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.
2 Evidence Regarding Wage Differentials THE EXISTENCE OF WAGE DIFFERENTIALS In the United States today, women earn less than men on the average and minorities earn less than nonminonties on the average. In 1978 women of all races who worked full time all year earned SS percent as much as white men, and black men earned 72 percent as much as white men Arable 1~. Moreover, although schooling has been consistently found to be closer, correlated with earnings, at every level of schooling women and black men have lower earnings than white men (Table 2~.- For example, black men with some college education have lower mean earnings than white men who are high school graduates and only slightly higher mean earnings than white men who have not graduated from high school; and both black and white women who are college graduates have lower mean earnings than white men with eighth-grade educations.' The difference in income between white men and '`black and other" men who work full time all year has tended to decline over the past two decades (see Table 3~. Between 1955 and 1975, for example, about 40 percent of the difference was eliminated. We do note, however, that ' These data are based on reports of what people with these characteristics say they actually earn. Full-time x ear-round work is defined as 35 or more hours per week. SO or more wocks per year. It is possible that within this category. if the amount of time actually worked by men and women were taken into account' the earnings ratio presented in Table 2 would be slightly different (see note 8~. 13
14 WOMEN. WORK. AND WAGES TABLE 1 Mean Earnings of Year-Round Full-Time Civilian Workers 18 Years Old and Over, 1978 Percentage of Earnings of White then Women Men Women All races S17~547 S9,939 97.7 55.3 White 17,959 9,992 100.0 5S.6 Black 12.898 9,388 71.8 52.3 Spanish origins 13,002 8~654 72.4 48.2 Persons of Spanish origin may be of any race. SOURCE: IS. Bureau of the Ccusus. l9~:Table 57. the gap between minority family income and nonminonty family income, after a period of decline during the 1960s, has remained constant or increased. These divergent trends reflect the effects of many underlying factors, of which the most important are the growing unemployment and declining labor force participation of minority men and the growing proportion among minority families of single-parent families headed by women. They highlight the dangers, especially for minorities, of as- sessing general progress by referring only to full-time year-round work- ers.2 By contrast, the difference in income between women and white men who work full time all year has failed to show any decline (see Table 3~. The aggregate pattern is a combination of the somewhat dissimilar experiences of white women and black and other women. During the late 1950s and early 1960s the income disparity between white women and white men was growing, and over the last 20 years this disparity has remained essentially unchanged. Over the same penod much of the racial component of the difference in income between black and other women and white men was eliminated, a fact that has reduced the overall gap but left these women in approximately the same position as white women. Since the mid-1970s, when black and other women achieved virtual panty with white women, the income dispanty vis-a-vis white men has not declined further. Charged with assessing methods for determining job comparability, our focus is properly on the earnings inequalities among workers who 2 Lucre is a substantial literature on differences in earnings by race. which discusses the measurement of trends and suggests that the process resulting in race differentials is somewhat different from that resulting in sex diffcrcotials (see, for example, Freeman, 1973; Welch, 1973; Smith and Welch, 19J7; "d Reich, 1981).
Is x of - x ~'. he ~ ~.- ~ is' =~ c . - ~ c Lo c 3 Cal - c Ct c U: ._ E sit I: ~3 ~ C ~ 3 _ ~ C 3 3 ~ C,) ~ E ~ ^ :e 3 c-, E ._ 3 ~ ~ C,) Cal . He LLi 3 a - o ~ ~ o C TIC ~ . . . . . . .. C ~ ~0` e' U) ~ ~- . C ~ ~ ~to of ~to \o if. ~ ~ - , - d ~- o oK oc ~ ~ ~ x ~ ~ ~ ~oo ' ~ -d v^d ~-d - do~ - - - ~ ~ ~ `& - d q~ \O - ~ x ~ oc ~o - ~oK - - - -d - -) - ~ ~#= c o ~ - - ~ ~o' - # oc ~ oc ~ ~ ok ~ o ~- ~ oc oK - - - - -- ~ ~ ~ ~ ~o~ - d - ] - - - - - ~ ~- e 8 ~ ~ ° E ~ t~ ° ~ ~ o ~ _ ~ . - ~ o ~ ~ ~S ~ _ . . o~ _ s o j
6 WOMEN. WORK. AND WAGES TABLE 3 Median Income of Year-Round Full-Time Workers by Sex and Race, 195~197B° . Percentage of Income of White Len Median Income Blaclc Black of and and White White Other Black Other Black Men Women Women Women Men Men 195~1959S 4.874 63.2 36.4 n.a.. 60.6 n.a.. 19019646,017 59.5 38.8 n.a.h 63.9 n.a.6 196~19697.697 57.8 42.8 n.a.6 6S.8 n.a.6 1970~197410.893 57.1 S0.4 49.3 70.7 68.3 197~197814.811 58.6 SS.8 55.0 75.3 72.9 _ . . . . Table refers to income sines earnings data are not available; income cats are not available separately for blacks prior to 1967. b Not available. SOURCE: Computed from V.S. Bureau of the Census. Current Population Reports Cries P-60, Numbers 60. 66. 75. 80. 85. 90. 97. 99. 107. 116. 120. 123. 125. actually hold jobs. We have further chosen to concentrate primarily on analyzing inequalities between male job-holders and female job-holders, for three reasons. First, as we note above, for full-time year-round workers the difference in earnings between men and women is greater than that between minorities and nonminor~ties, and the difference in earnings between men and women has not declined while the difference in earnings between minorities and nonminorities has declined. Second, as we note below, the extent of occupational segregation the degree to which different groups hold different rather than similar jobbers greater by sex than by race, and this segregation is the very situation that evokes an interest in methods for determining the comparable worth of dissimilar jobs. Third, most of the available research on comparable worth considers sex differentials. Nonetheless, despite the apparently greater immediate relevance of the comparable worth issue to women than to minorities, our analysis is applicable whenever substantial job segregation between different groups exists and whenever particular jobs are dominated by particular groups. How can one account for the difference in earnings between men and women? Two kinds of explanations have been proposed: those that focus on the characteristics of workers and those that focus on the characteristics of jobs. Of the first Icind are studies that attempt to relate pay differences between the races and the sexes to differences that are believed to affect productivity, such as training and experience. Of the
Evidence Regarding Wage Differentials 17 second kind are studies that explicitly recognize that earnings differ among jobs and focus on the substantial segregation of the labor force into different jobs on the basis of sex and race as a major explanation of the differences in earnings. ibis chapter reviews evidence regarding the sources of earnings dif- ferentials between men and women with only occasional reference, for the reasons given above, to racial and ethnic earnings differentials. We begin with studies that focus on the charactenstics of workers, then consider those that focus on the characteristics of jobs, reviewing the pnnc~pal findings of the relevant bodies of research. Because of the vast literature involved, our survey is selective rather than comprehensive. THE EFFECT OF WORKER CHARACTERISTICS ON DIFFERENCES IN EARNINGS In recent years a substantial amount of research has been done by economists and sociologists attempting to explain differences in earnings between men and women on the basis of differences in their personal charactenstics. Most of this research has been based implicitly or ex- plicitly on a "human capital" approach. The human capital approach denves from the neoclassical economic theory of wages, which treats wages, the price of labor, like all other prices and posits that, in the absence of discrimination, equilibrium wages wall be just equal to the marginal revenue product of labor. In noneconomic terms this means that in the absence of discrimination workers wall be paid an amount exactly equal to the value of their economic contribution to a firm. Hence, according to human capital theo~, it should be possible in principle to measure directly the extent of inequalities in earnings due to discrimination by comparing wage differences between men and women with differences in their economic contnbution, or "productiv- ~ty"; wage differences not accounted for by differences In productivity could then presumably be ascnbed to discrimination. There are a number of difficulties in such calculations. One difficulty Is that wages may not reflect the entire reward paid for a job. Another, more difficult problem is that, with the exception of a few jobs involving the production of physical goods (e.g., coal mining, button sewing), for which the amount produced is easily measured (and for which workers are often paid on a piecework basis), differences in productivity among jobs are virtually impossible to measure. When attempts have been made to measure productivity directly (see, for example, Materiel and Malkiel, 1973), the researchers tbemsclves usually acknowic~ge the un- ~atisfactory nature of the exercise. To get around this problem, rc
18 WOMEN. WORK. AND WAGES searchers using the human capital approach have attempted to estimate productivity indirectly by assuming that differences in productivity among workers derive from differences in their stock of "human capi- tal," that is, their education and training, work expenence, continuity of work history, effort or commitment, health, etc., all of which can be measured more or less directly (Schultz, 1961; Mincer, 1970; Becker, 197s).3 The basic procedure in human capital studies of earnings differences between men and women is to estimate what their average earnings would be if men and women received an equal return on their human capital and the only differences in their earnings were those due to differences in the amount of their human capital, which are considered to be proxies for differences in productivity. Such estimates are then used to decompose the total difference in average earnings into that part due to differences in human capital and, presumably, productiv- ity-and that part due to differences in the rate of return on investments in human capita~ohen assumed to represent discrimination. (A more detailed discussion of these procedures is presented in the technical note at the end of this chapter.) Some problematic features of this approach should be noted before reviewing the evidence based on it. First, the marginal productivity theory of wages is not universally accepted. Many argue that factors other than productivity, such as custom. union strength' and the eco- nomic friability of an industry or enterpnse, affect wages (e.g., Bibb and Form, 1977; Phelps-Brown, 1977; Piore, 19791. Second, human capital variables are of unknown quality as proxies for productivity differences. "Experience," for example, may reGect either productivity-enhancing, on-the-job learning or simply seniority: in observing a correlation be- tween wages and experience, one does not know to what extent it is higher productivity or greater seniority that is associated with higher pa)~.4 Third, the link between concepts and indicators is often quite 3 Various indicators of labor force commitment are often included in earnings equations because the degree of commitment could affect the quality of work effort and hence productivity. For example. it is sometimes argued that marriage reduces the productivity of women but increases that of men because married women adjust their work lives to accommodate their family obligations, while married men arc motivated by their family responsibilities to increase their earning power; marital status is therefore often included as an explanatory variable. 4 The idea that wages rise with cxpericncc because workers arc making on-the-job investments in productivity~nhancing human capital is not universally accepted. Some (e.g., Edwards, 1977) argue that cxpcnencc should be interpreted as a proxy for senionty rather than for productivity: that is. mirages rise with c~pcnencc as a result of seniority or tenure provisions that rcflcet normative c~spectations that older workers or long-time cmployocs should earn more regardless of their level of productivity.
Evidence Regarding Cage Differentials 19 tenuous. For example' even those who accept the idea that education enhances productivity do not necessarily accept "years of school com- pleted" as a good indicator of the quality and extent of job-specific skills learned in school. Finally, to interpret as discrimination all earnings differences between groups that are not accounted for by the variables explicitly studied (that is, the "residual difference") requires two very strong assumptions. The first assumption is that all relevant factors are measured: 'all relevant factors" includes all factors that underlie dif- ferences in productivity and that are distinct from all other factors.S The second assumption is that all factors are measured without error.6 In practice, however, these assumptions are virtually never completely sat- isfied; at best, there is no way of being certain to what extent they are satisfied. Hence, there is always a degree of doubt as to the validity of the empirical findings regarding discrimination. Let us now turn to the empirical literature.' Table 4 summarizes the findings from several studies based on data from national samples of the working population that attempt to account for differences in earn- tngs between men and women on the basis of the characteristics of workers. In most of the studies, worker characteristics account for very little of the difference in earnings; in fact, only two of the studies can explain more than one-fifth of the difference between men's and women's average earnings in terms of differences in worker character- istics. The two studies whose findings go furthest toward explaining the observed earnings gap, those by Mincer and Polachek (1974) and by Corcoran and Duncan (19~9), account for less than half the difference. The relative success of these two studies in explaining the earnings difference can be attributed largely to their use of a measure of actual labor market experience that is more complete than those usually avail- able. Me study by Corcoran and Duncan (1979) is perhaps the most thor S This assumption invites consideration in each specific case of what additional factors might create differences in productivity between women and men and thereby reduce the unexplained pan of the difference in earnings. It must be understood, however, that in order for such additional factors tO contribute to an explanation of the difference, they must be correlated with sex. correlated with earnings. and relatively uncorrelated with factors already in the equation; otherwise they will add little or nothing to the explanation of the difference in earnings that results from this type of statistical analysis. (Of course, omitting variables that are correlated with variables included in the prediction equation biases the coefficients of the included variables but does not particularly affect the prc- dictions, which arc the focus of interest here.) ~ Lee term "measurement error" here encompasses issues of reliability. validity, func- a;onal form. and error structure- all of which can acatc seriously misleading results with the regression procedures that are usually used in these studies. ~ Many of these studies have been summarized previously by Kohen (1975~.
20 o lo Cat ._ CC at: - to of a: · _ ._ Cal C) _e ._ x C) Cal - 3 8 6 ._ ~3 - C~ to O In .O - cn _ ·_ LL e' _ cell ~ V To Cal ~ lo.' a, ~ ~ ILL ~ _ K ~ 0 as ·a ~ es ~ . - ~ ~ E ~ ~ o ~ _ o o ~ ~ ~o ~ ~ ~ _ to =t O tic ~or 0 ~ ~ to Be G~ ~ ~- TO ~) - \0 ~ r- ID ~ - 1 1- ~ ~ ~TIC ~ ~ G · t_ ~ C ;^ . _ . ' O ~ ~_ ~ ~_ _ C~ 4~, ~ , ~, ~ ~ ~, , ·~ K .S ~ - - - - - - - - - ._ ., V) C ~ ~ Ct ~n eo ~L ~ ~ . . C,2 . . ~ · _ C C) ~e ~: o - - C ~OL :, ~e ·- U, ~: ~3 00 ~.' I: , _ U,o E ~ ~ o~ , eo - (V- ~r - o~ o o' ~, - _ _ . . V, Cd ~ o U~ _ _ _ .' .' .' _ _ _ ~, e ~ ~ ~ e ee ee . · ~ · _ 3 o e - U' ~ ~s ;^ o _ _ 3 0` 0\ ~_ _ _ ._ - . . 00 .c ,e :~` ~0 ° E c ~ ~ 2 ~ ~ a a~ E a it E ~ ~ ~ ~ ~ E ~, ,~ ~ ~ E ~ ~ .~c a Z ~ .~o,, ~ ~ ~ ~ [5 z ~ e ~ ~ ~y -- ~ . ~ ~ S~
21 et E A: a ._ - V. - et C) a. 11 .m in, - c ._ in, TIC C) - E - - ~C C) E - = ^ ._ _ ~, ~ C~J, ~ K 3 i .C ~ ~ Cm ~ en, v V ~ ~ ~ Ct ~ ~ ~ ~ ~ S S ~ .X - E C ~ ~ ~' ~ ~C ~ ~ o~ c _ I,) ~ ~_ _ o ~s c ~: -o ~_ ~o .._ os._ - ~'~ ' 8 so .2 ~ io C_ E ~ ~ ~ o ._ oE =: 3_ oo C °a, ~C~ K . ~ E ~ ~ 3 - ~- ~ · . 'o E s 2 · ~ i~ '~ u u t: o ~E OD ~ c ~ .=, .° ~ ~U, a ,u Y 5, i ~ ~ j, - } ~ O ~ ~ _ - ~ C ~ ~ ~ O - o - es VJ - ._ - ·_ - S 3o co = 3 v - v, ._ - s o ._ oo ._ v, - c~ c~ - _ g~ O ~ _ .C) ' c ~ ._ o f: ~ t._ ~ 3 z 11 11 11 o ~ ~
22 WOMEN. WORK. AND WAGES ough of this genre of studies' including detailed measures of educational attainment, work history, on-the-job training, and attachment to the labor force for a national sample of household heads and their spouses who were in the labor force in 1975.8 This study explicitly excluded occupation as an explanatory variable in order to provide a pure test of the human capital approach. According to human capital theory, individuals invest in human capital as long as they expect future returns to compensate them for foregone earnings and other costs of acquiring human capital. With perfect opportunity for mobility and perfect avail- ability of information' people should seek the highest return for their human capital in the labor market; thus, over the long run, market competition should equalize the returns on human capital across oc- cupations and industries. Differences in returns on human capital for those employed in different occuDations~must then brine to the - -a - - ~ ~ r ~ ·~ ~ God-111~ V ~11~ ~ . . _ theory, be taken as reflecting past or present market imperfections including institutional bamers and discriminatory practices. Table 5 shows Corcoran and Duncan's decomposition of the difference in earnings between white men and white women. According to their formulation, a little less than half (44 percent) of the difference in the mean earnings of men and women can be attributed to differences in education, work experience, and labor force attachment, and virtually all of this is due to differences in work history- women have less overall work experience, less previous experience with the current employer, and less on-the-job training. Interestingly, indicators of labor force at- tachment account for almost none of the gap, a finding that suggests to Corcoran and Duncan that although women may stay out of the labor force because of family obligations, losing valuable years of experience, when they do work their "dual obligation" (to the family as well as to the job) does not negatively affect their earnings potential (see also Corcoran, 1979~. Exactly how experience affects earnings and whether experience has the same consequences for men and for women are questions currently subject to considerable debate (see, e.g., Mincer and Polachek, 1974; Edwards, 1977; also see below). We have already noted alternative interpretations of the relationship between experience, productivity, and ~ In Corcoran and Duncan s sample. the hourly wages of women avcra~cd about 75 percent of those of men, a figure somewhat higher than the often 60 percent ratio in annual earnings of full-time year-round workers. It is unclear what accounts for this disparity, although one possibility is that Corcoran and Duncan's data are controlled for the number of hours that ' full-time" workers actually work. Women may work fewer hours than men, a possibility that. if not controlled. would exaggerate the size of the differcoce in their earnings. Another possibility is that differences in the nature of the samples (the Corcoran and Duncan data are from a sample of families rather than indi- viduals) create diffcrcoces in the earnings ratios.
Evidence Regarding Wage Differentials TABLE 5 Decomposition of the Wage Differential Between Employed White Men and White Women 23 Explanatory Variables Work History Percentage of 1975 Hourly Wage Gap Explained Ycars out of labor force since completing school Years of work experience before current employer (plus square) Ycars with current employer prior to current position Ycars of training completed on current job Ycars of post-training tenure on current job Proportion of total working years that were full time Indications of labor force attachment Hours of work missed due to illness Hours of work missed due to illness of others Placed limits on job hours or location Plans to stop work for reasons other than training Formal education (years of school completed) PERCENTAGE EXPLAT~'ED PERCENTAGE L'N'EXPLAINED TOTAL 6 3 2 11 3 2 44 56 100 -1 8 o -1 2 These percentages are derive ed b! Corcoran and Duncan for each variable by the formula c`= (AZ`- 6,,~m)/(lnU',,,,,- lnU',,~), where AZ, Is the difference In means between white men and white women on variable i, 0,,,,,, is the regression coefficient of variable i for white men, InW,,.m is the natural log of mean hourly gages for white men. and lnW,,~ is the same for white women. SOURCE: Adapted from Corcoran and Duncan. 1979. earnings. Here we add the observation that returns on experience are generally lower for women than they are for men, which accounts for part of the earnings differential; that is, the earnings differential is not due simply to the lesser experience of women. But the lesser experience of women does need to be explained. The conventional interpretation Is that women voluntarily limit their labor force experience because of the demands of their family, responsibilities. Some, however, would argue that the difference in labor force experience between men and women, particularly in the kind of experience that may be most relevant to earnings (on-the-job training), may itself renect discriminatory re- striction of occupational opportunities. For example, employers may be reluctant to hire or to train women because they assume that women will leave the labor force to bear or raise children (Sawhill, 1973~. Or the difference in the amount of on-the-job training between women and
24 WOMEN. WORK. AND WAGES men may be largely the result of institutional practices that tend to exclude women (Duncan and Hoffman, 1978~. To the extent that dif- ferences between men and women in the characteristics thought to affect income themselves result from discriminatory processes in employment and training, empirical estimates of the extent of discrimination will be too low. This downward bias may somewhat offset the tendency of such estimates to be too high due to the possible omission of additional ieg~timate determinants of earnings from estimation equations. In these studies, researchers have consistently found that 8 substantial part of the earnings difference cannot be explained by factors thought to measure productivity differences. Taken at face value, these results create a presumption of additional factors at work, possibly including institutional barriers and discrimination. Nonetheless, because of the many difficulties inherent in the human capital approach (discussed above), because the consistency of results from these studies may resect identical flaws in the research, and because the findings concerning discrimination are so indirect (other factors failing to explain fully the difference in earnings rather than discrimination being shown to explain directly the remaining difference), the committee concludes that these studies are suggestive rather than definitive. THE EFFECT OF JOB CHARACTERISTICS ON DIFFERENCES 11 EARNINGS It is not surprising that explanations focusing on the characteristics of individual workers leave a substantial portion of the earnings gap unexplained, since occupational differences in earnings are very large and the labor force is substantially segregated by race and by sex. A second category of studies that attempts to explain the earnings differ- entiaIs between men and women focuses on the characteristics of the jobs they hold. The central fact is that men and women tend to hold different types of jobs. We should note here that we often refer to job segregation rather than occupational segregation. As we argue below, even within finely defined occupations (e.g., lawyer, sales clerk, hair- dresser), jobs are frequently segregated by sex. Unfortunately, available data frequently permit comparisons only between occupations, but this empirical limitation should not be allowed to obscure the conceptual dist~ctions.9 ~ Jobs are specific positions within establishments or the economic activities of specific individuals. fancy cattail particular duties and responsibilities and involve the performance of particular tasks in particular settings. Examples of jobs include Headwaiter at Lion D'Or in Washington, D.C.~" 'wielder on Assembly Line 3 at the Ford Motors assembly
Evidence Regarding Wage Differentials 2S Table 6 shows the distribution of workers across broad occupational categories by sex and by race. Women are substantially more likely than men to work in clerical and service occupations and less likely than men to work in craft and laboring occupations. Black and other men and all women are less likely to hold managerial jobs than are white men. Black and other men. who are more likely to work in blue-collar occupations than are white men. are less likely to work as craft workers and more likely to hold jobs in the operative and laborer categories. Blacks and others of both sexes, especially women, are more likely to work in service occupations than are whites. Segregation indices calculated for the distributions across these broad occupational groups provide one way of measuring the extent of oc- cupational segregation. An index of segregation between two groups can be interpreted as the minimum proportion of one group that would have to be shifted for its occupational d~stnbution to be identical to that of the other.~° For example, in 1970, 44 percent of white women would have had to shift their occupational category for the distribution of white women across broad occupational groups to be identical to that of white men. Occupational segregation by sex has barely decreased at all among whites over the past several decades; it has decreased substantially among minorities. but it still remains high. Table 7 presents these indices as calculated by Treiman and Terrell (1975b:167~. The large change in plant in Los Angeles.'' Seller of }eather goods at a street stall on Fifth Avenue in New York." and "cardiologist in private practice in St. Louis.'' Occupations arc aggregations of jobs, grouped on the basis of their similarity in content that is. similarity in the tasks. duties. and responsibilities they entail and the conditions under which they arc performed. Such aggrc~ations ma; be more or less gross. depending on their purpose. '6 The segregation index. or index of dissimilarity. A. is given as where x, is the percenia'ge of one population (c.g., men) in the ith category of a classi- fication. and', is the percentage of the other population (c,g., women) in the ith category (Duncan and Duncan. 1955). ~ is then the percentage of either population that would have to shift categories to make its distribution exactly equal to that of the other popu" ration. Obviously. we would never expect one group to shift completely and the other group not at all (e.g.. for women to shift occupations to achieve a distribution exactly like the current distribution of men), since this would result in a major shift in the distribution of the total population over all categories (in this Cask. a shift in the distribution of the total labor force over occupational groups). It is more plausible to imagine the kinds of shifts that would be required by both groups tO create identical distributions without changing the distribution of the total. It can be shown, however, that the sum of the proportions in the two groups required to shift categories to achieve identical distributions in this latter instance is exactly equal to 6. Hcscc ~ is an awr~natc arc of the dissimilarity. or segregation. of the two groups.
26 y o _. K ~y e E D~ ~me E V A - lo o o .0 - D ._ - U, ._ - to e_ Cal ~0 CD ~ O ~ ~ 0\ ~ ~ ~-] ~] He ~) ~) ON OC) Ox f-) _ ~co ~ ~ ~ O ~ ~ ~ _ 0\ m) 00 OC ~ ~_ _ c, E 0 ~ ~ o ~ ~ ~ on to 0 ~ lo' 0 ~ ~ _ - , _ ~t_ ~ ~ ~ 0 o ~ ~ 1 _ o o ~ ~ to ~ ~ 0 to ~ - ~o ~ ~ ~ ~ ~ Do _ ~ _ _ to ~- _ ~ - ~ ~_ U) ~ 0 ~ o~ _ o~ ~ ~ ~ _ o - ~ ~ ~ ~ ~ ~ - C - ~ ~ ~ _ eY ~e ~V m O ~ ~ ~e Ze ~ 0U 3 ~ o.§ ~ r~ - _ U) o~ U) ·o ~ ~} ~ U-) t_ o~ _ l_ ~ ~o ~o o ~ ~ ~ `~, ~ ~ ~ _ _ _ ~ ~c ~ ~ 0 '> 0 ~ ~ ~ ~ ~ e ~ ~ ~_ ~ `0 ~ _ ~ ~ et_ | t_ ~· ~ - C K i i e i ~i e Iiti a7eiii,I'~.~e~ - _ z _e D ep - .0 _ ._ _ V, _ O 3 ~e ; -O E ~le · .
Evidence Regarding Wage Differentials TABLE 7 Occupational Segregation Indices 1940 1970 27 1940 1g50 1960 1970 Occupational segregation b) sex among: Whites 0.46 0.43 0.44 0.44 Blacks and others 0.58 0.50 0.52 0.49 Occupational segregation by race among: Men 0.43 0.36 0.35 0.30 Women 0.62 0.52 0.45 0.30 Notc: Indices are calculated for occupational distributions across 11 major census cate- goncs. The data from 1940 to 1960 are classified according to the 1940 census detailed occupational classification; the 1970 data are classified according to the 1960 census de- tailed occupational classification. SOURCE: Treiman and Terrell 1975b:167. Copyrights 1975 by Russcll Sage Foundation. Rcpunted by permission of the publishers Russcll Sage Foundation. the occupational distribution of black and other women occurred as they were able to enter clerical and sales jobs and service jobs other than in private households. In 1940, 75 percent of all employed black and other women worked as domestic servants or farm laborers (Treiman and Terrell. 1975b:160~; by 1979 only 7 percent worked in domestic service (Table 6~. Since broad occupational groups are made up by aggregating smaller detailed occupations, each of which may be dominated by a particular group, segregation indices that are calculated using detailed occupations are usually larger than those based on broad occupational categories (Iogically they cannot be smaller). Within the clerical category, for ex- ample, mail carriers are mainly (92 percent) men while stenographers are mainly (93 percent) women. Similarly, among craft workers, con- struction trade workers are virtually entirely (98 percent) men, while a majority of bookbinders, decorators, and window dressers are women (U.S. BUT 2.U of the Census, 1973: Table 1~. Me U.S. Commission on Civil Rigt:~- has calculated segregation indices across approximately 400 detailed census occupational categories. In 1976, the index of occupa- tional segregation between white men and white women was 66.] and between black women and white men was 69.3 (U.S. Commission on civil Rights, 1978:42~. Occupational segregation by race is also substantial, although it has declined considerably since 1940. Treiman and Terrell's csiculation (see Table 7), based on broad occupational categories, show Mat for 3970 the index of occupational segregation by race for both men and women
28 WOMEN. WORK, AND WAGES stood at 30. Me more detailed categorization used by the U.S. Com- mission on Civil Rights (for 1976) resulted in an index of occupational segregation between black and white men of 37.9 and between black and white women of 35.~. Regardless of the level of aggregation, oc- cupationa] segregation is more pronounced by sex than by race. There are 553 occupations with wage and salary earners included in the most disaggregated level of the 1970 U.S. census classification (U.S. Bureau of the Census, 1973:Table 24~: 310 of them (more than half) have at least 80 percent male incumbents' and another 50 (9 percent) have at least 80 percent female incumbents. Moreover, 70 percent of the men and 54 percent of the women in the labor force are concentrated in occupations dominated by their own sex." Not only do women do different work than men, but also the work women do is paid less, and the more an occupation is dominated by women the less it pays (Sommers, 1974~. For the 499 wage and salary occupations included in the 1970 census expanded occupational classi- fication (with values on all relevant vanables), the solid line in Figure 1 shows the relationship between the percentage female and the median wage and salary earnings for job incumbents of both sexes. For these data, each additional percent female in an occupation results in an average of about S42 less in annual income: overall, "women's work" pays about S4,000 less per year on the average than "men's work."'3 IT It is interesting to note that occupations filled mainly (80 percent or more) by men average many fewer incumbents than occupations filled mainly (80 percent or more) by women: 1(~,363 and 312~144. respectively. in 1970. This difference is sometimes taken to indicate the crowding of women into relatively few occupations; it is probably also a reflection of the propensity of designers of occupational classifications (including the census classifications to make finer distinctions among the occupations filled mainly by men than among those filled mainly by women. For example. although secretarial jobs are probably about as varied as managerial jobs. the census classification includes 36 subgroups within the category "managers and administrators. not elsewhere classified" sad only two subgroups within the category "secrctancs.'* Differences in the degree of inclusiveness of occupational categories render somewhat problematic discussions of the degree of occupational segregation. Of course. successive disaggregation of occupational categories can only result in increased (or unchanged) estimates of the degree of segre- gation. t2 Ibc earnings figures ~ adjusted to mtr~ for di~ctences in the estimated number of be worked per year: a=~iz-A medi-carding ~ median annual earnings x [2,-/ (mean hours worked last Aced x ~ wocks Aced last year)]. Zinc constant 2,()80 (40 x S2) is an estimate of the he of labor contnbuted by full-dine year-round workers. |3 When average earnings arc estimated separately for men and women from the percent female in occupations, the resulting equations are }as, = 8,324 - 29.6(W) and i,= 5.761 - 16.3(W), respectively, fibers W is the permst female in each occupation. and Y., and t, "e Me estimated An - =d Age "d Clam c~n~ of men and women. ~e" equations can be intc~prcted as indicating that on the average each additional percent female of an occupation costs male workers about S30 in annual income and female
Evidence Regarding Wage Differentials Sl O.O~ ",000 ._ ~ S6,000 ",000 - ' S2,000 _ Y - 7767 - 27 so(w) Y - 8185-42.39~/~ 40 60 80 tOO 0 20 Percent Female FIGURE 1 Relationship between percent female and annualized median earn- in~s of incumbents for 499 1970 census occupational categories. Solid line is simple regression of mean earnings on percent female. Broken }ins is regression of mean earnings on percent female controlling for six human capital and job characteristics see text for details. 29 worl~crs about Sib: men doing women's work'' can expect to earn nearly S3.000 less per year on the average than men doing "men's work"; women doing "men's work" can expect to earn about S1. - more Per year on the average than women doing '~women's work.'' Because women earn less on the average than men. it is not surprising that occupations dominated by women pa! less on the average (the correlation between percent female and mean annualized earnings the solid line shown in Figure lapis -.45); nonetheless, it is theoretical!, possible for the entire difference in average pay to be accounted for by differing distributions of men and women between low-wage and high-wage occupations, with each occupation paying men and women equally. Such a result would indeed be consistent with the finding that the higher the percentage female, the lower the average earnings of both male and female workers. In fact, however. men tend on the average to earn substantially more than women in the same occupation. For Pupations tenth any Even maletlemale ratio' men can, a great deal more on the average than womcn~hc expected difference ranging from more than SI.200 for occupations dominated by women to about S2.400 for occupations dominated by men (or, to cypress the relation in tc''~.s of proportions' women's expected earnings fall from 77 percent to 69 percent of those of anon as occupations become increasingly dominated by men).
30 WOMEN. WORK. AND WAGES The relationship between the sex composition of occupations and the earnings of incumbents cannot easily be accounted for on the basis of either the personal characteristics of incumbents or the requirements of the jobs. In a study using the census data described above (Hartmann et al., 1980), the staff attempted to predict the median earnings of incumbents of each occupation from seven vanables: mean years of school completed; mean years of postschooling labor force experience;~4 four measures of job requirements denved from data in the Dictionary of Occupational Titles (substantive complexity, motor skills, physical demands, and unfavorable working conditions; see Miller et al., 1980, Appendix F); and percent female. The dotted line in Figure 1 shows the relationship between percent female and annualized median earn- ings, holding constant the other six vanables.tS From the substantial similarity of the solid and dotted lines in the figure it is evident that the differences among occupations with respect to these factors account for relatively little of the relationship between percent female and median earnings (although, of course, they do account for a large portion of the differences among occupations in average earnings).36 On the basis of these data it appears that the sex composition of occupations, inde '' Extent of previous work cxpcnence is, unfortunately. not measured direct!`' by the census. For men an adequate proxy can be computed in a straightforward we`: mean work experience is estimated by mean age minus mean years of school completed minus 6. on the assumption that on the average men start school when they are 6 and work every year subsequent to completing their schooling. For women. by contrast. such a proxy is inappropriate because large numbers of women leave the labor force after mar. rims and still larger numbers leave during childbearing years. To achieve a reasonable estimate of the average labor force experience of women in each occupation. an estimation procedure was employed that involves viewing the average labor force experience of women in each occupation as a function of age-specific labor force participation rates and the average age of women in the occupation. Scc Hartmann ct al. (1980) for details on the construction of this measure. |5 The equation corresponding to this line in the figure was denved by substituting the means of all independent variables concept percent female into the equation shown m note 16. '6 The full equation relating earnings to the other seven characteristics for the 499 census occupations is Y= -3341 + 681(E) + 75.2(X) + 217(F,) + 82.4(F2) - 9.2S(F3) + 23.~F4) - 27.5~, where Yis annualized median earnings of incumbents, Eis mean years of school completed by incumbents, X is mean years of labor force cxpenencc of incumbents, Fl is the com- ple~uty of the occupation, F2 is J measure of the motor skills required by the occupation, F3 is the physical demands required by the occupation, F4 is 8 measure of undesirable worldling conditions. and W is the percent female among incumbents. The R2 associated with the equation is .7S1.
Evidence Regarding wage Differentials 31 pendent of other occupational characteristics and of average personal characteristics' has a strong effect on the earnings of incumbents.~' To assess the role of occupational segregation in accounting for the differences in earnings between men and women, we reviewed a set of studies analogous to the human capital studies reported above but that include the characteristics of the jobs that workers hold as well as the characteristics of workers. Measures of job characteristics include a Ninety of scales for which scores are available for each category of the census three-dig~t occupational classification (e.g., prestige; Duncan's socioeconomic index; the median income of male Incumbents, an in- dicator of which occupations are intrinsically high-paying or low-paying; authority exercised on the job; and percent female) as well as cIassifi- cations of jobs by major industry group, class of worker, major occu- pational group, and detailed occupational group. Insofar as differences in the sex composition of jobs are associated with differences in pay rates' as tines appear to be, the more finely disaggregated an occupa- tional classification is. the larger is the proportion of the total difference In earnings that can be attributed to the segregation of men and women In the labor force. In studies that adjust only for differences between men and women across large occupational groups, the vanables used do not in general account for a large portion of the difference in earnings.'8 For example, ''The technical basis for this conclusion is the fact that the net regression coefficient associated with percent female in the equation reported in note 16 is statistically significant £ .01~. Birnbaum (1979) has argued. however. that simple reliance on a coefficient for gender (or gender composition) to measure disenmination may be misleading since - it is possible to generate statistically significant coefficients rciating gender and earnings even under a single-factor model in which gender. earnings. and "merit" (legitimate determinants of pa`; differences) are all determined by an underlying "quality" dimension but arc otherwise uncorrelated. This is known as underestimation bias (Madansky, 1959; Cook and Campbell. 1979). He proposes as a test for discrimination that both relations hold: 8,~,, A, ~ O and EMS ~ ~ O. where Y and W arc defined as in note 16, M is a measure of "meet (in this case the expected earnings predicted by the equation in note 16 with W omitted). and the ~ s are standardized net regression coefficients. Under this cntenon discrimination is clearly present. since pYW M = - .2~ and pow Y - .201; that is, holding constant diffcrcoces among occupations in the educational attainment and c~pcnencc of incumbents and in the four occupational characteristics. the higher the percent female the lower the average earnings; and holding constant average earnings. the higher the percent female the higher the overall level of the factors that predict earnings. a One exception is a study, by Oaxaca (1973~. We job charactensti" he "cd~road census occupational category' class of worker, industry. and union mcmbership~account for ~ substantial portion of the difference in earnings. Human capital variables accounted for 20 percent of the male-female diffcrcnec in comings among whites and 6 percent among blacks, and the job descriptor variables accounted for an additional 37 percent for whites and 39 percent for blacks. Tic important variables, however. ~ctc industry and union membership; broad occupational group and class of worker explained little of the difference in earnings.
32 WOI~EN. WORK, AND WAGES TABLE ~ Annualized Median Earnings° by Sex for Census Major Occupation Groups, 1970 Women's Earnings as a Pcr centage Census Major Group Men Women of Men's Professional, technical. and kindred S9.701 S7,731 ?9.7 workers Managers and administrators. except 9,496 5,514 58.1 fan Sales workers 7.684 3,185 41.4 acncal and kindred workers 7.298 4,805 65.8 Craft and kindred workers 7,840 4,676 59.6 Operatives, except transport 6.544 3,936 60.1 Transport equipment operatives 6.351 4.064 64.0 Laborers, except farm 5,025 3.556 70.8 Fanners and fann managers 4,663 2~442 52.4 Farm laborers and foremen 2,364 1,495 63.2 Service workers. except private 5,117 3.032 59.3 household Private household workers 2.170 1,525 70.3 TOTAL 7,394 4,~3 62.3 Average within-category earnings - - 63.4 ration Adjusted to correct for differences in the number of hours worked per year. Annualized median earnings = median cannings x [2080/(mesn hours worked last week x median weeks worked last yearly. The constant (=40 x 52) is an estimate of the hours of labor contnbuted by full-time year-round workers. ~ Weighted average of within-category ratios, with weights equal to total labor force in each category. SOURCE: U.S. Bureau of the Census. 1973:Table 24. in a study by Bluestone and his colleagues (1973), education and a classification of occupations into broad census groups accounted for only 6 percent of the difference. Fuchs (1971), using a number of human capital vanables and the distinction between private and government wage and salary-workers, accounted for IS percent of the difference. As Table 8 shows, one reason for the lack of explanatory findings is that the differences in earnings within major occupational groups are on average nearly as large as that for the labor force as a whole. Me differences In mean earnings within major occupational groups arise in large part from the fact that, within each category, men and women are
Evidence Regarding B'age Differentials 33 substantially segregated by specific occupation, the earnings levels of which are often quite different. Table 9 show s a decomposition of the earnings differences of men and women into a part due to occupational segregation and a part due to within-occupation pas differences for three successively more detailed classifications of 197() census data. The decomposition is camed out in two ways to cope with the index problem (Oaxaca, 1973~; the alternate estimates are shown in column (A) and (B) under each classification. Starting at the left. we note that adjusting for gender differences in distribution over 12 major occupational groups accounts for not more than about 10 percent of the difference in earnings. When 222 categories are used (the middle columns). between 10 and 20 percent of the dif- ference is accounted for by occupational segregation. And when 479 categories are used occupational segregation accounts for about 3500 percent of the difference.'9 This exercise illustrates that further analysis of occupational segregation requires much more detailed data than are currently available from the census or from national sample surveys. Studies that use the job characteristics assailable for the more detailed occupational classifications in the census and in some national samples (summarized In Table 10) account for more of the difference in earnings between men and women than do those studies using only broad oc- cupational categories. With one exception. these studies account for 30 percent or more of the difference in earnings. The exception, the study by Featherman and Hauser (1976), uses Duncan's socioeconomic index as the occupational descrip or. This index, like prestige scales, in the aggregate gives similar scores to mends and women's jobs and hence would not be expected to account for much of the difference in earnings between men and women.20 Sanborn (1964) accounts for 71 percent of the difference in earnings primarily because he uses a detailed occu- pational classification: it is the different distribution of men and women 'I He portion of the earnings difference between men and women due to earnings differences within occupations would be expected to continue to decrease as the occu- pational classification used became finer. It would be ideal to have data on the segregation of the labor force into jobs (. job meaning a collection of positions within a firm involving similar tasks and requiring similar skills: see note 9). ~ Although the use of prestige scales or Duncan s socioeconomic index does not usually account for much of the difference in earnings betwen men and women. both the Suter and Miller (1973) and the Treiman and Terrell (197Sa) studies. which use these occu- pational characteristics. account for about two-f~fths of the difference. This is almost certain!! due to their use of a variable representing actual labor market c%periencc. a variable that generally accounts for a sizable portion of the difference in earnings between men and women because they tend to have significantly different amounts of employment experience (see the discussion of the findings of Corcoran and Duncan, pp. 1~22~.
34 At: a ._ es c ~ o I: ._ ._ _ ~ es _ 3 ~ Cat o I: ._ c: ~ E o o ·- C O ~ - _ ._ _ ~ ~ o Cat ._ _ _ ~ C ~ ~ Cal Cat on ~ ~ o ._ ~ I,e ' go g I: o E ·_ _ ~ ·- E.° U ~ ~ ~ ha a ~ ~ .~ 0 ~ ~- _ ~ [d ~ ~ o ._ ~ .0 ~ ~ - O ~`0 -) - c ~ ~ _ _ - - 2 8 ~ 52 2 8 ~ ~ ~ _ _ .~ 3 ~ o ~ ~ C~ ° ' E o E ~ E ~ · Y 8 y E E ~ _ _ .t L~ :=
35 ~ ~C =& ~ -o ~e _ ~ - o' O - o. ~ ~ C oo C ~ O ._ - 1: 5 O _ C ' ._ O O < _ _ _ CC _ _ i: _ - CJ ._ E 3 - - E - E E :' E :5 c - c . _ - .. - ._ ~: - ~C - - ._ ce ~; ~, _ c~. C ~ . ~ _,c X . r^, - t- _ o~ o C _ _ ~ ~.~ .- ~ ._ ~ E ,~, os ~ ~ 6 4~, ~ _ ~ oo ~ ' ·- C o _ _ ~, ~- ' t: ~ 1,,, . _ c`0 ~ s E e; .{e tJ w: ~ ~ C C) .' ' ,' ~ ~ ,,, a .~ E ~ ~c). Y c =_ 3C. _ C ~ a, o~ - . ~ ~ ~ re O ". ~ rv ~ ~: Q e~ ~ E, ~ ~ ~ C _ _ ~ , _ - ~ C ._ e~ ~ ~ ~ E ~v c - OD ~ X . 3 c. s E .~ ~, c~ c c~ ~ .c .E ~. :e ._ C oU C _ ~E .° ~ ~ ~s _ tc ~ _ _ - - ~L~ t~: - ;~` ~ lv O ~ o o . oc3 ~ · _ , ~ 4~, . 3 ~ O _ ~ x E c u: ie ~ S 3 0 ',, ~e E _ ~. ~ ~ 0 ~ - .~ ~ e ~ :' ' O LU x - o - E c 11 o e . . ~ t: ~ 3 ~ e e 0 8 ~ ,, 3 0 ° E E E E E o~ O _ ~ ._ ~ ~ ~ CL - oo 0 ~ ~ _ ,o ~ ~ _ 0 i~ ~ g CL e ~ ~ =,.`,.e ° . ·E-o ~ ·. ' ~ !, o S' 3 ,, -, L ~ .- ~ ~ ~ ~ . .~ - o ~ eo~ 1 oo ~ e oo ~ X ·UE 0 ~ _~ ~ 0 ~ 0 . _~ E .o- 8 . : Q ~ ~ - LL EE U, - . - , - .o - o ~C - . . - n o Y . ~ _ ~ O j ·c ·-] ~ O ~ _
36 Cal o 3 i_ to V, ._ U' Cal To ~ ~ ~ . _ ~ ee ~ Yin ~ ~ _ ~ :~ C LU -- -~ ._ ~ -a ~ C) .~ ~ ~ to E o ~ 3 ~ C) ._ ,~ O O ~ C.i Ct', ~_ o LU .~ ee . _ _ - LL] C} ._ x U. - O `^ ._ 6 A, ._ ._ - s I_, ~ o v O E ~ ' - t~ C ~ ._ - - l,, ._ _ CD _ Cal to ~ ~ - ,~ _ _ oe 0 `O~ ~ ~ d -d er u~d S ~- x o )d ~ ~ ~- d c o ~l ~c ~ - ~, - ) ~ - . c~ S - - ~- - - ~ ~ - - ~ u, i ~c ~u) ll ~ i~ Ct :~d ~O! LL ~: . - E c C} .u ~ _ Y o o _ C~ ~ - et ~ oo _ · _ ~, ~_ ~, o ~eo C~ _ ~_ 3 e Y ~. ~.c E ~C, ee 2 te ~_ o ~ . Ce ~ ' ~ [7 · 3 ~ ~ ~ E ~ ~E ~ ' 3 o O ~]- '. ~ ~ -y ~
37 ~_ - ~ ~ _ 3 _ E "- ~ ~ct. ~ C At;C ° v. - E C cv. ~ E ·, ~E C '~ ~e I _ I 141i _ , ~ _ ^ _ · , ~- _ e c ~ ~= ~ ~oL ~ .c C, ~_ ~O ~. _ ~_ at, . 'A E oE .o .' o _ 4, _ - -- E_ .c_ ° [' ~ _ ~5 Hi ~ ~ C .§ ~-^'~22, ·,,.-t7~,tj ·- ·_ E ~ {:i 6 ~ ~ ~ 6 ~ ~ ~ 5 ~ ~ ~ ~ ~ 0 6 ~ ~ i ~ O ~ O O a~ e, ~ c~ e-, ~ u~d ~ ~ eec ~ C ~ ~ -) ~ ~ ~) `0 ~ ~ ~ 0\ 0 - r~ ~ ~ ~ ~ 3 - - - - ' _ _ _ ~ _ _ t~ ~ ,~ -
38 WOMEN. WORK. AND WAGES across these more narrowly defined occupations that accounts for a large portion of the earnings gap. Sanborn's study, however, includes no variables that attempt to describe characteristics of jobs, and his study provides no information as to what it is about these occupations that accounts for earnings differences. One study is particularly informative about the explanatory power of the different types of variables. Roos (1981) found that variables of the human capital type accounted for about 20 percent of the difference in earnings; that the addition of Duncan's socioeconomic index and pres- tige variables did not account for any additional portion, nor did oc- cupational characteristics from the Dictionary of Occupational Titles; and that the addition of other occupational characteristics (industry, supervisory status, percent female, and median income of male incum- bents) accounted for an additional 11 percent of the difference. Roos's occupational characteristics can be thought of as of two types, "benign" and "suspect," with respect to the issue of comparable worth. Benign characteristics, such as ratings of job complexity, supervisory duties, and possibly industry (which might stand for the ability of an employer to pay), are, like the productivity differences that human capital vari- ables attempt to measure, generally regarded as legitimate bases for pay differentials. Suspect characteristics, such as percent female and median earnings of male incumbents, generally would not be considered legit- imate because they indicate the extent to which women's lower earnings are related to their being in jobs held mainly by women and in low- paying jobs. Roosts study of determinants of individual earnings is con- sistent wild the staff study reported above (Hartmann et al., 1980) on determinants of occupational earnings, which indicates that percent fe- male is an important determinant of earnings.2' On the whole, however, the studies of earnings differences that use job characteristics as explanatory variables do not constitute a definitive body of literature. There are simply not enough of such studies nor are they conclusive enough to show what it is about jobs held mainly by 2' In a different context, several studies by economists have related job characteristics to wages in an attempt to test the theory of compensating or equalizing wage differences originally suggested by Adam Smith. (The theory suggests that unpleasant job charac- tcristics should earn a premium.) Lucas (1977), using job characteristics from the Dic- ~ionary of Occupational Titles, found. for example. that positive coefficients are estimated for variables representing repetitiveness and unpleasant physical working conditions. sug- gesting that they can be viewed as compensating differences. Robert Smith (1979) sum- marizes the generally mixed results of a number of these empirical studies. These studies can be viewed as attempts to estimate the implicit prices of job characteristics from knowledge of the explicit pnces~thc wage rates of job~and are thus an example of hedonic price equations of the type discussed by Rosen (1974~.
Evidence Regarding Wage Differentials 39 men and those held mainly by women that accounts for their differences in earnings. These studies do confirm, however, the importance of job segregation by sex in explaining the difference in earnings between men =d women. Two kinds of detailed studies do provide a more complete under- standing of the effect of job segregation on earnings: studies of particular occupational groups and studies of workers in individual finds or or- ganuations. The fraction of the types of jobs covered by these studies is unknown. Still. they provide some sense of the extent and impact of job segregation on pay differentials by sex and by race. First, for man`, if not most, occupations a substantial portion of the difference in average earnings between men and women can be attr~b- uted to the fact that women are more likely than men to be employed in low-paying firms and less likely to be employed in high-paying firms. Blau s (1977) detailed study in three metropolitan labor markets of selected white-collar occupations filled by both men and women is par- ticularI, instructive on this point. Blau investigated the distribution of workers among firms and found that the job segregation of men and women is important in explaining wage differentials even when occu- pations are integrated. Using data provided by the Bureau of Labor Statistics' Blau demonstrates that within occupations that are integrated by sex, such as accounting clerk. men and women are not randomly distributed among firms: there is more segregation by sex across firms than would occur through random hiring processes. Moreover, a wage hierarchy exists among firms: those that employ more men in a given occupation than expected from a model of random hiring processes pay- the highest u ayes those that are integrated pay average wages' and those that employ more women than expected pay the lowest wages.22 BIau finds that more of the within-occupation wage differences be- tween men and women can be explained by differences in pay among firms than can be explained by differences in pay within firms. That is, the wage hierarchy among firms and the segregation of women into the low-wage firms account for the larger part of the differential in men's and women s wages. Blau finds that this wage hierarchy is consistent across occupations, so that those firms that pay higher wages do so in all of the occupations studied. Moreover, firms at the high end of the hierarchy hire fewer women across all occupations. These results lend support to an institutional explanation for wage differentials. Since Blau's study does not rely on particular attributes 22 These findings are similar to those reported by Buckley (1971) for eight office and two plant occupations each with significant numbers of jobholders of both scans. as well as those reported by McNulty (1967).
do WOMEN, WORK, AND WAGES of occupations to explain either differentials in pay or segregation itself, it points to those structural and institutional factors that can affect all jobs. Since the occupations she studied are very narrowly defined, they must have similar attributes regardless of whether they are performed by men or women, and the incumbents are likely to have similar qual- ifications and expenence. Occupational segregation by sex exists never- theless. A number of other studies have also shown that within occu- pations jobs are substantially segregated across Liens, always with the result that jobs held mainly by men are paid more than jobs held mainly by women.23 Occupational segregation also exists within firms and is widely known to be common' although precise measurement of its extent is difficult because publicly available data at the establishment level are rare. To a large extent, the occupational segregation observed within firms simply mirrors that observed in the labor market as a whole. For example, the secretaries that firms hire are pnmanly women because most trained secretaries are women; similarly. the accountants they hire are primarily men because most trained accountants are men. In many firms it is typical for managerial jobs to be dominated by white men; for professional jobs to be dominated by whites. although not so exclusively by white men as managerial positions; for clerical jobs to be dominated by women; for craft and laboring jobs to be dominated by men; for specific operative jobs to be dominated by one sex or the other and sometimes by one race or ethnic group; and for most service jobs to be dominated by women and minority men. An interesting example comes from the U.S. Office of Personnel Management (formerly the U.S. Civil Service Commission), which de- tennines personnel policies for the federal government, the nations largest single employer of white-colIar workers. The Office of Personnel Management uses a standardized occupational grading system (the "General Schedule,~' GS) consisting of 18 grades defined on the basis of the knowledge required for a position, the degree of autonomy, and a number of other factors (see Treiman, 1979:17-20, for a more detailed description). A GS level is assigned to each job in the federal civil service, which determines its pay range; hence, the GS hierarchy is also a pay hierarchy. Table 11 shows the distnbution of federal white-collar workers by GS level and sex as of November 1977. Women are over 23 Scc, for example, Bridges and Berk's study (1974) of nonsupervisory white collar employees in a~icago-arca financial institutions; Talbert and Bose's study (1977) of retail clerks in a metropolitan area; Allison's suldy (1976) of beauty salon operators; and the study by Darland et al. (1974) of salary differences between male and [cmalc college and uni~cr~ty faallty.
Evidence Regarding Wage Differentials TABLE 11 Percent Female by Occupational Grade (GS Level) in the Federal Civil Service for Full-Time, White-Collar Employees of Federal Government Agencies. 1977 41 GS Level 8 4 15 2-}3 9~11 S 8 ID TOTAL Percentage Female Percentage Distribution of All Employees 10 30 63 77 43 615.342 6 19 24 30 ~0 1(~0 ,429,645 Less than 0.5 percent. SOURCE: Barrett. 1979:Table 4. whelming!! concentrated in the lower grades, in what are. for the most part clerical jobs (see also Osterman's 1978 study of a large publishing firm). Job segregation by sex. whether within a firm or across firms, provides an important clue to the causes of the difference in earnings between women and men. vet leases open the question of why jobs and occu- pations are segregated and what the exact relationship is between job segregation and pay differentials. While it is clear that women are con-- centrated in jobs that pas less. it is unclear to what extent this is because women choose jobs with low pa, for reasons other than their sex com- position (for example. because tines tend not to penalize incumbents with intermittent labor force experience), to what extent women are restricted to such jobs. and to what extent some jobs pay less than others because they tend to be held by women. We take up these questions in the next chapter. CONCLUS10~3 In this chapter we have reviewed evidence on the extent and causes of earnings differentials between men and women. That such differ- cntials exist is not in dispute: among full-time year-round workers the earnings of women average less than 60 percent of those of men. What causes the difference in earnings is a matter of considerable dispute. The evidence suggests however, that only a small part of the earnings
42 WOMEN, WORK AND WAGES differences between men and women can be accounted for by differences in education, labor force experience. labor force commitment or other human capital factors believed to contribute to productivity differences among workers. The findings from studies attempting to explain the differences in earnings between men and women on the basis of such factors usually account for less than a quarter and never more than half of the observed earnings differences. The evidence reviewed on job segregation by sex suggests that an aaa~uona' part of the earnings gap results from the fact that women are concentrated in low-DavinQ lobs. Job segregation by sex is quite pro- nouncea ano snows tew signs of substantially diminishing. Women are concentrated in low-paying occupations and' within occupations, in low- paying fimns. The significant degree of job segregation by sex may be a consequence of a variety of institutional forces' discriminatory prac- tices, or other factors that operate in the labor market to depress market wage rates for women's jobs. In Chapter 3 we investigate the institu- tional features of the U.S. labor market that may depress woments wages and result in the persistence of an earnings differential between men and women over time. at . ~. ~. ~v ~- - - . . . ~. . - TECHNICAL NOTE Two logically similar statistical methods have been used in human capital studies of sex discrimination in earnings, sometimes in combi- nation: demographic standardization techniques and decomposition pro- cedures based on multiple-regression analysis. Since the latter are more common, we descnbe them in some detail. Let us consider a hypothetical regression analysis of the difference in the earnings of men and women. The conventional approach is to es- timate, separately for men and women, an ordinary least-squares regres- sion equation of the form Y= a ~ Ib'Xi, (1) where Y is the worker's earnings and the X' are the worker's human capital characteristics (e.g., years of school completed, years of labor force experience, amount of on-the-job training). First the equations arc estimated for each sex; then the mean values of each of the human capital characteristics (the X,) for one sex are substituted in the equation estimated for the other sex to derive an estimate of the average earnings expected if the only difference in average earnings between men and
Evidence Regarding wage Differentials 43 women was due to differences in the measured human capital charac- teristics. For example, suppose that for a sample of men the least-squares estimate of earnings from schooling and experience is Y = 1,000 ~ 500(years of schooling) ~ 200(years of experience) (2) and that their average level of earnings is S10,000, that their average amount of schooling is 10 years' and that they have an average of 20 years of labor force experience. Now, suppose further that the average level of earnings of women is S6,000, that their average amount of schooling is 11 years' and that they have an average of 10 years of labor force experience. Substituting the means for women into the equation for men (eq. 2~. Y= 1,000 + 500~11) + 200~10) = 8,500, (3) which implies that if women had the same rate of return on their edu- cation and experience as men they would earn 85 percent (8,500/10,000) as much as men. while in this example (and in the U.S. economy) women actually earn only 60 percent as much as men. On this basis, the total earnings gap is decomposed into a portion due to differences in human capital factors and a portion due to differences in rates of return on those factors. In this example, about two-thirds of the earnings gap (~85 - 601~100 - 601) could be attributed to the fact that women get a lower return on their education and experience and about one-third of the earnings gap to differences between men and women in the amount of their education and experience. Alternatively, the means for men on education and experience could be substituted into an equation for women. This does not usually produce an identical answer, but the two sets of answers tend to be similar (see Oaxaca, 1973, for a discussion of this issue. known as the index problem). Usually the difference in rates of return estimated from such equations (that is, the difference remaining after gender differences in the factors thought to affect productivity have been controlled) is taken to represent discrimination. This approach has a number of difficulties, which are discussed above. 1
