Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
1 Salaries, Salary Growth, and Promotions of Men and Women in a Large, Private Firm BARRY A. GERHART and GEORGE T. MILKOVICH Studies of differences in earnings between men and women have sought to adjust the raw differential for gencler-related cliffer- ences in factors thought to reflect individual differences in productivity. In practice, these studies have examined differences in pay between men and women controlling for (1) supply-side factors, such as investment in human capital (e.g., general and firm- specific work experience, education) and (2) demand-side factors intencled to represent the relative scarcity of specific types of labor in specific markets (e.g., occupation, in- clustry, geographic region). In most in- stances, this research has used national sur- vey data on inclivi(luals working in many different (and unknown) firms. In reviewing this research, Treiman and Hartmann (1981:19) concluded that "worker characteristics account for very little of the difference in earnings (between men and women]." Gen(ler-related differences in de- mand-sicle factors, mainly occupational dis- tribution, were found to explain a larger part of the earnings gap. Nevertheless, after ad- justing for both supply- and demand-sicle factors, a substantial portion of the earnings differential between men ancT women re- 23 mains. This remaining portion is taken as evidence of labor market discrimination against women and as being indicative of researchers' inability to identify, measure, and control for all aspects of worker productivity. At least two problems exist with pay discrimination research of the type de- scribed above. First, some variables fre- quently included on the right-hand side of the wage equation may themselves be en- dogenous to the process of discrimination (Blinder, 1973; Cain, 1986; Oaxaca, 1973~. Thus, for example, women may have less experience in the labor market than men because their work opportunities in the market have been less favorable than tho.s'? available to men having similar productive abilities. As another example, it may be that gender-basec] occupational (and firm) segregation (see Reskin and Hartmann, 1985) reflects not only different aspirations an work values of men ant! women, but also Differences that exist between men and women in terms of preferences for different types of occupations may also reflect discrimination, whether its source is differential socialization in families, schools, and other premarket institutions, or is market based.
24 unequal access to firms, occupations, and jobs. Fuchs (1971), in fact, contended that virtually the entire earnings gap between men and women could be explained if an occupational classification scheme having sufficient detail were used, but he noted that the question of why men and women had `different occupational distributions then had to be resolved (see also Sanborn, 1964~. Although it is clear that access to jobs and occupations is an important determinant of earnings, little research has examiner] the question of why such attainments slider ac- cor(ling to gen(ler.2 A second problem with marketwicle stud- ies of gender-based discrimination is that they provide no direct information on work- er productivity at the level of the firm. As one consequence, previous research has had to rely on indirect measures of productivity (e.g., experience, tenure, and education). Some evidence indicates, however, that such measures may converge poorly with mea- sures of productivity at the level of the firm (e.g., Brown, 1982; Mecloff and Abraham, 1981~. All firms attempt to assess the produc- tivity of their workers in one way or another. Many firms formalize this process by, for example, regularly conducting performance appraisals of employees. In these firms, compensation and internal-staffing decisions (e.g., promotion) are often explicitly based on such productivity measures.3 These mea- sures have a number of desirable charac- teristics. First, they are designed to assess worker productivity in a specific job in a specific firm. Second, such measures are 2This issue of differential attainment due to possible unequal access to firms, occupations, and jobs is some- times referred to as employment discrimination. 3Based on a survey of personnel and industrial relations executives, the Bureau of National Affairs (1983) concluded that performance appraisal results are used by 86 percent of firms for making salary in- crease decisions and by 79 percent of firms for making promotion decisions concerning their white-collar workers. PAY EQUITY: EMPIRICAL INQUIRIES influenced by human capital characteristics only insofar as the latter are useful in per- forming the particular job in question. In other words, firm-level productivity mea- sures potentially carry information on the quality of the match between the worker's abilities and what the job requires. Thirdly these job-specific productivity measures re- flect differences in worker motivation, which, together with the abilitybob requirements match, affect actual inclividual performance levels. Given the discussion above, three ave- nues of research on labor market discrim- ination would appear to be especially useful. First, firm-level research is neeclecl that examines differences in men and women's salaries adjusted for possible differences in job-specific productivity measures. Second, research on the determinants of occupa- tional and job attainments would also be helpful. Some evidence indicates that sub- stantial gender-based, within-firm job seg- regation may exist (BielLy and Baron, 1986), which suggests that job level may be an important determinant of gender-based sal- ary differences within firms. Although with- in-firm analyses have been rare in the lit- erature, they have indeed found the effect of job level on earnings differences to be quite large, especially relative to withinjob differences (Halaby, 1979; Malkiel and Mal- kiel, 1973; Rosenbaum, 19851. Third, lon- gitudinal data would permit stu(ly of changes in salary. Further, such data would aid in assessing the extent to which firms' pro- motion practices contribute to job-level dif- ferences. Again, it would also be useful to know what role job-specific performance measures play in decisions on promotion and salary increases. In this paper we examine possible gencler- based differences in attainment in a large, private firm. Specifically, we focus on three general issues. First, controlling for job level, performance ratings, and individual characteristics, is there a salary disadvantage for women? Second, given longitudinal data,
SALARIES, SALARY GROWTH, AND PROMOTIONS we can move beyond] cross-sectional anal- yses and examine models of salary change over time, addressing questions of the fol- lowing type: Do women's salaries grow at the same rate as men's? Do men and women receive equal salary increases for a given level of performance? Do men and women receive equal salary returns from promo- tions within the firm? Third, controlling for initial job level, performance, and individual characteristics, do men and women receive different numbers of promotions over time? Unlike much previous research, we measure not only whether a promotion has occurred, but also the number of promotions over time. In other words, are women, for ex- ample, held to higher promotion standards (Olson and Becker, 1983~? THE FIRM The data for this analysis are from the personnel information system of a firm that produces a highly diversified set of industrial and consumer products. The firm has rough- ly 100 manufacturing operations in more than 30 states. It enjoyed general financial success cluring the period of the study, as evidences! by revenues that grew faster than the Consumer Price Index and a return on equity that comfortably exceeded the me- clian for the Fortune 500. Exempt jobs are the focus of this study. Professional, managerial, sales, and tech- nical jobs are the major broac] categories. Examples of some of the most common job titles include engineer, senior engineer, sales representative, area sales manager, administrative assistant, technologist, su- pervisor, and maintenance supervisor. Compensation policies and practices of the firm are typical of those in the Fortune 500. The firm participates, for example, in over six annual salary surveys for jobs in- clucled in this study. The focus of these surveys varies some focus on a selected group of perhaps 10 to 15 product market competitors and others focus on labor mar- 25 Let competitors that employ persons with similar skills or have similar occupations. Statistical methods are used to combine the results of the surveys. Consiclerable jucig- ment is also exercised, however, because different degrees of confidence are placecl on the results of the various surveys. This is consistent with Rynes ant] Milkovich's (1986) argument that ad hoc judgments are typically made throughout the process. Strategy also plays an important role in determining pay level. During the time period of the study, a policy of "paying with the leaders" was follower! for the jobs we examined. In practice, this policy meant establishing the pay policy line at the fiftieth percentile of the group of pay leaclers. The pay structure of the jobs included in this study was maintained with the help of a single, national job evaluation system. The structure is defined by 15 job classes, or levels, each with minimum and maximum rates of pay roughly 20 percent below and above the midpoint. A Conference Boarc] (1984) survey of 557 major U. S. firms found that the median number of levels in exempt pay structures for all industries was 19. The compensable factors used in the job eval- uation system are education/knowledge re- quired, experience required, complexity of duties, working conditions, and responsi- bility. Of the 491 firms using formal job evaluation in the Conference Board study, 90 percent used a formal plan. Of this group, approximately 20 percent used a plan of the general type used by the firm we studied. An explicit pay-for-performance policy ex- ists for the determination of individual pay increases. The policy is implemented through the use of annual merit increase guides (see Milkovich and Newman, 1987, for exam- ples). These gui(les are designed to control the cost of annual pay increases, as well as to encourage a distribution of increases to employees. According to the Conference Board survey, 83 percent of firms used such guides. In the firm we studied, recom- mencled salary increase ranges were a func-
26 tion of (1) merit rating on one axis ant! (2) current position in the salary range on the other axis. Higher ratings and less pene- tration into the salary range at a particular level were associated with both larger and more frequent salary increases.4 Promotions are based on performance as well, and acIditional consideration is given to years of experience with the firm. A salary increase goes with promotions. Again, the corporate compensation department is- sues yearly guidelines that specify the size of promotional increases. The importance of examining the firm's promotion system is increased by the fact that the firm engages in a fairly strong practice of promotion from within. Thus, women's access to higher level jobs is most often governed by decisions made while they are current employees. Direct access from the external labor market is limited. Performance is assesses] through a formal, annual performance appraisal process.s The immediate supervisor rates each employee on a 4-point scale, with 4 being the highest performance level. Raters are instructed to consider not only how well job requirements were satisfied, but also the difficulty of the job requirements and the appropriateness of methods used to satisfy the job require- ments. The numerical rating given is sup- plemented by a written description of the subordinate's performance during the year. The complete appraisal is typically reviewed by a higher level manager. According to The Hay Group, the actual median increase nationally for 1982 ranged from 9 percent for "average performers" to 14.3 percent for "outstanding performers." By 1986, the corresponding figures were approximately 6 percent and 9 percent, respectively (The Hay Group, 1986~. The corresponding figures for the salary guides used in the firm were similar. Main (1986) has argued that supervisory ratings of performance are not "admissible" because they "might reflect discrimination." The empirical evidence does not support this hypothesis, however, despite the fact that a large amount of both laboratory and field research has been devoted to this question (see Dipboye, 1985, for a review). PAY EQUITY: EMPIRICAL INQUIRIES The metho(1 of appraisal, as well as its central role in making promotion and salary increase decisions, is consistent with the way firms typically operate with respect to their exempt employees. The review system is also a common feature of performance appraisal plans.6 In regard to equal employment oppor- tunity (EEO), the firm's standard training for its managers includes materials on EEO compliance with respect to staffing, access to training, compensation, and performance appraisal. The inclusion of EEO issues in training programs is typical of large firms.7 Corporate personnel monitor managers' ac- tions in these areas and encourage improve- ment in those displaying subpar perfor- mance. Several external events may have also had an elect on the firm's human resource prac- tices. First, as with many large firms, it faced EEO litigation during the 1970s. The major case involvecI a cIass-action suit filed by female hourly employees in a very small number of the approximately 100 plants. This suit was eventually settIe(1 out of court for a substantial sum of money. As part of the settlement, the firm stated that it would also develop a plan to enhance the hiring ancI promotion of female salaried employ- ees. The litigation, however, dicI not involve female salaried employees. Further, no for- mal or numerical goals for salaried women came about as a result of the litigation activity. A second external influence was the reces- sion of the early 1980s. Again, as with many other large firms, the firm we stuclied re- 6See the survey results reported in footnote 3 on promotion and salary increases. The same survey also indicates that rating scales and essays are the most commonly used methods of appraisal. Further, 97 percent of firms review appraisals at a higher level. iA Bureau of National Affairs (1985) survey found that among firms with over 1,000 employees, over 60 percent included EEO issues in their manager training programs. Further, EEO was the fourth (of 19) most commonly included issue in such programs.
SALARIES, SALARY GROWTH, AND PROMOTIONS duced the size of its white-colIar work force during this period. Much of this reduction came about through early retirement. As a result, the number of men in exempt jobs actually declined slightly over the course of the stucly. In contrast, the number of women in exempt jobs grew by over 50 percent cluring the same period. METHOD Sample ant! Measures This study involves two samples: exempt employees in job levels 1 through 7 (1) present in 1986 (the cross-sectional sample) and (2) present in both 1980 an(l 1986 (the Analyses longitudinal sample). The majority of ex- empt employees are employee] in these lev- els (approximately 84 percent of men, 97 percent of women).8 The cross-sectional sample comprises 2,412 women and 9,647 men. The longitudinal sample comprises 840 women and 5,550 men. The depenclent variables are annual sal- ary, salary growth, and promotions.9 The first set of independent variables, referred 8The firm did not provide data on former employees or on active employees in levels 8 through 15. 9The number of promotions is defined using salary increase codes, which indicate the reason for an in- crease (e. g., merit or promotion). The number of promotional increases between 1980 and 1986 is used as the definition. A promotion can occur without a change in job level. Similarly, job level can change in the absence of a promotion. Thus, the correlation between changes in job levels and the number of promotions is high but not perfect (r = . 731. Although some evidence suggests that women receive more within-level promotions (Flanders and Anderson, 1973; Stewart and GudyLunst, 1982), this correlation did not differ by gender, which suggests that changes in job level and the number of promotions, as defined by the firm, were related in a similar manner for both men and women. Moreover, the female/male ratio of change in job levels was virtually identical to the female/male ratio of the number of promotions (1.33 and 1.38, respectively). Thus, we chose to use the firm's definition of a promotion rather than attempt to draw inferences from changes in job level. 27 to as human capital (HC) variables in this study, are firm tenure, it job tenure (years at a particular job level), potential experi- ence (age minus years of schooling minus six), and education dummies for highest degree. Squarect terms for job tenure, firm tenure, and potential experience are also includecI. In the cross-sectional analyses, the most recent performance rating prior to the most recent salary change is used. In the longitudinal analyses, the average performance rating over the 1980-1986 time frame is used. The variables used are sum- marizec] in Appendix A. The following salary equation is estimated separately for the years 1980 and 1986: In(Si~) = XilBt + eat (1) where In(Si`) is a vector of the natural log- arithm of salaries for i persons cluring time period t; Xi~ is a matrix of observations on the exogenous variables contained in Ap- pen(lix A; B is a coefficient vector; and e is a disturbance term composed of all un- measured causes of salaries. Given the availability of longitudinal data, we also estimate the following salary growth equation: In(Si~986/Si~98o) = Xi~980B + ei~g80 (2) Thus, salary growth is definecl as the natural logarithm of the ratio of 1986 salary divided by 1980 salary. Finally, we estimate a similar equation for the number of promotions received clur- ing the 1980-1986 period: PROM = Xi~g80B + ei~g80. (3) infirm tenure is based on the date used for cal- culating benefits. It is important to note that this date can differ from the original hiring date. Therefore, this measure should give an accurate indication of the amount of actual time spent with the firm even for persons not continuously employed with the firm.
28 The salary growth and promotion equa- tions provide a unique opportunity to study the attainment of men and women over time. These analyses may help explain the process by which men and women reach the differential levels of attainment so wide- ly observed in cross-sectional research. Estimates of the three equations were obtained separately for men and women and used to decompose salary, salary growth, ant] promotion differences into two com- ponents (Blinder, 1973; [ones, 1983~: (1) differences in mean levels of endowments and (2) differences in coefficients or prices received for those endowments. Because the result of a decomposition varies as a function of which group is used as the standard (Cain, 1986), we report decom- positions using both the advantaged and disadvantaged group as the stanciard. In a(l(lition, we report corresponding "a(ljusted ratios" (Cain, 1986:746~. As ~liscussed by Blinder (1973), Cain (1986), Oaxaca (1973), and others, such decompo- sitions assume that the variables on the right side of the equation are exogenous to tithe inclusion of individual-specif~c intercepts in Equations (2) and (3) could be used to eliminate bias due to any lack of independence between X time- invariant components of e (Mundlak, 1978~. Given data at two points in time, the use of individual-specif~c intercepts is equivalent to a first-di~erencing model. There are at least two problems with this model, however. First, variables that do not change over time (e.g., firm tenure) must be excluded. Second, the model exacerbates any unreliability problems. As a result, differences in parameter estimates may stem from unreliability rather than elimination of the effects of nonindependence of X and e. In this analysis, for example, performance rating is a key variable. King et al. (1980) have estimated the upper-bound reliability of supervisory ratings to be approximately .60. In our analysis, the correlation between performance ratings in 1980 and 1986 is approximately .20. Using a formula given by Guilford (1954:394), the resulting reliability of the change in performance rating would then be approximately .50. Across adjacent years, the corre- lation between performance ratings is closer to .40, which results in a difference score reliability of .33. By averaging performance ratings and counting pro- motions over time, the reliability problem is reduced. PAY EQUlf Y: EMPIRICAL INQUIRIES gender. If not, additional equations for these right-sirle variables can, in theory, be a(lde(1 to the model. In the present context, for example, an equation for job level might be warrantecl. As Blinder argued, however, implementing this latter strategy is often clifficult in practice because of identification nrohlems. Consistent with this point, we found the determination of job level ant] salary to be so closely intertwined that iden- tification of a two-equation mode] was not possible. As an alternative, we follow the Blinder and Oaxaca approach of estimating a series of equations, introducing variables of more questionable enclogeneity in steps to the X vector. Thus, in the cross-sectional salary analyses, for example, we introduce job level last. This strategy permits an exami- nation of how the decomposition results change in response to different model spec- ifications. RESULTS Table 1-1 reports mean salaries for all employees active in 1986 and for employees active in both 1980 an(l 1986. The female/ mate salary ratio is somewhat higher based on employees active in both 1980 and 1986. One explanation may be the fact that this latter group does not include new entrants to the jobs. New entrants are more likely to begin in lower job levels and have less time accumulated at each level. Both factors contribute to Tower pay relative to higher tenure employees. The larger growth of women's employment in the firm's exempt jobs relative to that of men suggests that most new entrants were women. Table 1-2 reports mean 1980 and 1986 salaries and their ratio as a function of gender and 1980 job level. These results allow comparisons between men and women start- ing at the same level in 1980 (but not necessarily at the same level as of 1986~. Overall, the ratio of women's salary to men's salary is .84 in 1980 and rises to .88 in 1986,
SALARIES, SALARY GROWTH, AND PROMOTIONS TABLE 1-1 1986 Salaries of Men ancI Women, by 1986 lob Level 1986 Cross-Sectional Sample 29 1986 Longitudinal Sample 1986 Women Men Women Men Job Mean Mean Mean Mean Level N Salary N Salary W/Ma N Salary N Salary W/Ma All 2,412 $35,503 9,647 $42,049 .84 840 $40,004 5,550 $45,620 .88 1 859 29,451 1,777 31,875 .92 114 31,092 386 32,519 .96 2 412 32,870 896 35,222 .93 156 34,675 325 35,869 .97 3 521 36,209 2,151 37,795 .96 174 37,697 1,032 38,789 .97 4 162 40,745 641 42,935 .95 88 41,589 468 43,584 .95 5 286 43,925 2,184 46,307 .95 182 44,274 1,617 46,819 .95 6 158 50,568 1,859 53,410 .95 109 51,067 1,607 53,557 .95 7 58,968 14 55,415 138 59,002 .94 17 55,805 114 .95 aW/M = women/men ratio. thus eliminating 25 percent of the salary differential. Consistent with this narrowing differential, the mean salary for women in- creased by a greater percentage (61 percent) between 1980 and 1986 than did the mean salary for men (54 percent). Within 1980 job levels, similar trends emerge, although the salary differentials are much narrower. The ratios of women's mean salaries to those of men range from .93 to .95 in 1980. In all cases, these ratios in- creased between 1980 and 1986. Again, the decreasing salary differentials are consistent with the greater salary growth of women observed at each 1980 job level. A cursory examination of the 1980 job levels having sufficient numbers of both men an(l women (1 through 3) reveals no obvious relationship between job level and gencler differences in salary or salary growth. One possible concern with the numbers in Tables 1-1 and 1-2 is that there may be a selection process related to gentler and salary because the longituclinal sample in- cludes only employees still active in 1986. For example, one scenario is that men were observed to receive lower salary increases because men experiencing larger salary growth were more likely to have left the firm. Alternatively, discrimination against women could have resulted in all but the "cream of the crop" quitting, leaving us to TABLE 1-2 1980 and 1986 Salaries of Men and Women, by 1980 Job Level 1980 Women Men Women/Men Job 1986a 1986a 1986 Level N 1980 1986 1980 N 1980 1986 1980 1980 1986 1980 All 840 $24,786 $40,004 1.61 5,550 $29,606 $45,620 1.54 .84 .88 1.14 1 348 21,954 35,762 1.63 872 23,516 36,479 1.55 .93 .98 1.14 2 177 24,014 38,862 1.62 774 25,171 40,009 1.59 .95 .97 1.05 3 195 26,635 43,048 1.62 1,515 28,220 44,229 1.57 .94 .97 1.09 4 48 28,881 46,760 1.62 485 30,848 47,661 1.55 .94 .98 1.14 5 58 31,894 50,042 1.57 1,307 33,801 51,346 1.52 .94 .97 1.10 6 14 35,717 52,768 1.48 571 37,705 55,225 1.46 .95 .96 1.03 7 0 26 36,840 54,321 1.47 al986 salary/1980 salary. bWomen tsalary 1986/salary 1980] 1/men [salary 1986/salary 1980] - 1.
30 observe only the latter group, which re- ceived relatively large increases. Supplementary data suggest, however, that these are not likely problems. First, the voluntary quit rate averaged only about 2 percent per year over the course of the study. Second, anti more important, we examined salary data on all men and women active in 1980 or 1984. The average salary of men and women grew by 35 percent and 40 percent, respectively, during this period. Restricting the sample to only employees active in both 1980 and 1984 yielded growth rates for men and women of 46 percent and 53 percent, respectively. The key point is that the femaTe/male ratio of growth rates was approximately 1.14 in both cases, which suggests that focusing only on employees active in both years does not influence the observer] relative salary growth of men and women. Finally, note that the ratio of 1.14 is identical to that found in the first row, last column of Table 1-2. Decomposition of Salary Differences As Table 1-3 indicates, human capital variables alone explain up to 32 percent of the salary advantage of men in the cross section. Adding job level and performance rating to the mode} raises the explaine(l percentage to between 68 and 77 percent, which suggests that most of the gap is due to the fact that men tend to hold higher level jobs. The corresponding adjusted sal- ary ratios (AM and AW) suggest that equal- izing human capital would raise the salary ratio to .86 to .90. Equalizing job level and performance rating as well would raise it to .95 to .97. 12 Note that adding performance rating ac- tually results in a slight decrease in the explained percentage of the salary cliffer- i2Within levels, very little of the pay gap could be explained. Because, however, the raw ratios were high (see Tables 1-1 and 1-2), the adjusted ratios were also high. PAY EQUITY: EMPIRICAL INQUIRIES ential. Although not shown here, this find- ing was even stronger in analyses of specific job levels. The reasons are twofold. First, the mean performance rating of women is slightly higher than that of men (2.59 versus 2.52 overall in 1980~. Second, although women receive a slightly greater return for a given performance rating, inclu(ling per- formance rating in the mo(lel changes some of the other coefficients. 13 In summary, the cross-sectional results suggest that men, in general, receive greater returns to explanatory variables. These re- sults are consistent with previous cross- sectional analyses of male-female salary clif- ferentials (Cain, 1986~. Decomposition of Salary Growth Differences Although women's salary levels fell short of men's, women, as noted earlier, expe- rienced greater salary growth in percentage terms. Moreover, as the decompositions reported in Table 1-4 (remonstrate, this salary growth advantage cannot be entirely explaine(1 by the mo(lels. The parentheses in Table 1-4 in(licate instances in which (differences in coefficients favor women. Ap- pen(lix B reports the regression results. In the overall analyses, 40 to 49 percent of women's greater salary growth can be explaine(l by differences in human capital. Adding average performance rating raises the explained part to 50 to 62 percent. Finally, the inclusion of job level and num- ber of promotions raises this figure to 72 to 74 percent. Within specific job levels, there is a good clear of variance in the extent to which human capital, average perfor- mance rating, an(l promotions can explain the salary growth advantage of women. Table 1-5 shows the contribution of spe- cific factors to the salary growth (lifferential i3For example, adding performance rating to the equation having human capital variables increased the advantage realized by men in returns to job tenure.
SALARIES, SALARY GROWTH, AND PROMOTIONS TABLE 1-3 Decomposition of Salary Differences Decomposition Standard R2 Men Women Differ'?ntin 31 Variables in Equation Women Men Coeff.a Endow.b Coeff.a Endow.b Raw ARC AWE 1980 Longitudinal Sample Human capital (HC) .19 .29 .73 .27 .86 .14 .84 .88 .86 HC, perf. rating 80 (PA80) .21 .33 .76 .24 .88 .12 .88 .86 HC, job level .58 .71 .28 .72 .31 .69 .96 .95 HC, job level, PA80 .62 .73 .31 .69 .32 .68 .95 .95 N = 840 5,550 1986 Longitudinal Sample HC .29 .21 .80 .20 1.00 .00 .88 .90 .88 HC, PA86 .32 .28 .82 .18 1.01 -.01 .90 .88 HC, job level .79 .80 .21 .79 .29 .71 .97 .97 HC, job level, PA86 .82 .83 .23 .77 .30 .70 .97 .96 N = 840 5,550 1986 Cross-Sectional Sample HC .30 .31 .68 .32 .73 .27 .84 .89 .88 HC, PA86 .33 .37 .70 .30 .75 .25 .89 .88 HC, job level .81 .83 .25 .75 .22 .78 .96 .96 HC, job level, PA86 .83 .85 .27 .73 .24 .76 .96 .96 N = 2,412 9,647 aProportion of salary differential due to unequal coefficients or returns. bProportion of salary differential due to unequal endowments. CAdjusted salary ratio using men's coefficients as standard. Adjusted salary ratio using women's coefficients as standard. in the mode! containing only human capital variables. An important factor accounting for women's advantage is potential experi- ence. Because potential experience has a negative impact on salary increases, women benefit from having lower levels of potential experience and having a less negative coef- ficient. 14 Similarly, with respect to job level, women benef~tec] from having a lower mean and a less negative return. The addition of promotion to the mode} sharply reduces the 14The use of potential experience (age minus school- ing minus six), rather than actual experience, is prob- lematic for persons with intermittent labor force par- ticipation. Thus, the role of potential experience in explaining salary growth differences in our study may be partly artifactual. As an indirect test, we restricted the sample of women to unmarried women only. The resulting decomposition (of the full model) actually increased the importance of potential experience in explaining women's faster salary growth. importance of potential experience and job level in accounting for the differential. Another important factor is job tenure. 15 Women received larger percentage salary increase returns to job tenure. Recall that the firm's salary increase guicle recommends smaller percentage increases as employees progress within the salary range at a given level. Because this position in the salary range is not likely to be a perfect function of job tenure, the latter may not completely capture the effect of current position in the salary range. If men tend to be higher in the salary range than women, we would expect men to receive smaller increases than women. Women's higher salary growth was also MA more flexible functional forn1 (dummy variables for each year of job tenure) did not change this con- clusion.
32 PAY EQUITY: EMPIRICAL INQUIRIES TABLE 1-4 Decomposition of Salary Growth Differences, by 1980 Job Level Decomposition Standard R2 Men Women Differential Variables in Equation Women Men Coeff.a Endow.b Coeff.a Endow.b Raw AWC AMP All Levels Human capital (HC) .14 .21 (.51)e .49 ( 60) .40 1.14 1.07 1.06 HC, avg. perf. rating (AVGPA) .23 .30 (.38) .62 (.50) .50 1.09 1.07 HC, job level .15 .21 (.46) .54 (.53) .47 1.08 1.07 HC, job level, AVGPA .24 .31 (.25) .75 (.42) .58 1.11 1.08 HC, job level, AVGPA, promotion (PROM) .39 .43 ( 26) .74 (.28) .72 1.10 1.10 Level 1 HC .16 .29 (1.14) - .14 (1.02) - .02 1.14 .98 1.00 HC, AVGPA .23 .43 ( 51) .49 (.80) .20 1.07 1.03 HC, AVGPA, PROM .35 .57 (.67) .33 ( 77) .23 1.05 1.03 Level 2 HC .26 .30 (.50) .50 (1.10) - .10 1.05 1.03 1.00 HC, AVGPA .37 .34 - .13 1.13 ( 46) .54 1.06 1.03 HC, AVGPA, PROM .56 .45 - .28 1.28 ( 10) .90 1.06 1.05 Level 3 HC .14 .18 (.32) .68 (.52) .48 1.09 1.06 1.04 HC, AVGPA .31 .31 (.30) .70 (.58) .42 1.06 1.04 HC, AVGPA, PROM .53 .44 (.14) .86 ( 41) .59 1 OS 1 05 NOTE: N = 840 for women, 5,550 for men. aProportion of salary di~erentiai due to unequal coefficients or returns. bProportion of salary differential due to unequal endowments. CAdjusted salary ratio using women's coefficients as standard. Adjusted salary ratio using men's coefficients as standard. eParentheses indicate that coefficients favor women. partly a function of their higher average v ~ performance ratings. Nevertheless, the coefficient on average performance rating was smaller for women, which indicates they received a smaller payoff for performance. On the other hand, adding promotion to the mocle] eliminated this disadvantage. Decomposition of Promotion Differences Table 1-6 reports the mean number of promotions received by men and women between 1980 and 1986 overall an(1 as a function of 1980 job level. As in the case of salary increases, women had a (listinct promotion advantage. Of further interest, this advantage, like the women's salary growth advantage, does not decline at higher job levels. In fact, these simple descriptive sta- tistics suggest that the advantage may ne larger at higher job levels. Table 1-7 presents the results of the de- composition of the promotion cTi~erential under (li~erent model specifications. (Ap- pendix B presents the regression results.) A key finding is that the conclusion re- garding the final model depends on which coefficients are used as the stan(lard. When men's coefficients are used as the standard, the promotion advantage of women is com- pletely explained by the variables in the final motlel. Table 1-8 presents results of the decomposition into specific factors. In the mo(le} containing human capital vari- ables, the main advantage of women is again
33 o Cot C) C: an Cal o o Cat Cat V) Cat Cat 3 m C) a' of ._ 3 o 5- C) Cad t1 ~ to ._ V, o CO) a: o o ~ o Pa C) ~ C au ~ C) c) o Cat E-° ,- a; a.' of c 3 =~ ~ JO ~ . ~ o C) Cut o EM o 3 JO ~ o~ ~ m~ ~ - ; C) ~ o Cat E-° .= c~ ~ oo cM o ~ u) o c~ 'o cl 1 1 cM cM ~ o ;, 1 1 o 'o x o o oo ~ 1 ~ c~ o ~ ~ ~ x - c~ ~ c~ 1 1 c~ cN o ~ 1 '~ _ 1 1 ~ ~ ~ c~ ~ 1 °° 1 _ o co 1 ~ ~' 1 ~ C~ c~ 1 1 1 cO cs ~ o cO - - - ~ oO o x 1 ~ ~ c~ o o c~ - I=1 ~ 1 X 1 ~ 1 _ 1 oo o oo IO CD _ c~ 1 c~ 1 _ C~ o C~ _ o 1 _ u) CO CO _ o 1 '~ ~ oo C~ o ~o ~ ~ 1 '~ '= 1 CM CD O 1 ~ Xo ~ ~ ~ I ~ 8 1 ~ _ bC ._ C~ ~ _ Z ~ ° ~ ~ ~ V, - C~ ~o Ct cr o s~ ~ C~ ~ i C~ ~ V~ C c: o ._ c,, 3 o o ~ C~ o ~ v' ~ ~ o ~ o o c) c: ~ c, ~
34 TABLE 1-6 Mean Promotions for Men and Women, 1980-1986 1980 Job Women Men Level N Mean N Mean Women/Men All 840 1 348 2 177 3 195 48 58 14 1.29 1.29 1.38 1.34 1.34 .98 .57 5,550 .94 872 1.21 774 1.22 1,515 1.05 485 .94 1,307 .75 571 .32 26 .34 their combination of lower potential expe- rience and a less negative promotion return to potential experience. Perhaps the most interesting finding, however, is that women receive a much smaller promotion payoff for average per- formance rating. Specifically, an increase of 1 point in average performance rating is associated with an additional .45 promotions for men and .33 promotions for women. Effect of Percentage Female In addition to characteristics of employ- ees, characteristics of jobs, such as per- centage of female incumbents, may con- tribute to pay en cl promotion differences. For these analyses, we included persons employed in job title groups having 10 or more incumbents. Although information on job title groups was available for both 1980 and 1986, the 1986 information was more complete and of better quality than that available for 1980. Nevertheless, we were able to develop a sample of 171 women and 1,697 men present in both 1980 and 1986 in fairly homogeneous job title groups. In 1980, 91 job title groups had 10 or more incumbents. Of these 91 groups, 5 i6We again restricted the sample of women to those who were unmarried. In this case, the importance of potential experience was diminished somewhat, but remained important. PAY EQUITY: EMPIRICAL INQUIRIES hac] 50 percent or more female incumbents. Of the 171 women, 43 (25 percent) were employed in those 5 groups in 1980. Of the 5 predominantly female groups, 2 had no mate incumbents. These 2 groups (a(lmin- istrative assistant and secretary) employed 17 percent of all women in the sample. Of the remaining 86 job groups, which were predominantly male, 35 hacT no female in- cumbents. These 35 completely segregated croons emnloved 662 (39 percent) of the .37 .08 .14 .26 .47 .3 .78 =--¢- -- `- ~ ~ ~ total 1,697 men. Examples of some of the larger completely mate groups are tech- nologist, maintenance supervisor, designer, and senior engineer. Models of the type used earlier were estimate(1 separately for men and women. Rather than performing a decomposition, however, we acIdec] a measure of the per- centage of female incumbents in each job title group to each equation. As Table 1-9 indicates, among women, the impact of per- centage female on 1980 salary is not statis- tically significant in any model. In the 1986 salary equation containing only human cap- ital variables, the percentage female coef- ficient is negative and not statistically sig- nificant. However, the ad(lition of job level (and performance rating) results in the coef- ficient being statistically significant and changes the sign such that percentage fe- male has a positive impact on women's salaries in the final equation. Specifically, for each increase of 10 percentage points, salary is .7 percent higher. Thus, a change from a completely male to a completely female job title group would be associated with a 7 percent higher 1986 salary. In sharp contrast, men realized a salary penalty in both 1980 and 1986 for being in job title groups that had higher per- centages of women. In both years, the penalty was much larger without controls for job level. In the final models, each 10- point increase in the percentage female is associate(l with a .8 percent decrease in 1980 salary and a .5 percent (decrease in 1986 salary. Thus, a change from a com-
SALARIES, SALARY GROWTH, AND PROMOTIONS TABLE 1-7 Decomposition of Promotion Differences, 1980-1986 35 Variables in Equation Decomposition Standard R2 Men Women Differential Women Men Coeff. Endow. Coeff. Endow. Raw AWa AME Endow. Coeff. Endow. Raw All Levels Human capital (HC) .26 .27 (.38)c .62 (.49) .51 1.37 1.23 1.19 HC, avg. perf. rating (AVGPA) .28 .30 (.32) .68 (.47) .53 1.25 1.20 HC, job level .26 .30 (.09) .91 (.30) .70 1.34 1.26 HC, AVGPA, job level .28 .34 - .02 1.02 (.27) .73 1.38 1.27 Level 1 HC .27 .32 ( 34) .66 (.42) .58 1.08 1.05 1.05 HC, AVGPA .27 .38 - 1.15 2.15 (.22) .78 1.17 1.06 Level 2 HC .40 .32 (.47) .53 (.66) .34 1.14 1.07 1.05 HC, AVGPA .43 .35 ( 21) .79 (.43) .57 1.11 1.08 Level 3 HC .27 .25 (.29) .71 (.23) .77 1.26 1.18 1.20 HC, AVGPA .33 .30 (.28) .72 (.26) .74 1.19 1.19 NOTE: N = 840 for women, 5,550 for men. aAdjusted salary ratio using women's coefficients as standard. Adjusted salary ratio using men's coefficients as standard. CParentheses indicate that coefficients favor women. pletely male to a completely female job would be associated with an 8 percent lower salary in 1980 and a 5 percent lower salary in 1986. Parenthetically, we note that the similarity of results obtained in the 1980 and 1986 cross sections may re- (luce concerns regarding the quality of job title information in 1980. Results for the salary growth and pro- motion equations are also reported in Ta- ble 1-9. For women, percentage female appears to have no impact on either salary growth or promotions. In contrast, men's salary growth and promotion chances in- crease as percentage female increases. Re- call that earlier decompositions indicated that women had an advantage over men in terms of promotions and salary growth. These results suggest the possibility that part of that advantage may be due to the gender composition of the job. It is pos- sible that men receive some of the benefit that may be attached to predominantly female jobs. DISCUSSION We found that the ratio of female/male salaries rose from .84 to .88 over a 6-year period. Within job levels, salary ratios were higher (.93 to .95), but also increased over the period (.96 to .98 by 1986~. Consistent with the narrowing clifferentials, women received a greater number of promotions and larger percentage increases in salary between 1980 and 1986. As in numerous other cross-sectional studies, the salary advantage of men could not be completely explained by higher levels of endowments, even using a relatively com- plete mocle! that included job level and performance rating. More detailed analyses at specific job levels resulted in still less of the salary (lifferential being explained, al- though adjusted salary ratios were similar. Nevertheless, using only human capital variables, the adjusted salary ratios are among the highest in the literature (see Cain, 1986~. Further, the adjusted salary ratios for the
36 o 3 Cal a; so ._ o ._ o to o g ._ ._ V: o o Cal Go lo AS a; o o~ o ~ 3 sin ~ ~ to ~ o C) =5 a; C~ ~o C: c.;, ~o 3 C) ~ ~ ~o Ct o .= C~ X o X _ ~ ~ C~ IU) ~ C~ ~ ~ oo 1 _ 1 C~ 1 1 _ _ ~ ~ C~ 1 ~ o X o X oo 1 1 C: ~ ~ o ~ Io ~s oo C~ c~ 1 C~ oo C~ _ C~ c~ 1 _ C~ oo C~ o ~ ~ ~ o I X7 o _ 1 c~ 1 o o _ C~ ~S c~ ~ ~ 1 1 1 c~ ~ C: X CM o _ C~ C~ C~ C~ C~ O ~ ~ ~ ~ ~ C5, c~ ~ ~r 1 _ 1 1 - '~ ~ ~ ~ - c~ 1 ~ - - ~ o x o c~ ~ ~ o I C~ C~ CM C~ o 1 1 c~ _ bt ._ Ct - ~ o ~ ~ ~ s_ c, '~ ~ ~ 3 :~ ~ ~ ~2ct C~ o Ct o . :^ C~ C~ ~ Ct ~ ~ ,. . ~ 3 o o ~ C) o ~ ~ ._ Ct o ~ ~o o C~ 50 C: m ;: ~
SALARIES, SALARY GROWTH, AND PROMOTIONS TABLE 1-9 Regression Coefficients for Percentage Female 37 Women Men Variables in Equation B S.E. R2 B S.E. R2 1980 Salary Human capital (HC) .0078 .0046 .30 - .0384** .0040 .34 HC, job level .0049 .0041 .53 -.0097** .0027 .72 HC, job level, perf. rating 80 (PA80) .0023 .0044 .54 - .0084* .0026 .75 1986 Salary H C .0049 .0040 .40 - .0434* * .0044 .30 HC, job level .0123** .0023 .83 - .0054* .0024 .81 HC, job level, PASS .0072** .0023 .86 - .0050* .0023 .83 1986/1980 Salary H C .0058 .0045 .23 .0102 * * .0026 .23 HC, job level .0039 .0048 .24 .0089** .0026 .23 H C, job level, avg. perf. rating (AVGPA) .0025 .0053 .27 .0100* * .0024 .35 HC, job level, AVGPA, promotion (PROM) - .0017 .0049 .37 .0039 .0022 .47 Promotion HC .0293 .0259 .40 .1443** .0199 .33 HC, job level .0008 .0271 .44 .1075** .0200 .36 HC, job level, AVGPA .0134 .0303 .44 .1124** .0194 .39 NOTE: N = 171 for women, 1,697 for men. Percentage female is measured in units of 10 percent. *p < .05. **p < .01. models including job level exceed all such ratios summarized by Cain. Our estimates, however, may be somewhat inflated because of the exclusion of higher job levels (held mostly by men) and the use of salary rather than earnings. In contrast to men's cross-sectional salary advantage, women received both greater numbers of promotions and larger salary increases over a 6-year period. Other studies have found similar results with respect to unadjustecl differences in both promotions (Hartmann, 1987; Lewis, 1986; Stewart and Gudykunst, 1982; Tsui and Gutek, 1984) and salary increases (Megdal and Ransom, 1985; Tsui and Gutek, 1984~. Although the promotion advantage could be largely explained by differences in mean endowment levels, the same did not hold true for the salary increase advantage. Thus, if one chooses to interpret the cross-sec- tional results as showing salary discrimi- nation against women, one may be obliged to interpret the salary increase results as suggesting discrimination in favor of wom- en. The (lecomposition results were very sen- sitive to mo(le! specification. Human capital variables alone explained anywhere from 0 to 32 percent of men's cross-sectional salary advantage.Adding job level and perfor- mance rating raised this range to 68 to 77 percent. Human capital variables were bet- ter able to account for differences in salary growth (40 to 49 percent) an(1 promotions (51 to 62 percent). Again, however, adding job level and (average) performance rating increase(l this figure substantially (58 to 75 percent for salary increases; 73 to 102 per- cent for promotions). Of course, in the case of promotions and salary increases, the a(l- vantage to be explaine(1 belonged to women. Cain (1986) pointed out that the typical effect of including a(lditional variables in a salary equation was to increase the a(ljuste female/male salary ratio. Consistent with this general finding, Hildebrand (1980), Mil- kovich (1980), an(1 Milkovich and Newman
38 (1987), among others, have argued that the adjusted female/maTe salary ratio might be closer to unity if unmeasured differences between men and women and relevant labor markets could be better incorporated in salary equations. We were able to move in this direction by including, for example, several experience measures, job level, and performance rating in our moclels. Our results, however, indicate that at least in the case of performance rating, its inclusion actually led to a slight reduction in the adjusted female/male salary ratio. In the promotion and salary growth equations, in which the unadjusted ratios exceeded unity, the inclusion of average performance rating lee! to sometimes sizable increases in adjusted ratios. Thus, the common thread is that inclusion of performance rating does not help explain the raw salary advantage of men, but it does explain some of the raw advantage of women with respect to salary growth and promotions. Although women had an overall advan- tage in the salary increase and promotion process, they receive(1 a smaller payoff for their performance ratings. Based on the full promotion equation, for example, men re- ceived an average of .45 additional pro- motions for each additional point in average performance rating between 1980 and 1986. In contrast, women received an additional .33 promotions for each adclitional point. This finding is consistent with Olson and Becker's (1983) suggestion that women may be held to higher promotion standards than men. The fact that women experienced a lower salary disadvantage in the cross-sectional analysis yet received better salary increases and more promotions over time offers an interesting contrast. One implication may be that because men are not favored in the salary increase and promotion process, the explanation for women's salary level dis- advantage must be sought elsewhere. If so, one alternative avenue of investigation shoul PAY EQUITY: EMPIRICAL INQUIRIES perhaps be the recruitment and initial place- ment of men and women. Aside from possible differences in initial placement, it may be that women have not always been favored in terms of salary in- creases and promotions. In fact, the greater number of promotions and larger salary increases of women may reflect an attempt by the firm to reduce what was perceives! to be an inequitable salary and job level structure. Rosenbaum (1985), for example, found a reduction in the negative impact of percentage female on salary between 1965 and 1975. Further, he found that the impact of percentage female on promotions changed from negative to positive over the same period. In the case of the Rosenbaum stu(ly, the firm had implemented a "serious" af- firmative action program during the period of the study. The firm that we studied ha(l an ongoing affirmative action plan. It also covere(l EEO issues in its management training. Further, although EEO litigation pertaining to these issues among exempt employees does not appear to have been important, the firm did settle a case pertaining to such practices vis-a-vis hourly employees for a substantial amount. Although no formal goals or new practices resulted, one might speculate that this event enhanced or at least reinforced the vigilance with which progress toward affirmative action and EEO goals was mon- itore(l. Any progress in the affirmative action area may have been facilitated by the general financial success of the company during the periocl of our study. Research by Rosen- baum (1979) suggests that promotion op- portunities may be greater during periods of organizational ((lefine(1 as employment) growth. The general financial success an growth of the firm we studied may have facilitated affirmative action progress of fe- male employees. It is interesting to note, however, that the bulk of employment growth in exempt jobs seems to have been among
SALARIES, SALARY GROWTH, AND PROMOTIONS women rather than men. i7 As cliscussed earlier, this appears to have been partly due to a brief period of reductions in force during the early 1980s. The promotion and salary growth advan- tage of women implies that the salary gap would eventually remedy itself if past trends were to continue. Note, however, that even a small initial salary clisadvantage can take many years to be eliminated. As an example, a projection of salary growth rates over the 6-year period into the future indicates that the 1980 female/male salary ratio would not equal unity until the year 2003. Within job levels 1, 2, and 3, 1980 salaries would equalize in 1989, 1996, and 1992, respec- tively. Of course, as this equalization pro- cess works to its conclusion, women con- tinue to receive lower salaries. In addition to our focus on human capital, job level, and performance ratings, we also examined the possibility that percentage female in a job group was a structural factor contributing to attainment differences. Con- sistent with Rosenbaum (1985), controlling job level generally reduced the impact of percentage female by a substantial amount. Consistent with Hartmann (1987), percent- age female had a small, positive effect on women's salaries (in 1986~. Unlike Hart- mann's results, however, percentage female had a negative impact on salaries of men. With respect to salary growth, the lack of a negative impact of percentage female for both men and women was consistent with Hartmann's findings. Regarding promotion, although Hartmann found a negative impact of percentage female for women, Rosen- baum found a positive impact by the end of his study period. Our results suggest no stable impact of percentage female on wom- i'This is not to say that there were not also a substantial number of newly entering men as well. The net growth in the number of women seems to have been higher because there were fewer women in exempt jobs at retirement age, for example. 39 en's promotion rates. In contrast, a positive effect was found for men. The preceding discussion does not really provide support for the idea that percentage female is an important structural property that negatively affects women's (and perhaps men's) attainments. A better research strat- egy would be to examine its impact con- trolling for other characteristics of jobs or occupations that may be related to per- centage female. This strategy was demon- stratec] by Treiman and Hartmann (1981) using national survey data at the occupa- tional level. A stronger test, however, would make use of firm-level data wherein simi- larity of occupational titles is more likely to correspond to actual similarity of work con- tent. The effect of percentage female in different firms is another possible avenue of investigation (Pfeffer and Davis-Blake, 1987) in cases in which job content is stan- lar(lized. Given our use of data from a single firm, the study has a disadvantage relative to the coverage en cl external valiclity possible with marketwide or national surveys. Along these lines, replication studies would be necessary before attempting to answer the question of how typical our results are of other large firms' relative treatment of men and women. Nevertheless, the better coverage obtained using national survey data comes at the expense of not being able to measure pro- ductivity in the way that many firms actually measure it. Finally, recall that the policies anti practices of the firm we studie(l ten(le to be consistent with those reported in surveys of other large firms. Besides the job-specific productivity mea- sures, the results of our study were strength- ened by the following. First, the use of longitudinal (lata should have recluced the impact of any unobserved, constant incli- viclual differences in procluctivity. Second, firm-specific differences in (leterminants of salaries (e.g., pay policies and practices) are obviously not an issue. Finally, the use of
40 data from personnel records rather than self- reports typical of national surveys may have reduced the potential for reporting errors (Duncan and Hill, 19851. In their agenda for basic research on comparable worth, Hartmann et al. (1985:7) emphasizer] that "we need to understand better how wages are set within enterprises and how they are affected by other employer practices, such as job assignment...." They stressed the importance, especially in large firms, of internal labor markets and pro- motion from within as aspects of job as- signment. We hope that our research con- tributes to a better understanding of these processes. ACKNOWLEDGMENTS We would like to thank Donald Schwab, the other members of the Panel, the other authors, and our colleagues at Cornell Uni- versity for their helpful comments on earlier drafts of this chapter. REFERENCES BielLy, W. T., and J. N. Baron 1986 Men and women at work: Sex segregation and statistical discrimination. American Jour- nal of Sociology 91:759-799. Blinder, A. S. 1973 Wage discrimination: Reduced form and structural estimates. Journal of Human Re- sources 8:436-455. Brown, C. 1982 Estimating the determinants of employee performance. Journal of Human Resources 17:178-194. Bureau of National Affairs, Inc. 1983 Performance appraisal programs. Personnel Policies Forum, no. 135. 1985 Training and development programs. Per- Jones, F. L. sonnet Policies Forum, no. 140. 1983 Cain, G. G. 1986 The economic analysis of labor market dis- crimination A survey. Pp. 693-785 in O. Ashenfelter and R. Layard, eds., Handbook of Labor Economics. Amsterdam: North- Holland. PAY EQUITY: EMPIRICAL INQUIRIES Conference Board 1984 Pay and performance: The interaction of com- pensation and performance appraisal. Re- search Bulletin, no. 115. New York. Dipboye, R. L. 1985 Some neglected variables in research on dis- crimination in appraisals. Academy of Man- agement Review 10:116-127. Duncan, G. J., and D. H. Hill 1985 An investigation of the extent and conse- quences of measurement error in labor-eco- nomic survey data. Journal of Labor Eco- nomics 3:508-532. Flanders, D. P., and P. E. Anderson 1973 Sex discrimination in employment: Theory and practice. Industrial and Labor Relations Review 26:938-955. Fuchs, V. 1971 Differences in hourly earnings between men and women. Monthly Labor Review 94:9-15. Guilford, Joy Paul 1954 Psychometric Methods. New York: McGraw- Hill. Halaby, C. N. 1979 Job-specific sex differences in organizational reward attainment: Wage discrimination vs. rank segregation. Social Forces 58: 108- 127. Hartmann, H. I. 1987 Internal labor markets and gender: A case study of promotion. Pp. 59-97 in C. Brown and J. A. Pechman, eds., Gender in the Workplace. Washington, D.C.: Brookings In- stitution. Hartmann, H. I., P. A. Boos, and D. J. Treiman 1985 An agenda for basic research on comparable worth. Pp. 3-33 in H. I. Hartmann, ea., Comparable Worth: New Directions for Re- search. National Research Council, Com- mittee on Women's Employment and Related Social Issues. Washington, D. C.: National Academy Press. The Hay Group 1986 Trends and issues in total compensation. Compensation Quarterly October 6:3. Hildebrand, G. H. 1980 The market system. Pp. 79-106 in E. R. Livernash, ea., Comparable Worth: Issues and Alternatives. Washington, D. C.: Equal Employment Advisory Council. On decomposing the wage gap: A critical comment on Blinder's method. Journal of Human Resources 18:126-130. King, L. M., J. E. Hunter, and F. L. Schmidt 1980 Halo in a multi-dimensional forced choice performance evaluation scale. Journal of Ap- plied Psychology 65:507-516.
SALARIES, SALARY GROWTH, AND PROMOTIONS Lewis, G. B. 1986 Gender and promotions. Journal of Human Resources 21 406-419. Malkiel, B. G., and J. A. Malkiel 1973 Male-female pay differentials in professional employment. American Economic Review 63 693-705. Medoff, J. L., and K. G. Abraham 1981 Are those paid more really more productive? The case of experience. Journal of Human 1985 Resources 16 186-216. Megdal, S. B., and M. R. Ransom 1985 Longitudinal changes in salary at a large public university: What response to equal pay legislation? American Economic Review 75 271-274. Milkovich, G. T. 1980 Comparable worth: The emerging debate. Pp. 23-47 in E. R. Livernash ea., Compa- rable Worth: Issues and Alternatives. Wash- ington, D.C.: Equal Employment Advisory Council. Milkovich, G. T., and J. M. Newman 1987 Compensation. Plano, Tex.: Business Publi- cations. Mundlak, Y. 1978 On the pooling of time series and cross- sectional data. Econometrica 46 69-85. Oaxaca, R. L. 1973 Male-female wage differentials in urban labor markets. International Economic Review 14 693-709. Olson, C. A., and B. E. Becker 1983 Sex discrimination in the promotion process. Industrial and Labor Relations Review 36 624- 641. Pfeffer, J., and A. Davis-Blake 1987 The effect of the proportion of women on salaries: The case of college administrators. Administrative Science Quarterly 32 1-24. Reskin, B. F., and H. I. Hartmann, eds. 1985 Women's Work, Men's Work: Sex Segregation 41 on theJob. National Research Council, Com- mittee on Women s Employment and Related Social Issues. Washington, D. C.: National Academy Press. Rosenbaum, J. E. 1979 Organizational career mobility: Promotion chances in a corporation during periods of growth and contraction. American Journal of Sociology 85 21-48. Jobs, job status, and women's gains from affirmative action: Implications for compa- rable worth. Pp. 116-136 in H. I. Hartmann, ea., Comparable Worth: New Directions for Research. National Research Council, Com- mittee on Women's Employment and Related Social Issues. Washington, D. C.: National Academy Press. Rynes, S. L., and G. T. Milkovich 1986 Wage surveys: Dispelling some myths about the "market wage." Personnel Psychology 39 71-90. Sanborn, H. 1964 Pay differences between men and women. Industrial and Labor Relations Review 17 534- 550. Stewart, L. P., and W. B. Gudykunst 1982 Differential factors influencing the hierar- chical level and number of promotions of males and females within an organization. Academy of Management Journal 25 586- 597. Treiman, D. J., and H. I. Hartmann, eds. 1981 Women, Work, and Wages: Equal Pay for Jobs of Equal Value. National Research Coun- cil, Committee on Occupational Classification and Analysis. Washington, D. C.: National Academy Press. Tsui, A. S., and B. A. Gutek 1984 A role set analysis of gender differences in performance, affective relationships, and ca- reer success of industrial middle managers. Academy of Management Journal 27 619- 635. Appendixes A and B follow.
42 ,' . o C~ U: ;~ V2 o x C: O CD ~ X c~ c~ ~ LO CC X ~ OC cc X CM ~ O O ~ ~ O ~ CO ~ CO C~ G) ~ ~ 0 C~ o o 0 0 )~ ~ o CD O ~ O , CS) ~ 1 CC) . . _ CM ~ C~ ~ ~ C~ CM Z) C~ ~ _ ~ CO U: C~ U) o C~ C~ ~ C~ CO o o X ~ ~ C~ ~ ~ ~ ~ C~ _ ~ C~ ~ ~ _ o C~ ~ ~ C~ ~ ~ C~ C] ~ o _ O ~ CM O O O O C~ C~ 00 ~ X CD O CC C~ ........ ....... .. X ~ o o C~ o CM C>) C~ _ ~ C~ ~ ~ O C] co ~ U) CD ~ CO CD C~ oo ~ O ~ CC ~ O CM ~ o U: U) o o CM ~ o ~ ~ ~ X U: ~ CO ~ ~ ~4 0 ~ C~ O O O O ~ ~ c~ ~ 0 1 =4 ~ )~ U: [f: ~ oo ~ _ 1 ~ c~ c~ u:^ C~ X X o ~ C~ o U: CC C~ ·c ~ o ~ o a; ~ c~ co ~ ~ c~ 0 _ 0 ~ ~ cc c~ oo ~ _ c~ X ~ CD ~ ~ ?~ °. ° °. ? ~ ~ ° ~ ~ ~ ° 1 ~ ~ ~ X oo 0 ~ co LO _ ~ ~ · 1 c~ c~ ;> C~ C~ _ ~ ~ ~ C~ _ ~ ~ O CM C~ ~ U: CC CD O CD ~ C~ C~ O ~ C~ O C~ C~ O ~ ~ O O C~ .= ~ ~ o ~ c~ o 0 0 0 cs' ~ c~ ~ ~ c~ 0 1 ~4 U) 1 ~ a; x 1~ ~ c~ c~ X 1 c~ c~ 1 ~ ·~7 ~ _~_= ~ C~ CO C~ X X O ~ ~ CO ~ O ~ oo U) ~ C~ C~ O .- ~ ~ ~ ~ ~ ~ ~ O ~ O ~ ~ ~ O ~ ~ O bt O ~ ~ o _ ~ 0 0 0 0 ~ x 0 ~ 0 ~ 0 1 ~ ~ 1 o ~ =1 ~ ~ ~ ~ ~o ~ 1 ~ ~ 1 1 C) .' ~ .- C~ = V 5 ~ ~ ~ ~ C . V ~ ~ ~ ~ F _
43 .o~ ~ 2 * * * * * * * * * * * * * * * * * * U) ~ CS ~ ~ o ~ ~ o ~ ~ ~ o CO ~ ;o ~ CM ~ C~ ~ ~ ;o X ~ o ~ ~ Lr: ~ oo CD CO ~ X G: O c~ ~ O c~ O r: c5) ~ ~ O ~ 0 0 0 c~ c~ ~ o o o o o ~ o ~ o o o o o o o o 1 1 1 1 1 1 1 * ** ** **** * * * * * * * * * * X ~ ~ ~ ~ c~ ~ ~ C~ C~ 5) G~ X ~ r~ C~ ~ X ~ O ~ ~ G. C~ CD o CM ~ o C~ o ~ ~ o C~ o o C] o ~ ~ C~ ~, o o o o o o o o o o o o ~ o o o 1 1 1 1 1 1 * , * * * * C~ ~ ~ X ~ G) O ~ CM ~ CD O C~ C~ (= C~ ~ o ~ ~ CM C~ oo ~ o X U: o ~ o C~ ~ ~ ~ C~ ~ ~ o ~ ~ o C~ o o o o o o o U) o o o o o o 1 1 1 1 1 1 1 * -x- ~ CD CM * * * * * * * * * * * * * * * * * oo ~ ~ C~ o X ~ ~ o ~ C~ _ ~ ~ X cc x ~ 0 ~ c, ~ co c~ X ~ ~ ~ O r~ o o ~ ~ o CM ~ ~ o ~ o o o ~ C~ C~ o o o o o o o o o o o o X o 1 1 1 1 1 1 1 1 1 U: o ·~, U: ;~ bt o ·~, o ? ~ o ~? = C) ~ ~ o U: o^ o X ~ o X ~ X ~ ~ ~ ~ ~ X ~ ~ v ° O - * * * * * * * * * 0 ~ ~ ~ ~ ~ [Q ~ cc ~ ~ cs r~ ~ 0 0 c~ _ c~ _ ~ _ ~ c _ 0 0 0 0 0 0 0 ~ 0 0 c . . . . . . . . . . . . O 0 0 0 0 0 0 0 0 0 0 c 1 1 1 1 1 1 1 1 * * * * * * * * * * * 0 _ _ 0 0 0 ~ ~ c~ ~ c~ 0 0 0 0 0 0 0 c~: 0 0 0 0 0 0 0 0 c 1 1 1 1 1 1 * * G) CD O O _ _ o ~ O O oC CD 0 _ 0 0 0 0 0 0 0 c 0 0 1 1 * * * * * * * * * 0 x _ ~ c~ c: _ ,_ _ 0 0 0 L~ 0 0 x 0 0 0 0 c~ 0 0 0 0 . . . . . . . . . O 0 0 0 0 0 0 0 0 1 1 1 * * * * * * * * * u: ~ ~ ~ c~ _ _ ~ ~r ~ c~ c~ X ~ ~ ~ ~ x ~r 0 0 ~ 0 ~ cc ~ c~ c~ 0 _ ~ 0 0 0 c~ 0 0 0 0 0 0 0 ~ 0 0 0 0 _ 0 0 _ c~ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 * * * * * * * . * * * * * 0 ~ ~ ~ c~ cc ~ _ _ ~ ~ 0 c>; _ c~ 0 0 c~ c~ ~ 5, 0 0 0 c ~ 0 0 0 0 0 0 0 c~ 0 0 c 0 0 0 0 0 0 0 0 0 0 0 0 c 1 1 1 1 1 1 1 * ~ * * * 0 ~ co _ ~ c~ 0 u: 0 _ Lf: c~ cc cc 0 0 0 0 0 0 c . . . . . . . . O 0 0 0 0 0 0 c 1 1 1 1 1 c~ * .. * * * co ~ ~ c~ c~ c~ c~ c~ 0 c: ~ 0 ~ 5) 0 0 0 0 _ 0 _ 0 0 0 0 ~o 0 0 1 1 1 1 1 * * * * * * * * * * * * * * * * '3) ~ ~ ~ ~ ~ _ O CO _ ~ O ~ 53 G) c~ _ 0 _ c~ ~ ~ ~ c~ _ 0 0 0 c~ 0 cc 0 0 0 0 0 0 0 c~ 0 0 0 0 x 0 . . . . . . . . . . . . . . . O 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 .~ c~ . _ 0= 0 V V =, -x~ * *
Commentary CHRISTOPHER WINSHIP Gerhart and Milkovich provide an anal- ysis of sex differentials in salaries, raises, and promotions in a large, diversified firm. Their sample is confined to employees in the 7 lowest of 15 job levels for exempt personnel. The sample is further restricted to individuals working in the firm in 1980 who were still working with the firm in 1986. Their basic strategy is to estimate a large variety of regression models, almost exclusively of the single-equation type, with the aim of assessing what factors account for sex differences in salaries, raises, and promotions. The analyses here are quite straightforward. They carry out the usual decomposition procedures for determining how much of the sex difference in the cle- pendent variable, that is, salaries, raises, or promotions, can be explained by maTe- female differences in the in(lepen(lent vari- ables and how much can be explained by differences across sex in the effects of these variables on the depen(lent variable. A number of Gerhart and Milkovich's findings are of interest. The firm they study The author benefited from a conversation with Dale Mortensen in preparing these comments. 44 is unusual in that sex differences in salaries are relatively small. In 1980 the ratio of female to male salaries was .84. In 1986 the ratio was .88. As they point out, these ratios are among the highest in literature. They also find that over the 6-year period, wom- en, on average, received higher raises, at least on a percentage basis, than men. Over- all, female salaries increase(1 61 percent over the 6-year period whereas mates' increased 54 percent. Promotions show a similar sex (difference. From 1980 to 1986, women re- ceived an average of 1.29 promotions and men an average of .94. Put in other words, women received, on average, nearly 40 per- cent more promotions than men. Gerhart an(l Milkovich also carry out breakdowns by job level. As one might expect, within job level the female/male salary (differences are much smaller. For instance, in 1980 they range from .93 to .95. The overall (differences between women and men in size of raises and number of promotions hold within levels as well as across the population. An important impli- cation of this is that differences in the dis- tribution of men and women across levels does not appear to explain the larger raises
COMMENTARY ancl higher rates of promotion for women. Gerhart and Milkovich also examine the effect of sex composition of jobs on salaries. In many of their models they find that the effect of percentage female on the salaries of women is statistically insignificant. The sign of the effect also changes across models. In sharp contrast, however, men are pen- alized in both 1980 and 1986 for holding jobs that are predominantly hell] by women. For example, in 1980, they estimate that for a man, a change from a completely male to a completely female job would be as- sociatecI with an 8 percent lower salary. It seems to me that one of the advantages of studying a single firm is that it provides the opportunity to think in detail about what kinds of processes affect salaries, rais- es, and promotions. It gives us a chance to bring our general knowledge about how the world works in a specific situation to our statistical analyses. The authors, however, tell us almost noth- ing about the firm they are examining. In evaluating the analyses, I would have found it quite useful to know about the kinds of jobs in the sample and, in particular, the kinds of jobs that were associated with the seven job levels they analyze. It would also be quite useful to know something about the organization of the personnel system within the firm. lob level is a variable frequently used in the analysis, but at the end of the paper I was not quite sure what it meant. I was particularly puzzled when I read a footnote that an individual can be promoted without changing job level and also can change job level without being promoted. It would also be valuable to know something about the structure of the firm. Do the jobs in the sample fall into different divisions or departments? Are men and women clifferentially distributed across these units and does this have any effect on salary? These considerations might also refocus the analysis a bit. One major factor in de- termining sex differences in salaries is dif- ferences in the type of jobs men and women 45 hold, but this paper offers only a very partial analysis ofthis factor. Gerhart and Milkovich do examine the effects ofthe sex composition of jobs on salaries, but they go no further than that. Perhaps the most serious variable that is not even discussed is differences in hours worked, or how demanding a job is. The analysis here is in terms of salaries, not wages. The sample inclucles only exempt employees. As such, one might guess they are all expected to work a full week and that their weekly salary does not change even if they work more or less in a particular week. One wouIc] expect that, on average, some jobs have fuller work weeks than others or that some have more variability in hours worked in a week and that this would be reflected in salary differences. If this is in fact the case, it would not be surprising to fins! that women were dispro- portionately in jobs with fewer hours ant] with less variable hours and that this ex- plained part of the sex difference in salaries. Other types of job differences might also be important for salary differences anc] cor- relatec] with sex. ~ 1 ~ ~ ~ _ Let me turn to a different issuethe sample. Gerhart and Milkovich have se- lectecl their sample so that only inclividuals who were working with the firm in 1980 ant] were still working with the firm in 1986 are inclucled in their analysis. As such, they excluded individuals who were working with the firm in 1980 but left before 1986. It may well be that the exclusion of inclivicluals who have left the firm has only a negligible effect on the analysis. The question is not whether women or men are more or less likely to leave the firm, but rather what the interaction is among sex, exiting, and the depenclent variable, that is, salaries, raises, or promotions. It is easy to tell stories about how inclusion of former employees could change the sex (lifferential in raises and promotions. If men who left the firm were in(lividuals who would or (lic1 receive mul- tiple promotions and large raises and women
46 who left would or did receive no promotions and small raises, then the sex differential in raises anti promotions could disappear by controlling for this differential selection. My other criticism of the sample is the authors' decision not to inclucle more than averages of variables from the years between 1980 and 1986. One's salary at a particular time and the degree to which it has changed over the years is a result of year-to-year changes. A better model, for instance, would be one in which performance rating, and perhaps changes in performance rating, af- fected one's raise and one's chance of pro- motion in each particular year. It is the history of one's performance that is probably most important. The inclusion of average performance rating in predicting raises and promotions over the 6-year period strikes me as a misspecification of the model. This paper could perhaps make its great- est contribution in terms of the analysis of the performance variable. Most research (loos not inclu(le a measure of productivity, but they would need to consider a richer data structure in order to include perfor- mance in their models appropriately. With a richer set of data, lagged values of per- formance might also be appropriate. As they acknowledge in a footnote, their perfor- mance variable probably contains a consid- erable amount of measurement error. If measurement error in the performance vari- able was corrected, performance might have a much larger effect. This would be im- portant in analyzing the sex differences giv- en that women have somewhat higher per- formance measures than men. Of consi(lerable interest in the paper is the observation that women received higher raises en c] more promotions than men in this firm. As the authors note, other re- searchers have also reporte(l this fincling. Gerhart and Milkovich's findings indicate that nearly half of these differences can be explained by the lower levels of experience and job tenure of women. This, however, PAY EQUITY: EMPIRICAL INQUIRIES leaves a sizable unexplained portion. Ger- hart and Milkovich offer one explanation- that the firm has seen the light an(l is trying to make up for past inequities. The eco- nomics literature suggests five other pos- sible interpretations. All of them potentially have to do with childbearing, and several also have to do with the employer's inability to (liscern which women will stay with the firm Tong term and which will not. First, there is a very simple supply-side story. As women have children, their re- servation wage may rise over time. As a result, women in the lower paying jobs, women who expect to receive small raises and few promotions, will leave the firm. Those with better prospects remain in the firm and make other arrangements for child care. As a result, women's salaries rise faster than they would have if the lower pai women had not left the firm. A second selection story works with con- stant reservation wages, an(1 has to do with learning about the quality of the match of the job and the employee. (For a formal model of this nature, see the matching models developed by lovanovic, 19791. All that is needecl is for women to have higher reservation wages than men. This might be true because of their greater productivity in the home. Here, neither workers nor firms know how well they are matched. As time goes on, they learn how good the match is an(1 the poor matches (lissolve. As in the first example, this selection process increas- es the raises and promotions of women by selecting women who have higher values on their reservation wages. With respect to both selection stories, it is unclear whether enough women leave the firm to explain the residual sex differ- ence in raises and promotions. Recent anal- yses by Tope! (1986) an(1 by Altonji and Shakotko (1987) have found that selection (lue to matching in the above sense explains most of the growth in wages with job tenure for men.
COMMENTARY There are also a set of demand-side sto- ries. A straight human capital story would be that firms anticipate a higher quit rate for women than for men. As such, they might be leery of investing in tamale em- ployees for fear they would not recover their investment. As a result, they might adopt a salary schedule whereby women are paid less early on and more later. A related story is that firms cannot tell the difference between women who will stay with the firm and those who will leave after several years. They may adopt a salary schedule that pays women below their mar- ginal product initially and above their mar- ginal product later, so that only women who intent! to stay with the firm will take jobs there. This is the adverse selection mode! of Salop and Salop (1976~. Here, what hap- pens is that firms specialize. Firms with low turnover costs end up hiring women who are likely to move in and out of the labor force, and firms with high turnover costs hire only those women who are likely to remain with them. The hypothesis here is that Gerhart and Milkovich's firm is a firm with high turnover costs. This mode! is consistent with Gerhart and Milkovich's ob- servation that women are no more likely to leave the firm than men. There are also stories based on nonTin- earities due either to risk aversion or non- linearities in the production process. If there is more ex ante unobserved variability for women than for men in their productivity, firms, if they are risk averse, will initially pay women less than men until those pro- ductivity differences are revealed. This idea is presented in an early paper by Aigner and Cain (1977~. A relatecl idea of Roths- chiTcI and Stiglitz's (1970) is that if firms are worse at matching women to jobs than men, both in that they are likely to err by putting women in jobs that they are overquatifie(l for as well as in jobs they are unclerquali- fied for, then as a result, women will be less productive than men and will be paid less. , ~ . ., _ ~ . . 47 Finally, there is the possibility that wom- en may be more likely to shirk on the job than men. Perhaps this might be the case because child-care responsibilities are more of a distraction for women. As a result, firms may want to offer women contracts in which they are paid below their marginal product early on and above it later. Consi(lerably more about the institutional structure of Gerhart and Milkovich's firm would have to be known to evaluate the appropriateness of these different stories. Each of these stories implies some amount of job segregation, since it would probably be difficult for an employer to pay men an women different wages for the same jobs. These hypotheses could be studied using firm-levl! data. More than two periods of (lata would be nee(le(l, however, preferably a panel of yearly (lata. In addition, single- equation regression models would probably not be adequate. Selection out of the firm ant] unobserved heterogeneity wouicl have to be accounted for. There are now a considerable number of sophisticated models of wage growth in the economics literature. I would encourage the authors to examine them and to consider adopting a more structural model for their subsequent analysis. Doing so could create a greater un(lerstancling of why they see the empirical relations that they (lo and woul(1 enable them to interpret their results in terms of a conceptual mo(lel. Their work would help us all move forward in under- standing why men and women are paid differently. REFERENCES Aigner, D. J., and G. G. Cain 1971 Statistical theories of discrimination in the labor market. Industrial and Labor Relations Review 30:175-187. Altonji, J. G., and R. A. Shakotko 1987 Do wages rise with job seniority? Review of Economic Studies Uuly):437-45l.
48 Jovanovic, B. 1979 Job matching and the theory of turnover. Journal of Political Economy 87:972-990. Rothschild, M., and J. Stiglitz 1970 Increase risk I: A definition. Journal of Eco- nomic Theory 2:225-243. Salop, S. C., and J. Salop 1976 Self-selection and turnover in the labor mar- PAY EQUITY: EMPIRICAL INQUIRIES ket. Quarterly Journal of Economics 90:619- 628. Topel, R. 1986 Job mobility, search, and earnings growth: A reinterpretation of human capital earnings functions. Pp. 199-233 in R. G. Ehrenberg, ea., Research in Labor Economics. Vol 8., A. Greenwich, Conn.: JAI Press.
