The effects of comparable worth in the public sector on public/private occupational relative wages.
It is well known that women are paid less than men, and that this differential persists despite policies of equal employment opportunity.(1) Proponents of comparable worth attribute much of the wage differential to occupational sex segregation. The fact that many occupations have high proportions of either men or women contributes to the differential, it is argued, because of pay practices that place a higher value on jobs held typically by men than on jobs held typically by women. The resulting male/female wage differences cannot be remedied by a law like the Equal Pay Act of 1963 which only requires equal pay for workers on similar jobs. The remedy, according to comparable worth advocates, is to adjust wages in a way that eliminates the portions of occupational wage differences due to discrimination. This involves the use of a common standard to set the wages for those jobs eligible for adjustment. Typically, the standard takes the form of a model that translates job attribute measures into the appropriate wage.
This study is concerned with the way that comparable worth affects wage differences between sectors when only one sector--the public sector--is covered by the policy. This differs from the typical analysis which is concerned with the way that comparable worth affects male/female wage differentials within the covered sector.(2) Of particular concern is the way that public/private occupational relative wages are affected by comparable worth in the public sector. As detailed below, knowledge of how the level and structure public/private occupational relative wages will be changed is an important ingredient in any evaluation of the economic effects of the policy. In the context of the prevailing wage principle, which holds that the wages for public jobs should be equal to the wages for similar private jobs, the degree to which the public occupational wages conform more, or less, closely to private occupational wages has implications for economic efficiency.
This study uses Current Population Survey (CPS) data to simulate the effects of comparable worth in the state and local government (SLG) sector on public/private occupational relative wages. The results suggest that when comparable worth is implemented through special wage increases, public wages move further from compliance with the prevailing wage principle. This is because public wages are, on average, already on a par with private wages. However, comparable worth tends to give larger wage increases to occupations where public/private relative wages are low. And when the payroll budget under comparable worth is set at the same size as the budget before comparable worth, public wages may move closer to compliance with the prevailing wage norm.
The next section explains why comparable worth is likely to be confined to the SLG sector and explores the implications for government pay policies. Section III describes the alternative comparable worth scenarios. The data and model specification are discussed in Section IV. Section V provides details of the estimation and the results. Conclusions are presented in Section VI.
II. Comparable Worth Coverage and Public Pay Policies
Although early discussions on comparable worth envisaged a comprehensive national policy,(3) it now appears state and local governments are much more likely to implement the policy than private employers are.
In the case of the SLG sector, it is reasonable to presume that the momentum established during the 1980s, when a number of governments took steps toward comparable worth, will continue.(4) Support this view can be found in the apparent susceptibility of government pay policies to political influence, the fact that relatively high proportion of SLG workers are women, and the fact that formal occupational wage structures are already well-established in the public sector.
Currently, there is little likelihood of significant comparable worth coverage in the private sector. Due to competitive pressures, private employers are unlikely to voluntarily introduce comparable worth. And it is not likely that they will soon be legally obliged to install comparable worth. The courts have resisted attempts to redefine existing antidiscrimination laws (the Equal Pay Act of 1963 and Title VII of the Civil Rights Act) in a manner that would impose a general comparable worth standard on employers. A number of decisions, discussed by Milkovich and Newman (1990) and Smith (1988), have affirmed the legality of wage differentials for different jobs even though these jobs may be judged to be of comparable value.
In general, a comparable worth policy that covers only part of the economy presents interesting questions about the distribution of benefits. Smith (1988), for example, sought to determine whether a policy confined to large private firms and governments targeted those who most needed help--women who were low paid and/or subjected to greater discrimination.
When comparable worth coverage is restricted to the public sector, special questions about the efficiency of public payroll expenditures emerge. It is a basic premise of this study that the answers to these questions will depend on how the structure and the average level of public/private occupational relative wages are changed as a result of the policy.(5)
The prevailing wage principle provides the basis for evaluating the changes in public/private occupational wages. This principle, as discussed by Fogel and Lewin (1974) and Smith (1977), holds that the wage for a public-sector job should be equivalent to the wage for similar work in the private sector. It has enjoyed wide acceptance as a norm for government pay-setting. Payment of wages above those prevailing in the private sector is presumed to be inefficient because it implies an expenditure level that is greater than that required to retain a suitable workforce. Payment of wages below those prevailing in the private sector is also presumed to be inefficient because of the difficulties associated with retaining qualified workers in the long term.
Despite the acceptance of the prevailing wage principle, the available evidence shows that for a number of occupations at least, public/private wage parity does not exist (Fogel and Lewin 1974, Smith 1977, Hundley 1991). Therefore, the economic consequences of comparable worth in the public sector is an empirical question. Although many economists might argue that comparable worth in the private sector is inefficient a priori because it inevitably distorts wages relative to the standards set by freely operating private markets, public-sector wages are already distorted in this way. Consequently, comparable worth in the public sector could move public wages closer to, not further from, compliance with the prevailing wage standard.
The way that comparable worth affects the degree to which public wages comply with the prevailing wage standard depends on how the policy affects the distribution of public/private occupational relative wages. Here, analysts are likely to be interested in the effects of comparable worth on both the average level and the dispersion of public/private occupational relative wages--although in some cases the effects of the policy on the wage dispersion will be of singular concern.
The change in the overall mean public/private relative wage (measured as a ratio of the average public wage to the average private wage) has implications for the efficiency of payroll expenditures. A mean in excess of unity indicates that expenditures exceed the amount that is consistent with wage parity in all occupations. A mean that is less than unity implies that not enough resources are available for wage parity--no matter how occupational wages are structured. A mean public/private occupational relative wage (measured as the average of the public/private occupational relative wage ratios) that equals unity is a necessary condition for the prevailing wage standard to be met.
Since public employers hire workers from many occupations, and workers from different occupations are imperfect substitutes, the effects of the policy also will depend on how the dispersion of public/private occupational relative wages is changed. It is obvious that a distribution of public/private occupational relative wages with a mean that is not equal to unity can come closer to complying with the prevailing wage standard than does a distribution which has a mean closer to unity but is more widely dispersed. In general, comparable worth will bring the public sector closer to wage parity if public/private occupational wage ratios are, on average, brought closer to unity, that is, made less dispersed around unity.
Although the shift in the average level of public/private wages is of interest, it should be recognized that the revision of the occupational wage structure is the essence of comparable worth. Nothing about the policy permanently fixes public/private relative wages at a specific level. Consequently, given special but plausible assumptions about the size of the payroll under comparable worth, there are at least two cases in which analysts will focus on the effects of comparable worth on the dispersion of public/private occupational relative wages. The first case arises if it is assumed that decisions about the size of the payroll budget under comparable worth can be made separately from decisions about the public-sector occupational wage structure. Although it is often supposed that the changes in occupational relative wages within the covered sector are achieved through wage increases,(6) it is also appropriate to make these changes with a program that includes wage reductions.
If it is assumed that public officials can choose the size of the payroll budget under comparable worth, the main concern will be whether the distribution of public/private occupational relative wages is made relatively more or relatively less dispersed by the policy. If the dispersion is lessened, then it will generally be possible to set the payroll at a level that will ensure a post-comparable worth relative wage distribution that is closer to meeting the prevailing wage standard than any distribution that would have been possible before comparable worth. If the distribution is made more dispersed, then public wages after comparable worth cannot be brought as close to the prevailing wage standard as public wages before comparable worth could have been.
The second case in which the effects of comparable worth on the dispersion of public/private relative wages will be the dominant concern arises if it is believed that any increase in the general level of public/private relative wages that results from initial policy implementation will be temporary. Comparable worth plans, as proposed, do not fix average public/private pay levels in the long term. The evidence supports the idea that taxpayer resistance will, in the long term, exert downward pressure on higher than usual public/private pay levels. Although the overall public/private pay ratio has tended to fluctuate over time, state and local government wage premia have not persisted over long periods (Smith 1983, Freeman 1987). Consequently, consistent with the conjecture of Megdal (1986), the initial gains in public/private relative wages under comparable worth may dissipate as taxpayer resistance sets in and officials attempt to maintain established levels of service.
Because comparable worth does not fix average public/private pay levels, policy-makers will prefer a narrower distribution of public/private occupational wages to a more dispersed one. This is because as (or if) public/private relative wages tends towards parity, the more dispersed distribution will always represent a greater deviation from the prevailing wage standard. Indeed, the scope for long-term reductions in public/private relative wages, and perhaps even for the provision of the same level of government services with a payroll that is no larger than the one that would have existed in the absence of comparable worth, will depend on the public/private relative wage structure.
If public payroll budgets (and, ceteris paribus, average public/private pay) do tend to adjust back toward customary levels, then a comparable worth policy that reduces the relative dispersion of public/private relative wages means it will be possible to get closer to meeting the prevailing wage standard in the long run. Suppose, consistent with the evidence (Smith 1983, Moore and Raisian 1991) that the average public/private wage in the SLG sector is close to unity. Then, as budgets return to usual levels, the post-comparable worth public/private occupational relative wage distribution will be closer to meeting the prevailing wage standard than any distribution that was possible before comparable worth. The opposite holds if the public/private occupational relative wage distribution is made more dispersed by comparable worth.
III. Comparable Worth Wage Adjustment Methodologies
The analysis requires estimates of: (1) public/private occupational relative wages before comparable worth, (2) the adjustments to public-sector occupational wages under comparable worth, and (3) public/private occupational relative wages after comparable worth. Thus, three wage variables are defined: |W.sub.u~, the wage received by a public worker in the absence of comparable worth, |Mathematical Expression Omitted~, the expected wage of the worker in the private-sector, and |Mathematical Expression Omitted~, the public-sector wage under comparable worth.
Private-sector wages are assumed to be determined by the semilogarithmic wage equation:
(1) ln |W.sub.r~ = |X.sub.r~||Beta~.sub.r~ + ||Epsilon~.sub.r~,
where |X.sub.r~ is a matrix of relevant explanatory variables, ||Beta~.sub.r~ is a coefficient vector, and ||Epsilon~.sub.r~ is the disturbance. The subscript, i, for individual observations is suppressed. The estimates are used to predict the private wage that would be received by each public-worker:
|Mathematical Expression Omitted~
where |X.sub.u~ is the matrix of relevant variables for public-sector workers and |Mathematical Expression Omitted~ is the estimated residual variance from (1).(7) The public/private relative wage for occupation j prior to comparable worth is:
|Mathematical Expression Omitted~
where there are |n.sub.j~ public workers in the jth occupation.
Three policy scenarios are analyzed. In the first two scenarios, comparable worth wage adjustments are determined by a wage equation fitted to all or part of the public-sector sample. In the third scenario, which has two variants, wages are adjusted to remove those parts of occupational wage differences attributed to occupational differences in gender composition.
Each scenario is investigated under two alternative assumptions about the size of the payroll budget after comparable worth. First, it is assumed that the budget is set to the level that permits the changes in public/private relative wages required for initial implementation of the policy to be maintained in full. Second, it is assumed that the payroll budget remains at its pre-comparable worth level--consistent with either where comparable worth is implemented by reallocating the existing payroll or where, in the long run, the payroll budget returns to the size it would have been in the absence of comparable worth.
This method, proposed by Treiman and Hartmann (1981), uses the wage equation for male-dominated occupations to compute wage adjustments for workers in mostly female occupations. The model is assumed to be:
(4) ln |W.sub.um~ = |X.sub.um~||Beta~.sub.um~ + ||Epsilon~.sub.um~,
where ln |W.sub.um~ is the logarithm of the wage, |X.sub.um~ is the matrix of relevant individual and job attributes for employees in the male-dominated occupations, and ||Epsilon~.sub.um~ is the disturbance. The comparable worth wages for workers in female-dominated occupations are determined by moving the data points for members of the mostly female occupations to the regression line fitted for predominantly male jobs. Therefore,
|Mathematical Expression Omitted~
where |X.sub.uf~ is the matrix of relevant attributes in the female-dominated occupations. There is no wage adjustment for members of occupations that are not female-dominated. For these individuals: |Mathematical Expression Omitted~. The percentage change in the average wage for the jth occupation is:
|Mathematical Expression Omitted~
The public/private relative wage for occupation j after comparable worth is:
|Mathematical Expression Omitted~.
The effects of the policy depend on criteria for determining which occupations are male- or female-dominated. This study follows the precedent set by most previous applications, for example, Sorensen (1987). Occupations with more than 70 percent male workers are designated male-dominated and those with more than 70 percent female are designated female-dominated.
Here, a single regression is estimated from data for public workers in all occupations. The wage adjustments are made by moving the data points to the regression line itself. The wage equation is:
|Mathematical Expression Omitted~
The comparable worth wage is |Mathematical Expression Omitted~, and the wage adjustments public/private differentials are computed as in (6) and (7). Since comparable worth wages are predicted from a regression estimated from the entire sample, there will be wage reductions for all individuals represented by data points that lie above the fitted regression.(8)
This method, also advanced by Treiman and Hartmann (1981), and used by Aldrich and Buchele (1986) and Johnson and Solon (1986), purges the estimated effect of an occupation's gender composition from the wage for that occupation. The following wage equation is estimated for all public workers:
|Mathematical Expression Omitted~
where |F.sub.u~ measures the proportion of the occupation that is female. The coefficient, |c.sub.u~, according to Treiman and Hartmann (1981), "indicates the difference in pay rates that would be expected on the average between two occupations that differ by one percentage point in their sex composition, but are identical with respect to all other measured variables."
Two alternative methods of computing the comparable worth wage adjustments are used. In the first--Scenario IIIa--the comparable worth wage is predicted directly from estimates of (9) with |F.sub.u~ = 0, that is, |Mathematical Expression Omitted~. The comparable worth wage adjustment and the resulting public/private differential are calculated as in (6) and (7) respectively. If |Mathematical Expression Omitted~, most, if not all, occupations with some females will gain. Under the second method of adjustment--Scenario IIIb--the comparable worth wage is calculated by subtracting the wage effect due to the 'proportion female' from the actual wage of each worker, that is, |Mathematical Expression Omitted~.
Scenario IIIb provides an important check on the biases that can arise where, as with all other scenarios, comparable worth wages are predicted directly from a fitted regression. It is possible that wages estimated in this manner are purged of variations that would legitimately exist after the implementation of comparable worth and that this leads to biased estimates of the effects of comparable worth. Suppose, for argument's sake, that the public-sector equation excludes an attribute that is positively correlated with wages. Although there is no reason to expect that the estimate of the average public-sector wage for all workers will be biased, there may be biases in the estimates of the average wages of specific groups. In general, there will be a tendency to overpredict the comparable worth wage for groups with members who have relatively low amounts of the omitted attribute (and who would have tended to lie below the fitted line), and to underpredict the wage adjustments for groups with members high in the attribute (and who would have tended to lie above the fitted line). Biases in the estimates of the effects of comparable worth on the dispersion of public/private occupational relative wages are not inevitable, but will occur if the omitted variable is correlated with public/private occupational relative wages before comparable worth. Up to a point, the degree of relative wage dispersion after comparable worth will be underestimated if the regression underpredicts the wage adjustment in occupations with high public/private differentials before comparable worth.(9) Similarly, up to a point, the degree of relative wage dispersion after comparable worth will be underestimated if the regression line underpredicts the wage in occupations with low public/private differentials and overpredicts the wage in occupations with high public/private differentials.
Although it is not certain how much of the wage variation not predicted by the regression would remain after implementation of an actual comparable worth policy, it is likely that a substantial component (due to, say, unmeasured productivity effects) would be. On this basis, estimates of relative wage dispersion estimates from Scenario IIIb may be favored over estimates from the other scenarios. But since the policy could be implemented in a way that eliminates some or even all of the residual variation, Scenario IIIa (as well as Scenarios I and II) can be considered special, less likely, cases. And since the biases, if any, that result from predicting wages directly from the wage equation can not be determined a priori, a comparison of IIIa and IIIb may provide a general idea of the scale and direction of the biases under Scenarios I and II.
IV. Data and Model Specification
The first six months of the 1985 CPS were pooled to provide observations on 71,163 wage and salary workers in the private-sector, and 9,530 wage and salary workers in the SLG sector. Individuals employed in agriculture or domestic service, or aged less than 18 years, or who reported an average hourly wage of less than $1.50 were excluded. School teachers were excluded on the basis of Smith's (1988) arguments that teachers are not likely to be affected by more general comparable worth measures. The federal sector was also excluded. Summary statistics for all variables are in Table 1.
Estimates of job characteristics were derived from the Dictionary of Occupational Titles (DOT). Measures from the fourth edition of the DOT (U.S. Department of Labor 1977) were aggregated to the level of 1980 three-digit census occupations and merged with individual observations. The DOT provides a range of measures that correspond to job attributes that may affect earnings. A number of studies have chosen a priori DOT measures, notably specific vocational preparation and general educational development (Aldrich and Buchele 1986, Buchele and Aldrich 1985). This analysis uses four occupational characteristics scales derived from a factor analysis of the DOT scores by Miller et al. (1980). These scales: Motor Skills, Physical Demands, Undesirable Working Conditions, and Substantive Complexity, correspond reasonably well to three of the factors customarily found in the job evaluation plans: skill requirements, effort, and working conditions. Treiman et al. (1984) have shown that they explain a substantial proportion of the variance in occupational median earnings.
Estimates of |F.sub.u~, the proportion of state and local government workers in three-digit occupations who are female, were derived from 19,484 observations on SLG workers from all twelve months of the 1985 CPS.(10) Because a sizable number of occupations had small cell sizes, Census Bureau estimates of the proportion female in three-digit occupations for all sectors (U.S. Bureau of the Census 1983) was used to adjust the initial CPS estimates. Average hourly earnings (usual weekly earnings divided by usual weekly hours) is used as the dependent variable.
V. Estimation and Results
Estimates of the private-sector wage equation, (1), and the public-sector wage equations, (4), (8), and (10) are provided in Table 2.
Some idea of the validity of the estimates can be gained by comparing predictions from these models with similar estimates from other sources. The average comparable worth wage increase under Scenario I was found to be 12.1 percent for all females and .5 percent for males; and the wage average increase in occupations receiving adjustments was 17.6 percent. These figures are close to estimates TABULAR DATA OMITTED by Sorensen (1987) who applied a similar methodology to job evaluation data from five state or local governments. The overall mean public/private relative wage--the average public wage divided by the average estimated private wage, |Mathematical Expression Omitted~ is .989, and, as shown in Table 3, the mean of public/private occupational relative wage is .998. These figures fit well with Moore and Raisian's (1991) estimates of the public/private wage differential in late 1970s and early 1980s.
Two possible sources of bias in the estimated coefficient, |c.sub.u~, on the proportion female variable (|F.sub.u~) in the Scenario III wage equation are considered. First, there is the possible bias due to the exclusion of actual (rather than potential) work experience and seniority from the wage equation. If workers with less actual experience and less seniority have lower wages, and they are more heavily represented in predominantly female occupations, then variations in |F.sub.u~ are likely to capture some of the negative effects of the greater time spent out of the work force and of lower seniority. Thus, the estimate of |c.sub.u~ will be biased away from zero. Experiments with richer data available from the May 1988 CPS suggests that this bias may be small, however. A model in which the potential experience measures were replaced by measures of age, age-squared, and dummies for the number of children--a specification that was found by Mincer and Polachek (1974) to produce similar results to models that included actual work experience--yielded a coefficient that was slightly smaller and not significantly different from the coefficient from a specification similar to that reported in Table 2. When this model was augmented by a tenure variable (to capture seniority effects), the coefficient was further reduced, but was still not significantly different from the coefficient in the original model.(11)
TABULAR DATA OMITTED
TABULAR DATA OMITTED
The second area of possible bias in the estimate of |c.sub.u~ involves the absence of industry dummy variables from the SLG wage equation. Estimates of the proportion female coefficients from private-sector samples have been shown to shrink considerably when industry controls are included.(12) There does not, however, appear to be a case for including industry effects in the SLG model. The industry controls in private-sector models are justified by the argument that the firm-specific plans contemplated in the private-sector would not remove the interindustry components of wage differentials.(13) But comparable worth proposals for the public sector usually seek to apply the same plan across different functions and agencies; thus it is unlikely that workers in different functional areas will be covered by different plans.
It is noted that the coefficient estimate of -.33 in Table 3 is close to, though greater than estimates for single government employers: -.26 from 1983 State of Iowa data by Orazem and Matilla (1989), and -.3 from State of Washington data by Aaron and Lougy (1986). Since the wage equation does not control for the biases due to the exclusion of actual experience and seniority, as discussed above, |c.sub.u~ = -.33 may be closer to the upper (absolute) bound of likely values. Consequently, Scenario IIIb analyses were also conducted with |c.sub.u~ set to an arbitrary lower bound of -.27.
The estimated wages from each scenario are used in several different ways to explore the effects of comparable worth. The effects on public/private occupational wages are evaluated in three ways. First, the properties of the distributions of public/private occupational relative wages before and after comparable worth are compared. Second, estimates are made of the relationship between the comparable worth wage adjustments and the public/private relative wage adjustments before comparable worth. Third, the experiences of specific major occupational groups are examined. Finally, the effects of comparable worth on male/female relative wages are estimated.
A. Public/Private Relative Wage Distributions
Estimates of the properties of the distributions of public/private occupational relative wages before comparable worth, and under the three major scenarios are in Table 3. The mean of the sum of squared deviations from unity MSSDU = ||Sigma~.sub.j~|n.sub.j~|(|X.sub.j~ - 1).sup.2~/N, where |X.sub.j~ is the jth relative wage and N is the total number of SLG workers, summarizes the degree to which public/private occupational wages depart from the prevailing wage norm. Compliance with the prevailing wage standard exists when MSSDU = 0. Higher values of MSSDU occur when public occupational wages lie, on average, further from private wages for the same occupation. The coefficient of variation (where the standard deviation is divided by the mean) provides a measure of relative dispersion--it shows which of two distributions that have been scaled to have identical means has the greater variance. The percentile measures show the percentage of public employees in occupations with public/private relative wages equal to, or less than, the specified level.
The estimates show that, on average, public wages would be higher than private wages immediately after implementation of those comparable worth programs that require upward wage adjustments (Scenarios I, IIIa, and IIIb). The percentage increase in the average SLG wage under comparable worth, and hence the percentage increase in the payroll required to employ the same work force immediately after comparable worth, was computed as (d - 1).100, where |Mathematical Expression Omitted~, and found to be 5.9 percent under Scenario I, 18.5 percent under Scenario IIIa, and 17.8 percent under Scenario IIIb. Since the average public/private occupational relative wage before comparable worth is so close to unity, it is not surprising that these scenarios result in public occupational wages which, as indicated by MSSDU, are further from compliance with the prevailing wage standard. In the case of Scenario II, however, where the average SLG wage actually declines slightly, MSSDU indicates that the public wages are now closer to the prevailing wage standard.
Because comparable worth does not guarantee the maintenance of specific levels of public/private relative pay, the properties of the public/private occupational TABULAR DATA OMITTED relative wage distributions under alternative payroll budgets are of interest. To assess the effects of a fixed budget, it is assumed that the value of the payroll (relative to private-sector wages) is set to pre-comparable worth levels, and that the funds are used to employ a work-force identical to the one that was in place when comparable worth was implemented. Under this assumption, the overall public/private relative wage ratio is set to pre-comparable worth levels (g = .989). Therefore, in order to retain the public occupational pay structure fixed by comparable worth, all occupational wages are adjusted equiproportionately, that is, the public wage in occupation j is set to |W.sub.uj~(1/d).
Characteristics of the resulting public/private relative wage distribution are in Table 4. MSSDU is less under each scenario than it is prior to comparable worth. The reduction under Scenario IIIb when the proportion female coefficient is set at its upper bound (|c.sub.u~ = -.33) is very small, however. Because only Scenario IIIb controls against the possible biases in the estimates of relative wage dispersion due to the purging of residual variations that may legitimately exist after comparable worth, it cannot be definitively concluded that comparable worth with a fixed payroll budget will bring wages closer to compliance with the prevailing wage standard. If the direction and relative size of the bias are the same under TABULAR DATA OMITTED Scenario I and Scenario II as under Scenario IIIa, then the most reasonable conclusion is that comparable worth can produce a public/private occupational relative wage structure that, in terms of the quantitative measures used here, comes at least as close to compliance with the prevailing wage standard as the public/private wage structure before comparable worth did.
B. Wage Adjustments and Public/Private Relative Wages
To determine whether or not comparable worth provides greater benefit to those with comparatively low public/private relative wages, weighted least squares was used to estimate:
CWW|A.sub.j~ = ||Alpha~.sub.1~ + |a.sub.2~P|P.sub.j~
As shown in Table 5, ||Alpha~.sub.2~ is always negative and precisely estimated, indicating that comparable worth tends to give more help to those
public-workers who, according to the prevailing wage standard, are relatively underpaid.
Because comparable worth seeks to remove the wage differences attributable to occupational sex composition, an indication of the potential that this approach has for producing a wage structure that is closer to the prevailing wage standard is provided by a regression of public/private relative wage against the proportion female variable (|F.sub.u~). Weighted least squares estimates (including t-values) were:
|Mathematical Expression Omitted~
This result adds to Bridges and Nelsons' (1989) finding that employees of a state government in mainly female jobs were less successful in raising their wages relative to the private sector than employees in mostly male occupations were. It confirms that it may be possible to realign public-sector occupational wages in ways that bring governments closer to meeting the objectives of both comparable worth and the prevailing wage principle.
C. Major Occupational Groups
In order to assess the pattern of comparable worth wage adjustments across actual occupational categories, average public/private relative wages and average comparable worth wage adjustments were computed for 22 major occupations. The effects of comparable worth on the public/private relative wages for specific groups are of interest because, even if it is true that any departure from public/private wage parity is undesirable, policy makers may not be indifferent between a given deviation from the prevailing wage in one occupation and the same deviation in another. For example, the negative weight attached to overpayment may be less for occupations considered to be critical to government mission, or in which the nature of tasks permit superior workers (attracted by the wage premium) to contribute more to organizational success.
As shown in Tables 6 and 7, the pattern of relative wage gains (and/or losses) is, with a couple of exceptions, similar for all scenarios. The occupations can be placed into five major categories.
First, there are those occupations with comparatively low public/private relative wages before comparable worth that improve their position after the policy is implemented. These occupations have larger than average comparable worth increases under most or all scenarios. They include nearly all clerical and administrative support categories, "other" services, and sales workers--groups which constitute 25.8 percent of the total sample.
Second, there are occupations with above average public/private pay levels that have relatively small gains under the scenarios that involve general wage increases, or who would lose under fixed budget scenarios. These include public safety, cleaning service, and horticultural occupations--17.4 percent of the sample.
Third, there are the groups with below average public/private wages that suffer a further decline in their relative position after comparable worth. These include managers, scientists, engineers, physicians, communications and distribution workers, administrative supervisors, skilled trades workers, and operatives--32.7 percent of the sample.
Fourth, there are the groups that have relatively high wages before comparable worth and would have even higher rankings after the policy is implemented. These include "other" professionals (those who are not scientists, engineers, or health care professionals), health services, and food services--16.3 percent of the sample.
Finally, two groups--health treatment professionals (mostly nurses) and technical workers--7.8 percent of the sample--have lower than average public/private relative wages before comparable worth but the relative size of the comparable TABULAR DATA OMITTED worth adjustment depends critically on the specific comparable worth scenario. These occupations have substantial increases under Scenarios IIIa and IIIb which directly adjust wages for the 'proportion female' effect, but would lose ground (relative to other occupations) under Scenarios I and II where comparable wages are directly predicted from a fitted wage equation.
Although it is obvious that the wages for many public occupations will decline when the payroll budget is fixed, it is also possible that some occupations will have decreases when comparable worth is implemented through general wage increases. Such decreases could occur under Scenarios I and IIIa, but not under TABULAR DATA OMITTED Scenario IIIb where the comparable worth wage is determined by adding a non-negative amount to each wage.
Under Scenario I, where the data points for mostly female occupations are moved to regression line for mostly male occupations, the occupations that could lose are those that are more than 70 percent female (so that they are eligible for adjustment) and so highly paid relative to other public occupations that individual observations from these jobs lie, on average, above the fitted regression. Ten occupations, representing 4.6 percent of the SLG sample had wage decreases averaging 4.8 percent. Most (87 percent) of workers receiving decreases were in nursing and health care professional and technical occupations. These occupations could lie above the regression line because the model does not capture compensating differentials for onerous job attributes (such as shift work) that are characteristic of these jobs.
The occupations that are candidates for wage decreases under Scenario IIIa, where pre-comparable worth wages are purged of the proportion female effect plus all residual variation, are those that are predominantly male (and thus receive relatively small adjustments for gender composition) and highly paid relative to other public occupations |thus tending to be underpredicted by the regression coefficients, |Mathematical Expression Omitted~ from Equation (9)~. Occupations representing 15.6 percent of the sample had wage reductions which averaged 4.1 percent. Half the workers in occupations receiving reductions were in public safety jobs (police, fire, and corrections).
Under both Scenario I and Scenario IIIa, the occupations receiving wage decreases had higher than average public/private relative wage ratios before comparable worth (1.03 under Scenario I; 1.21 under Scenario IIIa). The ratio was reduced to less than wage parity under Scenario I (.97) but remained much higher under Scenario IIIa (1.16).
Overall, the pattern of occupational wage adjustments presents a mixed picture. Many of the female-dominated public occupations that gain from comparable worth have relatively low public/private relative wages before comparable worth. Some of these gains are accompanied by a reduction in the public/private relative wage position of male-dominated occupations. But there are many occupations where public/private relative wages move further from the prevailing wage norm. These include the highly skilled professional groups (scientists, engineers and doctors) and skilled trades workers.
D. Male/Female Relative Wages
An overall evaluation of comparable worth is likely to involve consideration of the way that gains and losses are shared among workers. Estimates of the percentage wage changes for males and females under each scenario, and of the male/female wage ratios for the SLG sector and for the whole economy are in Table 8. These estimates show that the wage adjustments under comparable worth are heavily weighted in favor of females, although the relative share between males and females differs across scenarios. It is of special interest that the relative share of males under Scenario IIIb (where none of the residual wage variation is purged from the comparable worth wage) is greater than it is under Scenario IIIa (where the residual wage variation is removed). This suggests that if a substantial portion of the residual wage variation is retained by an actual comparable worth policy, then Scenarios I, II, and IIIa (all of which purge the residual from the wage after comparable worth) may overstate the female share of wage gains.
A reduction in the overall male/female wage ratio shown in Table 8 implies a reduction in the degree of wage discrimination against women. Following Smith (1988), the amount of gender-based wage discrimination is defined as that part of the total male/female wage differential that cannot be explained by male/female differences in compensable attributes. The total male/female differential, T = ((|W.sub.m~ - |W.sub.f~)/|W.sub.f~) TABULAR DATA OMITTED comprises an explained portion, E = (W* - |W.sub.f~)/|W.sub.f~, and a residual attributable to discrimination, R = T - E = (|W.sub.m~ - W*)/|W.sub.f~, where W* is the wage that would be received by men if their attributes were valued in the same way as women's attributes are (that is, predicted from the female wage equation). The ratio E can be interpreted as the male/female wage differential that would exist if male and female attributes were valued equally. Wage equations estimated for females in the SLG sector and for the whole economy were used to arrive at estimates of W*, and the amount of wage discrimination in each sector before comparable worth. Assuming that E remains constant under comparable worth, the amount of gender-based wage discrimination under each scenario is computed as |R.sub.cw~ = |T.sub.cw~ - E, where |T.sub.cw~ is the total male/female wage differential under comparable worth.
Table 8 shows that comparable worth would produce sizable reductions in the overall male/female wage differential and in the estimated amount of gender-based wage bias in the SLG sector. This is not surprising, given that women receive much larger wage increases than men do. A comparable worth policy confined to the SLG sector would not have a major impact on male/female wage differences in the economy as a whole--due to the relatively small scale of SLG employment.
Within the SLG sector, the degree to which male/female differentials are reduced varies across scenarios. In all cases, the differential attributable to discrimination is at least halved, and under Scenario IIIa, estimated wage discrimination is eliminated. However, if comparable worth methodology is more likely to resemble IIIb than IIIa, the most reasonable conclusion is that the male/female wage gap due to discrimination may be reduced by up to about two-thirds. The percentage reduction in male/female wage bias in the SLG sector is much larger than that found by Johnson and Solon (1986) for a comparable worth policy covering the whole economy. But this discrepancy is to be expected if it is true that SLG comparable worth programs would eliminate some parts of "interindustry" wage differentials, whereas private-sector plans would not.
VI. Caveats and Conclusion
This study provides estimates of the outcomes if SLG employers were to fully implement a comparable worth program along the lines of one of the policy models proposed in the literature. Consequently, the estimates should not be interpreted as the best guess at the result of any announced intention to implement comparable worth. As Ehrenberg (1989) and Orazem and Matilla (1990) point out, political pressures are likely to mean that any program that is adopted will deviate from the idealized model, and the actual wage changes may be smaller and not as favorable to women workers. Indeed, because the different ways of adjusting relative wages and of funding the program can produce different results, the type of comparable worth policy to be adopted is likely to be a subject of contention.
The evidence shows that the implications that comparable worth has for compliance with the prevailing wage principle depend on the way the program is funded. Comparable worth programs that are implemented through special wage increases would entail--at least in the short term--movement away from the prevailing wage norm. This is because payroll budgets would be larger than what would be required to ensure that all public workers were paid wages equal to what they would receive for similar work in the private-sector.
When comparable worth programs are implemented by redistributing the current payroll budget, however, public wages will be at least as close to conforming to the prevailing wage standard, and perhaps closer. Under those scenarios that remove all wage variations that are not reflected by the variables in the wage equation, public wages after comparable worth will come closer to the prevailing wage norm. Under those scenarios where all sources of residual wage variation are allowed to remain, the reduction in the degree to which public wages deviate from the prevailing wage is less, but in no case is the degree of deviation increased. These results, together with the finding that comparable worth tends to provide bigger wage adjustments to occupations where public/private relative wages before comparable worth were relatively low, may be encouraging to the advocates of comparable worth.
The findings present comparable worth proponents with an interesting dilemma. It is possible to introduce comparable worth in a way that does not further distort wages relative to the standards set in private-sector markets, but this requires that budgets be constrained to their current levels; therefore, wage increases in those occupations receiving them will have to be paid for by wage reductions in others.
1. As shown in Gunderson (1989), the ratio of the hourly earnings of full-time female workers relative to full-time male workers has remained constant over much of the post World War II period. There is evidence that the male/female earnings gap adjusted for individual differences is declining (O'Neil 1985, Blau and Beller 1988), but a substantial differential that cannot be explained by male/female differences in observed attributes remains.
2. One set of studies, including Aldrich and Buchele (1986), Buchele and Aldrich (1985), and Johnson and Solon (1986), examined the effects of a national comparable worth policy on economy-wide wage differentials. Others, such as Ehrenberg and Smith (1987a), Evans and Nelson (1989), Orazem and Mattila (1989), and Sorensen (1986, 1987) assumed comparable worth to be confined to the government sectors and examined the effects on wage differentials among public workers. Empirical analyses of comparable worth have been reviewed by Ehrenberg (1989).
3. For example, Blumrosen (1979), Nelson et al. (1980).
4. Comparable worth initiatives by state and local governments have been summarized by Ehrenberg and Smith (1987b).
5. It is assumed that private-sector wages do not change as a result of changes in public wages. Although private-sector adjustments are theoretically possible, it should be noted that SLG workers account for only 13 per cent of employment in the typical labor market and cover many different occupations. Thus private-sector wage adjustments in response to public-sector wage changes are likely to be very small.
6. In the simulations by Aldrich and Buchele (1985), and Sorensen (1987), for example, all relative wage adjustments are achieved by payment of nominal wage increases to workers in previously under-valued occupations.
7. The term, |Mathematical Expression Omitted~ is used to adjust for the biases in the prediction of ln |W.sub.r~ that are due to the customary log-normal distribution of the dependent variable in the semi-logarithmic wage equation. (See, for example, Lee 1978, pp. 426-27).
8. When the dependent variable is defined in arithmetic terms, as in Treiman and Hartmann (1981), the properties of the least squares estimates ensures that the average public wage level remains unchanged and thus all wage increases are paid for by wage reductions on those jobs represented by data points that lie above the fitted regression.
9. The tendency for the dispersion to be underestimated under these circumstances has a limit, for it is possible to reach the point where the degrees of over(under)prediction of the wage adjustments for occupations with low(high) public/private pay will be so great that the estimated dispersion will begin to increase. Thus, the degree of bias in estimates of public/private occupational relative wages can not be determined a priori.
10. A regression of the initial estimates of |F.sub.u~ on the Census Bureau measures of proportion female (PFEM) and its square estimated for all SLG occupations with a minimum CPS cell size of 80 yielded the following equation (t-values in parentheses): |F.sub.u~ = 1.5064.PFEM (17.34) - .55645.PFE|M.sup.2~ (5.21). |R.sup.2~ = .98, n = 59. The model was used to predict the final estimates of |F.sub.u~ in all occupations. The omission of an intercept term from the regression stems from the logic that the absence of females among all private and public workers in a given occupation (PFEM = 0) implies there are no female SLG workers in that occupation. Moreover, the coefficient estimates which indicate that |F.sub.u~ will be higher (lower) than PFEM for lower (higher) values of PFEM is consistent with the idea that the greater emphasis given to affirmative action by public employers has produced a more balanced distribution of males and females among occupations in the public sector.
11. The estimated coefficients (with standard errors in parentheses) were -.328 (.026) from the original specification, -.321 (.026) where age, age-squared, a dummy for the presence of children, and number of children were included in the place of the potential experience measures, and -.305 (.027) when the tenure variable was added to the latter specification.
12. For example, Sorensen's (1990) estimates of the proportion female coefficient drops from about -.27 to -.20 when 42 industry dummies are added.
13. See, for example, Johnson and Solon (1986) Sorensen (1990).
Aaron, Henry J., and Cameran M. Lougy. 1986. The Comparable Worth Controversy. Washington D.C.: The Brookings Institution.
Aldrich, Mark, and Robert Buchele. 1986. The Economics of Comparable Worth. Cambridge, Mass.: Ballinger.
Blau, Francine, and Andrea Beller. 1988. "Trends in Earnings Differentials by Gender, 1971-1981." Industrial and Labor Relations Review 41(4):513-29.
Blumrosen, Ruth G. 1979. "Wage Discrimination, Job Segregation, and Title VII of the Civil Rights Act of 1964." University of Michigan Journal of Law Reform 12(3):379-502.
Bridges, William P., and Robert L. Nelson. 1989. "Markets in Hierarchies: Organizational and Market Influences on Gender Inequality in a State Pay System." American Journal of Sociology 95(3):616-58.
Buchele, Robert, and Mark Aldrich. 1985. "How Much Difference Would Comparable Worth Make?" Industrial Relations 24(2):222-33.
Ehrenberg, Ronald G. 1989. "Empirical Consequences of Comparable Worth." In Comparable Worth, Analyses and Evidence, ed. M. Anne Hill, and Mark R. Killingsworth, 90-106. Ithaca, N.Y., ILR Press.
Ehrenberg, Ronald G., and Robert S. Smith. 1987a. "Comparable-Worth Wage Adjustments and Female Employment in the State and Local Sector." Journal of Labor Economics 5(1):43-62.
-----. 1987b. "Comparable Worth in the Public Sector." In Public Sector Payrolls, ed. David A. Wise, 243-88. Chicago: University of Chicago Press.
Evans, Sarah M., and Barbara J. Nelson. 1989. Wage Justice: Comparable Worth and the Paradox of Technocratic Reform. Chicago: University of Chicago Press.
Fogel, Walter, and David Lewin. 1974. "Wage Determination in the Public Sector." Industrial and Labor Relations Review 27(3):410-31.
Freeman, Richard B. 1987. "How Do Public Sector Wages and Employment Respond to Economic Conditions?" In Public Sector Payrolls, ed. David A. Wise, 183-207. Chicago: University of Chicago Press.
Gunderson, Morley. 1989. "Male-Female Differentials and Policy Responses." Journal of Economic Literature 26(1):46-72.
Hundley, Greg. 1991. "Public- and Private-Sector Occupational Pay Structures." Industrial Relations 30(3):417-34.
Johnson, George, and Gary Solon. 1986. "Estimates of the Direct Effects of Comparable Worth Policy." American Economic Review 76(5): 1117-25.
Lee, Lung-Fei. 1978. "Unionism and Wage Rates: A Simultaneous Equations Model with Qualitative and Limited Dependent Variables." International Economic Review 29(3):415-33.
Megdal, Sharon Bernstein. 1986. "Comparable Worth: Some Issues for Consideration." Contemporary Policy Issues 4(1):40-51.
Miller, Ann R., Donald J. Treiman, Pamela S. Cain, and Patricia A. Roos. 1980. Work, Jobs and Occupations: A Critical Review of the Dictionary of Occupational Titles. Washington D.C.: National Academy Press.
Milkovich, George T., and Jerry M. Newman. 1990. Compensation. Plano, Tex.: BPI.
Mincer, Jacob, and Solomon Polachek. 1974. "Family Investments in Human Capital: Earnings of Women." Journal of Political Economy 82(2):S76-S108.
Moore, William J., and John Raisian. 1991. "Government Wage Differentials Revisited." Journal of Labor Research 12(1):13-34.
Nelson, Bruce A., Edward M. Opton, Jr., and Thomas E. Wilson. 1980. "Wage Discrimination and Title VII in the 1980s: The Case Against 'Comparable Worth'." Employee Relations Law Journal 6(3):380-45.
O'Neill, June. 1985. "The Trend in the Male-Female Wage Gap in the United States." Journal of Labor Economics 3(1):S91-S116.
Orazem, Peter F., and J. Peter Mattila. 1989. "Comparable Worth and the Structure of Earnings: The Iowa Case." In Pay Equity: Empirical Inquiries, ed. Robert T. Michael, Heidi I. Hartmann, and Brigid O'Farrell, 179-99. Washington, D.C.: National Academy Press.
Orazem, Peter F., and J. Peter Mattila. 1990. "The Implementation of Comparable Worth: Winners and Losers." Journal of Political Economy 98(1): 134-52.
Smith, Robert S. 1988. "Comparable Worth: Limited Coverage and the Exacerbation of Inequality." Industrial and Labor Relations Review 41(1):227-39.
Smith, Sharon P. 1977. Equal Pay in the Public Sector: Fact or Fantasy? Princeton: Princeton University Press.
-----. 1983. "Are State and Local Government Workers Overpaid?" In The Economics of Municipal Labor Markets, ed. Werner Z. Hirsch and Anthony M. Rufolo, 59-89. Los Angeles: UCLA Institute of Industrial Relations.
Sorensen, Elaine. 1986. "Implementing Comparable Worth: A Survey of Recent Job Evaluation Studies." American Economic Review 76(2):364-67.
Sorensen, Elaine. 1987. "Effects of Comparable Worth Policies on Earnings." Industrial Relations 26(3):227-39.
Sorensen, Elaine. 1990. "The Crowding Hypothesis and Comparable Worth." Journal of Human Resources 25(1):55-89.
Treiman, Donald J., and Heidi I. Hartmann. 1981. Women, Work, and Wages: Equal Pay for Jobs of Equal Value. Washington, D.C.: National Academy Press.
Treiman, Donald J., Heidi I. Hartmann, and Patricia A. Roos. 1984. "Assessing Pay Discrimination Using National Data." In Comparable Worth and Wage Discrimination: Technical Possibilities and Political Realities, ed. Helen Remick, 137-54. Philadelphia, Pa.: Temple University Press.
U.S. Bureau of the Census. 1983. "Detailed Occupation and Years of School Completed by Age for the Civilian Labor Force by Sex, Race, and Spanish Origin: 1980." PC80-S1-8. Washington, D.C.: GPO.
U.S. Department of Labor. 1977. Dictionary of Occupational Titles, fourth edition. Employment and Training Administration. Washington: GPO.
The author is an associate professor of management at the University of Oregon. The very useful comments of two anonymous referees are gratefully acknowledged. Data used in this study can be obtained beginning in December 1993 through December 1996 from the author at: Graduate School of Management, University of Oregon, Eugene, OR. 97403.
|Printer friendly Cite/link Email Feedback|
|Publication:||Journal of Human Resources|
|Date:||Mar 22, 1993|
|Previous Article:||Do community-based, long-term-care services reduce nursing home use? A transition probability analysis.|
|Next Article:||Gender differences in training, capital, and wages.|