Labor Market Discrimination Against Men with Disabilities in the Year of the ADA.

Johnson, William G.

Marjorie L. Baldwin [*]

William G. Johnson [+]

The Americans with Disabilities Act (ADA) provides civil rights protections to persons with disabilities, but the debate that preceded passage of the Act was not based on empirical estimates that could be used to measure its performance. This article estimates the extent of wage discrimination against men with disabilities in 1990, providing a reference that can be used to evaluate the impact of the ADA. The results show large productivity-standardized wage differentials between disabled and nondisabled men that are weakly correlated with the strength of prejudice against different impairments. Physical limitations explain part, but not all, of the wage differentials. The results also show that low employment rates are a more serious problem than wage discrimination for workers with disabilities.

1. Introduction

The Americans with Disabilities Act of 1990 (ADA) is the first federal legislation to provide civil rights protections to persons with disabilities. The Act's supporters promised that it would increase the employment and wages of persons with disabilities and reduce their dependency on public programs. The promises were not based on empirical estimates of discrimination, and there is no empirical base against which to compare the promises to the performance of the ADA. This article estimates the extent of wage discrimination against men with disabilities in 1990, providing a reference that can be used to evaluate the impact of the Act.

The ADA does not grant a right to employment to all persons with disabilities. It requires that persons with disabilities be able to perform the jobs they seek, subject to "reasonable accommodations" by employers. This article, therefore, considers men who are able to work even though their health limits the amount or kind of work they can perform.

The ADA's provisions assume that low wages and low employment rates for persons with disabilities are the result of discrimination caused by prejudice. Prejudice against persons with disabilities is well documented, but so are the limiting effects of physical impairments on workplace productivity (Hahn 1987; Margolis and Shapiro 1987; Clogston 1990). The core of the debate concerning the value of the ADA is whether the poor labor market outcomes of persons with disabilities are the result of discrimination or the limiting effects of health conditions. The relative importance of the two effects is a question that this article answers, within the limits of the methods used to study discrimination.

The fundamental assumption of the discrimination argument is that employers incur disutility if they hire persons against whom they are prejudiced (Becker 1971). We use measures of negative attitudes toward persons with disabilities to test the influence of prejudice on the size of productivity-standardized wage differentials. The measures are from recent studies that use preferences for social distance to measure attitudes, similar to the concepts Becker (1971) used to represent preferences for discrimination.

We estimate wage discrimination by decomposing between-group differences in mean offer wages into a part attributed to differences in productivity and a part attributed to discrimination and residual effects (Oaxaca 1973; Reimers 1983; Cotton 1988). Some of the elements used to control for differences in productivity between disabled and nondisabled workers are health-related limitations on physical functioning, such as lifting and carrying. Although the decomposition method has been used by most studies of discrimination, its estimates of wage discrimination include nondiscriminatory differences in wages, if significant variables are omitted from the wage functions. We address the problem in part by estimating alternative specifications of the wage functions and using the alternative results to represent a range of estimated discrimination effects.

Wage discrimination affects employment as well as wages. If labor supply curves are upward-sloping, some men who would have worked at a nondiscriminatory wage will not accept a lower discriminatory wage offer. We use Baldwin and Johnson's (1992b) estimator to measure the disincentive effects of wage discrimination on the employment of men with disabilities.

The results show that there are large productivity-standardized wage differentials between disabled and nondisabled men. The wage differentials are weakly correlated with the strength of negative attitudes toward different health conditions. Differences in physical limitations are an important factor explaining disabled-nondisabled wage differentials, but controlling for differences in physical limitations does not eliminate the wage differences.

The results also indicate that low employment rates are a more serious problem than wage discrimination for workers with disabilities. The disincentive effects of wage discrimination account for only a small part of the difference in employment rates between disabled and nondisabled men.

2. Definitions

Disability refers to activities, such as working, rather than attributes, such as gender or race. It is important to distinguish "disability" from "impairment" and "functional limitation," two terms that are often used incorrectly as synonyms for disability.

An impairment is a "physiological or anatomical loss or other abnormality." An impairment may or may not cause a functional limitation, that is, a restriction of sensory, mental, or physical capacities. A disability occurs when functional limitations restrict the ability to perform activities such as working or attending school. [1]

Consider, for example, a worker with a neurological impairment, such as epilepsy. Epilepsy causes a functional limitation, namely, the inability to walk and perform physical tasks during severe seizures. Seizures can restrict the person's ability to work, creating a work disability. If seizures are controlled by medication, a typical situation for most persons with epilepsy, the worker's job performance is not affected and the worker is not disabled. Workers with epilepsy are, however, still subject to discrimination.

Economic discrimination occurs when groups of workers with equal average productivity have different mean offer wages or different opportunities for employment. Discrimination in employment can be expressed as refusals to hire, differentially high rates of job terminations, or refusals to rehire workers following work absences. We use data from the 1990 Survey of Income and Program Participation (Bureau of the Census 1992) to estimate wage discrimination against men with disabilities, and the disincentive effects of wage discrimination on employment. The next section describes the data and methods.

3. Data and Methods

The data come from Wave III of the 1990 panel of the SIPP. Wave III includes questions on health and disability, including 18 functional limitations. One or more of 29 impairments are listed as the cause of the limitations. We define a person with a disability as someone with health-related limitations on work.

Defining Comparison Groups

We use Royal and Roberts' (1987) and Westbrook, Legge, and Pennay's (1993) studies of attitudes toward disabilities to rank the 29 impairments on our SIPP data by the intensity of prejudice they elicit. Each study uses social distance scales to rank attitudes from most acceptable to least acceptable. [2] The Royal and Roberts' study is older but includes a measure of visibility, which is an important influence on the extent of discrimination. [3] The Westbrook study is more recent and includes attitudes toward some conditions on the SIPP that are omitted by Royal and Roberts. The rankings of attitudes that appear in both studies are highly correlated, permitting us to take advantage of the unique features of each. [4]

The SIPP samples are too small to analyze each of the 29 impairment groups. Instead, we classify men with disabilities into two groups, namely, men with impairments that are less visible or subject to less prejudice (LP) and men with impairments that are visible and subject to more prejudice (MP). The groups are described in Table 1.

The scores for each impairment group are mean ratings from Royal and Roberts' (1987) survey, truncated to integer values. The Westbrook, Legge, and Pennay (1993) mean social distance ratings are reported in parentheses whenever they differ from the Royal and Roberts results. The Westbrook scores are used for heart disease, stroke, alcoholism, and AIDS, which are omitted from the earlier study.

The LP (less prejudice) group includes all conditions that are unranked on the attitudes studies, and six conditions with high acceptability rankings and/or low visibility rankings. The MP (more prejudice) group includes 13 conditions, most with acceptability rankings of 3 or less and visibility rankings of 2 or more. The exceptions are cancer, speech disorder, and deafness, which have visibility rankings of 1 (not at all visible). We classify these impairments with the MP group because an employer is likely to have knowledge of the impairment even though it is not literally visible. [5]

The sample consists of 11,708 nondisabled men (ND), 662 men with impairments subject to less prejudice (LP), and 240 men with impairments subject to more prejudice (MP). [6] The samples represent populations of approximately 54 million (ND), 3.1 million (LP), and 1.1 million (MP) men.

The use of attitude studies to test for the effect of prejudice on discrimination toward minority workers is unusual. It is surprising, however, that studies of discrimination never test the fundamental assumption that prejudice, or "tastes for discrimination," is the cause of discrimination. The power of the two comparison groups to test for links between preferences and discrimination is limited, but we hope it encourages others to develop better tests.

One of the important differences between disabled workers and other protected classes is the fact that productivity can be limited by the same health conditions that make disabled workers eligible for civil rights protection. The next section describes the methods we used to estimate the effects of health-related limitations on productivity.

Controlling for the Effects of Functional Limitations on Productivity

An ideal approach to measuring the effect of functional limitations on productivity would be to create a profile of the physical functions required to perform a job and then compare the profile to a worker's abilities to perform the required functions. Worker-job matching systems are sometimes used by vocational rehabilitation agencies and insurers but profiles of job demands are not included in the SIPP. We must, therefore, rely on the profiles of workers' functional limitations alone.

The SIPP data indicate, for each of 18 functions, whether a person has no difficulty performing the function, has some difficulty performing the function, or is unable to perform the function at all. [7] The responses to the limitation questions are appropriate measures of the effect of impairments on a person's ability to work, but there is uncertainty concerning the best method of combining individual limitations to represent the total effect of impairments on productivity. [8] Because of the uncertainty, we estimate two models, namely: a model in which a set of dummy variables controls for functional limitations, and a model in which limitations are measured by a set of factors created from principal components analysis. [9]

Dummy Variable Specification

We construct two dummy variables for each of 14 functional limitation categories. [10] The first dummy equals one if an individual has difficulty performing the function (otherwise zero) and the second equals one if the individual cannot perform the function (otherwise zero). Seven limitations are included in the models in the form of these two dummy variables. The limitations are: climbing, hearing, lifting, seeing, walking, getting in and out of bed, and personal care. The remaining limitations collapse to single dummy variables (where 0 equals no difficulty and 1 equals has difficulty or cannot perform the function) because there are some subgroups with no observations in the "cannot perform" category. These limitations are: bathing, speaking, doing housework, getting around, handling money, preparing meals, and using the phone.

The primary advantage of the binary variable specification is that the estimated coefficients are cardinal measures of the effects of limitations on wages and employment. The disadvantage is that the 21 binary variables are highly correlated, so the estimated coefficients tend not to be statistically significant.

Factor Analytic Specification

An alternative specification uses factor analysis to combine the individual limitations (Johnson and Lambrinos 1985; Baldwin and Johnson 1994, 1995). Different limitations can be highly correlated because one impairment, such as arthritic degeneration of the joints, can limit several functions (e.g., climbing, lifting, walking).

The results, shown in Appendix A, are intuitively appealing. The variables loaded most heavily on the first factor, which we interpret as a measure of mobility, include: "getting around inside and outside," "using an aid to get around," and the personal care variables. Variables loaded most heavily on the second factor, a measure of strength and endurance, include, "lifting and carrying," "walking a short distance," and "climbing stairs." Variables loaded most heavily on the third factor, a measure of cognitive and sensory limitations, include hearing, seeing, speaking, using the telephone, and handling money and bills. We use the variable loadings to construct three composite factors that are continuous, increasing measures of functional limitations. The three factors are included as control variables in the second specification of the wage and employment functions.

The advantage of the factor approach is that it allows correlations among the data to determine how the limitations variables will be combined. The primary disadvantage of the approach is that it is impossible to interpret the regression coefficients as cardinal measures.

Estimating Wage Discrimination

Log linear wage equations of the form

ln [W.sub.i] = [beta][X.sub.i] + c[[lambda].sub.i] + [[epsilon].sub.i] (1)

are estimated separately for the ND, LP, and MP groups. The wage equations are estimated for the 10,928 men who were employed at some time between June and December 1990.

The dependent variable in Equation 1 is the natural log of the hourly wage rate of the ith worker; [X.sub.i] is a vector of variables representing worker productivity and labor market characteristics; [[epsilon].sub.i] is a mean-zero, random disturbance term. [11] The sample selection variable, [[lambda].sub.i], is included to correct the bias that results because offer wages are not observed for nonworkers. Productivity is represented by years of education, the functional limitations variables, and three measures of work experience. [12] Labor market variables include union membership, race, and two occupational categories (professional or manager, laborer).

Substituting means and coefficient estimates of the variables in the wage equation into Equation 2, we obtain an expression for the offer wage differential between ND and LP men:

ln [W.sub.ND] - ln [W.sub.LP] - ([c.sub.ND][[lambda].sub.ND] - [c.sub.LP][[lambda].sub.LP]) = ([X.sub.ND] - [X.sub.LP])[d[[beta].sub.ND] + (1 - d)[[beta].sub.LP]] + [[X.sub.ND] (1 - d) + [X.sub.LP]d]([[beta].sub.ND] - [[beta].sub.LP]), (2)

with a similar decomposition for the nondisabled (ND) and more prejudice (MP) groups. The left side of Equation 2 represents the difference between mean offer wages for disabled and nondisabled workers. The first term on the right represents the difference in offer wages attributable to differences in average productivity, as represented by means of the variables in the wage equation; the second term represents the unexplained part of the wage differential attributed to discrimination and residual effects.

The second term cannot be decomposed into discriminatory and nondiscriminatory components because the difference in intercepts varies with differences in the units of measurement (Jones 1983). Thus, the second term is an unbiased measure of discrimination only if no significant variables are omitted from the wage equations and all variables are accurately measured. Our estimates must be interpreted with this constraint in mind. [13]

The weight d, valued from 0 to 1, represents the relationship of the nondiscriminatory wage structure to observed wages. We assume the nondiscriminatory wage structure is the observed wage structure for nondisabled men (d = 1) because workers with disabilities are a small fraction of the employed labor force. [14]

Estimating the Employment Effects of Wage Discrimination

Most studies of wage discrimination assume that labor supply functions are perfectly inelastic, implying that discrimination affects wages but not employment (Cain 1986). In reality, wage discrimination reduces minority employment because some workers who would accept a nondiscriminatory wage will not work at the lower, discriminatory wage offer. [15] We use the Baldwin--Johnson (1992a) technique to estimate these "disincentive effects of wage discrimination" on the employment of men with disabilities. The technique imposes no a priori assumptions about the elasticity of the labor supply curves.

The observed probability of employment for the average man in the LP or MP group, [II.sub.j], is derived from probit estimates of the likelihood function

log L = [[sigma].sub.i[epsilon]E] log F([gamma]'/[sigma][Z.sub.i]) + [[sigma].sub.i[epsilon]E] log[1 - F([gamma]'/[sigma][Z.sub.i])], (3)

where i [epsilon] E if a man is employed and i [epsilon] E otherwise. In Equation 3, F is the standard normal cumulative distribution function, [Z.sub.i] is a vector of variables that determine offer wages and reservation wages, and [gamma]/[sigma] is a corresponding vector of coefficients estimated separately for the LP/MP groups. Independent variables in the employment function include age, education, race, functional limitations, marital status, and six sources of nonwage income.

Using the coefficient estimates and means of the independent variables in Equation 3, the observed probability of employment is

[[pi].sub.j] = F[([gamma]/[sigma])Z], j = LP, MP, (4)

where ([gamma]/[sigma])Z represents the difference in the log offer wage and log reservation wage of the average LP/MP man, standardized by the adjustment factor [sigma].

Probabilities of employment in the absence of discrimination, [[pi].sup.*].sub.j], are obtained by correcting the log offer wage-reservation wage differential, ([gamma]/[sigma])Z, for the effect of discrimination in offer wages. Let ln [[W.sup.o*].sub.j] represent the mean log offer wage to LP/MP men in the absence of discrimination (computed by substituting the mean characteristics of the LP/MP group into the offer wage function for nondisabled men). Then,

[[pi].sup.*].sub.j] = F([gamma]/[sigma])Z - (ln[[W.sup.o*].sub.j] - ln[[W.sup.0].sub.j])/[sigma]. (5)

The number of jobs lost to men in the LP/MP group because of wage discrimination is estimated by multiplying the discriminatory differential in the probability of employment for the average man by the size of the group, [N.sub.j]. The loss of employment, [E.sub.j], is given by

[E.sub.j] = ([[[pi].sup.*].sub.j] - [[pi].sub.j])[N.sub.j]). (6)

The estimator of employment effects assumes that employers are free to express their tastes for discrimination as low offer wages and that the unexplained difference in offer wages measured by the wage decompositions is entirely attributed to employer wage discrimination. [16] To the extent that employers' wage offers are constrained by minimum wage floors or that important variables have been omitted from the wage equation, our estimates may understate or overstate the employment effects.

4. Results

Wage and Employment Functions

Means and estimated coefficients of the employment and wage functions are reported in Appendices B and C. Considering the employment function, the functional limitations variables are negative and significant in four of nine occurrences in the factor model. In the binary model, the functional limitations variables are significant in only one-fourth of occurrences and occasionally have a counterintuitive sign reflecting the collinearity among the variables. The coefficients of the non-health-related variables in the employment function have the expected signs and satisfy the usual significance tests. Likelihood ratio tests show that the structure of the employment function is significantly different for each of the three groups. [17]

In the wage equations, the functional limitations variables are significant in only one of nine occurrences in the factor model and in only eight of 63 occurrences in the binary model. Other variables in the wage equation are generally significant and have the expected signs. The wage model for nondisabled men is significantly different from the models for LP and MP men, but there is no significant difference between the models for the two disabled groups. [18]

The finding that functional limitations are more often significant as influences on employment than on wages is consistent with previous research (Reimers 1983; Johnson and Lambrinos 1985). The previous studies interpret the result to suggest that the primary effect of physical limitations is as an obstacle to employment. Persons with disabilities who obtain employment have convinced employers that they are physically capable of meeting job requirements. Thus, the effect of functional limitations on wages is, the studies suggest, likely to be small. Their interpretation is consistent with our results, but it cannot be tested empirically in this study.

Wage Discrimination

Average hourly wages are $13.35 for nondisabled men, $11.13 for men in the LP group, and $10.90 for men in the MP group. The observed wage differentials are converted to offer wage differentials by adjusting for sample selection bias. The offer wage differentials are then decomposed into a part attributed to between-group differences in productivity-related characteristics and an unexplained part attributed to discrimination and residual effects. [19] The offer wage differential between nondisabled and LP (MP) men is approximately 19% (28%) of the mean offer wage for nondisabled men. Decompositions of the offer wage differentials are reported in Table 2 for both the binary and factor models.

The results show that physical limitations depress wages, but the size of the effect varies greatly with the specification of the model. Limitations account for 16% (0.029/0.181) of the offer wage differential between ND/LP men in the factor model compared to 25% (0.047/0.190) in the binary model. The two models yield even greater differences for ND/MP men: limitations account for 10% of the offer wage differential in the factor model compared to 39% in the binary model. [20] There is no clear choice between the two specifications, although the binary model has a higher log likelihood in the employment function for both LP and MP groups, and a higher adjusted r-square in the wage equation for the MP group. The discussion that follows uses the more conservative estimates of wage discrimination from the binary model.

The explained differential, attributed to differences in human capital and job characteristics, is 17% of the offer wage differential for ND/LP men and 41% of the differential for ND/MP men. Disabled men's offer wages are lower in part because they are less well educated and work in less skilled occupations than nondisabled men. [21] These differences are somewhat offset by the fact that disabled men have more work experience, on average, than nondisabled men.

The productivity-standardized wage differential, attributed to discrimination and residual effects, is 83% of the offer wage differential for ND/LP men and 59% for ND/MP men. The relative sizes of the unexplained wage differentials for LP and MP men provide little support for the hypothesis that the men who are subject to more prejudice experience more discrimination. The unexplained wage differential is 15.7% of the nondisabled offer wage for LP men and 16.4% for MP men. Sensitivity tests based on alternative specifications of the model yielded larger estimates of the unexplained wage differential than the model discussed above. [22]

Employment Effects of Wage Discrimination

There are large differences in observed probabilities of employment for disabled and nondisabled men. Part of the employment differentials reflect the disincentive effects of unexplained differences in the offer wages of the two groups. In other words, some disabled men whose reservation wages fall between the observed offer wage and the productivity-standardized offer wage are deterred from working. We call this the "disincentive effect of wage discrimination," although we recognize that factors other than discrimination may contribute to the unexplained part of the wage differential.

As shown in Table 3, the estimated probabilities of employment ([[pi].sub.j] in Equation 4) are 75% for LP men and 62% for MP men, compared to 93% for nondisabled men. The disincentive effects of wage discrimination explain less than two percentage points of the 18 percentage point difference in employment rates between nondisabled and LP men and less than two percentage points of the 31 percentage point difference between nondisabled and MP men.

The estimated employment effects are small in terms of percentage points, but the losses of jobs and income to the disabled groups are not trivial. An additional 44,000 LP men and 11,000 MP men would be employed in 1990 if disabled men were offered productivity-standardized wages. If we assume these men would obtain jobs at the mean productivity-standardized wage rate for their group, the annual earnings loss to men in the LP group is $1.02 billion and the annual earnings loss to MP men is $0.24 billion. In total, the disincentive effects of wage discrimination cost not-employed men with disabilities $1.2 billion in lost earnings in 1990. It is probable that some of these men seek disability benefits from private or social insurance plans to compensate for their losses of income.

There still remains a large (approximately 17 percentage points for LP men and 30 percentage points for MP men) employment differential that is not explained by the disincentive effects of the wage differentials. Part of the differential represents differences in the productivity-related characteristics of disabled and nondisabled men, the most important of which are education and functional limitations. On average, disabled men have no formal schooling after high school, whereas nondisabled men complete one year of college (means reported in Appendix B). The education differential likely reflects the older ages of the disabled groups, rather than an outcome of disability, because relatively few persons are affected by disabling health conditions during childhood.

The probability of employment decreases as functional limitations increase, and mean values of the limitations variables increase as one moves from ND to LP to MP men. [23] Thus, even in the absence of employer discrimination, we would expect to observe lower employment rates for disabled men than for nondisabled men.

Discussion

The results show that workers with more functional limitations have lower wages, all else equal, than workers with fewer limitations, supporting the contention that disabled workers' wages reflect lower average levels of productivity. Even so, approximately 60% of the offer wage differential between disabled and nondisabled men is attributed to unexplained differences in productivity-standardized wages by the more conservative (binary) model. The remaining question is the extent to which the productivity-standardized wage differentials represent discrimination.

The great limitation of the decomposition technique is the inability to decompose the estimates of productivity-standardized wage differentials into discriminatory and nondiscriminatory components. It seems unlikely that the wage equations have not omitted productivity-relevant variables. It is equally unlikely that all the proxies for worker productivity are, as assumed, free from the effects of prior discrimination.

Interpretations of our results must, therefore, recognize the effects of the following assumptions and choices. First, we assume that education and work experience have not been affected by disability-related discrimination. The increase in the incidence of disabling health conditions with age suggests that the assumption is reasonably correct regarding education but not correct with respect to work experience. Second, we choose the binary model as the most appropriate specification of functional limitations. The part of the offer wage differential attributed to differences in productivity is four times larger in the binary model than in the factor model that has been used in previous studies. Thus, the choice of binary measures of functional limitations, and the assumption of no prior discrimination, contribute to conservative estimates of discrimination relative to other assumptions and choices. Nevertheless, no discrimination study can claim an exact measure of wage discrimination. The results that we hav e described must be interpreted as boundaries within which the true measure of wage discrimination is likely to fall.

The assumption that prejudice (preferences) is the root cause of discrimination is implicit in most studies of discrimination against African Americans and other minority groups but has not been tested directly. This article attempts to investigate the extent to which the assumption is valid.

The results show that persons with disabilities are subject to wage discrimination, but they do not conclusively demonstrate that discrimination is caused by prejudice. One source of uncertainty is the weakness of the test. A second problem is the sensitivity of the results to the specification of the functional limitation measures. The results from the binary model indicate that there is no difference in unexplained wage differentials for LP and MP men, but the factor results indicate that the unexplained wage differential is 50% larger for MP men (Table 2). The factor model, therefore, more strongly supports the presumption that wage discrimination is a reflection of prejudice.

Uncertainties regarding the interpretation of our results should not obscure the fact that a relatively small group of workers bears rather substantial losses because of discrimination toward their health conditions. The results suggest losses of wage income ranging from $10 to $12 billion in 1990 that are in part or in whole attributable to wage discrimination. [24] Were the true measures only half as large, the absolute losses to disabled workers would be substantial.

Income losses attributed to disability-related discrimination have implications for nondisabled persons as well. The primary income support program for disabled persons with Social Security coverage is Social Security Disability Insurance (SSDI). The program pays more than $28 billion annually to employment-aged individuals who are disabled and not working (GAO 1994). It is well known that decreases in employment rates generally lead to increases in SSDI applications (Kreider 1998). It is reasonable to expect that disabled men who are disadvantaged by wage discrimination find SSDI an attractive alternative.

5. Conclusions

The ADA was enacted in response to the political activism of persons with disabilities, who adopted the strategies used by African Americans and other minority groups in their fight against discrimination. The Act is based on the premise that prejudice creates obstacles to employment and limits wages for persons with disabilities.

The existence of prejudice against persons with disabilities is well documented, but the extent to which prejudice is translated into employment and wage differentials is not. Our results include empirical tests of the importance of prejudice as a cause of poor labor market outcomes for men with disabilities. The tests are only the beginning of a badly needed development of empirical tests of discrimination, for all minority groups, that reflect the theoretical propositions on which economic analyses of labor market discrimination are based.

Our results suggest that advocates of the ADA were correct in asserting that labor market discrimination is an important problem for persons with disabilities. In 1990, there were unexplained wage differentials between nondisabled and disabled men that reduced the wages of working men and discouraged others from accepting employment. The total income lost to men with disabilities is large: we estimate approximately $11 billion in 1990. The size of the estimate is conditional on our definition of the disabled population, which excludes men who are physically unable to work and subject to the caveat that unexplained wage differentials include unmeasured differences in productivity, as well as discrimination effects. Nevertheless, the size of the estimate suggests that if the ADA is successful in eliminating discrimination against men with disabilities, it has the potential to increase their incomes substantially and to decrease their reliance on public support programs.

Our comparisons of differences in employment rates between disabled and nondisabled men suggest that employment is an even more important problem than wage discrimination. Only a small fraction of the large differences in employment rates are attributable to the disincentive effects of wage discrimination, suggesting that the remainder is influenced by differences in employability and refusals to hire in cases where discriminatory offer wages fall below the legal minimum. Ultimately, the success of the labor market provisions of the ADA should be judged by their success in increasing employment rates among these workers.

The best chance for success appears to begin with a recognition that the majority of men with disabilities are able to work, but discrimination reduces their wages and opportunities for employment. It is equally important to recognize that impairments do limit productivity, contradicting the assertions of some disability rights activists that the only limits on the employment of persons with disabilities are the perceptions and prejudices of nondisabled persons.

Our tests for the importance of prejudice as a cause of wage differentials suggest that prejudice has a strong effect for a relatively small minority of men with disabilities. Prejudice appears to be relatively unimportant in determining wage differentials for a much larger group.

The results in this article establish a reference to which the effects of the ADA in the years since 1990 can be compared. Although the implementation of the ADA's provisions lagged the passage of the Act by more than a year, the data should soon be available to perform such a comparison.

(*.) Department of Economics, East Carolina University, A433 Brewster, Greenville, NC 27858-4353, USA; corresponding author.

(+.) School of Health Administration and Policy and Department of Economics, College of Business, Arizona State University, Tempe, AZ 85287-4506, USA.

The authors acknowledge helpful comments from Richard Burkhauser, David Salkever, participants in the "Work Disability" session of the 1994 AEA meetings, and two anonymous referees. Any remaining errors are our own.

Received January 1998; accepted April 1999.

(1.) Our definitions combine concepts from Nagi (1969) and the World Health Organization (WHO 1980). The ADA definition of disability is more comprehensive. It includes anyone with "a physical or mental impairment that substantially limits one or more of the major life activities," anyone with a record of having such an impairment, or anyone who is perceived as having such an impairment. Any individual who satisfies one of the three criteria is protected from wage or employment discrimination by the Act.

The distinctions between impairment, limitation, and disability are useful but do not completely eliminate ambiguities because certain illnesses or injuries create multiple impairments. We will, therefore, sometimes use the term "disabling condition" rather than "impairment" in our discussion.

(2.) The Royal and Roberts (1987) survey asks, "How much would you like to have this person as a friend?" Respondents rate each disability on a scale of 1 (least acceptable) to 5 (most acceptable). The Westbrook, Legge, and Pennay (1993) survey asks respondents to "consider what level of acceptance is most usually found for each disability ... circle the number that best describes the typical acceptance level" (Wcstbrook, Legge, and Pennay 1993, p. 617). The following scale is provided: (i) No acceptance (people would prefer a person with this disability to be kept in an institution or out of sight); (ii) Low acceptance (people would try and avoid a person with this disability); (iii) Moderate acceptance (a person with this disability would be acceptable as a fellow worker); (iv) High acceptance (a person with this disability would be acceptable as a friend); (v) Full acceptance (people would accept a person with this disability marrying into their immediate family) (Westbrook, Legge, and Pennay 1993, p. 617).

(3.) The survey asks "How easily could you tell that people have (disability name) just by looking at them?" Disabilities are ranked on a scale of 1 (not at all visible) to 5 (very visible) (Royal and Roberts 1987).

(4.) Westbrook, Legge, and Pennay (1993, p. 615) also report that their disability hierarchy is "remarkably similar to other hierarchies reported over the last 23 years," suggesting little change in attitudes toward disabilities over time.

(5.) The acceptability score for cancer is on the borderline in the Royal and Roberts (1987) study (4.00) but ranked considerably lower in the Westbrook, Legge, and Pennay (1993) results (3.75). We also report the results of sensitivity analyses when the three borderline conditions are moved to the LP group.

(6.) Men who receive Social Security Disability Insurance (SSDI) or who are institutionalized are excluded from the sample because they are unable to work. Applicants for SSDI must prove they have not worked for five months, are unable to work, and have no prospect of gainful employment in the future to qualify for benefits (USDHHS 1984). The exclusions restrict the severity but not the types of impairments included in the analysis.

(7.) Self-reported measures of functional limitations have been shown to be reasonably accurate correlates of the capacity to work (Stern 1989).

(8.) The scores for different functional limitations are correlated because one impairment can produce multiple limitations. The use of a dummy variable for each limitation in a wage equation typically produces results in which none of the individual limitation variables are statistically significant. In addition, some less common limitations do not occur in some samples, reducing the regression models to less than full rank for those samples. One solution to these problems is to use principal components analysis to produce a combined score. Principal components analysis reduces collinearity, but it is difficult to interpret the coefficients of the factors that replace the direct measures of functional limitations. In addition, some information is lost when 18 measures of functional limitations are reduced to three factors.

(9.) We thank a referee for suggesting the binary model.

(10.) We first constructed dummy variables for each of the 18 functional limitations reported on the SIPP. The variables equal zero if an individual has "no difficulty" performing a function and one if the individual "has difficulty" or "cannot" perform the function. Perfectly colinear variables are then combined into logically associated groups (e.g., limitations in dressing, eating, or toileting are included in a variable for "personal care"). The result is 14 limitations categories.

(11.) The system is identified by including the following variables in the employment function that are not included in the wage equation: nonwage incomes, marital status, age. In theory, all variables in the wage equation should also be included in the employment function; however, some of the variables in the wage equation, namely occupation, union membership, and work experience, are not observed for nonworkers. This may severely limit the selectivity bias correction in our model. Nevertheless, we prefer not to estimate a more restrictive wage equation because the residual is our estimate of discrimination. The sample selection variable, defined as the inverse of Mill's ratio (Heckman 1980), is constructed from coefficient estimates of the employment function described in the following section. The hourly wage rate is computed by dividing total wages for the four-month reference period by total hours worked.

(12.) The experience variables are "specific experience" (years worked for current employer), "general experience" (years worked for other employers), and "missing experience" (years in which a man was of labor force age but was neither employed nor in school). Specific experience squared is also included to control for declining investments in job-specific training over time.

(13.) The estimates of wage discrimination may, for example, be too large or too small because the occupation variables do not control adequately for interactions between job demands and functional limitations.

(14.) The nondiscriminatory wage structure for minority workers is not observable but is assumed to be bounded by the observed wage structures of the majority and minority groups. Most studies of discrimination use weights of one (the nondiscriminatory wage structure is the wage structure of the majority group), zero (the nondiscriminatory wage structure is the wage structure of the minority group), or 0.5 (Reimers 1983). Cotton (1988) argues that the weights should represent the proportions of majority and minority workers in the employed labor force. Our choice of d = 1.0 is consistent with Cotton's rule that would set d = 0.96 for LP men and d = 0.98 for MP men. There is an alternative weighting scheme suggested by Neumark (1988), in which the nondiscriminatory wage structure is obtained from a pooled regression over the majority and minority groups. Oaxaca and Ransom (1988) argue that the pooled regression approach is less attractive in models that control for sample selection bias because the estimated p arameters from the pooled sample are not a linear function of the estimated parameters from the separate regressions.

(15.) That is, labor supply schedules are upward sloping. Becker predicts discrimination will increase majority worker employment but reduce total employment. The net effect of discrimination on total employment depends on the elasticities of supply and demand for majority and minority workers (Becker 1971; Thurow 1975). Baldwin and Johnson (1992a) test the weighting schemes proposed by different studies of wage discrimination and find that Cotton's (1988) and Neumark's (1988) weights produce estimated employment effects consistent with Becker's model of discrimination.

(16.) A wage floor fixed by legislation or collective bargaining agreements can prevent this assumption from being satisfied (Gilman 1965). An employer whose tastes for discrimination can only be satisfied by offering wages below the legal minimum will refuse to hire minority workers. The extent to which discrimination is expressed by refusals to hire because of wage floors cannot be predicted a priori because it depends on worker productivity, workers' reservation wage functions, the strength of employers' tastes for discrimination, and the prevalence and structure of wage floors.

(17.) The values of the test statistic in the factor model are: ([[X.sup.2].sub.16] = 129.16) for ND/LP men, ([[X.sup.2].sub.16] = 145.70) for ND/MP men, and ([[X.sup.2].sub.16] = 24.54) for LP/MP men. The values in the binary model are ([[X.sup.2].sub.34] = 146.12) for ND/LP men, ([[X.sup.2].sub.34] = 158.91) for ND/MP men, and ([[X.sup.2].sub.34] = 48.71) for LP/MP men.

(18.) In the binary model, the test statistics are ([F.sub.3,10,724] = 1.36) for ND/LP men, ([F.sub.31,10,399] = 1.68) for ND/MP men, and ([F.sub.31,547] = 1.06) for LP/MP men, in the factor model, the test statistics are ([F.sub.31,10,760] = 2.56) for ND/LP men, ([F.sub.13,10,435] = 2.00) for ND/MP men, ([F.sub.13,583] = 1.04) for LP/MP men. We attribute the failure to find significant differences between the LP and MP wage equations to the loss of degrees of freedom when the test is restricted to the small samples of disabled men and to the large standard errors for some of the coefficients. When the wage equation is reestimated jointly for the disabled groups with a binary variable identifying MP men and interaction terms between MP and all other variables, the results show significant between-group differences in the intercept term and in the slope coefficients for the experience variables. We cannot, however, use a joint equation for the wage decompositions.

(19.) Estimates of wage discrimination against workers with disabilities have been reported by Johnson and Lambrinos (1985) using data from the 1972 Social Security Survey of Disabled and Nondisabled Adults (Bureau of the Census 1992) and by Baldwin and Johnson (1994, 1995) using data from the 1984 SIPP.

(20.) The alternative specifications of the limitations variables produce large differences in the measured effects of functional limitations on wages but have little effect on the coefficients of other variables in the wage equations.

(21.) Relative to nondisabled men, LP men are underrepresented in high-wage professional and managerial occupations, while MP men are overrepresented in low-wage laborer occupations.

(22.) The decompositions do not change substantively when we re-estimate the wage equations without the control for sample selection bias and decompose observed wage differentials. The results do change when we reestimate the model assuming that the wage structure for disabled men is the nondiscriminatory wage structure. The change in weights increases the unexplained component to approximately 95% of the offer wage differential for LP men and 120% for MP men. We prefer the model with d = 1 because we believe it represents the wage structure that would prevail in the absence of discrimination more accurately and because it yields more conservative estimates of discrimination.

We also reestimated the wage models with alternative definitions of the LP and MP groups to see whether our results are sensitive to changes in the cutoff point on the acceptability scale. In particular, we reclassified the less visible impairments (cancer, speech, hearing) into the LP group. The decompositions for the LP group are virtually unchanged. For the MP group, the positive component attributed to functional limitations increases, but the negative component attributed to experience decreases in absolute value. The changes are offsetting, so there is no change in the explained and unexplained components of the wage differential. The overall effect of the change in definitions is to move some impairment categories with relatively low levels of functional limitations and work experience to the LP group.

Finally, we reestimated the models excluding the controls for occupation because these variables may be endogenous to discrimination. When occupation is excluded, a larger part of the wage differential is explained by functional limitations and education, but there is no substantive difference in the results of the decompositions. Results are available from the authors.

(23.) The one exception is the factor measuring strength and endurance, which has a higher average value for LP men than for MP men. The extreme values of the limitations variables for the ND group are, however, within the range of values for the disabled men. This result is consistent with previous research showing that impairments of the same severity are not always equally disabling because of differences in compensatory technologies.

(24.) The estimates of income losses are computed as follows. We estimate mean offer wages in the absence of discrimination by substituting mean characteristics for the disabled groups into the estimated wage function for nondisabled men. We then compute income losses for working disabled men by multiplying the unexplained difference in offer wages, by 2000 hours per year, by the weighted population totals. We compute income losses for nonworking disabled men by multiplying the mean nondiscriminatory offer wage, by 2000 hours per year, by the number of job losses attributed to the disincentive effects (Table 3).

References

Baldwin, Marjorie L., and William G. Johnson. 1992a. A test of the measures of nondiscriminatory wages used to study wage discrimination. Economics Letters 39:223-7.

Baldwin, Marjorie L., and William G. Johnson. 1992b. Estimating the employment effects of wage discrimination. Review of Economics and Statistics 74:446-55.

Baldwin, Marjorie L., and William G. Johnson. 1994. Labor market discrimination against men with disabilities. Journal of Human Resources 29:1-19.

Baldwin, Marjorie L., and William G. Johnson. 1995. Labor market discrimination against women with disabilities. Industrial Relations 34:555-77.

Becker, Gary S. 1971. The economics of discrimination. Chicago: University of Chicago Press.

Bureau of the Census. 1992. Survey of income and program participation (SIPP). Wave III rectangular core and topical module microdata file. Washington, DC: The Bureau.

Cain, Glen G. 1986. The economic analysis of labor market discrimination: A survey. In Handbook of labor economics, volume I, edited by Orley Ashenfelter and Richard Layard. Amsterdam: North-Holland, pp. 693-785.

Clogston, J.S. 1990. Disability coverage in 16 newspapers. Louisville, KY: Avocado Press.

Cotton, Jeremiah. 1988. On the decomposition of wage differentials. Review of Economics and Statistics 70:236-43.

General Accounting Office (GAO). 1994. Social security disability: Most of gender difference explained. Report to the Special Committee on Aging, Washington, DC: U.S. General Accounting Office.

Gilman, Harry J. 1965. Economic discrimination and unemployment. American Economic Review 55:1077-96.

Hahn, Harlan. 1987. Advertising the acceptably employable image: Disability and capitalism. Policy Studies Journal 15:551-70.

Heckman, James J. 1980. Sample selection bias as a specification error with an application to the estimation of labor supply functions. In Female labor supply; Theory and estimation, edited by James P. Smith. Princeton, NJ: Princeton University Press, pp. 207-48.

Johnson, William G., and James Lambrinos. 1985. Wage discrimination against handicapped men and women. Journal of Human Resources 20:264-77.

Jones, F. L. 1983. On decomposing the wage gap: A critical comment on Blinder's method. Journal of Human Resources 28:126-30.

Kreider, Brent. 1998. Workers' applications to social insurance programs when earnings and eligibility are uncertain. Journal of Labor Economics. 16:848-77.

Margolis, Howard, and Arthur Shapiro. 1987. Countering negative images of disability in classical literature. English Journal 76:17-22.

Nagi, Saad Z. 1969. Disability and rehabilitation. Columbus, OH: Ohio State University.

Neumark, David. 1988. Employers' discriminatory behavior and the estimation of wage discrimination. Journal of Human Resources 23:279-95.

Oaxaca, Ronald. 1973. Male-female wage differentials in urban labor markets. International Economic Review 14:693-709.

Oaxaca, Ronald L., and Michael R. Ransom. 1988. Searching for the effect of unionism on the wages of union and nonunion workers. Journal of Labor Research 9:139-48.

Reimers, Cordelia W. 1983. Labor market discrimination against hispanic and black men. Review of Economics and Statistics 65:570-9.

Royal, George P., and Michael C. Roberts. 1987. Students' perceptions of and attitudes toward disabilities: A comparison of twenty conditions. Journal of Clinical Child Psychology 16:122-32.

Stern, Steven. 1989. Measuring the effect of disability on labor force participation. Journal of Human Resources 24:361-95.

Thurow, Lester C. 1975. Generating inequality. New York: Basic Books.

U.S. Department of Health and Human Services (USDHHS). 1984. Social security handbook. Washington, DC: U.S. Government Printing Office.

Westbrook, Mary T., Varoe Legge, and Mark Pennay. 1993. Attitudes towards disabilities in a multicultural society. Social Science and Medicine 36:615-24.

World Health Organization (WHO). 1980. International classification of impairments, disabilities and handicaps. Geneva: World Health Organization.

 Classification of Impairments
 Social
 Disability Category Distance
SIPP Impairment Category from Attitudes Studies Ranking [a]
Impairments that are less visible or
subject to less prejudice (LP)
 Back or spine problems -- --
 Broken bone/fracture -- --
 Head or spinal cord injury -- --
 Hernia or rupture -- --
 High blood pressure -- --
 Kidney stones or chronic kidney
 trouble -- --
 Stiffness or deformity of the foot,
 leg, arm, or hand -- --
 Thyroid trouble or goiter -- --
 Tumor, cyst, or growth -- --
 Learning disability Learning disability 3
 Stomach trouble Ulcer 4
 Lung or respiratory trouble Asthma 4
 Diabetes Diabetes 4
 Heart trouble Heart disease (4)
 Arthritis or rheumatism Arthritis 4
Impairments that are visible and subject
to more prejudice (MP)
 Missing legs, feet, arms, hands,
 or fingers Amputation 3
 Cancer Cancer 4 (3)
 Speech disorder Speech deficit 3
 Blindness or vision problems Blindness 3
 Deafness or serious trouble
 hearing Deafness 3
 Stroke Stroke (3)
 Epilepsy Epilepsy 3
 Paralysis of any kind Paraplegia 3
 Cerebral palsy Cerebral palsy 3 (2)
 Alcohol or drug problem Alcoholism (2)
 Mental or emotional problem Mental illness 2
 Mental retardation Mental retardation 2 (1)
 AIDS AIDS (1)
 Visibility
SIPP Impairment Category Ranking [a]
Impairments that are less visible or
subject to less prejudice (LP)
 Back or spine problems --
 Broken bone/fracture --
 Head or spinal cord injury --
 Hernia or rupture --
 High blood pressure --
 Kidney stones or chronic kidney
 trouble --
 Stiffness or deformity of the foot,
 leg, arm, or hand --
 Thyroid trouble or goiter --
 Tumor, cyst, or growth --
 Learning disability 1
 Stomach trouble 1
 Lung or respiratory trouble 1
 Diabetes 1
 Heart trouble --
 Arthritis or rheumatism 2
Impairments that are visible and subject
to more prejudice (MP)
 Missing legs, feet, arms, hands,
 or fingers 4
 Cancer 1
 Speech disorder 1
 Blindness or vision problems 3
 Deafness or serious trouble
 hearing 1
 Stroke --
 Epilepsy 2
 Paralysis of any kind 4
 Cerebral palsy 4
 Alcohol or drug problem --
 Mental or emotional problem 2
 Mental retardation 4
 AIDS --

(a.)Social distance rankings range from 1 (low acceptance) to 5 (high acceptance). Visibility rankings range from 1 (not visible) to 5 (very visible). Rankings are derived from Royal and Roberts (1987) mean acceptability and visibility ratings, and truncated to integer values. Rankings in parentheses are derived from Westbrook, Legge, and Pennay (1993) mean social distance ratings. The latter rankings are reported only when they differ from the Royal and Roberts results or when the impairment is not included in the earlier study.

 Decompositions of Wage Differentials Between
 Disabled and Nondisabled Men [a]
 Disabled (LP) Disabled (MP)
 Binary Model Factor Model Binary Model
Wage differential $2.22 $2.22 $2.45
Difference in log wages 0.153 0.153 0.228
Difference in offer wages 0.190 0.181 0.280
Components of the differential
 Functional limitations 0.047 0.029 0.109
 Education 0.026 0.026 0.030
 Race -0.006 -0.006 -0.003
 Union 0.001 0.001 0.001
 Experience -0.049 -0.049 -0.034
 Occupation 0.015 0.015 0.013
Explained differential 0.033 0.016 0.116
Unexplained differential 0.157 0.165 0.164
 Factor Model
Wage differential $2.45
Difference in log wages 0.228
Difference in offer wages 0.271
Components of the differential
 Functional limitations 0.027
 Education 0.031
 Race -0.003
 Union 0.001
 Experience -0.034
 Occupation 0.013
Explained differential 0.034
Unexplained differential 0.237
Source: Bureau of the Census (1992) Wave III 1990 panel.
(a.)Components of the differential equal
([X.sub.ND] - [X.sub.j])[[beta].sub.ND],
j = LP, MP.
 Employment Effects of Wage Discrimination [a]
 Disabled (LP)
 (Pop. = 3,118,864)
 Binary Factor
 Model Model
Estimated probability of employment
 ([[pi].sub.j]) 75.67 74.99
Estimated probability of employment in
 the absence of wage discrimination
 ([[[pi].sup.*].sub.j]) 77.08 76.50
Difference in probabilities of employ-
 ment ([[[pi].sup.*].sub.j] - [[pi].sub.j]) 1.41 1.51
Jobs not taken as a result of the disin-
 centive effects of wage discrimination 43,976 47,095
 Disabled (MP)
 (Pop. = 1,091,308)
 Binary Factor
 Model Model
Estimated probability of employment
 ([[pi].sub.j]) 62.70 61.51
Estimated probability of employment in
 the absence of wage discrimination
 ([[[pi].sup.*].sub.j]) 63.73 63.10
Difference in probabilities of employ-
 ment ([[[pi].sup.*].sub.j] - [[pi].sub.j]) 1.03 1.59
Jobs not taken as a result of the disin-
 centive effects of wage discrimination 11,240 17,352
Source: Bureau of the Census (1992) Wave III 1990 panel.
(a.)j = LP, MP. Probabilities expressed as percents.
 Appendix A
 Factor Analysis of Functional Limitation Variables
 Rotated Factor Pattern [a]
 Factor 1
Variable Mobility
Difficulty seeing 0.121
Difficulty hearing 0.007
Difficulty speaking 0.089
Uses an aid to help get around 0.355
Difficulty lifting and carrying something as
 heavy as 10 lbs. 0.299
Difficulty walking for a quarter mile 0.227
Difficulty climbing a flight of stairs 0.243
Difficulty getting around outside the house 0.523
Difficulty getting around inside the house 0.575
Difficulty getting in and out of bed 0.485
Needs help to prepare meals 0.566
Needs help with light housework 0.535
Difficulty using the telephone 0.092
Needs help to take a bath or shower 0.687
Needs help to get dressed 0.599
Needs help eating 0.268
Needs help to use the toilet 0.623
Difficulty keeping track of money and bills 0.211
Eigenvalues 3.139
Percent total variance 17.44
 Factor 2
 Strength Commun-
 and Factor 3 ality
Variable Endurance Sensory Estimates
Difficulty seeing 0.183 0.204 0.090
Difficulty hearing 0.092 0.267 0.080
Difficulty speaking 0.020 0.668 0.455
Uses an aid to help get around 0.297 0.067 0.219
Difficulty lifting and carrying something as
 heavy as 10 lbs. 0.515 0.123 0.369
Difficulty walking for a quarter mile 0.817 0.116 0.732
Difficulty climbing a flight of stairs 0.766 0.115 0.659
Difficulty getting around outside the house 0.327 0.176 0.411
Difficulty getting around inside the house 0.246 0.083 0.398
Difficulty getting in and out of bed 0.313 0.077 0.340
Needs help to prepare meals 0.188 0.222 0.405
Needs help with light housework 0.261 0.077 0.361
Difficulty using the telephone 0.024 0.662 0.448
Needs help to take a bath or shower 0.185 0.074 0.512
Needs help to get dressed 0.110 0.127 0.387
Needs help eating 0.051 0.234 0.130
Needs help to use the toilet 0.076 0.138 0.414
Difficulty keeping track of money and bills 0.076 0.379 0.194
Eigenvalues 2.080 1.383 --
Percent total variance 11.56 7.68 36.68

Source: Bureau of the Census (1992) Wave III 1990 panel.

(a.)The factor patterns are obtained from principal components analysis with varimax rotation. Three factors satisfied the selection criterion of preliminary eigenvalues greater than one. The communality estimates are the proportion of the variation of each variable involved in the factor patterns. The percent total variance is the percentage of variation among all the variables involved in each pattern.

 Appendix B
 Means and Estimated Coefficients of Variables in
 the Employment Function [a]
 Nondisabled
 (N = 11,708)
 (Pop. = 54,125,633)
Variable Mean Coefficient
Employed 0.89
Demographic characteristics
 Education 13.2 0.082 [*] 0.082 [*]
 Race (1 = white) 0.86 0.441 [*] 0.443 [*]
 Age 35.7 0.028 [*] 0.027 [*]
Functional limitations
 Factor 1 (mobility) -0.05 0.107
 Factor 2 (strength) -0.13 -0.121 [*]
 Factor 3 (sensory) -0.03 0.023
Functional limitations (binary)
 Difficulty seeing 0.01 -0.189
 Cannot see 0.00 -0.087
 Difficulty hearing 0.03 0.052
 Cannot hear 0.00 0.282
 Difficulty speaking 0.00 -0.241
 Difficulty lifting 0.00 -0.376
 Cannot lift 0.00 0.159
 Difficulty climbing 0.01 -0.509 [*]
 Cannot climb 0.00 -0.373
 Difficulty walking 0.01 -0.208
 Cannot walk 0.00 -0.781 [*]
 Difficulty with phone 0.00 6.819
 Difficulty getting around 0.00 6.356
 Difficulty getting in/out bed 0.00 3.802
 Disabled (LP)
 (N = 662)
 (Pop. = 3,118,864)
Variable Mean Coefficient
Employed 0.71
Demographic characteristics
 Education 12.4 0.133 [*] 0.127 [*]
 Race (1 = white) 0.87 0.488 [*] 0.497 [*]
 Age 41.6 0.003 0.002
Functional limitations
 Factor 1 (mobility) 0.12 -0.080 [*]
 Factor 2 (strength) 0.15 -0.126 [*]
 Factor 3 (sensory) 0.17 -0.011
Functional limitations (binary)
 Difficulty seeing 0.07 -0.380
 Cannot see 0.00 -0.311
 Difficulty hearing 0.14 0.267
 Cannot hear 0.00 -1.006
 Difficulty speaking 0.01 2.274 [*]
 Difficulty lifting 0.13 -0.282
 Cannot lift 0.06 -0.512 [*]
 Difficulty climbing 0.15 -0.037
 Cannot climb 0.08 0.164
 Difficulty walking 0.15 -0.447 [*]
 Cannot walk 0.09 -0.539 [*]
 Difficulty with phone 0.01 -0.736
 Difficulty getting around 0.04 -0.187
 Difficulty getting in/out bed 0.03 -0.538
 Disabled (MP)

 (N = 240)
 (Pop. 1,091,308)
Variable Mean Coefficient
Employed 0.58
Demographic characteristics
 Education 12.3 0.120 [*] 0.091 [*]
 Race (1 = white) 0.83 0.667 [*] 0.736 [*]
 Age 41.5 -0.012 -0.012
Functional limitations
 Factor 1 (mobility) 0.56 -0.003
 Factor 2 (strength) 0.69 -0.090 [*]
 Factor 3 (sensory) 1.23 -0.001
Functional limitations (binary)
 Difficulty seeing 0.15 -0.144
 Cannot see 0.05 -1.127 [*]
 Difficulty hearing 0.13 0.859 [*]
 Cannot hear 0.04 0.132
 Difficulty speaking 0.09 -0.373
 Difficulty lifting 0.03 -0.312
 Cannot lift 0.07 -0.699
 Difficulty climbing 0.11 0.814 [*]
 Cannot climb 0.08 -0.388
 Difficulty walking 0.10 -0.683 [*]
 Cannot walk 0.06 0.145
 Difficulty with phone 0.05 0.442
 Difficulty getting around 0.10 -1.353 [*]
 Difficulty getting in/out bed 0.02 2.173 [*]
 Cannot get in/out of bed 0.00 5.439 0.05 0.247
 Difficulty bathing 0.00 5.156 0.02 0.145
 Difficulty with personal care 0.00 4.679 0.02 0.251
 Cannot personal care 0.00 6.084 0.01 -0.546
 Difficulty with money 0.00 0.315 0.02 -0.983
 Difficulty with meals 0.00 6.398 0.01 0.864
 Difficulty with housework 0.00 -0.801 [*] 0.02 -1.473 [*]
Log Likelihood -- -3278 -3287 -- -297 -311
 Cannot get in/out of bed 0.05 0.694
 Difficulty bathing 0.05 1.059
 Difficulty with personal care 0.02 -1.866
 Cannot personal care 0.02 -0.305
 Difficulty with money 0.09 0.440
 Difficulty with meals 0.07 0.065
 Difficulty with housework 0.06 -0.034
Log Likelihood -- -118 -130
Source: Bureau of the Census (1992) Wave III 1990 Panel.
(a.)Population totals equal sum of the sample weights.
An (*.) identifies variables significant at the 10% level or better.
Models also include controls for nonwage incomes and three marital
status dummies. Complete results available from the authors.
 Appendix C
 Means and Estimated Coefficients of Variables
 in the Wage Equation [a]
 Nondisabled
 (N = 10,319)
 (Pop. = 47,967,041)
Variable Mean Coefficient
Wage $13.35
Worker characteristics
 Education 13.3 0.045 [*] 0.045 [*]
 Race (1 = white) 0.87 0.147 [*] 0.146 [*]
 Union member 0.22 0.134 [*] 0.133 [*]
 Disabled (LP)
 (N = 467)
 (Pop. = 2,214,022)
Variable Mean Coefficient
Wage $11.13
Worker characteristics
 Education 12.8 0.044 [*] 0.043 [*]
 Race (1 = white) 0.91 0.039 0.050
 Union member 0.21 0.207 [*] 0.226 [*]
 Disabled (MP)
 (N = 142)
 (Pop. = 636,541)
Variable Mean Coefficient
Wage $10.90
Worker characteristics
 Education 12.6 0.053 [*] 0.072 [*]
 Race (1 = white) 0.89 0.120 0.132
 Union member 0.22 0.233 [*] 0.300 [*]
Functional limitations
 Factor 1 (mobility) -0.05 -0.011 -0.001
 Factor 2 (strength) -0.13 -0.028 [*] 0.86
 Factor 3 (sensory) -0.03 -0.006 0.16
Functional limitations (binary)
 Difficulty seeing 0.01 0.023 0.05 -0.089
 Cannot see 0.00 -0.366 [*] 0.00 -0.785
 Difficulty hearing 0.03 -0.034 0.13 0.124 [*]
 Cannot hear 0.00 -0.078 0.00 0.269
 Difficulty speaking 0.00 0.037 0.02 0.046
 Difficulty lifting 0.00 -0.052 0.11 -0.048
 Cannot lift 0.00 0.154 0.03 -0.036
 Difficulty climbing 0.01 -0.095 0.13 -0.064
 Cannot climb 0.00 0.095 0.06 0.079
 Difficulty walking 0.01 -0.058 0.12 -0.027
 Cannot walk 0.00 -0.101 0.07 -0.043
 Difficulty with pone 0.00 -0.104 0.01 -0.018
 Difficulty getting around 0.00 -0.767 [*] 0.02 -0.014
 Difficulty getting in/out bed 0.00 0.158 0.02 0.167
 Cannot get in/out of bed 0.00 -0.261 [*] 0.04 0.082
 Difficulty bathing 0.00 0.031 0.02 0.018
 Difficulty with personal care 0.00 0.194 0.01 -0.011
 Cannot personal care 0.00 0.149 0.01 -0.109
 Difficulty with money 0.00 -0.243 0.01 -0.089
 Difficulty with meals 0.00 -0.078 0.00 -0.149
 Difficulty with housework 0.00 -0.057 0.01 -0.052
Adjusted r-square -- 0.37 0.37 -- 0.34
Functional limitations
 Factor 1 (mobility) 0.001 0.36 -0.005
 Factor 2 (strength) -0.003 0.46 0.017
 Factor 3 (sensory) 0.007 1.08 0.005
Functional limitations (binary)
 Difficulty seeing 0.15 0.082
 Cannot see 0.04 -0.434
 Difficulty hearing 0.15 0.097
 Cannot hear 0.05 -0.192
 Difficulty speaking 0.07 -0.399 [*]
 Difficulty lifting 0.03 0.119
 Cannot lift 0.04 -1.218 [*]
 Difficulty climbing 0.13 0.243
 Cannot climb 0.04 2.200 [*]
 Difficulty walking 0.08 -0.088
 Cannot walk 0.04 -0.983
 Difficulty with pone 0.04 0.484 [*]
 Difficulty getting around 0.06 0.195
 Difficulty getting in/out bed 0.02 -0.197
 Cannot get in/out of bed 0.05 0.193
 Difficulty bathing 0.02 0.624
 Difficulty with personal care 0.01 0.091
 Cannot personal care 0.01 -0.101
 Difficulty with money 0.08 -0.039
 Difficulty with meals 0.06 0.117
 Difficulty with housework 0.03 -0.802
Adjusted r-square 0.35 -- 0.53 0.49

Source: Bureau of the Census (1992) Wave III 1990 panel.

(a.)Population totals equal sum of the sample weights. An (*.) identifies variables significant at the 10% level or better. Models also include controls for work experience, two occupational dummies, and the sample selection variable. Complete results available from the authors.