred because random effects are not consistent. Fixed effects is preferred for this exercise for several reasons: 1. The data set is structured as a population rather than a random sample 2. The number of groups is small (4 plants and 21 lines) 3. The assumption of normality may be suspect given the limited data points per line The fixed effects group used will be production line allowing for a different intercept to account for differing technical and structural differences by production line. Another approach which will be used in the fitting process is to fit all the functions in a natural logarithmic form. The Translog production function is already in a logarithmic form. Cobb-Douglas will need to be linearized by converting to logarithmic form in order to use least squares regression. Constant Elasticity cannot be linearized and will need to be solved using nonlinear estimation. Because the 52 Cobb-Douglas and Translog forms will be solved using logarithmic formats it is important that all of the equations do so. The estimates, residuals and the R squared results cannot be compared if some equations have the dependent variables as a natural log transformation and others do not. Wooldridge (2009) explains the importance of this by showing that the exponent of the predicted log Ŷ will systematically underestimate the expected value of Y. The statistical description of the data used for these empirical fittings is shown in Table 3.7. Given the statistics for Kurtosis and Skewness we will not reject the hypothesis that the data follow a normal distribution. 3.3.2.1 Cobb-Douglas Functional Fit The Cobb-Douglas is perhaps the most common production functional form used. It provides for a complementary relationship between capital and labor and Table 3.7 Descriptive Statistics for 10 Years of Production Data VOLUME SOC HEADCOUNT Mean 27.5 39.4 115.6 Standard Error 1.3 1.9 4.3 Median 29.0 36.9 121.0 Mode 33.3 13.8 136.0 Standard Deviation 17.7 26.9 61.1 Kurtosis -0.3 -0.2 0.2 Skewness 0.4 0.7 0.3 Range 72.8 100.0 304.0 Minimum 0.7 2.9 3.0 Maximum 73.5 102.9 307.0 Count 199 200 199 53 will provide an output of zero if either capital or labor is zero. However, the Cobb- Douglas form forces the function to have a constant elasticity of substitution of 1. Because there is no reason to assume that this restriction is appropriate for our dataset, this constraint may limit the results of this model from achieving the best possible fit. The regression model for Cobb-Douglas is fit using logs of all data because of the nonlinear structure of the format. Ln (output) = Ln(technology) + αLn(Capital) + βLn(Labor) + ε The resulting Fixed Effect equation converted back into standard form is as follows:  The technology factor is .226 (average of fixed effects)  The exponents add up close to 1 which implies constant returns to scale. This can be tested as follows: F( 1, 154) = 0.58 Prob > F = 0.4492 as a result the hypothesis that the coefficients equal 1 cannot be rejected.  R squared is 90%  The statistical significance of the exponents and intercept are high, above 99%.  The coefficients and t statistics are shown in Table 3.8 Another key test for the fixed effects model is to determine if the model with fixed effects is more significant than without. This significance can be confirmed with the Table 3.8 Cobb-Douglas Estimates Coefficient Estimate t statistic Capital exponent 0.698 8.15 Labor exponent .457 9.04 Intercept - technology .226 4.18 54 F test of the hypothesis that all fixed effects are zero: F(22, 174) = 9.4 Prob = 0.0000. Clearly the Fixed Effects add explanatory power to the model. 3.3.2.2 Constant Elasticity of Substitution Functional Fit Volume The Constant Elasticity of Substitution functional form is also commonly used. It has fewer constraints than the Cobb-Douglas because elasticity of substitution is not restricted to 1. However, it is required to be constant throughout the entire function. This restriction may also prevent us from achieving the best possible fit. The Constant Elasticity of Substitution Function exhibits constant returns to scale if B3 in the function above is equal to 1. The Constant Elasticity form also contains several well know functional forms within it. If B3 and B2 are both equal to 1 then the isoquant is a straight line defining a linear production function. If B2 approaches zero then the form will take a Cobb-Douglas form (Varian, 1992). If B2 approaches negative infinity then the form will take a Leontief form of perfect complementary inputs. Using a nonlinear iterative function which minimizes the sum of squared residuals the following equation was derived for the Constant Elasticity Production Function. Ln version of CES: ln volume ln ln RESULT: ln volume ln ln L Converted into standard format: Volume 55 The important elements in this function include the following:  The scale factor .1518.  The weight of factor inputs - here capital is given a heavy weight of 91.2 while labor has only a 8.8 weight.  The coefficient B2 used to calculate the constant elasticity of substitution factor is 1.141. This would imply an elasticity of substitution of  A negative elasticity of substitution is meaningless and results from the fact that this equation assumes concavity in the function and in this case the function is convex (Mos-Colell, Whinston, & Green, 1995). Because this factor is close to 1 (1.141) we have an almost flat plane in three dimensional production space. However, there is some curve, it is just small.  If B3 were equal to 1 then the function would have constant returns to scale. Because B3 is more than 1 (1.32) we have increasing returns to scale.  Despite the previously mentioned objectionable theoretics, this function has a very good fit, as identified by a 99% adjusted R squared which means 99% of the deviation from the average output is explained by the estimated equation.  The t statistics for the derived coefficients also provide a high level of confidences (99%)  The coefficients and t statistics are shown in Table 3.9 56  Table 3.9 Constant Elasticity Estimate Coefficient Estimate t statistic Scale Factor .1518 18.25 Factor Weight .912 32.24 Rho 1.141 4.95 Adjuster 1.32 4.91 3.3.2.3 Translog Functional Fit The Translog functional form is a generalization of the Cobb-Douglas form; it does not assume constant elasticity or returns to scale. The Translog functional form is very flexible as it has both linear and quadratic terms along with a cross product. This form permits the elasticity of substitution between inputs to vary. Also the elasticity of scale can vary with output and factor proportions permitting the long run average cost curve to take the traditional U shape (Weins, 2010). Because of its flexibility and lack of constraining assumptions the translog production form has become popular in current econometric work. The translog production function in standard two input format is as follows: 1Ln(Capital)+B2Ln(Labor)+B3 Ln2 (Capital) + B4 Ln2(Labor) + B5Ln(Capital) ×Ln(Labor) The results of the regression are as follows: (Capital)-.537Ln(Labor)+.0227 Ln2 (Capital) + .206 Ln2(Labor) -.233Ln(Capital) ×Ln(Labor)  The overall fit provides an R squared of 92%. 57  The F test (5,171) that all coefficients are equal to zero gives a probability of .0000.  Most, but not all of the t statistics on the coefficients are significant. The coefficients and t statistics are shown in Table 3.10. In order to have constant returns to scale the conditions outlined in Table 3.11 must be met (Bairam, 1998): This fit does not yield constant returns to scale. Also, this fit has an unreasonable area of the curve where if capital is very high and labor is very low the marginal product of labor will be negative. While this does not happen normally in our current data set, it does not seem to be a particularly meaningful portion of the curve. Table 3.10 Translog Estimate Coefficient t Statistic Capital 1.669 4.16 Labor -.5378 2.95 Capital squared .0227 .38 Headcount squared .206 4.95 Cross product -.233 2.83 Constant -1.247 2.05 Table 3.11 Translog Constant Returns to Scale Condition Results F Test probability 1+B2 = 1 1.669 -.5378 = 1.13 is close not equal to 1 73.3% - B3=B1 -.0227= is not equal to 1.669 0.00% -B4 = B2 -.206 = is not equal to -.5378 2.97% 58 3.3.2.4 Additive Production Function Previously the additive production function had been eliminated for logical reasons (see Choosing a Production Function in this chapter ). At this point that decision will be reconsidered. The resulting curve for the Constant Elasticity of Substitution, which has a very good fit (R squared 99%) is very flat. While clearly for these processes Capital and Labor are complements, and one could not produce output without both of the factor inputs, in the short-term, it is possible they can act as pure substitutes. For example if a line has a bottleneck or a breakdown, in the short-term the problem can often be solved with either capital or labor fixes. Because of this short-term effect, the Additive Functional form will be included in the study, and to be consistent with the other results the form will be expressed with natural logs. Base form Volume = B0 + B1*Capital + B2*Labor Log form ln(volume) = ln(B0 + B1*Capital + B2*Labor) The resulting fit ln(volume) = ln(-2.896 + .67*Capital + .02*Labor) or in standard form volume = -2.896 + .67*Capital + .02*Labor  The adjusted R squared is very high: 99%  The t statistics are significant for all coefficients - above 1% level.  The additive form is very restrictive, being a flat plane it has both constant returns to scale and infinite elasticity of substitution.  The coefficients and t statistics for the Additive form are shown in Table 3.12 59 Table 3.12 Additive Estimates Coefficient t Statistic Constant -2.896 6.47 Capital .669 25.36 Labor .020 4.3 3.3.2.5 A Comparison of the Four Models All four models provide good fits and a reasonable estimation of the coefficients as determined by errors and t statistics and logic (size and signs of coefficients). The only troubling element is the increasing returns to scale found in the Constant Elasticity model. As previously identified the key way to judge the goodness of fit is to measure the size of the residual errors resulting from the estimations. The two most common methods are to use the root mean square error and the mean absolute percent error. These comparisons are shown in Table 3.13. Under both of these measures the Constant Elasticity Function shows a better empirical fit. However, the other three production functional forms still provide very good fits and should not be summarily dismissed. Table 3.13 A Comparison of the Goodness of Fit METHOD Cobb-Douglas Constant Elasticity Trans Log Additive RMSE .284 .020 .313 .033 MAPE 9% 7% 10% 9% 60 However there are several concerns that should be noted, first the Cobb- Douglas, Constant Elasticity and Additive functional forms include some restrictive assumptions which force a certain shape. Also the Constant Elasticity of Substitution has increasing returns to scale which are unreasonable, and the shape is very flat. The Translog function presents a concern by not approaching constant returns to scale which should be expected across a mature stable production process, and under certain conditions results in a negative marginal product of labor. A further comparison of the three models can be found in the statistics listed in Table 3.14. Using a midpoint of the data range (50 for capital and 150 for labor) the functions would estimate the following output shown in Table 3.15. These are incredibly consistent results and are reflective of the extremely good fits across the various estimated functional forms. Table 3.14 Comparison Statistics Marginal Product of Labor Elasticity of Substitution Cobb- Douglas .103 ln ln Constant Elasticity Note: A negative σ is not meaningful and is due to the fact that the function is convex. Translog ln ln Too complex to be shown here Additive .02 Infinite 61 Table 3.15 Comparison of Outputs METHOD Cobb- Douglas Constant Elasticity Trans Log Additive Volume Output 34 34 35 34 While one might argue that one of the forms is superior to the other based on either the consistency with engineering design of the lines, or the goodness of fit, it is clear that they all represent very good fits with the data as is shown in Table 3.16. There is no need to eliminate forms at this time. The purpose of the empirical fitting is to ensure that we have reasonable forms on which to test the impact of wage premiums. We can use all these forms to test the efficiency wage hypothesis and if we get consistent results in all circumstances it will serve to strengthen the findings. A graphical representation of each of the fitted production functions is shown in Figures 3.8-3.11. These shapes show a generally flat area for the middle portion of the curve where most of the data exists, but show significant differences around the edges of the curves. 3.3 Calculating the Wage Premium The wage premium is calculated as a percentage of pay for Mars associates over the average pay for similar jobs in the local market. Mars associate pay is described in a previous section of this chapter and is a per hour wage average for all Table 3.16 Comparison of R squared and MAPE METHOD Cobb-Douglas Constant Elasticity Trans Log Additive R squared 90% 99% 92% 99% MAPE 9% 7% 10% 9% 62 nonexempt associates. This will include the appropriate mix of technicians, high level operators and lower level servicers employed. The market pay to compare this with is obtained from the U.S. Government Bureau of Labor Statistics (BLS). The BLS has been calculating statistics related to pay and employment for over 60 years and is the recognized source of market level data. They provide many statistics but the reports of interest to this research are the hourly pay by job within a market. The hourly pay by certain markets has been captured by the BLS since before 1970. The data needed for the 1990s are nothing new to the BLS as it is a continuation of surveys that have been completed for many years. In 1991 and 1992 the BLS prepared what they call Area Wage Surveys. These were completed every year3 for 120 markets (geographies) and show, by job classifications the hours worked, number of employees and average and medium pay. In 1992 the name of this report was changed to the Occupational Wage Survey but it kept the same geographies and job classifications. The data were compiled by the BLS through surveying establishments with over 50 employees and reaching a sample size (in all cases included in this research) of over 50% of the market. The data were then projected by the BLS to represent all establishments in the area. In 1997 the BLS again changed their survey to the National Compensation Survey. Over the next few years they adjusted some of the market areas to better represent changes in population patterns. They also changed the classification of jobs or occupations at this time. 3 For certain markets limited to every other year. 63 Figure 3.8 Resulting Cobb-Douglass Functional Form Midpoint Range Capital 50 Labor 150 64 Figure 3.9 Resulting Constant Elasticity of Substitution Functional Form Midpoint Range Capital 50 Labor 150 65 Figure 3.10 Resulting Translog Production Functional Form Midpoint Range Capital 50 Labor 150 66 Figure 3.11 Resulting Additive Functional Form Midpoint Range Capital 50 Labor 150 67 The data collected were from the following sources identified in the reference section. 1. Area Wage surveys- BYU Lee Library Government Documents 2. Occupational Compensation Surveys- BYU Lee Library Government Documents 3. Occupational Compensation Surveys- Utah Marriott Library Government Documents 4. Metro Area Cross Industry- BLS Web Site Archived Documents The BLS data do present some problems. First geographies can change. For markets used in this research two geographies change. The data for the 1993 and 1994 survey for Waco include nearby Killeen Texas but in 2000 and 2001 Waco is surveyed separately. The other change is within Chicago. The borders of the Chicago metro area are large and hard to define. During some of the years different suburbs are included in the definition of the Chicago metro area. Neither of these market issues is considered significant for the pay data used in this study. The other data problem is that surveys are sometimes conducted every other year. For certain key years data will be missing for some statistics. This can be solved by simply averaging the data from the year before and the year after. While not precisely accurate it will give a reasonable measure as pay changes moderately over time. Another data challenge is errors in the survey. At times we see very unreasonable jumps in the wages for a given year and then they decline to previous levels the following year. Jobs are classified by individual survey participants and 68 one year they might include a worker as a fork lift driver and the next year as a warehouse specialist even though the worker's job did not change. This along with the sample selection and participation rate of establishments included in the survey can cause an occasional obvious error. When errors are obvious, a similar job classification that appears more stable during the period in which the error occurred will be used to stabilize the numbers. A final related problem is the fact that job classifications changed in 1997. The classifications used after 1996 are much more specific and detailed and do not necessarily match up with jobs pre-1997. This is addressed by looking at the market for which the most detail exists (Chicago) and bridging jobs between 1996 and 1997 the year of the change. This bridge, shown in Table 3.17, shows a reasonable transition between job classifications and provides encouragement that the jobs groupings used before 1996 and after 1997 are consistent. The wage is weighted overall by the number of Mars employees in the high, medium and low job categories. The weighted average pay between these years increased by 2.1%, providing a reasonable bridge between the job classifications. Table 3.17 Job Classification Bridge for Chicago JOB Pre 1996 WAGE 1996 JOB after 1997 WAGE 1997 High Maintenance Mechanic 20.07 Heavy equipment Mechanic 19.48 Medium Warehouse Specialist 11.18 Machine Operator 11.58 Low Shipping and receiving 10.15 Handlers/Labor 10.64 Weighted AVG 12.38 Weighted AVG 12.64 69 Specific market occupations need to be selected and weighted together to provide an appropriate match for the Mars average wage. This is done by grouping jobs into high pay, medium pay and low pay groupings. This roughly corresponds to the pay zones that Mars operates within. The highest pay zone is for hourly technical associates, which include electrical technicians, and machinery maintenance mechanics. Also in this category would be team leaders and quality technicians. The middle pay group would be the experienced machine operators and the lower group would be newer associates who have not yet learned key firm specific skills. The low group is used for hand packaging, material movement and general labor. For the early years of the study the BLS job classification had a limited number of job classifications with actual data presented. These included several maintenance jobs, and warehousing and material movement jobs. No production job categories were included in the earlier job classifications. The criteria used to pick BLS occupations to match to the Mars jobs were: 1)the occupation was consistently available across most of the required years, 2) the occupation was consistently available across the geographic markets 3) the data included a large number of survey participants so that Mars data could not drive the numbers, and 4) the occupation was viewed as reasonable market match with the Mars jobs by the researcher. For the later years of the study the jobs were picked with the same four criteria and then matched through the 1996-1997 bridge using Chicago data. The occupational data used to match the Mars zones are listed in Table 3.18. 70 Table 3.18 BLS Occupations Used Pre-1997 Post-1996 HIGH PAY Maintenance Mechanic Heavy equipment Mechanic MEDIUM PAY Warehouse Specialist Machine Operator LOWER PAY Shipping and receiving Handlers/Labor An average wage comparison was obtained by creating a percentage weight for the Mars job categories by location and then using these percentages to weight the BLS wages. These weights are shown in Table 3.19. With these percentage weights the BLS data can be weighted into an average to compare with the Mars average wages. A wage premium can then be calculated as a percentage increase over the market base. Table 3.20 shows the market wage from the BLS calculations and the average Mars wage. This comparison results in the following wage premiums shown in Table 3.21 and graphically in Figure 3.12. Table 3.19 Employee Weights by Zone CHICAGO CHATTANOOGA 1993 2000 1993 2000 Technician 17% 21% Technician 20% 20% Operator 53% 51% Operator 42% 43% Servicer 30% 28% Servicer 38% 37% WACO ALBANY 1993 2000 1993 2000 Technician 23% 22% Technician 16% 19% Operator 49% 45% Operator 49% 42% Servicer 28% 33% Servicer 35% 39% 71 Table 3.20 BLS and Mars Wages Table 3.21 Wage Premiums These tables graphically straight-line estimates between 1994 and 2000. The calculated wage premiums show a premium level starting at a range of 16% to 39% and increasing to a range of 41% to 53%. Two considerations, which partially offset each other, should be considered relative to the size of these premiums. The first is the fact that neither the BLS nor Mars payroll data includes benefits. Mars had an extremely rich benefit package during this period with fully Albany Chattanooga Chicago Waco Albany Chattanooga Chicago Waco 1993 $ 10.25 $ 10.04 $ 12.96 $ 9.82 $ 14.22 $ 13.24 $ 15.06 $ 13.22 1994 $ 10.10 $ 10.23 $ 13.60 $ 10.17 $ 15.21 $ 14.25 $ 16.21 $ 13.56 2000 $ 11.41 $ 11.21 $ 14.38 $ 10.67 $ 18.29 $ 16.80 $ 20.76 $ 15.87 2001 $ 12.00 $ 11.56 $ 15.11 $ 11.39 $ 18.39 $ 17.46 $ 20.77 $ 15.62 2002 $ 11.78 $ 11.65 $ 14.63 $ 10.97 BLS Mars Albany Chattanooga Chicago Waco 1993 38.8% 31.9% 16.2% 34.7% 1994 50.7% 39.2% 19.3% 33.3% 1995 52.2% 40.9% 23.1% 35.8% 1996 53.8% 42.7% 27.1% 38.3% 1997 55.4% 44.4% 31.2% 40.8% 1998 57.0% 46.2% 35.4% 43.4% 1999 58.6% 48.0% 39.8% 46.0% 2000 60.3% 49.9% 44.3% 48.7% 2001 53.3% 51.0% 37.4% 37.1% 2002 52.9% 49.0% 41.2% 42.4% Wage Premium 72 Figure 3.12 Premium Pay paid insurance and a top defined benefit pension plan. This would tend to increase the possible premiums. The second consideration is the tenure of the Mars employees. Because Mars turnover was very low and pay was indirectly tied to length of service (see 3.2.2 Payroll Data) Mars employees would have an above market average length of service and as a result enjoy a premium wage versus the market based on tenure only. These two partially offsetting considerations are expected to remain constant throughout the length of this study and therefore have minimal impact on the econometric tests. 73 3.4 Testing the Wage Premiums All four production models can be run with wage premiums included. The models will then be assessed in two ways. The first is for overall enhancement: Does the model including wage premiums have better accuracy and fit? The second is for individual assessment of the wage premium itself: Does the coefficient of the premium make logical sense and is it statistically significant? These two assessments will allow us to conclude that wage premiums do or do not help us explain production output. Because wage data were only obtained for 4 years the original models which were run on all 10 years of data do not provide a valid bench mark to measure the impact of the wage premium. It is entirely likely that eliminating the 6 years of data will have a significant impact thereby confusing the effect of including the wage premium. To properly measure the impact of the wage premium an identical base model is required so the only difference will be the inclusion of the wage premium. The four accepted models from the first fitting will need to be rerun with only the 4 years of data which include wage data. This will significantly reduce the number of observations.4 However there will still be 80 observations for each of the tests, enough to reasonably calculate comparisons. The same model structure determined most appropriate from the full 10 year fitting will be incorporated into the new base models. All models will be run in logarithmic format and fixed effects will utilized 4 An additional model maintaining the full 10 years of production data with straight lined estimates for missing wage premiums will be presented as an alternative test in Appendix C. 74 The descriptive statistics for the data which will be used in this regression test are shown in Table 3.22. These are slightly different from the statistics shown in Table 3.7 because they only include the 4 years 1993, 1994, 2000 and 2001. However the means are similar and while standard deviation increases the measures for Kurtosis and Skewness are still roughly within two standard deviations and as a result are accepted as normal. 3.4.1 Cobb-Douglas Test The base model covering the 4 years of data has an R squared of 89% and an overall F test of 20.66. The calculated coefficients and significance are shown in Table 3.23. The model run with three factor inputs including is structured as : Ln (Volume) = Ln(technology) + αLn(Capital)+βLn(Labor)+μLn(Wage Premium) + ε or in traditional form Output = Volume = tKαLβPμ ε Table 3.22 Descriptive Statistics For 4 Years with Wage Premiums VOLUME SOC HEADCOUNT PREMIUM Mean 27.1 38.5 117.7 1.39 Standard Error 2.0 3.0 6.7 0.01 Median 28.1 32.1 128.0 1.39 Mode 3.4 14.0 142.0 1.50 Standard Deviation 18.3 27.0 60.0 0.12 Kurtosis -0.3 -0.1 -0.6 -0.55 Skewness 0.5 0.8 0.0 -0.39 Range 72.8 99.4 241.0 0.44 Minimum 0.7 2.9 8.0 1.16 Maximum 73.5 102.3 249.0 1.60 Count 80 80 80 80 75 Table 3.23 Cobb-Douglas 4-Year Base Coefficient t statistic Capital .78 4.05 Labor .52 5.26 The Cobb-Douglas model has the following results: The model including wage premium has an R squared of 91% and an overall F test of 14.24. The calculated coefficients and significance are shown in Table 3.24: From an overall perspective the model is not improved. The reduction in the overall F test shows that there is a slightly higher chance that the model is not significant. However from the individual assessment of the premium we learn that most. The negative coefficient is counter intuitive. It implies that as wage premium increases then output declines. The coefficient is also not significantly different from zero. From the Cobb-Douglas comparison we cannot conclude that wage premiums significantly influence output levels as the efficiency wage theory proposes. Table 3.24 Cobb-Douglas 4-Year Wage Test Coefficient t statistic Capital .916 4.02 Labor .491 4.69 Premium -.146 1.11 76 3.4.2 Additive Model Test The additive model assumes that factor inputs are perfect substitutes. In the efficiency wage model which we are testing wage premiums are not a substitute for labor. Instead they enhance the labor and encourage additional effort. Rather than adding on wage premiums as a separate independent variable, labor will be multiplied by the wage premium percent. This will show a labor amplified by wage premiums as the substitute for capital. This technique for showing the efficiency of labor is an accepted approach (Mankiw, 2009). The additive model augmented to include wage premiums is as follows: Ln(Volume) = Ln(Constant + αCapital + β(Labor*Wage Premium))5 The revised base for the additive model run on the 4 years of data for which we have wage premiums results in: An adjusted R squared of 99.02% and a residual deviation of 31.94. The resulting coefficients are shown in Table 3.25: After including the premium wages the model results in: An adjusted R squared of 98.99% and a residual deviation of 34.84. The resulting coefficients are shown in Table 3.26: Table 3.25 Additive 4-Year Base Coefficient t Statistic Constant 3.185 3.62 Capital .661 12.96 Labor .0248 2.8 5 Some readers have suggested that wage premium be added as an input to production, substitutable with labor and capital. This format has been run as an additional test and yields a negative (counter intuitive) coefficient for wage premium. 77 Table 3.26 Additive 4-Year Wage Test Coefficient t Statistic Constant 3.175 3.36 Capital .678 12.83 Labor .013 2.21 From an overall perspective the model is not improved, with slight negative adjustments to the residual deviation and the adjusted R-squared. And while there is no specific coefficient for premium wage the coefficient for headcount (which now includes premium wage impact) is smaller and less significant. Using the Additive model comparison we cannot conclude that wage premiums significantly influence output levels as the efficiency wage theory proposes. 3.4.3 Translog Test The translog model is a more complicated model with squared terms and cross products. To include wage premiums will require a three-factor model which is expressed as follows: Ln(Volume) = Constant + B1Ln(Capital)+B2Ln(Labor)+ B3 Ln(Wage Premium) +B4Ln2 (Capital) + B5Ln2(Labor) + B6Ln2(Wage Premium)+ B7Ln(Capita)×Ln(Labor) +B8Ln(Capita)×Ln(Wage Premium)+B9Ln(Labor)×Ln(Wage Premium) The base model, run as before but with only the 4 years of data results in the following: An overall adjusted R squared of 89.69% and an overall F Test of 10.91. The resulting coefficients are shown in Table 3.27. After including the premium wages the model results in: 78 Table 3.27 Translog 4-Year Base Coefficient t Statistic Log Labor -.759 -1.58 Log Capital .239 .22 Log Labor squared .220 2.05 Log Capital squared .204 1.31 Cross product -.168 .89 Constant 1.05 .57 An overall adjusted R squared of 86.91% and an overall F Test of 3.27. The resulting coefficients are as shown in Table 3.28. From an overall perspective the model is not improved, with slight negative adjustments to the R squared and the overall F Test. And while there is no single coefficient for premium wage, the coefficients are all now statistically insignificant. An F test on the four coefficients which contain wage premium gives F (4,48) = .64 with a probability of .637. As a result we conclude that there is no statistical significance to including the wage premium into the equation. Table 3.28 Translog 4-Year Wage Test Coefficient t Statistic Log Labor -.604 -1.13 Log Wage premium 1.223 1.03 Log Capital -.166 -.06 Log Labor squared .194 1.74 Log Wage premium squared .409 1.47 Log Capital squared .237 .73 Capital X Labor -.166 -.84 Capital X Wage Premium -.0028 -.01 Labor X Wage Premium -.04 -.21 Constant 2.396 .47 79 Using the Translog model comparison we cannot conclude that wage premiums significantly influence output levels as the efficiency wage theory proposes. 3.4.4 Constant Elasticity of Substitution Test The Constant Elasticity model also requires a slightly different form with three factor inputs. The model including wage premium is expressed as follows: ln volume ln ln The base model, formatted as previously derived, but with only the 4 years of data results in the following: An adjusted R squared of 99.40% and a residual deviation of -7.6229. The calculated coefficients and statistical significance are shown in Table 3.29. After including the premium wages the model results in: An adjusted R squared of 99.39% and a residual deviation of -7.62844. The calculated coefficients and statistical significance are shown in Table 3.30. The models are almost identical. The very minor changes in R squared and residual deviation imply a model moving in the wrong direction. The coefficients do not show much change but the significance of the constant and weight for capital take a steep drop. The significance of the scale and Rho show only the smallest Table 3.29 CES 4-Year Base Coefficient t Statistic Constant -1.98 10.74 Scale 1.34 2.62 Weight to capital 88% 16.33 Rho .925 2.64 80 Table 3.30 CES 4-Year Wage Test Coefficient t Statistic Constant -1.8 .765 Scale 1.34 2.59 Weight to capital 76% .53 Weight to headcount 10% 2.63 Rho .928 2.63 incremental drop, but again imply a model moving in the wrong direction. Using the CES model comparison we cannot conclude that wage premiums significantly influence output levels as the efficiency wage theory proposes. 3.4.5 Test Conclusion It is inopportune that payroll data were limited resulting in the number of observations being necessarily dropped when conducting this final test of payroll impact. However, given the consistent results across all four tests we are forced to rely as much on the preponderance of evidence as on the depth of a more satisfying individual test. The conclusion of the tests is that this specific data sample does not provide any support for the idea that premium wages increase effort which increases output. Table 3.31 summarizes the statistics for the overall fit. Given the reduced number of payroll observations versus production information an alternative test has been prepared. The entire test above has been repeated using the full 10 years of production data and using wage premiums with the missing data filled in as a straight line between the 1993/1994 and 2000/2001 data points. These estimates are shown in Table 3.32. This test is presented in 81 Table 3.31 Summary of Overall Fit with Wage Premiums Table 3.32 BLS and Mars Data with Extrapolated Trends Base W/Premium Cobb-Douglas Adjusted R Squared 0.892 0.906 F Test 20.66 14.24 Additive Adjusted R Squared 0.9902 0.9899 Residual Dev. 31.94 34.84 Translog Adjusted R Squared 0.8969 0.8691 F Test 10.91 3.27 CES Adjusted R Squared 0.994 0.9939 Residual Dev -7.6229 -7.62844 Albany Chattanooga Chicago Waco Albany Chattanooga Chicago Waco 1993 $ 10.25 $ 10.04 $ 12.96 $ 9.82 $ 14.22 $ 13.24 $ 15.06 $ 13.22 1994 $ 10.10 $ 10.23 $ 13.60 $ 10.17 $ 15.21 $ 14.25 $ 16.21 $ 13.56 1995 $ 10.30 $ 10.39 $ 13.72 $ 10.25 $ 15.69 $ 14.64 $ 16.90 $ 13.92 1996 $ 10.52 $ 10.55 $ 13.85 $ 10.34 $ 16.18 $ 15.05 $ 17.61 $ 14.29 1997 $ 10.73 $ 10.71 $ 13.99 $ 10.42 $ 16.68 $ 15.47 $ 18.35 $ 14.67 1998 $ 10.96 $ 10.87 $ 14.12 $ 10.50 $ 17.20 $ 15.90 $ 19.12 $ 15.06 1999 $ 11.18 $ 11.04 $ 14.25 $ 10.59 $ 17.74 $ 16.34 $ 19.92 $ 15.46 2000 $ 11.41 $ 11.21 $ 14.38 $ 10.67 $ 18.29 $ 16.80 $ 20.76 $ 15.87 2001 $ 12.00 $ 11.56 $ 15.11 $ 11.39 $ 18.39 $ 17.46 $ 20.77 $ 15.62 2002 $ 11.78 $ 11.65 $ 14.63 $ 10.97 $ 18.02 $ 17.36 $ 20.66 $ 15.62 BLS Mars 82 Appendix C and importantly confirms the results of the original test presented here. In Table 3.32 years 1993, 1994 2000, 2001 and 2002 are actual data points and other years are straight line interpolations between the actual data points. In this alternative test the original production models can serve as the comparison base. 3.4.6 Possible Reasons for These Results Given the popular support for the efficiency wage argument and the bulk of articles and test supporting the hypothesis it is a small surprise that we find absolutely no evidence of wage premiums' impact on output. While this research cannot determine the exact reasons why, ideas about the reasons for a lack of significance should be proposed. One must first consider that, at least in this case, the efficiency wage hypothesis is simply not valid. This is a specific case study for one company and four manufacturing locations. The efficiency wage hypothesis could be reality in many situations but not be a universal maxim, thus allowing for differences in specific situations such as the one in this research. The results of this study apply specifically to the particular data and do not extend to other firms . A corollary of this point is that this case study may be an exceptional situation. And indeed this firm has many institutionally established cultures that could overpower other influences. The specific cultural institutions of this firm are addressed in Chapter 4. Second, limited data or errors in the data may prevent a true result. Given that the data were sourced directly from payroll and operational reporting systems we believe that the data are as robust as business performance data can be and are much more reliable than the overall averages and macrodata used in the other 83 referenced studies of efficiency wage (see Chapter 2). The limited data extends across four plants each with multiple production lines over 10 years. The specific limits to the data are fully incorporated in the degrees of freedom used in the statistical tests of significance. The type I error that a population characteristic which actually exists will not be found is small (for every test less than 5%). Even if the efficiency wage hypothesis is valid, over time, in a team production environment the impact may diminish. This may explain why no wage premium impact on output was detected in this study. Efficiency wage may have diminishing effects over time and eventually be replaced with a sense of entitlement. When an employee first starts she accepts the job understanding a wage premium in comparison to her alternative options. The wage premium can attract, motivate and retain per the efficiency wage hypothesis. In a team environment production output per associate is not visible. However, each manager and employee will form a subjective opinion of each team member's contribution. After many years of working side by side with other employees who make the same wage, she will notice that some put forth less effort and are rewarded the same. As a result she will feel entitled to the wage given that she feels she puts forth an average effort. At the locations in the case study there are many long time employees and this would be an interesting employee attitude to test for and understand. This of course could be compensated for if reward, including wage, were differentiated and awarded by effort. In our case study wages of all team members are the same and only the rare promotion opportunity would provide for individual recognition of effort. 4. TESTING INSTITUTIONALISM AND POWER RELATIONSHIPS 4.1 Mars Culture Mars, Incorporated is known for its unique, strong and pervasive culture. From the early days during the great depression until now many of the values and principles of the company have remained unchanged and uniquely Mars. Most family businesses do not last beyond the second generation, as divided ownership and interest lead to public sale. Mars, Incorporated is now on its fourth generation ownership. This is a testimony to the strong and pervasive culture developed within Mars. This also accounts for the strong institutions that have developed within the culture. These rules of the game, which provide constraints and structure to the employment relationship, all favor premium pay for employees. The first element of this unique culture comes from being run by family owners. "As a family business we base our activities on family values." (Mars, Incorporated, 2005, p. 19). The family owners worked tirelessly side by side with their employees. The owners traveled extensively to all sites around the world, showing up at 6:00am ready for work. The family has always been very concerned about privacy, shunning personal publicity and did not even allow portraits of the family on display. Because of the family ownership sometimes the profit 85 maximizing rule would not always hold. As an example, one of the family members had responsibility for a factory that burned down when he was young. As a result, he insisted on over-the-top fire protection at all sites. One engineering contractor was asked to sign a confidentiality agreement and said, " I'm happy to sign it but why do you care - no one else would build like this!" A second element of the unique culture was an egalitarian and paternalistic approach. From the beginning there were no private offices, no private parking spaces, only a few broad jobs. Employees were called "associates" and everyone from the president on down was required to clock into a time clock each morning. Because the owners worked side by side with associates throughout the world their personal exacting standards became a third element of the culture. In terms of stated company values this was referred to as "Quality - the consumer is our boss, quality is our work and value for money is our goal" (Mars, Incorporated, 1983, p. 1). However, the exacting standards were more than a product quality expectation. They included personal commitment . One company lawyer ended up being called in to work on the day of her wedding and she came. Finally one of the owners (an engineer) had a love of engineering technology. He believed that the only sustainable competitive advantage was a technical one. He insisted on highly mechanized plants and the latest and best technology. This cultural tenet became important in labor policy as higher skilled workers who were willing to learn and progress with technology were required. Several nonformal institutional practices result from these cultural elements. 86 The first is increased termination costs. Because of the intense private nature of the owners a threat of publicity was associated with termination. This threat increases the implicit cost of firing a worker or even the voluntary departure of an unhappy associate. Any adverse legal action was avoided at all possible costs. Second, this firm had a need for employment longevity. Because the owners worked side by side with associates they required a personal trust to be built up over time. They would prefer to work with someone they knew and trusted rather than the latest business savant from outside. As such they needed to encourage employment stability. Third is a difficult work environment. Because the owners were known to be demanding and because there were no private offices, associates were expected to endure difficult public treatment. If a company wants to be able to berate associates publicly, and change personal decisions and directives at a whim, a company must compensate the employees to offset these negatives associated with the job. Fourth is the threat of unions. Because the company was a private family owned company the threat of unions was not a simple business issue. The threat of unions was taken as a threat to the family ownership and control