Discrimination by gender and disability status: do worker perceptions match statistical measures?

• Author: Hallock, Kevin F.

1. Introduction

   This paper is motivated by the growing literature on measured gender discrimination and the study of persons with disabilities. We use a survey of university graduates to explore whether female (or male) perceptions of gender discrimination are related to ordinary statistical measures of discrimination, and whether persons with disabilities perceive disability discrimination as it is ordinarily measured.

   Kuhn (1987, 1990) studies two data sets from Canada and the United States. He finds that there is a negative and insignificant overall correlation between reports and measures of discrimination, which is largely driven by the fact that young well-educated women are both more likely to report discrimination and face the smallest measured wage gaps.

   Barbezat and Hughes (1990) study discrimination and perceptions of discrimination using a sample of university faculty. They find evidence consistent with Kuhn (1987) but suggest an interesting interpretation for the findings of a negative correlation between measured discrimination and the chances that women report being discriminated against. They write that "Employers are more likely to indulge in discrimination when there is a low probability that such action will be detected. Thus, a high level of measured discrimination will be accompanied, in equilibrium, by a small number of reports and vice versa." We suggest an additional explanation that is also consistent. Perhaps those who perceive discrimination feel that it occurs along other dimensions than pay, such as promotion or training opportunities.(1)

   Hampton and Heywood (1993) is similar to Kuhn (1987), but instead of studying the relationship between wage residuals and perceptions of discrimination, they study how much a woman feels underpaid. They estimate the relationship between perceived underpayment and measured discrimination. Somewhat contrary to Kuhn (1987), they find "a strong, positive correlation between women's perceptions of gender income differences they were experiencing and econometric estimates of those differences."

   For part of this work, we will follow Kuhn (1987) and Hampton and Heywood (1993) and explore female perceptions of underpayment and discrimination and study the connection between these and measured discrimination. Our paper also adds the study of the disabled following Hendricks, Schrio-Geist, and Broadbent (1997) (hereafter HSB) who examine wage discrimination for a sample of disabled and nondisabled graduates from the University of Illinois. Below, we use some of the data from HSB to study the connection between perceptions of discrimination and measured discrimination for the disabled (and for nondisabled men and women).

   Our paper is unique in at least three ways. First, it documents the frequencies of perceptions of discrimination for many types of workers. Second, it studies perceptions of "discrimination" and of underpayment on the same sample of workers. Finally, it studies these issues not only for nondisabled men and women but for the disabled as well. We have several findings. First, we report the frequency of perceptions of discrimination for several groups of workers. We find that the majority of the disabled respondents report feeling at least some discrimination due to their disability and that the majority of women report feeling at least some discrimination because of their gender. We also find that a surprising number of men report at least some gender discrimination (14%). However, we do not find a strong link between perceptions of discrimination and measured discrimination (whether it be gender discrimination against women, gender discrimination against men, or disability discrimination against the disabled). The findings for gender are consistent with the findings of Kuhn (1987) reported above. One explanation for this result is that there is evidence consistent with the fact that those who perceive discrimination feel that it occurs along other dimensions than pay. Second, and consistent with Hampton and Heywood (1993), we find there is a connection between whether an individual feels his or her income is inadequate and measured discrimination.(2) This is true for each of our main groups: men, women, and the disabled.

   In section 2 we describe the survey and explore the data. In section 3 we run a series of simple least-squares wage regressions by group (all, men, women, and disabled) to study wage determinants by demographic group. Section 4 more carefully describes what we mean by measured discrimination, outlines and implements the empirical strategy, and documents the link between perceptions of discrimination and measured discrimination. Section 5 studies issues of timing of discrimination. Section 6 examines sample selection and whether we have erroneous results due to studying only currently employed persons. Given that we and Kuhn (1987) find such a weak link between perceptions of discrimination and measured discrimination, we study nonwage types of discrimination in section 7. Section 8 concludes.

2. Data

   The University of Illinois at Urbana-Champaign has had a Division of Rehabilitation Education Services since 1948. Since then, approximately 1200 students who used the Division have graduated from the University.(3) The data for this study are from a seven-page survey conducted during the winter of 1993 that initially examined 865 of the 1200 graduates for whom addresses were located. Valid questionnaires were returned by 301 of the graduates of the program for a response rate of 35%. Because each disabled respondent was classified by primary field of study, gender, and year of graduation, we created a stratified sample of nondisabled students using records from the alumni office. Five matches were drawn for each of the 301 returned questionnaires from the disability group and 1505 questionnaires were sent to nondisabled graduates of the University. Valid questionnaires were returned from 339 of the nondisabled sample for a response rate of 23%.

   Table 1 outlines summary statistics for many of the variables we study. There are several significant differences worthy of note. Women earn slightly more than 60% of what men earn ($39,074 vs. $64,838) and, on average, the disabled earn significantly less than the nondisabled ($51,883 vs. $60,772). Although men and women have similar years of experience prior to the current job, women have significantly less seniority in the current job.(4) On the other hand, the disabled have significantly less precurrent-job experience than the nondisabled but have similar current seniority. Table 1 also documents that the disabled are more likely to report that a disability limits the amount of work they can do,(5) are less likely to be married, and are more likely to work in the public sector than the nondisabled. Also, women are more likely to work part-time(6) and in the public sector than men, but men are more likely to be married and self-employed.

3. Wage Determinants by Group

   As a first step to studying the effects of measured discrimination on discrimination perceptions, we follow Kuhn (1987) by exploring the wage structure for each of the demographic [TABULAR DATA FOR TABLE 1 OMITTED] groups. In our empirical model of wages, we regress the log of annual compensation on a set of characteristics:

   [w.sub.i] = [summation over j] [[Beta].sub.j][X.sub.ij] + [[Epsilon].sub.i]. (1)

   Table 2 documents some of the results from this analysis. The table is arranged in eight columns, two each for all respondents, men, women, and the disabled. Within each of the four categories we present results from two regressions. In the odd-numbered columns the independent variables, X, include four indicator variables for educational attainment, M.A., Ph.D., J.D., M.D.; the number of years of experience prior to the current job; the number of years of seniority on the current job, and its square; an indicator for whether the person is white; variables indicating whether the individual is male, married, a public sector employee, self-employed, a part-time worker, or has a health limitation that limits the amount of work he or she can do; the months since graduation from college; and a disability indicator. The even-numbered columns include these variables (except for the disabled indicator) but also include indicator variables [TABULAR DATA FOR TABLE 2 OMITTED] of the primary disability for all persons with a disability (their functional limitation groups or FLGS).(7)

   Columns 1 and 2 present results for the full sample. Many of the results in Table 2 are consistent with those from HSB. Several of the advanced degree indicator variables are positive and significant as expected.(8) The experience variables also yield expected signs. Those with more prejob experience earn more and those with longer current-job seniority earn more but at a declining rate. Women, part-time workers, the self-employed, and public employees all earn less than their respective counterparts. Finally, the point estimate on the indictor for disabled is -0.065 but is not significant. HSB find a point estimate of -0.087 on disabled in their baseline regression. Although we expect the results to be consistent with HSB, we should not expect the results to be identical for two reasons. The first is that HSB use several years of data on each individual. We, however, only study the most recent job for each respondent. The second difference is that we select individuals who have valid responses to many of the discrimination perception questions that are the focus of this paper and were not explored in HSB.(9)

   Column 2 of Table 2 controls for the 12 FLGS. As in HSB, this does not change any of the estimates reported in column 1 in any meaningful way. All variables that were significant remain so after controlling for the FLGS. The remainder of Table 2 reports regression results for men, women, and the disabled separately. The results for these...