Rudolf Winter-Ebmer. Josef Zweimüller. Intra-firm Wage Dispersion and Firm Performance

Transcription

1 Rudolf Winter-Ebmer University of Linz, Austria Centre for Economic Policy Research, London Josef Zweimüller University of Zurich Institute for Advanced Studies, Vienna Centre for Economic Policy Research, London Intra-firm Wage Dispersion and Firm Performance This research has been supported by a grant from the Austrian Central Bank (Jubiläumsfonds #4674) and from the Austrian Science Funds (J01082, Winter-Ebmer). Thanks for valuable comments by G. Akerlof, G. Baker, C. Brown, J. Leonard, D. Levine, S. Raphael, R. Schnabel L. Ulman and seminar participants at Berkeley and at the Society of Labor Economists, Chicago. Markus Gusenleitner provided excellent research assistance.

2 Abstract Personnel economics has suggested conflicting arguments about the impact of increased wage dispersion within firms on workers' productivity and firm performance. Besides giving more advancement incentives, bigger wage differentials might also give rise to less cooperation and more politicking amongst workers resulting in worse outcomes. We try to answer these questions using panel data for Austrian firms. As indicators for firm performance we use employment growth and standardized wages. For white-collar wages the following picture emerges: more dispersion leads to higher earnings up to some point where the relation changes its direction. For blue-collar wages we find a positive association between dispersion and standardized wages between firms, but no relation within firms over time. For employment growth the results are ambiguous. 2

3 1. Introduction The last years saw a generally rising interest in the topic of income distribution. This was, of course, caused to a great extent by rising income disparities in many countries, first of all in the U.S. Inequality has increased for the population at large, within specific industries, between groups of workers defined by educational attainment as well as within firms. The latter received much media attention as CEO's compensation grew rapidly relative to the salaries of the general work-force. The appropriateness of high executive compensation schemes or a very steep earnings profile within the firm, can be questioned on grounds of morality or fairness. There is not much economists can contribute to this discussion. On the other hand, there are theories that relate productivity in the firm to the firm's organization, especially the compensation structure. Incentive arguments - most evident in the form of tournament models - argue for a higher spread in order to motivate workers to aspire to superior positions. Other theories stress that cooperation between workers and general fairness considerations act as a constraint on overly dispersed pay schedules: cooperation between workers might suffer, politicking activities to influence the distributional process in the firm might increase and, thus, ultimately destroy the positive incentive effect of higher wage dispersion. To test these theories we use a data set for Austrian firms, where we observe wages for the whole work-force over several years. Although we cannot observe the financial performance of the firms, we do construct some performance measures: standardized wages for different groups and employment growth over time. We find, that the earnings of white-collar workers react in a hump-shaped way to wage inequality in the firm: at first, more inequality furthers earnings, but around the mean the relation reverses. More inequality above the mean of our empirical observations is in fact detrimental to the productivity of white-collar workers, as far as their productivity can be inferred from their wages. In the case of blue-collars, the standardized wage increases with wage dispersion. No consistent picture has been found for employment growth in the firms. The plan of the paper is as follows. The next section discusses theories on inequality and firm performance and reviews the relevant empirical literature. Section 3 presents our data and derives our indicators for firm performance as well as wage dispersion. In section 4 fixed effects panel estimates for firm performance are performed. As fixed effects estimators only concentrate on the effect of wage dispersion on performance within a firm over time, we proceed in section 5 to estimates between firms to investigate the impact of the long-run wage structure. Section 6 concludes. 3

4 2. Wage dispersion and firm performance Wage dispersion or compression becomes a decision variable for the firm once relative comparisons are made. Workers might view the appropriateness of their pay relative to the compensation of any reference group and act correspondingly. Employers, in turn, might use relative compensation schemes. One example of the latter is the theory of rank-order tournaments (Lazear and Rosen, 1981). Workers are not rewarded according to their absolute performance, but according to whether they are better or worse than their fellow workers. The better player receives the premium which may be thought as a promotion. This compensation scheme leads to higher equilibrium effort by both workers the higher the wage spread between the good and bad jobs. High wage dispersion enhances productivity. On the other hand, fierce competition between workers may be detrimental to the firm. A worker can increase his chances of winning a contest by increasing his effort, but also by trying to reduce the output of his contestant through means of sabotage or non-cooperation. Lazear (1989) shows that some wage compression is optimal in this situation. He distinguishes more aggressive, and thus sabotage-prone, workers ("hawks") from less aggressive ones ("doves"). The firm would like to employ only doves but the types cannot be identified. The optimal reaction is to introduce a more equitable wage structure, if one is found with many hawks in the firm. One implication of such an environment is a positive correlation between mean productivity and wage levels in a cross-section of firms. Similar arguments relate to influence and rent-seeking activities in firms (Milgrom, 1988, Milgrom and Roberts, 1990). Workers can engage in costly rent-seeking instead of productive work, if there is much potential for redistribution at the discretion of a supervisor. Remedies to limit these costs are to compress the firm's overall wage structure, thus limiting access of employees to the decision making process or reducing management discretion over decisions that are of little importance to the organization, but have potentially large redistributive consequences. Although some pay compression emerges as an efficient outcome, there are no clear testable relations between compression and firm outcome. A further strand of the literature relies on efficiency wages and fairness. Based on equity theory (Adams, 1963), Akerlof and Yellen (1990) postulate that workers withdraw effort if their actual wage falls short of the wage the workers themselves consider fair. This fair wage, in turn, can be determined as a weighted average of wages of a reference group and the respective market-clearing wage. The structure of pay in equilibrium exhibits wage compression due to considerations of fairness. A related argument is put forward by Levine (1991). In participatory firms where work groups are important, cohesiveness will further productivity. As pay compression increases group cohesiveness, wage compression augments overall productivity. Levine notes also, that although egalitarian policies might be advisable, firms will face difficulties, because top performers tend to leave the firm. 4

5 There may also be factors other than incentive considerations or (non-) cooperation issues linking earnings inequality to firm performance. In particular, there might be an underlying factor affecting both. One such factor is unionization or worker organization. On the one hand, unionized firms seem to have lower wage dispersion among observationally equal workers (Freeman, 1982). This may be due to the fact that pay compression enhances organizational strength and solidarity. Alternatively, union members may prefer more equality. On the other hand, unionized firms pay better. Other things being equal we would therefore expect to see less dispersion in firms with higher wages. 1 Another factor that may determine both firm performance and within firm wage dispersion is plant size. It is a well established fact that larger firms pay higher wages (Brown and Medoff, 1989). There may exist a systematic relationship between inequality and employer size. For instance, due to the use of more standardized technologies larger employers might offer more homogenous jobs leading to less variation in pay. Similarly, larger firms are more likely to be unionized so that the above reasoning applies. Early empirical evidence on the equity-productivity nexus stems mainly from social psychologists. Akerlof and Yellen (1990) as well as Levine (1991) review experiments along these lines, which generally speak to the importance of relative remuneration and fair wages. Additional evidence comes from surveys conducted with personnel managers concerning their daily practices. Blinder and Choi (1990) and Agell and Lundborg (1995) use a similar questionnaire in New Jersey, respectively in Sweden. They find that employers are particularly afraid of having an "unfair wage policy": 2 they fear that turnover will rise, effort will fall and the quality of future applicants will suffer. Likewise, most employers dismiss wage cuts because of relative wage considerations. Similarly, Levine (1993) finds that compensation executives are very reluctant to change the internal pay structure in response to different market wage developments for parts of their work-force. This was especially so, if these groups, facing different outside opportunities, worked closely together in the firm. Empirical tests of the above hypotheses are rare. This is mainly due to a lack of firm level inequality data. 3 Macroeconomic papers looking at inequality and growth or productivity usually use only economy-wide inequality indicators. One exception is Hibbs and Locking (1995). The authors study the Swedish experience from 1963 to In an aggregate production framework, they look at the relation of wage dispersion within and between firms on aggregate output. They find that higher wage dispersion between firms tends to reduce output, but output is furthered by higher dispersion within the firms. The overall effect of inequality on output is slightly negative. This result is in contrast to theories of within-firm worker cohesion and 1 This seems to less a problem in the case of Austria. Union contracts are generally extended to all workers, regardless whether they are union members or not. 2 Unfortunately no clear definition of such an unfair wage policy is given in the questionnaire. 3 The papers by Baker at al (1994a,b) use data with detailed information on the compensation of workers at various positions within one firm. However, their objective is not primarily in studying the impact of wage compression. 5

6 fairness considerations. The highly aggregated nature of this data set restricts its applicability. The results can be driven by the large wage compression in Sweden in the 1960s and 1970s, a development which was substantially reversed in recent years. The authors themselves entertain the possibility that "wage leveling in Sweden went far beyond the (unspecified) magnitudes that advocates of 'fairness, morale and cohesiveness' theories have in mind" (1995, p22). Cowherd and Levine (1992) come closest to our approach. They look at a specific measure of firm performance: customer-assessed product quality. Using data for 102 business units in the US and UK they relate product quality to two measures of withinfirm inequality: i) the pay relation of managers to hourly paid employees and ii) the pay differential within higher and lower management positions. For both inequality indicators they find a negative association with firm product quality. The authors attribute this finding to the impact of pay equity on employee commitment to managerial goals, effort and cooperation. The use of quality instead of quantity of production (productivity) may lead to another interpretation of the findings. If workers receive performance pay based on the quantities they produce, they might tend to neglect other - less observable - tasks like quality (Milgrom and Roberts, 1992). Another interesting paper is Leonard (1990). He was unable to find any effect of inequality on firm performance. His measure of performance was the return on investment, the inequality measures related only to managers. Pay differentials and the number of hierarchies within management positions had no relation to return on investment or the change thereof over time Data and indicators for performance and inequality To assess the impact of the wage structure on firm performance we use a panel of Austrian firms, which covers the period We use firms with more than 20 employees and restrict the panel to those firms for which at least four data points are available. This results in a sample of 130 firms. The data originate from social security records. We sampled each 50th firm and collected all available information of the workers of these firms. In that sense, all information on firms - besides location and industry - is constructed from the worker's characteristics. 5 This procedure has the advantage that the wage structure is (almost) perfectly observable, whereas no financial performance indicators are available. Most of the variables used in typical wage regressions are available, except for education. Only monthly earnings are recorded. Earnings are top-coded, affecting approximately 9% of the sample. Concerning wage compression within firms, two basic concepts can be distinguished: (i) wage differentials between unequal workers and (ii) differentials between "observationally equal" workers. The first concept refers to the male/female or the 4 Pfeffer and Alison-Blake (1992) look at the effect of wage dispersion on labor turnover, but they use only a very special data set for college administrators. 5 See Winter-Ebmer and Zweimüller (1996) for a detailed description of the original data set. 6

7 blue/white collar pay gap, or to general measures of inequality, like Gini indices, Atkinson indices and the like. These indices do have appeal, when it comes to the influence of CEO's pay on work morale of the rest of the work-force, or the relations between different work-groups, as discussed by Akerlof and Yellen (1990). For our purpose these measures are less practical. Owing to our top-coding problem, no information on executives is available 6, so looking at the influence of managers' compensation is impossible 7. On the other hand, many theories like tournaments rely on the interaction of similar workers. A pay spread as an incentive for being promoted is only effective for those who compete for the superior job, i.e. workers with a similar job situation. Likewise, industrial politics, non-cooperation, and sabotage is more easily established between workers at an equal job level. Following these lines, the pay structure should, therefore, account for different observable productivities, and only the residual inequality should matter. As suggested by Lazear (1989) we compute a conditional wage differential, conditional on observable characteristics of workers. For each firm and each year a separate tobit earnings regression is run in the following way: (1) ln Wijt * = a jt + b, Z, + ε k k jt k ijt ijt lnwijt = ln Wijt * if earnings not top-coded lnw = ln T if earnings are top-coded at T t. ijt t j indexes firms and t time, i is the index for individuals. The explanatory variables include age, age squared as well as dummies for sex, blue-collar, foreigner status and two tenure dummies. The standard errors of these regressions are taken as the conditional wage differential sigma for the firm j in time t. This gives us a measure of wage compression for the firm controlling for observable differences of the workforce. Table 1 Table 1 presents summary statistics for the data. The variable sigma corresponds to the variance of ε ijt in (1). Since regression (1) is run for each firm and for each year sigma varies not only between firms but also over time within a given firm. Of all firm/year observations the mean of sigma is equal to with a standard deviation of As we can only observe monthly wages and no weekly hours, we might get biased inequality indicators when some workers are only employed part-time. A firm with more 8 part-time workers - carrying lower monthly wages - would erroneously look like a firm with high wage dispersion. As only a negligible percentage of men work less 6 Using the methodology of Fichtenbaum and Shahidi (1988) it is possible to construct a Gini coefficient in the presence of top-coding. For each firm the upper tail of the wage distribution is assumed to follow a pareto distribution. Doing this, the unknown earnings can be approximated and inequality measures can be calculated. Results along the lines of tables using Gini or Atkinson coefficients - were generally insignificant. 7 See Leonard (1991) for manager compensation and firm performance. 8 Of course, the relation would be vice versa, if the majority of workers in a firm is employed part-time: observing more part-time workers would tend to lower measured wage dispersion. 7

8 than 40 hours, we use as a second inequality measure the standard error of an individual wage regression (1) using only male workers in the firm. Figure 1 shows that more than one half of all firm/year observations (n=1236) lie in the range between 0.15 and 0.25, with a fairly symmetric pattern beyond those values. The mean of sigma compares to a value of about 0.50 of the variance of log earnings in the underlying individual data. 9 This suggests that a considerable part of the overall wage variation is due to within-firm wage dispersion. While the total variance of the log individual earnings has increased considerably over the analyzed period, withinfirm wage dispersion stayed roughly constant during (see Figure 2). This suggests that the increase in earnings inequality is more likely attributable to changes in between-firm wage inequality and/or to changes in (or the returns to) individuals observable characteristics. Figures 1,2,3 Figure 3 shows how the wage levels and the inequality measure sigma are related to employer size. Since the size distribution of Austrian firms is skewed towards smaller firms compared, for instance, to the U.S. we report summary statistics on 4 relatively small size classes: (i) 20-49, (ii) 50-99, (iii) , and (iv) 250 and more employees. Figure 3 shows that sigma is increasing with employer size whereas the average wage level stays roughly constant. 10 This result is in contrast to the findings of Davis and Haltiwanger (1991) for the U.S. who conclude that within firm wage dispersion is roughly constant across size classes, whereas the average wage is increasing with plant size. To measure firm performance, unfortunately no direct productivity or profitability measures are available in the data. Instead we will focus on two different indicators. First, we look at wage levels - a likely proxy for firm productivity. Second, we consider firm growth, i.e. employment growth, over time. As long as firms pay their employees according to their marginal product, the firm wage level can be used as a measure of efficiency. However, this must not necesarily be the case. Consider, for instance, a situation where firms and workers split the firm s rents in a bargaining framework or according to some sharing rule. Whenever the pie becomes larger due to higher productivity, we would also expect workers to be paid better; even if they do not earn the marginal product 11. Employment growth can be considered a good medium-term productivity indicator, whereas short-term fluctuations might be caused by different factors. The two measures in combination give us further information. We would certainly not interprete rising wages caused by more equality "a rise in productivity" when at the same time, employment goes down. 9 For an analysis of the changes in Austrian earnings inequality see Gusenleitner (1995). 10 It should be noted, however, that a standard individual earnings regression including firm size yields a positive size employer wage differential (see Winter-Ebmer and Zweimüller, 1996). 11 See e. g. Blanchflower, Oswald and Garrett (1990) for the impact of internal profitability on pay in the U.K. Similar evidence for the U.S. is produced by Blanchflower, Oswald and Sanfey (1996). 8

9 A natural choice for the earnings indicator would be the mean or median income. Unfortunately, these measures are heavily influenced by the composition of the workforce and the pay dispersion itself. Therefore, we use two standardized earnings indicators: expected earnings of a typical white-collar and a typical blue-collar worker. For these measures we use the coefficients from the firm-specific wage equations and calculate expected earnings for a male white-collar (blue-collar) worker, 40 years old with Austrian nationality and 3 years of tenure in the firm. These standardized wages were deflated using a consumer price index. Wages could also be determined by an implicit life-cycle contract, above productivity at the beginning and below productivity at the end of the career. In this case, our standardized wage picks up a point in the middle of the career, where the deviations from productivity should be minor. The data show a substantial pay difference between the typical blue collar and white collar worker. The average pay gap amounts to somewhat less than 40 % (see the difference between average log wages in Table 1). Since the individual wage regressions (1) do lack information on schooling, the blue-collar/white-collar gap also reflects differences in formal education. For both, the typical white-collar and the typical blue-collar worker the standard deviation of log wages is somewhat larger than 0.40, which seems to be relatively high. It should be noted, however, that this variation is not only due to between-firm wage differences, but also due to wage changes over time. All subsequent firm-level wage regression will therefore account for fixed time-effects. Employment growth is measured only for surviving firms. Nonetheless the variability in our sample (see Table 1) seems to be rather high. Apart from possible reporting errors 12, all forms of organizational change at the firm level might influence employment levels dramatically. This is especially relevant for mergers, the closing down of parts of the firm or subsidiaries and the like, as long as the firm keeps the same reporting entity to the social security administration. Unfortunately, changes in organizational form cannot be inferred from the data. On the other hand, there is no prima facie argument to restrict firm growth to a sort of "organic growth", excluding these organizational changes. We will therefore produce results for the whole sample as well as for a sample, where we exclude "outliers", being those firms with an employment growth below -50% and over +100%. Apart from plant size, the remaining variables which will be used in the subsequent analyses describe the structure of the work force of firms. We account for the age distribution of employees by using the share of younger and older workers in employment, the share of blue-collar and foreign workers, the percentage female as well as the share of workers employed in the current firm for longer than 3 years. 12 There seems to be a minor error in the construction of the current raw data. Employment levels are always counted as of the 31 of may of each year. These figures do not include those workers who incidentally called sick at these particular days, thus producing some spurious volatility in the level of the work force over time. We are currently trying to review this procedure. 9

10 4. Within-Firm Results In this section we present ordinary least squares and fixed effects estimates for standardized blue-collar and white-collar wages as well as for employment growth. We use a two-way fixed effects panel estimator controlling for firm effects as well as for time (years) effects Wage effects The basic specifications for standardized wages are contained in Table 2. Within-firm inequality (sigma) enters in a quadratic form to allow for a non-linear pattern. The pattern for wages is, in general, hump-shaped. Blue-collar wages rise with inequality, but decline again after a peak of 0.29, which is around three-half standard deviations above the mean of sigma. Thus, in general, the picture is positive. The case is different for white-collar wages: Again, wages rise with inequality, but they already start to decline at a value of sigma of 0.23; shortly after the mean (0.205). Too much inequality is detrimental for productivity, as far as productivity is associated with wage levels. The quantitative effects of wage dispersion are as follows: Bluecollar workers' wages rise by 3.4% if the worker is employed at a firm with dispersion one standard deviation above the mean. In the case of white-collars, wages would fall by 2.1%. These results refer to the OLS regressions, which do not account for firm-specific effects. The second and fourth column of Table 2 present the results once we introduce firm fixed effects. In both cases, this reduces the impact of wage inequality on standardized pay strongly. However, while for blue-collar workers the effect vanishes completely, we observe a significantly positive (albeit comparably small) relationship between firm inequality and the white-collar wages. Since the fixed effects estimator relies on within-firm variation in both wage levels as well as inequality indicators, this might be a short-run phenomenon. Since in the longer term firm effects may be considered as being variable, the difference between OLS estimates and fixed effects may reflect differences in the inequality/earnings relationship between the short and the long run. Below we will discuss the between-firm effects more carefully by running group means regressions. Table 2 Control variables include firm size as well as several demographic characteristics of the work-force in the firm. Note that the dependent variables are standardized wages, so the coefficients should not pick up structural effects of the composition of the workforce. Most OLS-coefficients are highly significant, but the fixed effects are not. The fixed effects model is able to control for unobserved firm characteristics. On the other hand, most demographic variables will not vary much over time, and, thus, by differencing out firm effects, existing measurement errors in the data might be 13 Random effects models are rejected by a Hausman test in most cases, for some specifications a usual Hausman test could not be calculated, because of problems of invertability of the covariance matrix. 10

11 aggravated. Therefore, we should expect less precise estimates in the fixed effects model. The age structure in the firm serves as a proxy for the experience of the work-force. The percentage of prime-age workers in the firm fosters productivity; more youngsters as well as more elderly workers reduce standardized wages for white- and blue-collar workers alike. Interestingly, a high share of blue-collar workers is furthering wages for both groups. The same applies for a higher share of foreigners in the firm. The latter observation can be accounted for by a bargaining situation, where primary workers (natives) profit from the presence of secondary underpaid foreigners 14. A higher share of females reduces wages considerably. All these demographic variables in the OLS model can be seen as picking up structural components of firms, e.g. lower wages and productivity in women-dominated industries, like textile, etc. However, the negative female coefficient remains also in the fixed effects specification. Contrary to prior reasoning of possible positive external effects of human capital, a more tenured workforce has a somewhat negative impact on standardized wages. 15 Firm size is insignificant for blue-collar wages, but negative in the case of white-collar wages. Table 3 Table 3 performs some robustness checks for the wage regressions. As the inequality indicator may be contaminated by different hourly working time, we present in Panel 3A estimates using a sigma, calculated using only male employees in the firms. The results corroborate the previous findings. For blue-collar wages (OLS) a positive pattern emerges: for the most part of the empirical distribution of sigma, higher dispersion is leading to higher wages. For white-collar wages (OLS and fixed effects), wages rise first, but fall again after a value of sigma of As can be seen in Figure 3, the potential for wage dispersion is bigger in larger firms. A simple way to tackle this problem is to regress sigma on firm size and use the residual e sigma - the part of sigma orthogonal to firm size - as the new dispersion indicator 16. Panel 3B presents results using e sigma, and Panel 3C uses e sigma for the sigma calculated only from data for male workers. The results are very similar to the ones presented above: positive 17 or insignificant dispersion effect for blue-collars, humpshape with peak around the mean for white-collars. An interpretation of the results according to the aforementioned theories is somewhat difficult. Lazear (1989) would predict a rising dispersion-wage schedule, if different types of firms, respectively workers, are observed. This pattern turns out only for bluecollars. The fact that no such relation was found for white-collars, would mean that the potential for hawkish behavior is more important for blue-collars. Therefore, it would pay more for firms to separate hawks and doves among the lower ranks of the firm. 14 See Winter-Ebmer and Zweimüller (1996) for a model. 15 See Winter-Ebmer (1994) for an empirical examination of external effects of human capital accumulation. 16 The same procedure is used by Cowherd and Levine (1992). 17 Note that e sigma varies around zero, not around

12 This is somewhat at odds with the general implications of the model, i.e. hawks are more common in the higher ranks of the firm. 18 The pattern for white-collar wages seems, thus, to be more in line with the cohesiveness and fairness arguments of Levine(1991) and Akerlof and Yellen (1990) Employment Growth In this section analogous results for employment growth are presented in Tables 4 and 5. Here, the pattern of wage dispersion is U-shaped with a trough at a level of sigma of That means for the most part of the empirically observed wage dispersion, firm growth falls with rising inequality. Again, this clear-cut picture is washed out, once fixed effects are introduced. Table 4 Except for mean wages in the firm, the results in Table 4 use the same control variables as in Table 2. Mean wages have a big impact on firm growth. A firm with a mean wage of ATS 1,000 (approximately US$ 100) above the mean would grow 1.2, respectively 2.6 percentage points in the fixed effects specification, faster than usually. Growth exhibits significant regression to the mean: larger firms grow significantly slower. The quantitative effects of the demographic variables are smaller: firms employing more elderly workers grow slower, as well as firms with more females. Firms employing a larger share of long-tenured personnel tend also to grow slower. In Table 5 several specification checks are performed. At first, outliers - firms with employment growth lower than -0.5 and higher than are eliminated from the sample. Panels 5B, 5C and 5D present results using a dispersion indicator calculated only from data for male workers in the firms and 2 variants of e sigma, the component of sigma orthogonal to firm size. The results are very alike: in the OLS-regressions wage dispersion exhibits a negative impact on firm growth over the biggest part of the distribution of sigma, the panel fixed effect regressions show no impact of wage dispersion. 5. Firm Performance and Wage Inequality between Firms With the exception of compensation of white-collar work, the fixed effects estimates in the previous section have revealed a limited impact of temporary changes in withinfirm wage dispersion on firm outcomes. As the compensation system of a firm cannot be expected to change very much over time, measurement errors might lead in the fixed effects model to a worse signal to noise ratio, and therefore to downward-biased coefficients. It is suggestive to ask whether differences in the structure of wages in the longer run do have an impact on firms fortunes. 18 One possible explanation for the positive influence of wage dispersion on average pay for blue-collar workers might be the (unobserved ) prevalence of piece rates. If some firms use piece rates whereas others don't, we might expect to see a positive dispersion-earnings relation (Lazear, 1996). 12

13 We will do this by running group means regressions of the performance indicators on all variables used in the regression of Table 2. To account for technological differences and other effects of particular industries, we also include the log average real wage of the industry in which the respective firm is operating. Since our sample is unbalanced this variable is systematically related to the growth path of real incomes over the period under consideration. If, for instance, a particular firm is present only over the first couple of years, our compensation measures as well as the log industry wage will be smaller than for a firm present during the whole period. Consequently, in the group means regressions the industry wage should pick up not only industry-specific, but also time-effects. Table 6 Table 6 presents the results of the between-firm regressions. The estimates are weighted by the number of periods a firm is observed in the sample. It turns out that sigma has a significant impact on both the standard white collar wage as well as the standard blue collar wage with point estimates quantitatively much larger than for within-firm effects. For white collar workers the estimates imply that firms with little inequality on average also pay relatively low wages. Similarly, very high inequality seems to be detrimental for white collar compensation. Also here, we find that firms with a long-run wage dispersion close to the average, that is for a value of sigma equal to 0.236, pay the highest wage. This result remains qualitatively unchanged once we confine the analysis to observations where different inequality indicators are used (Table 7). The effect of sigma on the standardized white-collar wage remains large and highly significant, with exactly the same hump-shaped form. For blue-collar wages the results are similarly hump-shaped and in most cases highly significant. But here, the turning point is much later: at a value of sigma of 0.28 in Table 6. This means, that for the most part of the empirical distribution of sigma, a positive dispersion-wage schedule can be observed. This result is very similar to the OLS-regressions in Table 2. The general tone of the findings does not change, if other indicators for inequality are used (Table 7). It is even more pronounced, when wage dispersion only within males is used (Panel 7B and 7D). Summing up, we are able to find for the most part a positive association between inequality and blue-collar wages between firms, but not within firms over time. The lack of such a finding might be due to the relative constancy of the wage structure over time in most firms. Table 7 Column (3) of Table 6 shows the results of the group means regression using mediumrun firm growth as the dependent variable. Just like for the within-firm effects, wage dispersion does not determine the longer run growth (or decline) of firms. Of course, we cannot conclude from these results that the employment dynamics of firms is unrelated to the wage hierarchy. In fact, a sensible hypothesis would be that gross worker turnover is systematically related to the structure of earnings within a firm. We leave this question to future research. Furthermore, it is not even clear, if employment 13

14 growth is the right indicator for corporate performance. The result of "downsizing" is intended to be efficiency-enhancing. This, in turn, may allow firms to pay higher wages. Let us take a look at the impact of explanatory variables on compensation. The agestructure seems to have some power to explain differences in pay. Again, notice that the dependent variable is the average wage of workers with standardized characteristics. So, the impact of the various variables listed in Table 6 are not compositional effects. In particular, a high share of the very young workers depresses the white-collar wage. For blue-collar workers a larger share of older workers lowers the expected compensation but the effect is of lower significance. Observed differences in pay between firms cannot be attributed to differences in firm size. This confirms what we have seen from the rather crude description of the data in Figure 3. Compared to evidence from other countries, this is a striking result. One possible reason might be that the structure of the Austrian economy relies heavily on small and medium-sized firms, the performance of which is comparably high relative to other countries. Firms with a larger share of blue collar workers have been able to pay better. This holds for both, the typical blue-collar and the typical white-collar worker. One possible reason might be that those firms with a more efficient organization of non-production work are more profitable and able to pay higher wages. The percentage female has a modest effect while the percentage of alien workers has no impact on pay. Finally, we observe that the industry wage is a very strong predictor of firms wage levels. The elasticity of the firm wage with respect to the aggregate industry wage is not statistically different from unity for white-collar workers. Bluecollar compensation is somewhat less (but still highly) affected by the industry wage. The high variation in employment growth which is observed in the data (see Table 1) cannot be explained by characteristics of the firm related to the size and the composition of the work force. There are only two exceptions to this result. The first is that the longer-run rate of change in firm size is negatively related to the share of older workers. Similarly, a larger share of tenured workers seems to be detrimental to firm growth. 6. Conclusions Pay structure is an important determinant of firm performance. Although there has been a lot of research on optimal incentive schemes in firms, the empirical literature is relatively sparse. This paper is one of the first attempts to evaluate the impact of wage dispersion within firms on firm productivity. In the lack of financial variables, we use standardized wages as well as employment growth as our performance indicators. In the case of white-collar wages the relationship is found to be non-monotonic. For low levels of wage dispersion more inequality seems to be beneficial for firm earnings. If dispersion grows too high, however, this is detrimental in terms of compensation. While it is difficult to give these results a clear interpretation in terms of theoretical 14

15 arguments brought forth on the efficiency of compensation structure, we want to mention that our findings are consistent with the hypothesis that too little inequality is harmful for outcomes because incentives are too small, whereas on the other side of the scale there could be a fairness constraint leading to more inefficient outcomes if inequality becomes too large 19. For blue-collar wages the situation is different. For the most part of the observed range of inequality in the firms, standardized wages rise with wage dispersion. This picture is valid only between firms, but not within firms over time. No consistent pattern in the case of employment growth has been found. These findings are only a first step. Further research should concentrate directly on financial performance, on the one hand, and on the implications of wage dispersion on the individual behavior of workers in the firm, on the other hand. 19 See Falkinger (1995) for an interesting model from a public finance perspective, deriving a range of possible income distributions which are feasible in terms of incentives and social stability. 15

16 References Adams, Stacy J.: Towards an Understanding of Inequity, Journal of Abnormal and Social Psychology 67, 1963, Agell, Jonas and Per Lundborg: Theories of Pay and Unemployment: Survey Evidence from Swedish Manufacturing Firms, Scandinavian Journal of Economics 97, 1995, Akerlof, George A. and Janet L. Yellen: The Fair Wage-Effort Hypothesis and Unemployment, Quarterly Journal of Economics 105, 1990, Baker, George, Gibbs, Michael and Bengt Holmstrom, The Wage Policy of a Firm, Quarterly Journal of Economics, 109, 1994, Baker, George, Gibbs, Michael and Bengt Holmstrom, The internal Economics of the Firm: Evidence from Personnel Data, Quarterly Journal of Economics, 109,1994, Blanchflower, David G., Oswald, Andrew J. and Peter Sanfey: Wages, Profits and Rent-sharing, NBER WP #4222, Blanchflower, David G. Oswald, Andrew J. and Mario D. Garrett: Insider Power in Wage Determination, Economica 57, 1990, Blinder, Alan S. and D. H. Choi: A Shred of Evidence on Theories of Wage Stickiness, Quarterly Journal of Economics 105, 1990, Bowles, Samuel and Herbert Gintis: Productivity-enhancing Egalitarian Policies, International Labour Review 134, 1995, Brown, Charles and Medoff James: The Employer Size Wage Effect, Journal of Political Economy 97, 1989, Cowherd, Douglas M. and David I. Levine: Product Quality and Pay Equity between Lower-level Employees and Top Management: An Investigation of Distributive Justice Theory, Administrative Science Quarterly 37, 1992, Davis, Steve J. and John Haltiwanger: Wage Dispersion between and within U.S. Manufaturing Plants, , Brookings Papers on Economic Activity, Microeconomcs, 1991, Falkinger, Josef: Social Stability and the Equity-Efficiency Trade-off, 1995, mimeo, University of Regensburg. Fichtenbaum, R. and H. Shahidi: Truncation Bias and the Measurement of Income Inequality, Journal of Business and Economic Statistics, Freeman, Richard B.: Union Wage Practices and Wage Dispersion within Establishments, Industrial and Labor Relations Review 36, 1982, Gusenleitner, Markus, Trends in der Entwicklung der Einkommensungleichheit, Eine Analyse der Einkommensverteilung österreichischer Arbeitnehmer in den Jahren , University of Linz. Hibbs, Douglas A. Jr. and Hakan Locking: Wage Dispersion and Productive Efficiency: Evidence for Sweden, mimeo, Trade Union Institute for Economic Research, Stockholm, Lazear, Edward P.: Pay Equality and Industrial Politics, Journal of Political Economy 97, 1989, Lazear, Edward P. and Sherwin Rosen: Rank-Order Tournaments as Optimum Labor Contracts, Journal of Political Economy 89, 1989,

17 Lazear, Edward P.: Performance Pay and Productivity, mimeo, Hoover Institution, Leonard, Jonathan S.: Executive Pay and Firm Performance, Industrial and Labor Relations Review 43, 1990, 13S-29S. Levine, David I.: Fairness, Markets, and Ability to Pay: Evidence from Compenstion Executives, American Economic Review 83, 1991, Levine, David I.: Cohesiveness, Productivity, and Wage Dispersion: Journal of Economic Behavior and Organization 15, 1991, Milgrom, Paul R.: Employment Contracts, Influence Activities, and Efficient Organization Design, Journal of Political Economy 96, 1988, Milgrom, Paul and John Roberts: The Efficiency of Equity in Organizational Decision Processes, American Economic Review, Papers and Proceedings 80, 1990, Milgrom, Paul R. and John Roberts: Economics, Organization, and Management, Prentice Hall, Pfeffer, Jeffrey and Alison Davis-Blake: Salary Dispersion, Location in the Salary Distribution, and Turnover among College Administrators, Industrial and Labor Relations Review 45, 1992, Winter-Ebmer, Rudolf: Endogenous Growth, Human Capital, and Industry Wages, Bulletin of Economic Research, 1994/4, Winter-Ebmer, Rudolf and Josef Zweimüller: Immigration and the Earnings of Young Native Workers, Oxford Economic Papers,

18 Table 1: Summary Statistics Mean Standard deviation N employment growth log (income blue-collar) log (income white-collar) sigma mean wage in firm in 1000 ATS % age under % age over log(firm size) % blue-collar % foreigners % women % tenure over 3 years

19 Table 2: Firm-Level Wage Regressions blue-collar wage white-collar wage ols fixed effects ols fixed effects sigma (4.044) (0.48) (8.95) (2.12) sigma (3.26) (0.70) (9.14) (1.31) % age under ( (0.69) (10.07) (3.41) % age over (5.24) (1.18) (4.44) (0.05) log(firm size) (1.00) (0.95) (5.07) (0.67) % blue-collar (13.7) (0.98) (6.00) (1.21) % foreigners (2.22) (1.06) (0.37) (2.11) % women (2.59) (4.84) (4.98) (1.43) % tenure over 3 years (1.24) (2.78) (0.76) (2.85) N adj R F-Test *** *** 5.45*** autocorrelation F-Test for joint significance of sigma and sigma 2. 19

20 Table 3: Robustness checks for wage regressions blue-collar workers white-collar workers ols fixed effects ols fixed effects 3A sigma (3.50) sigma (2.15) (1.12) (1.28) (8.95) (9.01) (1.91) (2.12) F-Test N B e sigma (4.52) 2 e sigma (3.33) (0.26) (0.69) (2.42) (7.53) (2.91) (1.33) F-Test N C e sigma (5.12) 2 e sigma (2.33) (0.69) (1.45) (3.81) (5.75) (0.62) (1.78) F-Test N Notes: Panel 3A: sigma calculated using only male workers in the firm Panel 3B: e sigma residual from regressing sigma on firm size Panel 3C: sigma calculated using only male workers 20

21 Table 4: Employment Growth ols fixed effects sigma (2.61) sigma (2.01) mean income in 000 ATS (2.04) % age under (1.14) % age over (2.41) log(firm size) (1.42) % blue-collar (1.06) % foreigners (0.92) % women (0.63) % tenure over 3 years (3.16) (0.74) (0.71) (2.00) (0.84) (0.61) (13.27) (1.22) (0.38) (2.21) (0.07) N adj R F-Test autocorrelation F-Test for joint significance of sigma and sigma 2. 21

22 Table 5: Robustness checks for employment growth regression 5A 5B ols fixed effects ols fixed effects sigma (1.83) sigma (1.23) (1.30) (0.94) (1.94) (1.39) (0.25) (0.11) F-Test N C 5D ols fixed effects ols fixed effects e sigma (2.94) 2 e sigma 1.19 (1.51) (0.90) (0.86) (2.19) (0.56) (0.22) (0.05) F-Test N Note: Panel 5A: as Table 4 without outliers, employment growth rate <-0.5 or >2.0 Panel 5B: sigma calculated using only male workers in the firm Panel 5C: e sigma residual from regressing sigma on firm size Panel 5D: sigma calculated using only male workers 22

23 Table 6: Between Firm Effects (Group means regressions; weighted least squares) white-collar wage blue-collar wage employment growth sigma (4.51) sigma (4.31) % age under (3.68) % age over (1.02) log(firm size) (1.38) % blue-collar (4.09) % foreigners (0.13) % women (1.21) % tenure over 3 years (0.36) (3.11) (2.48) (1.45) (1.58) (0.04) (6.94) (0.54) (1.65) (0.46) (0.66) (0.40) (0.87) (1.72) (1.46) (1.03) (1.09) (0.54) (3.00) firm wage in 000 ATS (1.23) log (industry wage) (31.15) (19.09) (1.10) N (groups)

24 Table 7: Robustness checks for between-firm effects (Group means regressions; weighted least squares) white-collar wage blue-collar wage employment growth 7A sigma (5.59) sigma (5.24) (1.83) (0.99) (1.20) (0.84) N (groups) B e sigma (2.14) 2 e sigma (3.96) (3.52) (2.02) (1.20) (0.30) N (groups) C e sigma (3.13) 2 e sigma (3.44) (3.09) (0.58) (1.39) (0.18) N (groups) Notes: Panel 7A: sigma calculated using only male workers in the firm Panel 7B: e sigma residual from regressing sigma on firm size Panel 7C: sigma calculated using only male workers 24