Professional Documents
Culture Documents
A Closer Look at the Employment Impact of the Americans with Disabilities Act
Julie L. Hotchkiss
Abstract
This paper replicates recent findings that employment among disabled people has declined
since the ADA. A closer look indicates that this decline results from a drop in the labor force
participation rate among those classified as disabled. Further analysis indicates that this labor force
participation rate decline, however, was not the result of disabled individuals fleeing the labor
market, but, rather, more likely a result of the re-classification of non-disabled, non-participants, as
disabled. The unconditional employment probability among disabled people (taking selection into
the labor market into account) has not declined, and may have actually improved for certain
disability classifications.
Julie Hotchkiss is Associate Professor, Department of Economics, Andrew Young School of Policy
Studies, Georgia State University; and Associate Policy Advisor, The Federal Reserve Bank of
Atlanta. Support from the W.E. Upjohn Institute for Employment Research is gratefully
acknowledged. Useful comments and suggestions were much appreciated from David Autor,
Christopher Bollinger, Richard Burkhauser, Thomas DeLeire, Robert Haveman, Kevin Hollenbeck,
Allan Hunt, Bruce E. Kaufman, Douglas Kruse, M. Melinda Pitts, Robert E. Moore, Ben Scafidi,
Stephen Woodbury and participants of seminars held at Georgia State University; the University of
Virginia; the Bureau of Labor Statistics; and the W.E. Upjohn Institute. Jack McNeil and Vic
Valdisera of the Census Bureau, Bob McIntire of the Bureau of Labor Statistics, Barry Hirsh,
David Macpherson, Robert Margo, and Walter Oi, were also instrumental in the completion of this
project. In addition, the research assistance and contributions of Ludmila Rovba are without
measure. The views expressed here are those of the author and do not necessarily represent the
views of the Federal Reserve Bank of Atlanta or the Federal Reserve System. The data used in this
article can be obtained beginning June 2005 through June 2008 from Julie Hotchkiss, Georgia
I. Introduction
One of the goals of the American's with Disabilities Act (ADA), which was passed in 1990
and phased-in over the next four years, was to improve the labor market experience of workers with
disabilities. While a worker's labor market experience has multiple dimensions, employment levels
have been by far the most popular way to measure the impact of the ADA on the labor market
experience of disabled people. The conclusions of earlier analyses have not been good news for
advocates of the ADA; employment levels among all disabled people have steadily declined
throughout the 1990s (for example, see DeLeire 2000 and Acemoglu and Angrist 2001). These
previous studies, however, have failed to control for selection into the labor market and have
The purpose of this paper is to look more closely at the employment outcomes among
disabled people and to fairly evaluate whether potential barriers to employment have been impacted
by the passage and phase-in of the ADA. The analysis will make use of multiple strategies, evaluate
whether the results are robust across data sets and sub-groups, and determine whether outcomes
differ across disability type. The main conclusion is that the unconditional probability of
employment among disabled people (after controlling for selection into the labor market) has not
A common challenge when one evaluates behavior of disabled people is that the most
comprehensive data sets available typically rely on an individual's self report of their disability
status. This can be problematic if the incentive to report a disability is correlated with the behavior
being evaluated.1 It is widely believed that the incentive to report a disability is correlated with the
decision to enter the labor market. If this is true, the importance of abstracting from (or controlling
for) the labor supply decision when evaluating employment outcomes is even that much more
crucial in obtaining an unbiased picture of the determinants of those outcomes. The empirical
Employment levels of any population group are affected by both labor supply and labor
demand issues. Individuals suffering from a disability will experience a greater cost to entering the
labor market as greater effort or sacrifices must be made relative to non-disabled workers. As a
result, the reservation wage for disabled individuals will be higher than for non-disabled individuals
and fewer disabled people will choose to enter the labor market, ceteris paribus. In addition, a
person's disability will be more likely to render him or her less productive than an otherwise
identical non-disabled person. Consequently, a disabled labor force participant will be less likely to
qualify for a given job and therefore less likely to be hired. Note that merely a perception of lower
productivity, or greater difficulty predicting a disabled worker's productivity, will make him or her
less likely to be hired. So, for both supply and demand reasons, we would expect the employment
levels of disabled people to be lower than non-disabled people. Figure 1 presents evidence from the
Current Population Survey (CPS) from 1981 through 2000 consistent with this supposition.2 The
percent of working aged (ages 21-58) disabled individuals employed in any given year is at least 46
percentage points lower than the percent of non-disabled working aged individuals employed in that
year. In addition, the employment percentage among non-disabled people has made a fairly steady
climb over the entire period compared with the relatively stagnant, then declining, employment
Legislation that potentially affects the costs of either labor force participation or of hiring a
group of workers can be expected to impact the employment levels of that group. The ADA,
through its required accommodations, can be expected to reduce the cost to a disabled individual of
entering the labor force, thus increasing labor force participation. It might also be argued that
greater accommodation of a disabled worker's limitations will result in increased productivity of that
disabled worker, thus increasing the likelihood of employment. If those required accommodations,
however, are "binding" in the sense that the employer would not undertake them in the absence of
the ADA, the employer must believe that the increased productivity of the disabled worker will not
HOTCHKISS, page 4
offset the cost of implementing those accommodations.3 This may result in decreased employment
probabilities of disabled workers, since the cost of hiring a disabled worker has increased. In
Figure 1, there doesn't seem to be any noticeable, or permanent change in the employment
percentages for disabled workers around the time of passage of the ADA, although there is a
Figure 2, however, yields a preview of the potential problem in looking at an outcome that
reflects both labor demand and labor supply influences. Figure 2 plots the percent of disabled and
non-disabled labor force participants that are employed between 1981 and 2000 (ages 21-58). This
figure shows that among labor force participants, the percent of disabled people that are employed
continued to rise through the 1990s (with the exception of the recession), and that the difference in
percent employed between disabled and non-disabled labor force participants became smaller over
the time period (from a difference of 11 percentage points in 1981 to a difference of 7 percentage
points in 2000). Of course, this figure represents conditional employment outcomes, and the main
outcomes that control for the labor supply decision to be in the labor market in the first place.
The combined Current Population Survey (CPS) Annual Earnings files for the months of
March, April, May, and June, for the years 1981 through 2000, are used to obtain demographic
information, employment status, earnings, details related to the respondent's job, and location
information to control for local labor market conditions. These CPS Annual Earnings files are
matched with the March CPS survey for each year to obtain information on disability status, other
sources of income, and labor market information available for the previous year. 4 This matching
strategy results in a sample four times larger than any single month of current labor market
statistics, yielding greater confidence in the reliability of the results.5 The sample used for the
analyses here was limited to individuals between (and including) the ages of 21 and 58. This is the
age range of the sample used by Acemoglu and Angrist (2001). DeLeire (2000) included all
HOTCHKISS, page 5
individuals 18 to 64 years of age; all results and conclusions drawn in this paper are the same when
The earliest year available in the CPS for analysis of outcomes among disabled people is
1981; prior to 1981 identification of a disability in the CPS was made only in the context of why a
respondent was not working. Table 1 presents means for the CPS sample used for the analyses in
this paper. Over the entire sample period, there are 69,906 disabled individuals (28 percent of
whom are employed); there are 972,161 non-disabled individuals (79 percent of whom are
employed). Seven percent of the sample is disabled. Disabled individuals tend to be slightly older,
have lower education levels, have less attachment to the labor market, are more likely to be nonwhite,
and are more likely to be in a single household. The samples of disabled and non-disabled
individuals are similarly distributed across geographic regions and contain similar percentages of
women.
The Survey of Income and Program Participation (SIPP) is used to test robustness of the
results across data source and to disaggregate the results by nature of disability. The SIPP topical
modules on Work Disability History (Wave 2) allow for classification of type of disability (in
addition to other labor market information), and are available for the years 1986 through 1997.
While providing more detail related to the respondent's disability, the SIPP does not provide as long
or as large a data set as the CPS. The samples from the SIPP have been constructed to match those
from the CPS as much as possible (for example, regarding variable definitions, etc.).6 The nature
of a person's disability is placed into one of 30 different categories. In order to be able to include
controls for type of disability, these categories were combined to correspond to the groupings used
Aggregation was necessary due to limitations in category representation; the four categories are: (1)
musculoskeletal systems and special senses, (2) internal systems, (3)!neurological systems and
Table 2 presents sample means from the SIPP data set by disability and employment status.
HOTCHKISS, page 6
Over the entire SIPP sample period, 37,741 people are disabled (39 percent of whom are
employed); there are 359,441 non-disabled observations (81 percent of whom are employed). 9.5
percent of the SIPP sample are disabled. Demographic differences across disability status are
similar to those seen in the CPS. A higher percent of the SIPP sample than the CPS sample is
employed. This is likely because employment status in the CPS is determined by activity in one
week, and the employment status in the SIPP is in reference to activity over a four month period of
time. In addition, the higher incidence of disability in the SIPP is likely because SIPP respondents
are given two opportunities to answer a disability question positively and because the respondent is
reminded if he or she indicated a disability in the previous interview, increasing the chances of a
The largest group by type of disability contains those with musculoskeletal and special
senses disabilities, neurological and mental disorders is the second largest group, internal systems
is the third largest group, and "other" is the fourth largest category. Over time (not shown in the
table) one can observe a slight upward trend in the neurological and mental disorders classification,
while internal systems has declined slightly, and musculoskeletal and special senses has remained
The labor market provisions of the ADA were motivated by a desire to eliminate barriers to
disabled individuals that might exist in the labor market. Consequently, an appropriate assessment
of the success of the ADA in this endeavor would involve evaluation of whether there has been any
progress in employment outcomes for a disabled person drawn from random, controlling for the
likelihood that he or she would be a labor force participant. The resulting probability of interest is
an unconditional probability of employment. An alternative question, that has been the source of
recent concern about the employment impacts of the ADA, is whether there has been any progress
in employment among all disabled people. This second question involves evaluation of a joint
outcome: the probability of entering the labor force and being employed. Consideration of this joint
HOTCHKISS, page 7
outcome (or, employment among all disabled people) confounds conclusions regarding the
employment impact of the ADA with labor supply decisions. Clearly, however, any documented
impact of the ADA on labor force participation must be taken seriously, as well.
The focus in this paper on the unconditional employment outcome is analogous to the
typical estimation of, say, a wage equation. Clearly, wages are only observed for those who are
working. In the interest of being able to generalize the regression results to the population, care is
taken to control for selection into employment (see Heckman 1974 and 1979); in order to control
for the possibility that those observed with wages may be systematically different than those not
observed with wages. The net result, from estimating a wage equation on a sample of workers only,
having controlled for selection into employment, is a set of parameter coefficients with which one
can make unconditional predictions regarding wage outcomes for the population as a whole. The
issue is the same here. To be able to evaluate employment outcomes and make predictions relevant
to the population, one must estimate such an outcome on a sub-sample of labor force participants
only, controlling for selection into the labor force. Estimating employment outcomes on
and non-workers together; the result leads to a biased estimate of the determinants of the outcome
When interested the unconditional probability, one must control for unobservable
characteristics that might both affect the labor force participation decision and the employment
outcome. If this potential self-selection is not controlled for, any differences measured in the
affect the labor supply decision of disabled and non-disabled persons. And if these characteristics
change in a systematic way over time, the problem is magnified. A bivariate probit model with
be able to compare these with joint employment and labor force participation outcomes similar to
The first analysis is designed to establish trends in unconditional employment and joint
outcomes. A bivariate probit with selection model is estimated separately for each year of the data,
1981-2000, using the CPS. This section will also look more closely at the trend in labor force
The following model defines the relationship assumed between person i's propensity to be
in the labor force ( LFPi* ), person i's propensity to be employed ( EMPi* ), and individual
characteristics of person i that are believed to affect the labor force participation decision (X1i) and
DISABLEi is equal to 1 if person i is disabled, 0 otherwise. Since LFPi* and EMPi* are not
To account for the fact that EMPi is not defined unless LFPi* >0 (or, LFPi = 1), the relationship
between these two outcomes is specified as a bivariate probit with selection; e1i and e2i are
equal to r.
Several options exist for identifying the parameters of this model. Because of it's non-
linearity, the parameters are identified even if X1 and X2 contain identical regressors. Unfortunately
HOTCHKISS, page 9
relying on the non-linearity of the model does not yield very precise estimates and the model can be
difficult to estimate (Maddala 1983: 281). Alternatively, the parameters are identified if X1 contains
at least one regressor not included in X2 . This "exclusion restrictions" approach is the one most
commonly used (for example, see Averett, et al. 2002 and Mellor 1998). For the model at hand, a
natural exclusion variable, non-labor income, exists. Non-labor income is expected to influence the
labor supply decision but to not be influential in the employment outcome among labor force
participants.8 X1i and X2i both include age; age squared; marital status; state unemployment rate;
dummy variables to indicate female, nonwhite, education, region, central-city resident, worked last
calculating the probability of interest for each person (using the estimated parameter coefficients),
varying the disability index between zero and one, then averaging the difference across the sample.9
Separate specifications are estimated for each year and the marginal impact of having a work
limiting disability is calculated separately for each year. Note that this model specification allows a
comparison of results with earlier studies through calculation of the joint probability.10 Figure 3
reflects the impact of having a work limiting disability on the predicted joint participation and
The impact of having a work limiting disability on this joint outcome intensifies (becomes
more negative) rather dramatically in 1994, corresponding to the second phase of the ADA. The
marginal effect increases from an average of –17 percentage points prior to 1994 to an average of
–20 percentage points between 1994 and 2000. In other words, having a work limiting disability
decreased an individual's joint probability of being in the labor force and employed by three
percentage points more after 1994 than it did prior to 1994. This dramatic relative decline in the
joint probability among disabled people is the result on which both DeLeire (2000) and Acemoglu
Breaking the joint probability into its employment and labor force participation components,
HOTCHKISS, page 10
however, it becomes clear that this dramatic drop in relative employment probability among all
disabled people is driven by a decrease in the labor force participation rate. Figure 4 plots the
predicted employment and predicted labor force participation probabilities for the entire sample as if
each person was disabled using the same parameter estimates from the bivariate probit estimation
that generated Figure 3. After a fairly steady increase in predicted labor force participation, it
declines in 1994 and stays below 1987 levels. At the same time, and with the exception of the
recession years between 1991-1993, the predicted unconditional employment probability for a
Again, in evaluating the barriers facing disabled workers, change in the unconditional
employment probability is a more appropriate measure than the change in the joint labor force and
employment outcome. Consequently, the condemnation by DeLeire (2000) and Acemoglu and
Angrist (2001) of the employment impact of the ADA may be misplaced, since both of these
analyses confound their evaluation of employment changes with changes in labor supply decisions.
The results in Figure 4 show that the decline in employment levels among all disabled people is
labor supply driven and does not reflect an increase in employment barriers for individuals with
disabilities. One may argue that disabled people have decreased their labor supply in response to a
but the fairly encouraging predicted unconditional employment probabilities are not consistent with
this view.12
It may be the case, however, that the condemnation of the ADA by earlier studies should
potentially be aimed at its apparent impact on labor force participation. For example, if the ADA
resulted in lower wages for disabled workers (their employment has now become more costly due
to required accommodations), it is possible that the wage will fall below the reservation wage of a
significant number of disabled labor force participants, causing them to drop out of the labor
market. It could also be the case that a coincidental non-ADA-related event resulted in the
measured decline in labor force participation rates. 13 These possibilities are explored below.
HOTCHKISS, page 11
Although the ADA was not designed as a policy to necessarily increase labor force
participation among disabled people, a precipitous drop in their labor force participation rate, even
remotely attributable to the ADA, is considered by many as undesirable. The disabled and non-
Non-Disabled A B
Disabled C D
Cells A through D contain a given number of people at any given time period. A decrease in the
labor force participation rate among disabled people (LFPRd) corresponds to a decline in the ratio
C/(C+D). This ratio can decline if C decreases and/or if D increases. If C falls, these people have
most likely gone either to A (stay in the labor force but change their identification to non-disabled),
or to D (keep their identification as disabled, but leave the labor force). It is this later possibility
The LFPRd may also fall, however, as a result of an increase in D. An increase in D will
most likely result from people moving from C (disabled leaving the labor force) or from B (non-
disabled, non-labor force participants change their identity to disabled). The later (movement from
B to D) is what might result, for example, from (non-participant) welfare recipients changing their
identification from non-disabled to disabled in order to move off welfare programs and into
disability programs; this movement following the reform of welfare has been documented by the
Lewin Group (1999). Greater effort to be classified (and identified) as disabled might also result
from increased generosity of disability programs themselves (see Autor and Duggan 2001; and
So, are disabled people moving out of the labor force (from C to D) or are more non-
participants identifying themselves as disabled (from B to D)? One way to get an indication of the
HOTCHKISS, page 12
movement across these cells is to evaluate the trends in the percents represented in each cell.16
These trend results are depicted in the chart below. The percent in each cell represents the growth
(+) or decline (-) experienced within that cell between 1990 and 2000. The cells exhaust the
* => sig. at the 99 percent conf. level, + => sig. at the 95 percent conf. level
The largest net change in the cells was movement of non-disabled individuals out of the
non-participant category (cell B). The coefficient indicates that the non-disabled, non-participant
percent declined an average of 0.26 percentage points per year between 1990 and 2000. Even if the
entire increase in cell A (non-disabled labor force participants) came from cell B, that still means the
overwhelming bulk of the increase in cell D (disabled non-participants) came from cell B as well,
not cell C.18 In fact, the smallest net cell change was among disabled labor force participants. This
result provides strong evidence that the observed decline in the LFPRd was not the result of disabled
people fleeing the labor force, but most likely the result of re-identification of some non-labor force
explanation for the decline in LFPRd, this movement from cell B to cell D is a reminder of the
criticism of using self-reported disability status in statistical analyses. This also suggests that
endogeneity will be less of a concern for analyses that focus exclusively on labor fore participants
in evaluating the labor market experience of disabled people (using a self-reported measure of
disability). In other words, there is less movement across disability status among labor market
participants than among non-participants.19 In addition, these results indicate that the observed
decline in the LFPRd should not be of great concern in evaluating the labor market impact of the
ADA.
HOTCHKISS, page 13
This section presents several analyses across time to help quantify the trends identified in
the previous section. The strategy used to accomplish this is to estimate a cross-sectional, time-
series bivariate probit model with dummy variables representing whether the observation shows up
in the data pre-ADA or post-ADA and whether the observation is a disabled or non-disabled
person. These dummy variables are also interacted to determine whether being disabled had any
greater impact on employment after the ADA than before the ADA, relative to the experience of a
non-disabled person.20 While this type of pooled, cross-sectional analysis has been applied by
many researchers (for example, Card 1992, Gruber 1994 and 1996, and Hamermesh and Trejo
2000), the technique also has its critics (for example, Heckman 1996). The primary criticism of this
approach is that it is impossible to control for unobserved changes in the environment that occurred
at the same time as the event of interest. For example, the second phase of the ADA occurred in
1994. This was also the time when the CPS underwent a major overhaul and there is no way to
disentangle these two events. In addition, the economy began its longest running expansion in
recent history at the same time that the ADA was being phased in, an event that could potentially
confound any measurable impact of the ADA through this estimation strategy. One advantage of
the analysis here is that the CPS survey changes should not have a differential impact on disabled
and non-disabled respondents (as the changes did not affect measurement or classification by
disability) ,21 and general business cycle activity should essentially impact disabled and non-
disabled workers in relatively the same proportions.22 Nonetheless, use of the SIPP in addition to
the CPS will help dispel concerns about confounding survey design issues and the state
unemployment rate is included as a regressor in order to capture any general business cycle
influences.
The empirical model looks just like that describing labor force and employment propensities
for each year (equations 1 and 2), except with the additional time-period dummy variables and
interaction terms:
HOTCHKISS, page 14
Again, dichotomous variables are defined as in equations (3) and (4) and the model is estimated as a
bivariate probit with selection since EMPi is not observed unless LFPi* >0 (or, LFPi =1).
described earlier.
In this framework, the affected group (disabled individuals) is controlled for by a dummy
variable indicating whether the individual has a work limiting disability (DISABLE), and the time
period is controlled for by a dummy variable indicating whether the ADA had been implemented yet
or not (POST). The difference in the impact of having a work limiting disability on employment
across the two time periods can be calculated by evaluating the probabilities of interest for each
person, varying the DISABLE and POST dummy variables, then taking the difference between these
The decision of when one would expect the ADA to have its strongest impact (in other
words, how to define POST) is debatable. One might expect some impact when the ADA was
enacted (1990). However, employers were not required to respond until July 1992 (for employers
with 25 or more employees) and July1994 (for employers with 15 or more employees). The year
1993 was chosen for defining post since that is the first full year of implementation of the ADA.
The results presented in Table 3 are consistent with the conclusions drawn from Figure 4.
Namely, labor force participation among disabled people declined significantly after implementation
of the ADA relative to labor force participation among non-disabled people (the coefficient on
while all other regressors contribute significantly to explaining employment (all but three at the 99
HOTCHKISS, page 15
percent confidence level), being disabled after ADA implementation is not one of them; disabled
people are no more or less likely to be employed than non-disabled people, post-ADA relative to
pre-ADA. In other words, the ADA has not changed the relative employment probability between
disabled and non-disabled workers. Translating these parameter coefficients into partial derivative,
the labor force participation rate declined by about five percentage points more among those
classified as disabled than among non-disabled people post-ADA.24 In addition, the employment
probability for a random disabled person, relative to a non-disabled person, declined by at most 0.3
of a percentage point post-ADA, although the coefficient is not significantly different from zero.
The analysis above is repeated for separate samples of men and women; the results for the
key parameters are reported in Table 4. For both men and women, labor force participation among
those classified as disabled declined significantly post-ADA relative to the labor force participation
among non-disabled people. The parameter coefficients (along with the separate sample
characteristics) translate into roughly a 4.7 percentage point decline for both men and women. The
upper bound on the employment decline (although neither is significantly different from zero) is
0.4 of a percentage point for men and 0.8 of a percentage point for women.
In order to test the robustness of the CPS results, equations (5) and (6) are re-estimated
using the sample obtained from the SIPP for the years 1986 through 1997. Table 5 reports the
coefficients of interest. The results reported in Table 5 mirror those in Table 3, with one difference:
the employment probability for a random disabled person increased more post- versus pre-ADA
than did the employment of a non-disabled person. This positive 0.0535 coefficient on
DISABLE*POST translates into 0.5 of a percentage point higher employment probability for a
disabled person relative to a non-disabled person. In addition, the relative, significant decline in
labor force participation among those classified as disabled is found using the SIPP data, as well.25
is allowed to vary by type of disability. The estimating equations are analogous to equations (5)
and (6) with the indicator variable, DISABLEi, being replaced by four indicators of disability
Selected estimation results presented in Table 6 from the SIPP data set are also consistent
with the conclusions drawn using the CPS data: labor force participation declined more for all
classifications of disability (except MENTAL and OTHER), relative to non-disabled people, post-
versus pre-ADA. Employment probabilities (controlling for labor force participation) increased
significantly more for the disability classification of OTHER than for non-disabled people, post-
versus pre-ADA. Relative employment probabilities did not change significantly for the other
classifications of disability. Since there is no indication of what condition might lead to a disability
unclear. The conservative interpretation of the results from using the SIPP data, therefore, is
directly parallel to that from the CPS results: employment outcomes neither improved nor
Finding that the ADA did not have an impact on employment among disabled people may
not come as a surprise to those familiar with the fact that by the time the ADA was passed, all states
had passed some form of legislation addressing discrimination against disabled workers (see Table
7).27 In other words, the environment in which the ADA was passed was arguably already a post-
ADA environment. Given this, one might suggest that the ADA was redundant; the states were
already addressing the problem of discrimination against disabled workers and there was no need
for federal legislation. And this may be why no impact of the legislation is measured at the national
level.
federal ADA, the bivariate probit with selection estimation procedure detailed earlier is repeated on a
sub-sample of states that enacted disability legislation between 1981 (the beginning of available
data) and 1991 (the last year before the ADA took effect). The condition for being included in the
sub-sample was whether the worker resided in a state that adopted protection for disabled workers
during this period, so that each state would have some pre- and post-legislation observations.28 In
order to be consistent with the provisions of the ADA, the date at which legislation covering both
public and private employers was adopted was used to distinguish pre- and post-time periods. This
provides an environment similar to that which existed at the national level, since federal contractors
were already covered by disability anti-discrimination policies prior to the ADA. Some additional
regressors (log population and real gross state product) were added to the employment specification
Table 8 contains the parameter estimates from the bivariate probit with selection estimation
of labor force participation and employment probabilities. The first thing to notice from this table is
that nearly all of the parameter estimates are of the same sign and magnitude as those in Table 3
which correspond to the national sample.29 The implication of the similarity across the national and
state-level analyses is that the observations in this sub-set of states are not at all far from the norm
and that the results on the regressors of interest (those related to disability status) should be
Two important results stand-out from the state-level analysis. While, overall, disabled
DISABLE*POST is insignificantly different from zero). And, like the national analysis, in a
regression where most regressors are significant at the 99 percent confidence level, this is notable.
The second result of particular interest is related to the determination of labor force
participation. Recall that in the national analysis there was a dramatic decline in labor force
participation rate among those classified as disabled, post-ADA. If indeed the ADA legislation
HOTCHKISS, page 18
caused disabled people to flee the labor market, similar legislation at the state level should result in
similar behavior. The results in Table 8, however, indicate that the state-level disability legislation
had no such impact on the labor force participation among disabled people; the coefficient on
DISABLE*POST in the labor force participation equation is not significantly different from zero.
And, again, this is in an estimation where nearly all the other regressors are significant at the 99
percent confidence level. This result supports the earlier contention that the drop in the labor force
participation rate that occurred in 1994 at the national level can not be attributable to the ADA and is
likely the result of some other confounding factor (for example, modifications in the welfare and
The purpose of this paper was to evaluate the relative employment experiences of disabled
and non-disabled individuals. A pooled, cross-sectional analysis determined that the joint labor
force and employment probability among disabled people decreased significantly by roughly four
percentage points more after implementation of the ADA relative to a non-disabled person's joint
probability. This joint outcome, however, is influenced by both supply and demand factors and
therefore confounds the employment experience of a disabled person with labor supply influences.
It is found that the unconditional employment probability among disabled people, relative to among
non-disabled people, did not change significantly after implementation of the ADA. This lack of
deterioration in the employment condition facing disabled workers suggests that the rather dramatic
declines in measured labor supply among disabled people is not likely the result of feedback effects
or fear of negative outcomes. In fact, evidence is provided to indicate that the reduction in the labor
force participation rate among those classified as disabled (which drives the negative joint outcome)
is not the result of disabled people leaving the labor force, but, rather, a reclassification of non-
disabled labor force non-participants as disabled. This phenomenon likely occurred as a result of
more stringent welfare reform requirements and more generous federal disability benefits. The fact
that there was no parallel decline in the labor force participation rate among disabled individuals in
states that had enacted disability legislation prior to 1991 provides additional evidence that the ADA
HOTCHKISS, page 19
was not the source of the labor force participation rate decline post-1993.
Analysis of the SIPP data provided confirmatory evidence of the CPS results and allowed a
closer evaluation of employment probabilities by type of disability. It was found that those with
disabilities classified as "other" experienced a positive employment impact of the ADA. All other
classifications of disability did not experience any different employment probability growth than
While the results of this paper challenge the current conventional wisdom that the ADA had
a detrimental, unanticipated impact on employment outcomes among disabled people, they also do
not point to any substantial improvements in those outcomes either. Rather than act as a catalyst for
change, the labor market provisions of the ADA may prove to be more useful in providing an
environment in which more narrowly targeted policies can succeed in improving labor market
References
Acemoglu, Daron and Joshua Angrist. 2001. "Consequences of Employment Protection? The
Case of the Americans with Disabilities Act." Journal of Political Economy 109(5): 915-
957.
and State Compliance with Employment Protections and Architectural Barrier Removal A-
Americans with Disabilities Act of 1990 (July 26 1990) 42 US Code 12101, Sec (2). Also see
<http://www.eeoc.gov/laws/ada.html>.
Autor, David H. and Mark Duggan. 2001. "The Rise in Disability Recipiency and the Decline in
Technology.
Averett, Susan L., Daniel I. Rees, and Laura M. Argys. 2002. "The Impact of Government Policies
Black, Dan, Kermit Daniel, and Seth Sanders. 2002. "The Impact of Economic Conditions on
Participation in Disability Programs: Evidence from the Coal Boom and Bust." American
Bound, John and Timothy Waidmann. 2002. "Accounting for Recent Declines in Employment
Rates Among Working-Aged Men and Women with Disabilities." Journal of Human
__________. 1992. "Disability Transfers, Self-Reported Health, and the Labor Force Attachment
of Older Men: Evidence from the Historical Record." Quarterly Journal of Economics
57(4): 1393-1419.
Burkhauser, Richard V., Mary C. Daly, Andrew J. Houtenville, and Nigar Nargis. 2001. "How
Working-Age People with Disabilities Fared over the 1990s Business Cycle." In Ensuring
HOTCHKISS, page 21
Health and Income Security for an Aging Workforce, ed. Peter P. Budetti, Richard V.
Burkhauser, Janice M. Gregory, and H. Allan Hunt, 291-346. Kalamazoo, Mich.: W.E.
Card, David. 1992. "Do Minimum Wages Reduce Employment?" Industrial and Labor Relations
Carrington, William J., Kristin McCue, and Brooks Pierce. 2000. "Using Establishment Size to
Measure the Impact of title VII and Affirmative Action." Journal of Human Resources
35(3): 503-23.
Chay, Kenneth Young. 1996. "An Empirical Analysis of black Economic Progress Over Time."
Chirikos, Thomas N.1991. "The Economics of Employment." In The Americans with Disabilities
Act: From Policy to Practice, ed. Jane West, 150-79. New York: Milbank Memorial Fund.
DeLeire, Thomas Charles. 2000. "The Wage and Employment Effects of the Americans with
DuMouchel, W.H. and G.J. Duncan. 1983. "Using Sample Survey Weights in Multiple
78(383): 535-43.
Ginther, Donna K. and Kathy J. Hayes. 2003. "Gender Differences in Salary and Promotion for
Gruber, Jonathan. 2000. "Disability Insurance Benefits and Labor Supply." Journal of Political
Hale, Thomas W. 2001. "The Lack of a Disability Measure in Today's Current Population
Hamermesh, Daniel S. and Stephen J. Trejo. 2000. "The Demand for Hours of Labor: Direct
Haveman, Robert and Barbara Wolfe. 1984. "The Decline in Male Labor Force Participation:
Heckman, James J. 1979. "Sample Selection Bias as a Specification Error." Econometrica 47(1):
153-61.
__________. 1974. "Shadow Prices, Market Wages, and Labor Supply." Econometrica 42(4):
679-94.
__________. 1996. Comment. In Tax Policy and The Economy, ed. James M. Poterba, 32-8.
Kaye, H. Stephen. 2001. "Improved Employment Opportunities for People with Disabilities."
Paper presented at the conference on the Persistence of Low Employment Rates for People
Kirchner, Corinne. 1996. "Looking Under the Street Lamp: Inappropriate Uses of Measures Just
Because they are There." Journal of Disability Policy Studies 7(1): 78-90.
Kreider, Brent and John V. Pepper. 2002. "Disability and Employment: Reevaluating the Evidence
Kruse, Douglas and Lisa Schur. 2003. "Employment of People with Disabilities Following the
Kubik, Jeffrey D. 1999. "Incentives for the Identification and Treatment of Children with
73(2): 187-215.
Lewin Group, The. 1999. Policy Evaluation of the Overall Effects of Welfare Reform on SSA
Long, J. Scott. 1997. Regression Models for Categorical and Limited Dependent Variables.
Manski, C.F. and D. McFadden. 1981. Structural Analysis of discrete Data with Econometric
Mellor, Jennifer M. 1998. "The Effect of Family Planning Programs on the Fertility of Welfare
Recipients: Evidence from Medical Claims." Journal of Human Resources 33(4): 866-95.
Parsons, Donald. 1980. "The Decline in Male Labor Force Participation." Journal of Political
Stern, Steven. 1989. "Measuring the Effect of Disability on Labor Force Participation." Journal of
__________. 1996. "Semiparametric Estimates of the Supply and Demand Effects of Disability
Endnotes
1 This issue of potential endogeneity of self-reported disability status is discussed in greater detail
in endnote 4.
2 Details of the data set construction are presented in the next section.
3 It has also been suggested that persons with disabilities entering the labor force after ADA will
have more severe disabilities than those employed prior to ADA, making the potential for "binding"
accommodation requirements that much more likely and expensive (Chirikos 1991).
4 While some have questioned whether self-reported disability status (as in the CPS) suffers from
endogeneity (for example, see Parsons 1980, Haveman and Wolfe 1984), Stern (1989) finds that,
"any bias due to potential endogeneity is small" (p. 363). Of course, endogeneity may be more of a
concern since the passage of the ADA. And, as will be addressed later, endogeneity among the
population as a whole may be more of a problem than among only labor force participants.
Additional practical matters related to using the disability indicator in the CPS March income
supplement are detailed by Hale (2001). These issues are further complicated by the matching
undertaken to expand the number of observations. It is for these reasons that confirmatory
evidence of the CPS results are sought from an additional data source, as described below. Further
considerations of a disability identifier are proffered by Kruse and Schur (2002). They find
different labor market outcomes depending on the definition of disability used. While Kruse and
Schur argue that the definition that includes "work limitation" (as used in this paper) may not be the
definition most appropriate regarding ADA coverage, it is likely the group for which we will see the
5 Details of the matching procedure are available from the author upon request. Due to the
complicated matching across one to four months of the CPS, all analyses are performed un-
weighted. According to Wooldridge (1999), "stratification based on exogenous variables does not
cause any problems: estimators that ignore the stratification are consistent and asymptotically
normal, and the usual variance matrix estimators are consistent" (p. 1386). Since stratification in
HOTCHKISS, page 25
the CPS sampling design is based on exogenous variables (geographic and demographic), and the
attrition that results from the matching procedure is likely un-systematic, the use of weights would
produce inefficient parameter estimates (for further evidence on this point, also see DuMouchel and
Duncan 1983 and Manski and McFadden 1981). In addition, any effect of stratification on the
estimation can be accounted for by including indicator variables that correspond to the strata
(Ginther and Hayes 2001), so the inclusion of demographic variables (including disability status)
should control for any observable effect sampling based on those characteristics might have (either
initially or through attrition of matching). It should also be noted that any systematic attrition or
sample loss due to unobservables will not be accounted for, but also can not be corrected using
weights, either.
6 Details of the construction are available from the author upon request.
7 Details of the SSA listing of impairments for disability status purposes can be found on the
internet at <http://www.ssa.gov/OP_Home/cfr20/404/404-ap09.htm>.
8 As pointed out by a referee, it is possible that non-labor income could contain income sources
determined by labor force participation decisions (such as disability income), making that regressor
endogenous to the labor force participation equation. The inclusion of such income sources,
however, are crucial to capture the potential impact of growing program generosity on labor force
participation over the time period studied in this paper. Estimation of the model excluding disability
and public assistance income from non-labor income did not change the results. In addition, this
model was dominated (based on a likelihood ratio test) by the model with the more inclusive
9 This method of calculating the marginal effect of a change in a dummy variable is referred to as a
measure of discrete change and is described in greater detail by Long (1997, 135-8).
10 The comparison is not exact, of course, since earlier studies employ different estimation
techniques and data. DeLeire (2000) estimates a univariate probit model to determine the impact of
having a disability on the probability of employment for a sample of all disabled and non-disabled
persons, using the SIPP. Acemoglu and Angrist (2001) estimate a linear model of weeks worked
HOTCHKISS, page 26
during the year among all disabled and non-disabled persons, using the CPS. Both of these
analyses result in estimates comparable (at least in sign and significance) to the prediction of the
12 In addition, Stern (1996) presents empirical evidence that labor supply decisions of disabled
people are driven more by labor supply factors than by labor demand factors.
13 For example, Kaye (2001) offers one non-ADA reason why disabled people might be leaving
the labor force; he claims that the severity of disabilities has been growing among disabled people
14 Kirchner (1996) identifies the potential movement from C to A as a sign that the ADA is
"working;" providing a reason (better job opportunities) for the disabled to re-classify themselves
15 Acemoglu and Angrist (2001) dismiss this theory by showing that controlling for receipt of
disability benefits only marginally impacts their results. They fail to point out, however, that actual
receipt of benefits will reflect only a fraction of the actual desire to receive benefits (see Kubik
1999). Consequently, the actual impact of growing program generosity on the disability status
change for non-participants could be much larger than that measured by growing recipiency.
16 Of course, a precise measure of the exact number of people in each cell re-classifying
themselves requires the use of panel data and is beyond the scope of this paper. The examination
here is merely suggestive, but is consistent with what others have found.
17 These trend coefficient estimates for each cell were obtained from simple linear regressions of
the percent of people represented in that cell as a function of a time trend corresponding to the
period 1990-2000 (the period was chosen in order to focus on post-ADA changes).
18 Again, as Kirchner (1996) and Hale (2001) suggest, some of the increase in A may have come
19 This finding is consistent with evidence provided by Black, et al. (2002) who find that the
generosity of disability policy has more of an impact on low-skill labor force participation than on
HOTCHKISS, page 27
labor force participation among those with promising employment prospects. In other words, those
who are not in the labor force are more inclined to take advantage of generous disability programs
by changing their status, than those who are in the labor force. Also see Bound and Waidmann
(1992) for results on this issue for older men and Kreider and Pepper (2002) on similar results for
20 The strategy described here can be likened to the popular "differences-in-differences" (DD)
21 Acemoglu and Angrist (2001, Appendix A) show that results are fairly consistent across a
variety of sample restrictions based on differences between the 1993 and 1994 samples (crossing
the survey modification time period). Consequently, it is not expected that the results reported here
23 Footnote 8 provides a general idea of this strategy, although an additional variable (POST) is
25 The CPS analysis was repeated using the SIPP years (1986-1997). The results were nearly
identical to the full CPS results, suggesting that the positive coefficient on DISABLE*POST from
the SIPP results are specific to the SIPP data set, rather than specific to the years for which the
27 Also see Advisory Commission on Intergovernmental Relations (1989) for further details.
28 Methodologies that take advantage of differing legislative statuses across states (or, more
generally, across observations) have often been referred to as "natural experiments," and have been
applied by a number of researchers. For example, see Chay (1996) and Carrington, et al. (2000).
HOTCHKISS, page 28
29 The exceptions to this are the Nonwhite dummy variable in both equations, the College Grad
dummy variable in the LFP equation, the advanced degree dummy variables in both equations, the
Central City dummy variable in the EMP equation and some of the regional dummy variables.
30 In addition, the ADA analysis was repeated for this sub-set of states for the full-CPS time
period resulting in estimates nearly identical to those for the full national sample.
HOTCHKISS, page 29
Percent 90
Employed
80
Non-disabled
70
60
50
40
30
Disabled
20
10
0
1981 1983 1985 1987 1989 1991 1993 1995 1997 1999
(a) (b) (c) Year
(a) ADA Enactment
(b) ADA Phase I
(c) ADA Phase II
Figure 1
Percent Employed of Disabled and Non-disabled: Individuals 21-58 Years Old, CPS 1981-2000
HOTCHKISS, page 30
Percent 100
Employed
95
Non-disabled
90
85
Disabled
80
75
1981 1983 1985 1987 1989 1991 1993 1995 1997 1999
(a) ADA Enactment (a) (b) (c)
Year
(b) ADA Phase I
(c) ADA Phase II
Figure 2
Percent Employed of Disabled and Non-disabled Labor Force: Labor Force Participants 21-58
Years Old, CPS 1981-2000
HOTCHKISS, page 31
Difference
in -0.12
Probability
Across -0.15
Disability
Status
-0.18
-0.21
-0.24
1981 1983 1985 1987 1989 1991 1993 1995 1997 1999
(a) ADA Enactment (a) (b) (c) Year
(b) ADA Phase I
(c) ADA Phase II
Figure 3
Impact of Disability on the Joint Labor Force Participation and Employment Probabilities, CPS
1981-2000
HOTCHKISS, page 32
0.95
Probability
0.9
Predicted Probability
0.85
of Employment
0.8
0.75
0.7
0.65
Figure 4
Separate Predictions of Employment and Labor Force Participation Probabilities, DISABLE=1,
CPS 1981-2000
HOTCHKISS, page 33
Table 1
Not Not
All Employ- Employ- All Employ- Employ-
ed ed ed ed
Age 42.96 40.19 44.03 37.27 37.41 36.75
(10.48) (10.28) (10.37) (10.34) (10.11) (11.12)
Table 2
Not Not
All Employ- Employ- All Employ- Employ-
ed ed ed ed
Age 42.35 40.72 43.41 37.25 37.44 36.44
(10.30) (9.99) (10.36) (10.03) (9.82) (10.84)
Type of Disability:
Table 3
Labor Force Participation and Employment Bivariate Probit with Selection Results,
Labor Force
Regressor Participation Employment
Equation Equation
Intercept -2.7130* -0.2116*
(0.0272) (0.0777)
Midwest = 1 0.0216* +
0.0143
(0.0057) (0.0071)
Rho 0.1464*
(0.0334)
Log-likelihood -416,463
significant at the 95 percent confidence level; ^ => significant at the 90 percent confidence level.
Table 4
Selected Labor Force Participation and Employment Bivariate Probit with Selection Results by
Men Women
Notes: Standard errors are in parentheses. Additional regressors included age; age squared; state
unemployment rate; nonwhite, education, regional, central city, and worked-last-year dummy
variables; marital status; and non-labor income (labor force participation only). See notes to Table
Table 5
Selected Labor Force Participation and Employment Bivariate Probit with Selection Results,
Labor Force
Participation Employment
Regressor Equation Equation
DISABLE = 1 -1.0736* -0.2086*
(0.0105) (0.0199)
Rho -0.5618*
(0.0256)
Log-likelihood -208,130
Notes: Standard errors are in parentheses. Additional regressors included age; age squared; state
unemployment rate; female, nonwhite, education, regional dummy variables; an indicator for SMSA
residence; marital status; and non-labor income (labor force participation only). See notes to Table
Table 6
Selected Labor Force Participation and Employment Bivariate Probit with Selection Results by
Labor Force
Participation Employment
Regressor Equation Equation
MUSCULOSKELETAL = 1 -0.9302* -0.2357*
(0.0146) (0.0255)
Rho -0.5593*
(0.0257)
Log-likelihood -207,794
Notes: Standard errors are in parentheses. See notes to Table 5 regarding additional regressors.
See notes to Table 3 for significance levels. Sample contains individuals ages 21-58.
HOTCHKISS, page 40
Table 7
Source: Author's determination from original legislative sources for each state.
a Exact original coverage not available; classification based on current coverage definitions.
HOTCHKISS, page 42
Table 8
Labor Force Participation and Employment Bivariate Probit with Selection Results, State Analysis
1981-1991
Labor Force
Regressor Participation Employment
Equation Equation
Intercept -2.7222* -0.3876
(0.0865) (0.3551)
Midwest = 1 + -0.0529
-0.0553 (0.0373)
(0.0225)
South = 1 0.0137 0.1007*
(0.0183) (0.0245)
Rho 0.1949^
(0.1046)
Log-likelihood -44,700
Notes: Standard errors are in parentheses. See notes to Table 3 for significance levels. States
included in the analysis are Alaska, Arizona, Delaware, Idaho, Massachusetts, North Carolina, North
Dakota, South Dakota, Texas, and Wyoming. Sample contains individuals ages 21-58.