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.
Pay Equity: Assessing the Issues ROBERT T. MICHAEL and HEIDI I. HARTMANN Despite the progress economists and so- ciologists have made in recent years in understanding wage determination and the occupational structure of the labor market, large unexplained] differences in wages be- tween women and men remain. Differences in skill, experience, effort, labor force at- tachment, and many other variables that have been studied do not account for all the earnings differences observed. On the face of it, the unexplained gender differ- ences in wages are consistent with the wide- ly held belief that there is substantial dis- crimination against women in the labor market a systematic bias in wage payment that favors men over women. The inability of social accountants to "explain" the gender gap in wages is often joined with the wide- spread social suspicion of sex bias, and the former is viewed, at least indirectly, as evidence of the latter. "Comparable worth" or"pay equity" has been proposed, along with equal employ- ment opportunity and affirmative action, as a strategy to eliminate gender bias from the labor market, particularly in the determi- nation of wages. Comparable worth or pay equity strategies generally rely on the use 1 of objective criteria to value the content and requirements of jobs job evaluation) in a way that eliminates gentler as a com- pensable factor. One's assessment of comparable worth as a prescription for social ill (legends partially on one's understanding of the reasons for the observed gender difference. If that dif- ference in wages is attributed to legitimate market forces just not yet well unclerstoocl or not yet well measured in studies, then the prescription is probably viewed as a poor one: It imposes restrictions and bias on a worIcl that is working fine, albeit not well understood. In this view there is no social ill, so there is no rationale for any medicine. If, on the other hand, the gentler difference is attributed to systematic bias in the labor market, then there is a social ill, and a need for some medicine. In this case, if that prescription is comparable worth policy, it becomes necessary to employ "ob- jective criteria" for setting wages in a way that eliminates gentler bias. If comparable worth is prescribed, there is a need to assess its side effects as well as its potency. Extending the metaphor of illness and a proper prescription one step further, com-
2 parable worth is not designee! to cure all possible labor market discrimination against women. It is a prescription for a specific illness that has to clo with certain jobs being undervalued. In particular, if jobs held dis- proportionately by women are undervalued according to some objective criteria partly because they are held disproportionately by women, then comparable worth is a med- ication a wise cloctor would consider pre- scribing. If instead a different illness exists, one that is related to limited access to certain jobs for women, or to limited onnortunitv for advancement by women, or to lower pay to women for the saline work (as distinct from comparable work), then other medi- cation, such as equal employment oppor- tunity legislation, remedial affirmative ac- tion, or traditional equal pay remedies, would be more appropriate. In sum, the logic that would lead one to conclude that comparable worth is a wise social policy requires the following: (1) there is a gender difference in wages that is not explainecT by legitimate market forces; (2) the gender difference is linked to the un- dervaluation of jobs held disproportionately by women; (3) the jobs can be objectively evaluated such that an appropriate level of compensation can be determined by some mechanism other than competitive labor market forces (or that removes the effects of gender bias from market forces); and (4) performing the evaluation ant] implement- ing the implied appropriate wage structure is on balance preferred both to eliminating that wage difference by any other means and to not eliminating it (i.e., any adverse si(le effects from its implementation are overcome by the benefits of implementa- tion). These are the issues addressed by the papers contained in this volume. They are empirical studies by a wide spectrum of social scientists. The researchers were se- lected for funding by the Pane! on Pay Equity Research because each stucly acI- dresses key issues of fact that are important PAY EQUITY: EMPIRICAL INQUIRIES to assessing the appropriateness of compa- rable worth strategies: What determines in- dividual ant! occupational wages? How are wages set ant] how do firms and agencies structure their pay plans? How are decisions about promotions and new hiring made? How have workers fared as a result of comparable worth implementation ant! how did they react? Careful descriptive studies can contribute to our unclerstanding of many of these issues, and more analytic stuclies can address issues of causation. Nevertheless, the clifficulties of drawing causal inferences from the nonex- perimental ciata used in the social sciences must be noted here. The simultaneous op- eration of many factors in the real worIc! and the inability to devise perfect measures and controls make it very difficult to identify causes with much certainty. As described in the preface, our Panel on Pay Equity Research selected eleven empirical studies of aspects of the compa- rable worth debate through a competitive proposal process. Several additional experts were asked to comment on these papers at a workshop. The papers and selected dis- cussant comments in this volume address three questions of fact: 1. To what extent is the gender difference in wages in the United States today ex- plained by personal differences in skill, ef- fort, experience, and other characteristics that might be legitimate determinants of wages? Although essentially a factual issue, there are many ways to measure that fact, so it is not a trivial task to answer this question. The papers by Gerhart and Mil- kovich, by Sorensen, and by Nakamura and Nakamura address this question using data on earnings of inclividuals. The essay by Subich, Barrett, Doverspike, and Alexander adds psychological perspective by reviewing the literature that considers gender differ- ences in socialization and their potential impact on individual life outcomes. 2. Since job or occupational difference appears to be so intricately related to gender
ASSESSING THE ISSUES differences in wages in the United States today, how should job or occupational seg- regation by gender ancT differences in the average wages of female and male workers in occupations be understood? The papers by Baron and Newman, by Parcel, and by Filer address this question, with the job or the occupation as the basic unit of their analyses. 3. Since there are examples of the im- plementation of"comparable worth" plans, what is the evidence regarding their impact? Are such side effects as job loss or structural change significant? These are the questions addressed by the three papers by Orazem and Mattfla, by Evans and Nelson, and by Gregory, Anstie, Daly, and Ho. THE EMPIRICAL INQUIRIES Gender Differences in Wages: Wage Determination for Indivicluals Male-Femate Salaries and Promotions in a Large, Private Firm The Gerhart and Milkovich paper inves- tigates gender differences in wages ant] labor market treatment controlling for personal characteristics. The strategy in this paper is to study one large, private, unnamed, highly diversified firm and investigate cle- tails of salary and employment (lynamics (promotions and salary adjustments). The authors study workers in administrative and professional jobs, examining patterns of wages and wage changes for employees who were with the firm continuously from 1980 through 1986. The primary data set includes 5,550 men and 840 women. The strengths of this study include the following: (1) much is "held constant" in an investigation of behavior within a single firm, (2) the study has an unusually goof] independent measure of each employee's job performance (a 4-point scale), on which the firm's compensation policy is explicitly based, and (3) the measures include job tenure on that job, an especially important 3 factor for an investigation of gender differ- ences in wages. Two problems with the study, discussed by Winship, are (l) a stucly of only one firm cannot yield generalized findings we do not know whether the find- ings here apply to other firms or other periods of time, and (2) the focus on em- ployees who were continuously employed by this firm over the 6-year period] under scrutiny imposes a censoring of the data- employees who left the firm may have tract systematically different characteristics or ex- periences in the firm. A section of this paper does compare findings for the censored set of workers with an uncensored set. A specific question addressed by Gerhart and Milkovich is what, if any, is the salary disadvantage for women compared with men in this firm? In 1980, overall, the answer is that a woman received a salary that was only .84 of a man's salary, before adjusting for human capital differences, an(l was about .88 after adjusting for schooling, job tenure, tenure with the firm, ant] a measure of other potential labor market experience. (The main influence here is job tenurethe men had been at their specific jobs a good while longer than the women.) The salary (liffer- ential was not much affected by including in the analysis the 4-point job performance scale, but if"job level" is accounted for, the women's pay increaser] to about .96 of the men's pay. (job level is defined in terms of status and authority within the firm, which was measured by a 7-point scale. ) So we learn, that in this firm, in 1980, for these categories of employees, adjusting for all these skill measures and job assignments, there remained a 4 percentage point dis- advantage for women within job level, but a large 12 percentage point difference related to job level. It is not obvious what we should make of the finding that within job level the sex differences in salaries are as small as 4 percentage points, while across job categories the differentials are far greater. Interpreting the 4 percent residual re- mains a dilemma that is a plague of a re-
4 search strategy that leaves the crucial ele- ment in the residual: If we can remove the differential we might attribute it to the variable that achieved its removal, but if we cannot remove it from the residual all we can say is that it is still there. Interpreting the larger differential across job levels re- quires an answer to another question. Why are the women disproportionately in the lower salaried jobs? Gerhart and Milkovich's data cannot tell us. If it is because of choices men and women make about the type of jobs they want, we would not want to think of it as discrimination by this firm. If, how- ever, it is because of restrictions imposed on women, then we would want to attribute it to discrimination or put more cautious- Ty, we could not rule out the possibility that it is attributable to discrimination. For a small subset of men and women in their sample who worked in job titles with 10 or more incumbents, Gerhart and Mil- kovich find that the higher the percentage who are female in that job, the lower is the salary of the men, but not women, in that job. They conclude that this subset of their sample "does not really provide support for the idea that percentage female is an im- portant structural property that negatively affects women's (and perhaps men's) attain- ,, meet. A second important finding in Gerhart and Milkovich's study is that the wage dis- advantage of women declined slightly be- tween 1980 and 1986the overall relative wage of women to men rose from .84 (in 1980) to .88 (in 1986), and adjusted for skill, performance, and job level, it rose trivially and surely insignificantly (statistically) from .96 to .97. Gerhart and Mflkovich suggest that the firm may now be compensating women for past inequalities. Gerhart and Milkovich address two other questions in their paper: Do men and wom- en receive equal salary increases? Do they receive the same promotion opportunities over time? The answer to each of these questions is no women fare better than PAY EQUITY: EMPIRICAL INQUIRIES men. Women's salaries rose more rapidly than men's, and over the period 1980-1986, women "had a distinct promotion advan- tage," say Gerhart and Milkovich. Men had, on average, .9 promotions while women had 1.3. Moreover, the advantage women had in terms of promotions did not decline at higher job levels. For men, their greater experience in the labor market appears to be a major factor that helps explain their lower level of job promotion, because pro- motions come more frequently early in one's career. To respond to the concern about cen- soring in their sample, Gerhart and Mfl- kovich also looked at salaries for all women and men employed in 1980 or in 1984 without conditioning for continuous em- ployment. They then compared the average salary growth for the two separate groups with salary growth for the subset who were employed in both 1980 and 1984. They found that the relative growth rate of women compared with men was the same in the two cases: Women's salaries grew by 114 percent of the growth of men's salaries. Apparently, focusing only on those with continuous employment in this firm did not create a biased picture. In his critique of the paper, Winship stresses two additional points: (1) the results are potentially sensitive to the functional form of the equation used to adjust for skill and job level, and a less restrictive functional form might have been better and (2) many interpretations can be given to the findings in this paper. Winship elaborates several alternative stories based on women's childbearing behavior and the employer's screening devices that could explain the findings. Occupational Segregation and E. arcings Sorensen uses micro-level data from the May and June 1983 Current Population Survey to investigate the influence of oc-
ASSESSING THE ISSUES cupational segregation by sex and race on hourly earnings. She estimates regression equations on hourly earnings separately for white men, white women, minority men, and minority women, where minority in- cludes blacks and Hispanics (other minority groups are excluded from her study). The data contain information about tenure on the current job, but other job market ex- perience is measured by the convenient and frequently used device of age minus years of schooling minus six, approximating the number of years since the person left school. We call this the potential experience index. The measure may be a better indicator of labor market experience for white men than for white women, because women tradi- tionally have spent more of their adult life- time outside the labor market. After controlling for personal character- istics and attributes of the occupation and industry, Sorensen focuses on the propor- tion of the occupation that is female, a variable discussant Smith called the com- parable worth variable. That variable is sys- tematically related to lower wages for all four of the groups Sorensen studies. For each of the four groups white men, white women, minority men, and minority wom- en the wage is about 2 percent lower for someone working in an occupation with a 10 percentage point higher proportion of women. The finding seems to hold up when various alternative ways of estimating the equations are compared. For white men and women, Sorensen looks separately at three sectors: public, manufacturing, and nonmanufacturing (mainly the service sector plus construction and mining). The proportion female in the occupation lowered the wages of men and women in the public sector relatively strong- ly (by about 4 percent for white men and 2 percent for white women for each 10 percentage point increase in the proportion female). In the nonmanufacturing sector, the effect was less strong (by 3 percent for white men and 2 percent for white women), 5 and in the manufacturing sector it was least strong (by 2 percent for white men and insignificantly for white women). In her discussion of Sorensen's paper at the work- shop, Malveaux noted that the lack of im- portance in manufacturing could be due to the importance of industrial or firm seg- regation within manufacturing (e.g., men work in durable goods manufacturing and women in nondurable goods). Sorensen also investigates the impact of the percentage of the occupation that is minority (black and Hispanic) and finds a statistically significant impact only for white men a 10 percentage point increase in the proportion minority is associated with a 4.9 percent lower wage for the white men in the occupation. That effect is substantially smaller (-.8 percent) for white men in another specification of the model, and it appears to be present only in the nonmanu- facturing sector of the economy. Comparing the wages of white men and white women, Sorensen concludes that dif- ferences in jobs and personal productivity account for about 25 percent of the observed difference in wages overal14$3.32), and in- dustrial and regional differences account for another 15 percent. Occupational segre- gation by sex accounts for an additional 20 percent on average, which leaves about 40 percent unexplained by any of the mea- surable factors. For minority men compared with white men, the job and personal skill variables account for about half of the ob- served difference in wages overall ($2.11), but the other half is unexplained that is, the occupational segregation by sex or race and the industrial and regional differences in jobs held do not explain any of the observed differences between white and . . minority men. Malveaux raised the issue of whether personal productivity characteristics really are related to productivity (or are simply inexpensive screening devices) and whether they are free of race and gender bias. Dif- ferences in educational attainment or in
6 courses of study, for example, can result from "tracking" or other factors. The unex- plained residual might then understate the extent of discrimination, Malveaux con- tends. Overall, Sorensen concludes that as much as 20 percent of the national femaTe-male earnings disparity could in principle be eliminated by a policy that eliminates] oc- cupational sex-based differences in wages, ignoring all the other complications that might arise. The wage ratio for women to men could be increased, say, from about 64 percent to 72 percent, which would reduce the size of the wage gap from $3.32 to $2. 66. As Malveaux pointed out, this means comparable worth is a limited strat- egy, though by no means an insignificant one. Labor Market Crowding and Earnings of Women Nakamura and Nakamura provide a rather different study of individual wage deter- mination. They argue that it is important to understand how wages are cletermined in female labor markets, and they refer to a substantial empirical literature that sug- gests that there are distinct male and female labor markets. The Nakamuras use data from the 1980 U. S. census to investigate the wages of employed women 20 to 24 years of age. That data set is one of the few with sufficient numbers of observations to permit examination by occupation and other subgroups of interest. The authors focus on crowding in the labor market, by which they mean a relative abundance of women offering their labor in a particular market. They measure crowding by the number of women in the entry age bracket (20 to 24) compared with another age (25 to 29) and by the employment rate of women of that age compared with the other age. Nakamura and Nakamura suggest that there are two reasons why we should expect women with relatively low levels of eclu- PAY EQUITY: EMPIRICAL INQUIRIES cation, those who are black, and those with children to be especially vulnerable to labor market crowding: (1) barriers to entry to better jobs (such as schooling requirements) may protect the more e(lucated women but not the less educated and (2) effective labor bargaining can secure concessions from em- p-Ioyers through contractual agreements that rely on seniority and promotions from within the firm, so that those already employee} can protect themselves from crowding. This suggests that crowding would have an ad- verse effect on the wages of lower skilled and black women an(1 on those who are mothers, but not on the wages of higher skilled en cl white women. This indeed is what Nakamura and Nak- amura find tentative evidence of in their regressions of the log of wages on personal characteristics and state-level measures of crowding and unemployment. The relative population size an(l employment rate of women aged 20 to 24 tend to (depress wages of women in occupations that have fewer welI-educated women personal service, other clerical, secretarial, and salesbut not so in occupations requiring more school- ing managerial, health, and professional/ technical occupations. Similarly, the crowc3- ing effects are (liscernible for women in the sample sorted by less education, by the presence of children, and by race (black), but not so for women with more than 12 years of education, with no children, and who are nonblack. In his comments on the Nakamura ancT Nakamura paper at the workshop, Ehren- berg raised two cautionary notes. First, he suggested that the evidence of crowding is cloucled by inadequate control for job ex- perience and by a mismeasurement of the crowding variable. Census data do not re- veal how much labor market experience the women have, and the conventional mea- sure, years of potential experience (age mi- nus years of schooling minus six), probably overstates the experience for blacks, moth- ers, and the less eclucated, for reasons he
ASSESSING THE ISSUES articulated. The crowding measures, Eh- renberg thinks, would be better suited if they were occupation specific instead of state specific. He further cautions that if the crowding is the product of voluntary choice by women, the case for public policy intervention is not strong. He urges sub- sequent research looking into the process by which women make their occupational choices. The studies of individual earnings by the Nakamuras, Gerhart and Milkovich, and Sorensen substantiate that the pay received by a woman is less than the pay received by a man, when skill and other relevant factors are accounted for. Of the three pa- pers, perhaps Sorensen s paints the bleakest picture since a substantial gender gap re- mains in the Current Population Survey data after adjustment for measured char- acteristics. One of the important variables negatively affecting earnings, she finds, is the proportion of an occupation which is female. Although Gerhart and Milkovich s data on one firm exhibit large gender dif- ferences in job assignment, the proportion female of a job floes not seem to affect wages, salary growth, or promotion. The evidence over the 6 years following 1980 indicates that women who had less initial experience have had more promotions ancT raises and that the gender gap in wages was smaller, although not eliminated, by 1986. Nakamura and Nakamuras paper looks to the national job process and identifies three subsets of women who, they believe, are easily vulnerable to labor market crowding. Sex-Role, Occupational Choice, and Salary A very different orientation to incliviclual wage determination is represented by Sub- ich, Barrett, Doverspike, and Alexander. Their paper discusses a set of issues about psychological differences in men and women and whether those differences might par- tially be responsible for observed wage dif- 7 ferences. This paper provides a survey of literature on psychological research on fac- tors identified as related to occupational behavior and its outcomes. The factors in- clude male-female differences in knowledge of salaries in various occupations, in self- confidence and personal expectations in the marketplace, and in risk-taking behavior. Subich and her colleagues indicate, for example, that the literature supports the notion that risk taking is a masculine attri- bute, that men are bol(ler than women and more venturesome physically and with fi- nancial decisions. This gencler-role differ- ence, we are told, may carry over to men being more likely to gamble by asking for a raise. Subich and colleagues conducted two pil- ot studies with college students, the results of which are reporter! in their paper as illustrative of psychological gender differ- ences. The studies found that when asked about salaries in their inten(le(l occupations, both genders overestimated salaries sub- stantially, but men di(l so to a greater extent than women. There was, they report, no clear evidence of a gender difference in confidence about ones own occupational success. Men, however, did seem to be more prone to risk taking. Both fin(lings might contribute to salary differences be- tween women and men. Subich and colleagues remind us that there are subtle personal factors that affect expectations and performance in the labor market. These factors suggest alternative remedies to reduce male-female (liffer- ences. At the workshop, discussant Hu(lis suggested a future research strategy to iden- tify some of these factors. Though, as she noted, the subjects in the pilot studies were college seniors with, presumably, some in- terest in the job market, data from workers would be more fruitful to analyze. Risk- taking behavior by female and male em- ployees- taking a risky overseas assign- ment, for example could be directly ex- amined in a large firm, where actual salary
8 data, including salary history, would be available. Hudis further suggests that within a firm equal employment opportunity policy could be influenced by knowledge of wom- en's and men's risk-taking behavior. If risk taking pays off, then women should be encouraged to try it, and information about the rewards of various career opportunities should be more widely shared. Jobs and Occupations as the Unit of Analysis Three papers in this volume study the relationship among occupation-based wage rates (average wages for women and men in an occupation), gender, and various fac- tors that could explain how and why oc- cupational wage rates differ. The interest in jobs and occupations as the unit of analysis in comparable worth studies has several bases. Most important, perhaps, is the com- parable worth claim itself: Female-domi- nated jobs and occupations are underval- ued not indiviclual nurses, but the nursing profession itself is paid less than it is worth. A theoretical framework for evaluating the reasonableness of the comparable worth claim had already been establisher] in so- ciology and economics with the study of institutional labor markets and occupational structure. That body of literature has also contributed to the comparable worth stud- ies. Given that employers do treat howlers of particular types of jobs similarly (as group members rather than as individuals) and given that many occupational groups exhibit stable relationships with each other, it fol- lows that female-dominatecl occupations may exhibit some differentiating characteristics. Thus, the "percent female" of a job or occupation has become a variable of note. In the papers that use the job as the unit of analysis, Baron and Newman find that both female dominance ancT minority dom- inance of jobs in the California civil service system lower the wage rate for those jobs; Parcel also finds negative effects for percent PAY EQUITY: EMPIRICAL INQUIRIES female for male, but not for female, em- ployees; and Filer finds the effect small and insignificant for both genders. Effects of Demographic Composition on Pay Rates for lobs Baron and Newman study how the pay rates for specific jobs are affecter] by the demographic characteristics of the people who hol(l those jobs. They consider the state of California's civil service system- over 3,000 separate jobs an(l nearly 125,000 incumbents. The time perioc] they consi(ler is 1979 through 1985; some of the analysis considers the annual cross sections anc] some considers changes over the 6 years. Their dependent variable is the prescribed start- ing pay for a job, not the earnings of those in the job, so their measure is not directly affected by any sex or race differences in skill, seniority, or productivity. They study how that authorized starting salary is af- fected by factors like percent female or percent black, and they hoIc] constant in various levels of detail the job's content as measured by educational and experience requirements or by occupational cIassifi- cations that purport to reflect the (difficulty, or value, of the job. Their results are striking. No matter how many controls they introduce to take ac- count of the job characteristics, significant and sizable effects of sex composition and race composition on those starting pay rates remain. "jobs dominates] by men pay con- si(lerably more than otherwise comparable jobs (dominated by women," they conclu(le. Their Table 5-4 shows that the regression- estimated penalties apparent in female- and minority-dominated jobs are dramatic. Con- sider a nonsupervisory clerical job in "office or allied services," for instance, a job re- quiring 13 or more years of schooling and no more than 4 years of experience. If it had the demographic composition of the average fulI-time white male's job (which is 61 percent white male, 13 percent white
ASSESSING THE ISSUES female, 5 percent black male, 3 percent black female, etc. ), the starting 1985 month- ly salary would be $2,230. If it had the demographic composition of the average full-time white female's job (which is 18 percent white male, 47 percent white fe- male, 3 percent black male, 9 percent black female, etc. ), the starting 1985 monthly salary would be only $1,860. Baron and Newman also compare jobs in 1979 and 1985 and investigate whether the changes in the composition of incumbents are related to changes in starting salaries. They estimate that over the 6 years stucTied, the penalty on the starting salary associated with the presence of blacks and male His- panics increased. Typical of this finding is the estimate that a 10 percent increase in the percentage of black males lowered the starting salary of the job by 2.6 percent in 1979 but by 3.9 percent in 1985. The adverse effect on the starting salary of white and Hispanic females, on the other hand, seemed to be reduced: a 10 percent increase in white females lowered the salary by 3.3 percent in 1979, an elect that was weakened to 2.7 percent by 1985. They also find that more recently created jobs ones that were not in the system in 1979have less severe penalties than older jobs for female- and minority-dominated jobs. This, they con- tencT, is related to the fact that a dispro- portionate number of the new jobs were in high-skill, high-paying occupations, not to an across-the-board increase in equity in starting salaries. Overall, Baron and Newman conclude that their results show that "the entry of females and minorities into positions de- values them." The penalties against female with information from the Dictionary of anciminority-dominate~jobs appear severe, Occupational Titles (DOT) and has 503 oc- and the underpayment associated with these cupations in her study. Filer merges data workers is greatest in jobs that have many from several ancillary sources, including the incumbents. DOT, and, in order to obtain appropriate Baron an(l Newman's careful analysis of the job and pay structure of a single large employer is useful. Because they study start- ing salaries of jobs (rather than actual wages 9 of male or female incumbents), the "resid- ual" problem is less severe. With indivicI- uals, there might always be some unmea- sured characteristic, such as motivation, that might have an elect. With jobs, the im- portant requirements are more likely to be states! and therefore known to the research- ers. As Ross points out in her comment, Baron and Newman's study could be rep- licatec! at many public agencies (and possibly private firms as well). Of course, a study that is not about actual wages received leaves certain questions unanswered. The effect of this gender- and race-biased struc- ture of job salaries on actual salaries received by women, men, and minorities is not ex- plore(l here. Also, as with most statistical studies, the wage setting process and the employer's intent are unexplored. Ditl the employer lower job salaries when women and minorities entered them? Were women and minorities recruited because of a short- age of white men and/or because job re- quirements were changing? Baron anc! New- man's analysis controls for skill changes in the stated job requirements, but those skill requirements may lag or leacl changes in actual practice. Occupational Differences and Earnings Both Parcel and Filer use the detailed occupation (1980 census) as the unit of anal- ysis an(l study the occupation's average earn- ings. Parcel acljusts the earnings of all work- ers to a full-time equivalent level, while Filer uses data on only full-time, full-year workers. Parcel augments the census data matching, uses 430 occupations. Using factor analysis based on the DOT, Parcel identifies five distinct attributes of the occupations: the "substantive complex-
10 ity" of the occupation, the "physical dex- terity/perceptual ability" required in the occupation, its "physical activity/working conditions," and two others. Parcel imbeds these attributes of the occupation in a the- oretical context combining supply and de- mand variables with measures of social or- ganization. She has measures of the average educational and experience levels of the occupations' incumbents, the labor market conditions of the occupation (e.g., the re- serve labor pool), and characteristics of the incumbents (e. g., the percentage offemales, blacks, Hispanics, and Asians, and the per- centage of men and women who are mar- ried). Mindful of the deficiencies in the traditional measure of potential experience for women, Parcel attempts to improve the measure by adjusting for race and marital status, as described in her paper. Typical of occupational-level analyses when estimated for men and women combined, the percent female in an occupation is found by Parcel to have a sizable negative effect on the annualized earnings in the occupa- tion a 10 percentage point increase in the proportion female is associated with a $710 reduction in the average earnings in the occupation. Many of the other factors also display their usual effects. Parcel summa- rizes, "female-dominated occupations are low in earnings, experience, percent males married, unionization, and the job content measures of physical activities. They have high reserve labor pools, are urbanized, and have high black and Asian concentrations." When Parcel estimates the effect of per- cent female separately for men and women, however, she finds a significant negative effect for men, but no effect for women. She argues that "percent female is but one aspect of occupational market social orga- nization" that affects female earnings. Other social dimensions of labor market organi- zation that affect earnings include minority concentrations, extent of unionization, and proportions of males and females married. Smith finds Parcel's work skillful and sen- sible, but he questions the whole line of PAY EQUITY: EMPIRICAL INQUIRIES inquiry that includes as an explanatory vari- able the occupation's percent female. He argues that such a variable does not add to our knowledge about how wages are set in labor markets or whether there is or is not discrimination. He thinks it does no more than verify that the wage distributions for men and women differ; it does not help us understand why they differ. If the negative coefficient on percent female offers evidence of discrimination against women, he asks, then does Parcel's positive coefficient on "percent Asian" imply the existence of dis- crimination in favor of Asians Or are there other unmeasured factors? Filer's analysis considers numerous fac- tors from a variety of data sets. Although he also uses DOT information, he uses a wide array of very specific occupational de- scriptors rather than a condensed and syn- thesized (factor analyzed) set of five features of each occupation. At one level, Filer con- firms Parcel's finding. His Table 7-1 reports results on hourly wages for a change of 100 percentage points in the percent female. If we reduce the impact to a change of 10 percentage points and express it in annual earnings, we find his estimate of the de- crease in earnings (due to a 10 percentage point increase in the proportion of the work- ers who are female) to be between $626 ~ = $3.13 x 0.1 x 2,000 hours), controlled only for demographic and skill factors and unionization, and $270 (=$1.35 x 0.1 x 2,000 hours), controlled in addition for ef- fort, responsibility, and working conditions. Parcel's finding was $710 relatively uncon- trolled, and $500 to $574 with various con- trols. Smith, in his comment, points out that Table 7-1 also suggests that 20 percent of the wage gap is attributable to the "com- parable worth variable" (the proportion fe- male), a figure identical to Sorensen's. But Filer argues that these figures are misleading, because they are "inherently incapable of addressing comparable worth issues," defined by Filer to be a concern for raising wages in jobs or in occupations heavily filled by women. His argument has
ASSESSING THE ISSUES similarities to the point made by Smith in his discussion of the Parcel paper. If women are paid, say 75 percent as much as men, for any reason, then the average wage in an occupation will automatically be lower the higher the proportion of women in that occupation, even though the "percent fe- male in the occupation" has no effect what- ever on any inclividual's wage. The corre- lation between the average wage and the percent female is just a reflection of one wage schedule being below the other. It tells us nothing about why those schedules differ. Filer, therefore, argues that one should investigate separately men's and women's wages across occupations if one is interested in seeing whether the proportion female in an occupation has any effect per se on wages. (Note that Sorensen did this in her indi- vidual-level analysis and that Baron and Newman's study of starting salaries in jobs cloes not suffer from this problem. Both authors found large, significant differences in earnings due to differences in occupa- tional gender composition.) When Filer conducts an inquiry on men's and women's average occupational earnings separately, he reports perhaps the most controversial finding in this volume. When a large number of controls are used, re- flecting demographic characteristics, indi- vidual productivity factors (aggregated to the level of the occupation), unionization, and the usual job content factors used to assess comparability (effort, responsibility, ant] working conditions), there is no evi- dence that the percent female in the oc- cupation has an influence on either the wages of women or the wages of men. What appears to be an effect in other formulations, "results from the lower wages for women within each occupation," which Filer con- tencis can be corrected, if desired, by ap- plication of the equal employment laws, and would be `'immune to comparable worth . ,, remet lest Smith expresses reservations about Filer's approach, claiming that the inclusion of so 11 many separate variables (over 225) makes interpretation of the coefficient of each near- ly impossible and leads to questioning "the believability ofthe entire exercise." Another factor related to having a large number of variables, as Filer cloes, may also be of more substantive importance. As panel member Blau pointed out during the workshop, some of Filer's variables may be proxies for gender itself rather than indicators of substantive factors that couIcI reasonably be linked to productivity differences. Summarizing these findings at the job or occupational level of analysis, the authors find that all three papers confirm that wom- en's wages are less than men's wages at the occupational level. Baron and Newman's strategy does not suffer from the compo- sitional effect about which Filer and Smith warn, ant! Baron and Newman do find a systematic tendency for jobs held dispro- portionately by women to have lower start- ing pay than apparently comparable jobs held disproportionately by men. Filer's strategy for adjusting for average productivity differences between the oc- cupations and the universe from which his data are drawn are very different from Baron ant! Newman's, and his conclusion is dif- ferent as well. He finds no evidence of a systematic tendency for occupations hel(l disproportionately by women to have lower average full-time salaries than comparable occupations held disproportionately by men. Filer does confirm that within an occupation women earn less, but not because it is an occupation dominated by women. This dis- tinction may be subtle, but the potential vali(lity of a comparable worth policy may hinge on it. Filer's findings, however, are weakened by the weak rationales for some of the many variables in his analysis (some of which may be correlated with percent female rather than with compensable job factors). Parcel has employed a technique (factor analysis) that is clesigned to reduce a large number of variables to a few theoretically coherent and more easily interpretable major factors.
12 When her regressions are run separately for men and women, she, like Filer, finds no effect for percent female on the earnings of women, but unlike Filer, she finds a significant and sizable negative effect of percent female on the earnings of men in the occupation. Implementation of Comparable Worth Policies Comparable worth policies have been im- plemented in some private firms and gov- ernmental jurisdictions of various sizes. Three papers in this volume address the effect such policies have had. Two of the papers investigate the impact of state-imposed com- parable worth legislation on the state-wide government pay schedules in Iowa (intro- duced in 1985) and in Minnesota (passed in 1982~. The third studies the effects of a national policy of pay equity introduced in Australia ant] Britain in 1975. lowa's Comparable Worth Plan Orazem and Mattila study the case of Iowa, a state that hired a consulting firm to evaluate the 800 job classifications in the state employment system and, according to the authors, instructed the firm to "ignore market wages in conducting its analysis" and in making its recommendations about changes in wage structure. The firm used a point system to evaluate the attributes of the job or its requirements, using skill level, effort, responsibility, and working condi- tions to determine the "worth" of the job. As Orazem and Mattila describe it, the recommendations of the firm were modified in the political process of implementation, in which the employee unions and state political leaders figured prominently. In ear- ly 1985 the new system went into effect, at an estimated wage-bill cost to the state of about $19 million annually roughly $1,000 per employee. Orazem and Mattila take the state pay PAY EQUITY: EMPIRICAL INQUIRIES schedule of December 1983 (before com- parable worth) as the benchmark for their study. They use a 20 percent sample of the personnel files of the state's employees, gathering information on the individual's personal characteristics and experiences as well as his or her job and pay. For the 3,734 persons on whom 1983 actual biweekly earn- ings are known, Orazem and Mattila cal- culate two additional earnings figures: (1) the earnings associated with the consulting firm's pay recommendations, based on the comparable worth study (the "recommend- ed" earnings), and (2) the earnings associ- ated with the compromise plan actually im- plemented in 1985 (the "compromise" earnings). These latter two biweekly earn- ings figures are counterfactual estimates, not the actual earnings of employees. Or- azem and Mattila contend that this esti- mation scheme gives them a clearer picture of the effect of the new scheme, without confusing it with the many other factors that may also have affected wages between December 1983 and the introduction of the actual plan some 15 months later. Orazem an(l Mattila then perform several regression analyses of the log of biweekly earnings, using each of the three earnings figures separately. The authors compare the effects of personal characteristics and job attributes on the wages actually paid in 1983 to their effects on the recommen(lecl wages and to their effects on the compromise wages. Nearly half the employees in the state's wage system were women, and Orazem and Mattila found that the biweekly wage of the women initially was about 78 percent that of the men, unstandarclize(l for anything. By comparison, the recommended plan would have raised that raw proportion to 86 per- cent, an(1 the compromise plan that was actually implemental would have raised that proportion to 82 percent of the males' wages. After adjusting for human capital variables, Orazem and Mattila estimate, by one technique, that the women's biweekly
ASSESSING THE ISSUES initially was 94 percent that of the wage men s wage, and that the recommended plan would have raised that proportion to complete parity (100 percent); the compro- mise plan would have raised the proportion only to 96 percent. (Other techniques of estimating these figures yielded somewhat different results, but the qualitative con- clusions here are robust.) They suggest that the women's biweekly wage was raised through the compromise plan by about $50 and the men's wage was raised by about $30, for an average increase of about $40, which translates into a $1,000 annual earn- . . 1ngs Increase. Orazem and Mattila detail in their paper the major factors determining the actual 1983 wages ant] discuss the changes in the effects of those factors implied by the rec- ommencled and compromise plans. The compromise plan resulted in a tiny reduc- tion in the dispersion of biweekly earnings, compared with the 1983 actual distribution. It is interesting to note that the recom- mended plan dicI in fact completely elim- inate the statistical significance of the vari- able "percent female" as a determinant of the wage, thus eliminating a strong negative 14.6 percent effect on the actual 1983 wages. The compromise plan, by contrast, restored (or retained) a small gender differential of 5.8 percentage points. The recommended plan would have involved pay cuts for 7,300 workers and increases for 10,750, but one element in the compromise was that no one's wage would be lowerecI. Regarding the factors determining wages, Orazem and Mattila point out that from a human capital perspective, the "measured discrimination against women is very slight" in the sense that measures ot Skill and market conditions appear to explain nearly all of the variation in wages (R2 = .815 in their Table 8-3, including the human capital variables but exclucling the proportion fe- male of a job). But they also point out that from the perspective of a comparable worth advocate, their comparable worth mode} 13 implies that "large discrepancies in pay exist between men and women because women are concentrated in jobs that are paid below the value placed on comparable male jobs." (I.e., R2 = .769 in their Table 8-4, the mode! including percent female on a job.) As so often is the case, one's perspective on the finding can dramatically influence the interpretation it seems to support. In her discussion at the authors' work- shop, panel member Schoen pointed out that readers of the paper could more ade- quately form their own interpretations if the authors had provided more description of the institutional factors at play in Iowa. From her own experience with job evalu- ations, unions, and comparable worth, Schoen believes outcomes will vary substantially from state to state. Although protection against lower wages is a common outcome, she stressed that wage protection is often not accomplished by altering factors and weights in the job evaluation scheme, but by protecting current workers. An uncler- standing of the particular economic an political situation the unions and state lea(l- ers faced wouIc] help the reader evaluate the reasonableness of the outcome. Pay Equity in Minnesota Evans and Nelson study the case of Min- nesota, which passed pay equity legislation in 1982 for its state employees. The new policy was implemented over 4 years be- ginning in 1983. Minnesota had since 1979 had a job evaluation system based on a point factor scheme. The pay equity legislation of 1982 built on that scheme, requiring a single job evaluation system for all job clas- sifications in the state employment system. The evaluation measured the skill, effort, responsibility, and working conditions off each job and yielded a composite score for each job. All job classifications with the same score were then considered to have equal value and, hence, to command equal pay. Evans and Nelson stress that con-
14 versing this score into pay level was un- ciertaken in Minnesota using the white mate's 'wage as the norm. The vast majority (86 percent) of Min- nesota state employees are represented by unions, including a large majority of women working for the state. Neither the unions nor the state government aggressively a~l- vertisec] or notified employees of the impact of the pay equity legislation. "Changes in one's paycheck formed the major 'notifi- cation' of pay equity, a notification that did not distinguish between regular pay raises of approximately 3 to 4.5 percent per year and the additional increment due to pay equity raises." Over the 4 years of imple- mentation (1983-1986), Evans and Nelson report, about 8,500 of the state's 34,000 employees received pay equity raises, an of those 90 percent were women. The raises added about 3.7 percent to the state's wage bill. Evans and Nelson report findings from a telephone survey in June 1985 of about 500 state employees. The survey asked the re- spondent about his or her "support for, knowledge about, receipt of, and reactions to pay equity." The employees were rela- tively well eclucated (e.g., 37 percent had a bachelor's degree or more), and a majority had worked for the state for more than 7 years. The average salary of state employees in 1984 was $22,500. While Orazem and Mattila consider the economic impact of the pay equity legislation in Iowa, Evans and Nelson focus on the psychological effects in Minnesota in terms of the attitudes and knowledge of state workers about the new scheme. The survey indicates that the employees overwhelmingly supported the concept of pay equity; support for the concept ap- peared to be strong at both ends of the political spectrum and both ends of the occupational ladder. Likewise, the survey indicates the actual policy of pay equity was well known to the respondents: 82 percent of them ha(l heard of pay equity or com- PAY EQUITY: EMPIRICAL INQUIRIES parable worth legislation. Evans and Nelson characterize the specific understanding of the (letaits of pay equity as "quite knowI- edgeable," based on the respondents' an- swers to the questions in the survey. The most intriguing findings in the sur- vey, as Evans and Nelson stress, involve a comparison of whether the respondent thought he or she received a pay equity raise comparer] with whether he or she actually received one. The authors had in- formation on actual raises from the state employment records and could compare those facts against the telephone responses to questions about whether a raise was received. Recall that, for reasons the authors describe in their paper, neither the unions nor the state employment office made a major effort to inform the employee about his or her pay equity raise. The fincling is striking: Of those who actually received a pay equity raise (nearly one-third of the survey sample had received a raise), 56.9 percent knew they receive(l one, 21.6 per- cent reported not receiving one, and 21.6 percent never had heard of the pay equity policy. As Evans and Nelson say, "the social movement potential of pay equity is cer- tainly unfulfilled if 43.2 percent of the peo- ple who benefit from the policy are unaware of their benefits." About half the sample correctly reported that they received no pay equity raise. The accuracy of the reporting was greater at higher levels of education and salary. Evans and Nelson discuss the role of the union in supporting the implementation of pay equity and its strategy of avoiding pub- licity about its implementation. They con- clude that the strategy dampened both op- position to and support for the pay equity policy. In fact, of those surveyed 36 percent reported that they believer! that pay equity policy causerl many problems in the work- place, despite the overwhelming support of it as a concept. In commenting on the Evans and Nelson paper at the authors' workshop, panel mem-
ASSESSING THE ISSUES her Waite noted that a single cross-sectional telephone survey cannot elicit information about how the change in wages changed job satisfaction or attitudes toward pay eq- uity The survey offers only a static view, as the authors acknowledge. Waite also com- mented on one of the unique factors at play in the Minnesota case, as described by Evans and Nelson: The job evaluation hac] been done prior to the adoption and im- plementation of the comparable worth pol- icy. Thus, the general realignment of jobs and pay that often results from a new pay plan was not part of the comparable worth process. The comparable worth realignment was allowed to be a more specific, limited event. Waite suggested that the strong con- sensus in favor of comparable worth in Min- nesota may not be easily achieved in other states, where the job evaluations and re- sulting wage realignments are more directly occasioned by the comparable worth policy itself. Women's Pay in Australia, Great Britain, and the United States Gregory, Anstie, Daly, and Ho provide a very different empirical inquiry from oth- ers in this volume. Their study provides a two-decade perspective on the relative earn- ings and employment of women in three nations. They point out that Australia and Britain have experienced substantial in- creases in the female-male earnings ratio over the past 20 years but that same ex- perience has not been shared by workers in the United States. In their paper, Gre- gory and his colleagues address three ques- tions about that experience and attempt to synthesize the evidence from the three countries. In Australia, wages, or minimum wage rates, are awarcled by an official network of governmental tribunals for every occu- pation in the nation, for both the private and public sectors. For the period from 1950 to 1969, Gregory ant] colleagues tell 15 us that the official wage setting boards ex- plicitly market! down the wage in all oc- cupations clominated by women to 75 per- cent of the wage received by men. The wage levels set were explicitly lower for female occupations than for male occupa- tions. Over the 6 years from 1969 to 1975 that official practice was eliminated and the average wage ratio of awarded female to male wages rose accordingly from 72 percent to 92 percent, a dramatic change in a very short time span. That historic experience, mirrored in somewhat mutec] form in Brit- ain, makes the three-country comparison quite informative. The first question a(ldressed by Gregory and colleagues is why the relative earnings of women compared with men are so dif- ferent in the three countries. In 1981 in Australia full-time average earnings of wom- en were 79 percent as much as men, while in the United States and Britain women earned only about 60 percent and 64 percent as much as men, respectively. The authors use a conventional human capital moclel approach to attempt to provide an expla- nation. Their data consist of weekly full- time wage and salary earnings from house- hold survey data from each of the three countriesa 1981 survey in Australia and in Britain, and the March 1982 Current Population Survey in the United States. A standar(l (lecomposition analysis is per- formed to see if the observed differences in weekly earnings of full-time workers are attributable to differences in the human capital endowments of men and women, that is, to differences in schooling, job ex- perience, marital status, and the presence of children. Although the statistical mode} for each country performs reasonably well, and to a similar degree, as an explanation of the variation in earnings among men and women, it does not explain why women earn so much more relative to men in Aus- traTia. They conclude that the human cap- ital endowments of women relative to those of men, seem to be much the same in each
16 of these countries." Since large differences do not exist between women and men in one country compared with another, those human capital differences cannot explain the differences in relative earnings. The second question Gregory and col- leagues consider is why the pay ratios have changed so dramatically in Australia and in Britain but not in the United States in recent years. The answer, they argue, lies in in- stitutional considerations. In Australia, the governmental tribunals simply changed the acceptable relative wage from one that was substantially lower for women than for men to one that reflected "equal pay for work of equal value" without regar(l to the sex of the employee. In Britain, too, the authors describe a predominantly regulated wage structure in which national agreements in- volving large unions set rates of pay for a wide range of workers. Explicit discrimi- nation against women in pay rates charac- terized the British labor market, say Gre- gory anti his colleagues, until the Equal Pay Act of 1970, which became effective in December 1975. Table 10-3 in their paper shows the dramatic rise in the relative wages of women between the passage of that act and its implementation. In Australia and in Britain, Gregory and colleagues contend, "it was relatively easy to remove that which was identified as pay discrimination and, as a result, to affect dramatically the pay relativities between the sexes." In the United States, the federal legislation designed to achieve "equal pay for equal work" was passed earlier than in the other two countriesas early as 1963 or 1964. Its effect, however, is not nearly so evident in the aggregate time series data on relative wages, and the authors offer several conjectures about why that is so. They note that the large-scale institutions in the Australian and British labor markets (the minimum wage tribunals and collective bargaining agreements) had ma(le the dis- crimination implicit in market wages ex- plicit; the same large-scale institutions coul PAY EQUITY: EMPIRICAL INQUIRIES correct the explicit discrimination. In the Uniter! States there are no comparable large- scale institutions, and wage changes thus depend on the decisions of many actors in the labor market. The third question acldressed by the au- thors has to do with potential side effects from the comparable worth medicine, spe- cifically potential employment loss. They ask how the dramatic change in female earnings rates in Australia and Britain has affected the employment rate and the un- employment rate of women. The answer is a surprising one: The effect seems to be very slight. The female share of total hours worked rose over the period 1970-1984 in all three countries (by 25 percent in Aus- tralia, by 27 percent in Britain, and by 31 percent in the United States), but the rel- ative wage of women rose substantially more in Australia relative to the other two coun- tries. The small employment responses to the sharp changes in relative wages of wom- en in Australia and in Britain are surprising; they imply, the authors contend, a very low substitutability of men for women in the productive processes of the country. Their "cursory glance 'at unemployment rates also suggests only a slight impact in the relative (remand for female workers in Australia. Ehrenberg, in his comment, calls atten- tion to the virtues of bringing an interna- tional comparative perspective into the cle- bate about the policy of comparable worth in the Unitecl States. He argues, however, that the authors have not "pushed their empirical analyses as hard as they might have," and consequently, they may have drawn some inappropriate conclusions. For example, Ehrenberg notes that the coeffi- cients on human capital variables differ from country to country, but that insufficient explanation is offered. Both Ehrenberg and the authors note that such differences could be attributable to either real phenomena, such as differing labor market structure, or measurement errors. The reasons need to be further explored. Ehrenberg would also
ASSESSING THE ISSUES like to see a more thorough analysis of the relationship between changes in the relative wage of women and changes in employment and unemployment levels. One of the more intriguing implications of the paper by Gregory and his colleagues, as Ehrenberg notes, is that it may be easier to raise the relative wage of women in a country where wages are centrally set and where there has been explicit discrimina- tion. In the United States, where the labor market is highly decentralized and where discrimination in wage setting is unlikely to take such an overt form, the circum- stances may prove more difficult to change. CONCLUSION No single paper or volume can resolve major social issues like the one addressed here. The papers collected in this volume contribute to a better understanding of sev- eral dimensions of wage differentials and the comparable worth remedy. First, they substantiate differences in wages between women and men, even after measurable productivity-relatec] variables are taken into account. Second, they explore the role of occupation in the wage determination pro- cess, investigating the particular role played by the female dominance (percent femaTe) of an occupation. Third, they examine em- pirically the results of implementing com- parable worth or comparable worth type policies in several real world situations. In none of these areas are tong-standing de- bates resolved, but the papers do contribute to consensus on several important issues. Research Consensus The papers substantiate the fact that wom- en earn less than men after adjusting for measurable factors that might affect labor productivity. Several of the papers focus on estimating the components of these wage differences (Sorensen and Gerhart and Mil- kovich at the indiviclual level, and Parcel 17 and Filer at the level of average occupational wages). Others investigate mechanisms by which that fact comes about (Nakamura and Nakamura, Subich and colleagues, Baron and Newman, and Gerhart and Milkovich). None of the studies disputes the existence of a difference in wages for men and women, although Filer contends it is not related to the female dominance of a given occupation. The role of percentage female is not re- solved, though consensus has emerged on the proper way to assess its effects. A re- lationship between percentage female ant] average wages of an occupation (the weight- ed average of the male ant] female wages) could simply reflect a compositional effect of more or fewer women if women are paid less than men in each occupation. To iden- tify an effect on wages of the female (lom- inance of an occupation per se, all other things being equal, the wages of women and men must be examined separately (or normative wages rather than actual wages can be used, as in the Baron and Newman study). Except for Filer (using 1980 census data) and Gerhart and Milkovich (in a single firm), the studies reported here do fine] a significant net effect on wages of percent female in an occupation, when other factors, such as productivity differences and job requirements are taken into account. The Baron and Newman study of listed starting salaries of jobs in the California civil service provi(les perhaps the most dramatic results: When women or minorities enter occupa- tions the starting salaries fall, everything else, including job requirements, being equal. Such a finding suggests that jobs may be devalued by employers when women and minorities do them supporting a premise that lies behind the comparable worth rem- ecly. Alternative explanations, however, are also possible for example, that wages fall in response to changed conditions and then women and minorities take jobs that white men no longer fief] attractive. Consensus also emerged on the effects of comparable worth policies. The dramatic
18 turnabout in the nationally administered or regulated wage setting environments in Britain and, especially, in Australia have had little negative impact while moving the wage structure dramatically closer to gender equality. In Australia, the actual wage ratio increased from 59 to 74 percent and in Britain from 60 to 71 percent between 1964 ant] 1979; comparable worth type policies in the two countries eliminated 37 and 28 percent of the wage gap, respectively. In the United States, Sorensen estimated the maximum proportion of the national wage gap that couIcT be eliminated by comparable worth at 20 percent, and in the two actual cases reported here (Iowa ancl Minnesota)- both, not surprisingly, involving plans that resulted from political compromise, the re- ductions amounted to 18 percent and 15 percent of the respective wage gaps. Al- though the size of these effects suggests that comparable worth policy is not as rev- olutionary as some might have hoped, it nevertheless amounts to a substantial im- provement for women workers, without ap- parently causing negative side effects. Though the outcomes of comparable worth policies have varied according to the locale, positive effects and minimal negative side effects have generally been reported in the three cases presented here. The three papers on comparable worth implementa- tion taken together attest to the significant impact public policy can have on wages. The impact was large in Australia, where labor market institutions are centralized, and smaller in the United States, where labor markets are far more decentralized. Research Needs While there is consensus on some issues, many questions remain unanswered. The studies reported here suggest several new directions for research. Nakamura and Nak- amura's investigation of the crowding pro- cess finds that in states where there are more young women, relative to others, their PAY EQUITY: EMPIRICAL INQUIRIES wages are lower and that women with fewer years of eclucation and more children are more affected by crowding than others. Their study could be replicated for occupations (rather than states) to see which occupations are more susceptible to crowding. The paper by Subich and her colleagues reports pilot studies of students that investigate whether their attitudes ant! expectations might con- tribute to lower earnings for women. The pilot studies suggest that risk-taking be- havior of employees in firms, where salary history data exist, might be a fruitful area for further research. Several of the studies report evidence of improvement in the relative position of women in the past few years: Gerhart and Milkovich, in their single-firm analysis, find that in recent years women have received more salary increases and promotions than men. Baron ancl Newman, despite their generally negative finclings, do find that starting salaries in new jobs are less affected by gentler and race/ethnicity bias than are starting salaries in older jobs. Further in- vestigation of the extent of change and the reasons for the change wouic! be useful. The introduction of comparable worth legislation in the states of Iowa and Min- nesota has improved the relative economic position of women civil service workers without having ha(l major adverse effects on the state budgets or having engenclerec! political tensions, as reporter] by Orazem and Mattila and by Evans an(l Nelson. In neither state, however, was the relative wage of women to men raised by more than 8 percentage points (in Iowa, the relative wage went from 78 percent to 82 percent and in Minnesota from 74 percent to 82 percent). Neither of these papers addresses effects beyond the civil service labor market within each state. It would be of interest to know if any effects are felt by private employers or local governments, and wheth- er they are positive or negative. Further research on the mechanisms through which the earnings of women and
ASSESSING THE ISSUES men are made to differ also seems warranted. The papers in this volume provide evidence of the salience of gender in the labor market, both in terms of wage differences and sex segregation. Many, but not all, of the papers find that the proportion female of an occu- pation Towers its wages. Several of the Caners also find that percent female has a negative effect on wages for men, but not for women, within an occupation. The interpretation and policy implications that follow from these findings deserve more attention. Whether women choose female-dominated lobs. ner- haps because there are compensating non- wage differentials or because women's pref- erences differ, on average, from men's; whether they are tracked into them; whether women are discriminated against whatever their choices; whether men are discriminated against within female occupations; or whether other (as yet unmeasured) factors are important, we still do not know. Several of the papers suggest research (Erections that may be especially promising. Ffler's results suggest that the more sig- nificant portion of discrimination may occur within occupations rather than between them. This in turn suggests that differences be- tween firms or industries in their "treat- ment" of occupations might be important and that the practices of individual em- ployers should be examined further. Ger- hart and Milkovich's finding that job as- signment "explains" sex differences in wages suggests that the process of job assignment within the firm should be examined. What motivates individuals, both employers ant] employees, in job assignment, pay setting, job selection, and wage acceptance is sug- gested as a useful area of study by several of the papers, especially the one by Subich and colleagues. Further historical and in- stitutional studies of how things "came to be" are also warranted. The papers taken 19 together also suggest that further research on measurement issues is important, in- cluding research on the variables that belong in the list of legitimate contributors to ex- plaining the wage gap. As we suggested above, further research on the general equi- librium consequences of comparable worth implementation is also warranted. Both the potential spill-over effects from one sector of the economy to another and the general influence on the labor market have not been adequately explored. There are numerous labor market ant] comparable worth studies that will be use- ful in answering the basic questions "why are women pair] less than men" and "what should be done about it, " but we The editors of this volume, although not necessarily the members of the full panel] suggest that an acI(litional fruitful line of inquiry in the near future may be investigation of the relation- ship between gender ant] social behavior more generally. How do the social expec- ~ons of men and women generally in and out of the labor force affect their earnings and opportunities? How are female earnings and the (listribution of family in- come relate(l? Differences in the roles of women and men in regard to the important social responsibility of raising children may have significant labor market implications. The expectations for the genders in the conduct of familial and household duties and in political, religious, and sexual be- havior are examples of some of the areas that need to be better stuclied for their impact on labor market outcomes. The economic realm ant] labor earnings in particular do not exist in isolation from other aspects of the genclerec! division of social life more generally. Both research and policy intervention will be more suc- cessful if they are pursued in this broader context. . . .
