J
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/consequencesofemOOacem
m-^.^B
working paper
department
of
economics
«iiiiK_
CONSEQUENCES OF EMPLOYMENT PROTECTION?
THE CASE OF THE AMERICANS WITH DISABILITIES ACT
Daron Acemoglu
Joshua Angrist
September 1998
massachusetts
institute of
technology
50 memorial drive
Cambridge, mass. 02139
WORKING PAPER
DEPARTMENT
OF ECONOMICS
CONSEQUENCES OF EMPLOYMENT PROTECTION?
THE CASE OF THE AMERICANS WITH DISABILITIES ACT
Daron Acemoglu
Joshua Angrist
98-13
September 1998
MASSACHUSEHS
INSTITUTE OF
TECHNOLOGY
50
MEMORIAL DRIVE
CAMBRIDGE, MASS. 02142
MASSAC
~
C^
JAN 2
-STlTint
-
5Y
8 '959
LIBRARIES
First Version;
December 1997
Revised: June 1998
Consequences of Employment Protection?
The Case of the Americans with DisabiHties Act'
Daron Acemoglu and Joshua Angrist^
ABSTRACT
The Americans With
DisabiHties Act
(ADA)
workers and outlaws discrimination against the disabled in hiring,
ADA
The
was meant to increase employment
of the disabled,
ADA
had a
negati\-e effect
women under
to profits.
is
ADA
age 40. The effects appear to be larger in
ADA. The
men
ADA
most important and how
ADA.
CPS
size firms, possibly
effects
is
because
where there have
on hiring and
affect separations.
Finally, there
suggest that the
working ages and disabled
effects are also larger in states
reduced hiring of the disabled but did not
a pure firing-costs interpretation of the
are
of all
medium
been more ADA-related discrimination charges. Estimates of
the
and pay. Although the
Empirical results using the
on the employment of disabled
small firms were exempt from the
firing,
also increases costs for employers.
it
net theoretical impact turns on which provisions of the
responsive firm entry and exit
accommodate disabled
requires employers to
firing suggest
This weighs against
httle evidence of an
impact on
ADA
have been
the nondisabled, suggesting that the adverse employment consequences of the
limited to the protected group.
'We thank Lucia Breierova and Chris Mazingo for outstanding research assistance, and Patricia Anderson,
Peter Diamond, Alan Krueger, Paul Oyer, James Poterba, Steve Pischke and seminar participants in the
N.B.E.R. Labor Studies Meeting, the 1998 North American Meeting of the Econometric Society, M.LT. and
U.C.L.A. for helpful comments. Special thanks go to to Leo Sanchez for help with data from the Equal
Employment Opportunity Commission and to Greg Weyland at the Census Burea for consultations regarding
CPS matches and the CPS redesign. The authors bear sole responsability for the contents of this paper.
^Department of Ikonomics, Massachusetts Institute of Technology.
E-mail: datronSmit edu.angristSmit edu
.
.
Introduction
I.
Government
Pay Act
efforts to eliminate
and
of 1963
employment and wage discrimination date back to the Equal
Title VII of the Civil Rights
The consensus among
crimination on the basis of race and sex.
rights legislation
and related
Act of 1964. These laws prohibited
labor economists
is
that
improvement
in
economic
federaJ regulations led to a substantial
conditions for Black Americans (see, for example, Freeman, 1981;
Heckman and Payner,
or Leonard, 1990), while the evidence on the impact of antidiscrimination policy
less clear
civil
1989;
on women
is
cut (Leonard, 1989).
The most
recent pieces of federal antidiscrimination legislation are the Americans with Dis-
Act (ADA) and the
abilities
dis-
to offer reasonable
disabled in
Act of 1991 (CRA-91). The
Civil Rights
accommodation
wage determination,
and
The
firing.
requires employers
and bans discrimination against the
to disabled employees,
hiring,
ADA
ADA
seems to be more far-reaching
Mian CRA-91, which essentially modified existing antidiscrimination statutes.^
ADA
propo-
nents hope the act will improve the labor market fortimes of disabled workers. Critics of the
ADA
have argued that adapting the workplace to the disabled can be expensive and that the
costs of
accommodation and ADA-related
may have
significant negative
employment
Rosen, 1991; Oi, 1991; Weaver, 1991; Epstein, 1992; Olson, 1997).
effects (e.g.,
The
litigation
first
objective of this paper
economic conditions
for the disabled.
is
A
to evaluate
study of the
whether the
ADA
ADA
has in fact improved
also of broader interest, however.
is
Altliough high-profile reasonable accommodation cases have attracted the most media attention, the
majority of
ADA
charges are for wrongful termination.
ADA acts as a form of employment protection, similar
ADA primarily affects a specific group, the consequences
the
therefore possible that
It is
to
European
of
employment protection may be
firing costs. Since the
easy to detect in this case. Moreover, contrary to early experience with the Civil Rights Act
of 1964, in recent years,
most discrimination charges under
wrongful termination (EEOC, various years).
this
change
in civil rights
law enforcement
the disabled, a possibiUty raised by
The theory
If firing
may
all
statutes have been related to
costs have
ultimately
disemployment
effects,
then
harm protected groups other than
Donohue and Siegelman
(1991, 1993).
section of the paper uses a standard competitive
model
to highlight the distinc-
tion between hiring
and
disabled workers.
Although the reasonable accommodation provision creates an incentive to
firing costs
due to the threat of lawsuits and the costs of accommodating
employ fewer disabled workers, the introduction of hiring and
sis.
If
^See
firing costs
comphcates the analy-
the threat of ADA-related litigation encourages employers to increase the hiring of the
Abram
(1993).
Oyer and SchaefFer (1998) explore the labor-market impact of CRA-91.
and
disabled,
may
if
the
increase the
most charges are
the
ADA
The
number
of employers
employment
not very responsive to profits or costs, the
is
ADA
of disabled workers as
ADA
proponents had hoped. But when
wrongful termination and costs of reasonable accommodation are high,
for
hkely to reduce disabled employment.
is
empirical analysis looks at the emplojonent and wages of disabled and nondisabled
workers using data from the March Current Population Surveys (CPS) for 1988-1997. These
data are useful
ers,
for
and because the March
and
theoretical discussion,
the impact of the
matching
CPS
ADA
in the
ADA's compliance and
rotation groups. Finally,
men, with no
men, the
women aged
results for
effect
sanction provisions.
EEOC
we use
To
investigate
data on discrimination charges by state
ADA
activity.
had a negative impact on the employment
on wages. These results are unchanged by controlling
and Supplemental Security Income
women
We
are mixed.
ADA
The
women aged
impact on nondisabled workers
consequences of the legislation
in this case
for pre-
40-58.
in
exempt from the
ADA
(SSI). In contrast to the results
appears to have had a negative effect on
We also
find
no evidence that the
ADA
any age group, suggesting that the adverse
have been limited to the protected group.
check the basic findings in a number of ways.
cmplo\'mcnt of the disabled declined more
ADA.
work-
a variable that figures in our
market variables with the incidence of ADA-related charge
21-39, but no effect on
liad a ncgati\-e
are
size,
identifies disabled
trends in disabled employment and for the increase in the fraction of people receiving
Disability Insurance (DI)
for
has information on firm
empirical results suggest that the
of disabled
ADA
CPS
supplement
on turnover, we construct measmes of separations and accessions by
to connect changes in labor
The
CPS income
our purposes because the
in
First,
there
is
some evidence that
medium-size firms, possibly because small firms
and large firms can more
easily absorb the costs
imposed by the
Second, effects are generally larger in states where there have been more ADA-related
discrimination charges.
Finally, there
is
some evidence
of a negative
impact on hiring of the
disabled, but little evidence for a reduction in disabled separation rates. This suggests that the
negative effects of the
ADA may
be due more to the costs of reasonable accommodation than
to the threat of wrongful termination lawsuits.
The paper
is
organized as follows. Section
II
gives
some background and discusses related
literature. Section III provides a theoretical analysis of provisions that protect disabled workers.
Section IV describes the data and oxn empirical strategy. Section
findings
and Section VI concludes.
V contains
the main empirical
II.
A.
ADA
Provisions and Coverage
The
ADA
was signed
Background
into law in July 1990
and came into
Jnly 1992.
effect in
Previously,
there was no federal law dealing with the employment and wages of disabled workers in the
private sector, although the Rehabilitation Act of 1973 covered disabled workers employed by
A
the federal government or working for federal contractors.
number
of states also
had laws
protecting disabled workers, but the coverage and effectiveness of these laws varied. Title
the
ADA
initially
covered
all
employers with at least twenty
was extended to employers with
fifteen or
more employees.
five
Title
I
I
of
employees. In 1994, coverage
requires employers to provide
"reasonable accommodation" for their disabled workers. Examples include enabling wheel chair
equipment
access, purchasing special
for disabled
employees, and job restructuring to permit
disabled employees to work part time or from home. Title
the disabled in wages, hiring,
firing,
I
also
bans discrimination against
and promotion. For example, a disabled employee should
be paid the same amount as a nondisabled worker in the same job, and firms are not allowed
to consider disability in hiring
Enforcement of
(EEOC) and
and
firing decisions.^
ADA provisions is left to the Equal
the courts.
Employment Opportunity Commission
Disabled employees or job applicants
discriminated against can
file
a charge with the
EEOC, which
who
believe they have been
will investigate
If
sue on behalf of the charging party,
issues a letter of permission to sue
party
is
free to litigate at his or her
hiring, reinstatement, promotion,
payment
it
for firms
more employees (EEOC,
The underlying
and court
also available
(see, e.g.
much worse than
1991).
ADA
are
much
The
the nondisabled
what nondisabled workers
less likely to
does not
and the charging
remedies that include
costs.
if
As a consequence of CRA-91,
intentional discrimination
is
found.
1995, p. X-8).
logic of the
Kemp,
some
with 100 or fewer employees to $300,000 for firms with 500
seems to be that employers incorrectly perceive the
disabled to be less productive, or are unwilling to
them
for
in
back pay, front pay, and reasonable accommodation, and for
of attorney's fees, expert witness fees,
These range from $50,000
is
own expense. The law provides
compensatory and punitive damages are
or
the charge
EEOC
not resolved and the
cases try to resolve the charge or sue.
and
is
make modest adjustments
fact that the labor
not in dispute.
(see, e.g.,
accommodate
market fortunes of the disabled are
The
disabled earn 40 percent or less of
earn. Their labor force participation rates are
be employed
to
much
lower and they
Burkhauser and Daly, 1996, or our
statistics,
^xitle II covers discrimination in public programs, and Titles III and FV refer to public accommodations
(busmesses) and telecommuiucation. Title V contains technical information related to enforcement (see
EEOC
1995).
ADA
below).
proponents believe the law
induce companies to
will
make the investments and
modifications necessary to employ disabled workers, and reduce imjustified discrimination. In
recent years, interest in the labor market performance of the disabled has also been fueled by
efforts to
reduce the number of Disability Insurajice recipients
From July 1992
various years).
Of
September 1997, the
to
these, 29 percent
mention
EEOC
(i.e.,
our interpretation of the
B.
We
t's
How
Costly
is
the
The majority
Leonard, 1991)
ADA
received 90,803
"failure to provide
percent are for discrimination at the hiring stage.
wrongful termination
(see, e.g.,
charges
accommodation", while
as providing a
9.4
of charges, 62.9 percent, are for
discharge, failure to rehire, suspension, or layoff).
ADA
(EEOC,
This motivates
form of employment protection.
ADA?
have not foimd representative data on the costs of accommodation, though the Presiden-
Committee on Employment
of People with Disabihties has surveyed
some employers who
contacted them for help accommodating their disabled workers (Job Accommodation Network,
This survey shows an average cost of $930 per accommodation since October 1992.
1997).
ADA
regulations
litigation.
Since July
Other costs of accommodation include time employers spend dealing with
and reduced
An
efficiency
due to a forced restructuring.
important component of
ADA
costs results
1992, over 11,000 of the charges brought under the
from the threat of
ADA
were resolved by the
EEOC, and
employers paid over $174 million in settlements (EEOC, various years). This figiue does not
reflect
administrative costs, lawyers
we do not have data on
ADA
fees,
suits alone,
and private settlements
Condon and Zolna
in or
out of court. Although
(1997) report that employees
file
over 40,000 cases each year with state and federal courts, the majority related to discrimination,
and win almost 60 percent of the time. They estimated an average award of over $167,000 and
defence costs of over $40,000
(less
than the $80,000 estimated by Dertouzos, 1988, for wrongful
termination suits in California).
The
cies for
ADA may also have been a factor in
Employment
lawsuits.
the development of a
new insurance market,
poli-
Practices Liability Insurance (EPLI), which covers the costs of employee
The EPLI market
started in late 1990 and has since
grown
rapidly, with
minimum
premia ranging from $4,500 to $20,000 a year (Clarke, 1996). This suggests that the costs of
the
ADA
are indeed a real concern for employers.
Related Literature
C.
DeLeire (1997)
aware
is
and
first
He used the
of.
Income Dynamics
fore
the
after the
to
ADA
systematic empirical study of the impact of the
Siurvey of
that
we
are
Income and Program Participation and the Panel Siu-vey of
compare labor market outcomes
ADA. Our approach
for disabled
and non-disabled workers be-
and a number of empirical
uses different data
strategies
not explored by DeLeire, but we also begin with similar comparisons of changes in outcomes
ADA
by disability status. Since the
work on the impact of
and
creates firing costs, our paper
Europe
firing costs in
Nickell, 1997). Related theoretical
(e.g.,
is
also related to empirical
Lazear, 1990; Addison and Grosso, 1996;
papers include analyses of labor market regulations by
Mortensen (1978), Summers (1989), Lazear (1990), Bertola and Bentohla (1990), and Hopen-
hayn and Rogerson (1993), and the analysis of optimal
Slieshinski (1995). Finally, our
mandated
work
benefits (e.g., Gruber, 1994;
analyses of the impact of earlier
on the impact of
also related to the empirical hterature
Ruhm,
1996; Waldfogel, 1996) and, as noted above, to
civil rights legislation.
Consequences of Protecting Disabled Workers: Theory
III.
The
is
by Diamond and
disability benefits
theoretical consequences of the
ADA are explored using a standard competitive model with
two types of workers. Nondisabled workers supply labor according to the function na{wa) and
the disabled supply labor according to nii{wd), where Wa
workers and
wage
factor
wj^ is
rate for
P <
i
=
the wage rate for disabled workers.
All workers are infinitely lived
a.d.
There are
I.
M firms
in the labor
potential firms that can enter at cost F. This
both a market characterized by
is
fixed
{M >
and F
—
free entry
is
We
assume that
and
is
is
and a large number of
a convenient formulation enabling us to discuss
(when
M —^
0)
and one where the number of firms
t.
may be
assume F{Lt,Dt)
and have access
of nondisabled workers
We
assume that with probability
also
productivity of a worker at his or her current firm
Lt
falls
to zero,
and
e
<
e
1,
to the production
and Dt the number of
s
every period, the
though productivity elsewhere
due, for example, to match-specific learning as in Jovanovic, 1979).
=
^
m+
eD^)^
,
where a <
1,
and
<
p
and Dt include only employees who have not received adverse match
parameter
and have a discount
number
the
disabled workers employed at time
We
exit,
increasing in the
is
oo).
function F{Lt,Dt), where Lt
imaffected (this
n,
risk-neutral,
market which never
All firms are risk-neutral, discount the future at rate P,
is
the wage received by nondisabled
<
1.
The
quantities
specific shocks.
captures the relative productivity of disabled workers. For example,
when p
The
=
1
disabled and nondisabled workers are perfect substitutes, but the disabled have
a lower marginal product than nondisabled workers. This formulation cdso nests the case in
which firms discriminate against the disabled
The
most often
raised
issiie
ADA
in
for taste reasons as in
charges
Becker (1971).
wrongful termination, followed by
is
fail-
ure to provide reasonable accommodation and discrimination in hiring. Suppose disabled job
apphcants who are not hired sue with probability pd at expected cost
and lawyer
and
this
happens with probability pa and has cost
ha
=
Pa^a-
is
We
the firm incurs
therefore hd
refer to ha
when
with probability qd
and damages are
are fd
so h
—
and / act
benefit
it
for
like
and hd as hiring
costs,
though more
A
We
disabled worker
expected costs of
firing
do not get transferred
We
begin with the simple case where
also
is
fired sues
costs are pure waste,
(1
—
to sue, but
is
we
< Wa and
/„
/?)
We
"accommodate" disabled workers,
imposed on employers probably
(1
—
/5)
<
/^
Wd, which implies that firms always
employees wlio receive adverse match-specific shocks.
Dp
treat
a reasonable
defer a discussion of this case.
an excess number of applicants
and Lp are nondisabled.
like
for
may
Obviously, a fraction of these costs do accrue to
to disabled workers.
off the fraction s of its
assume there
all
and insurance companies. This seems
as lawyers
them a reason
assume that
wants to lay
who
a disabled and nondisabled worker
starting place since a significant fraction of the litigation costs
workers, giving
precisely, these are costs that
a tax from the point of view of the worker and the firm, though suits
some other party such
is
For a nondisabled worker, the corresponding probabilities
(pd-
0a, so the
qa4>a-
cost of not hiring a dis-
while the corresponding cost for a nondisabled worker
Pdi'd,
damages
=
fa
=
The expected
Ua-
decides not to hire an applicant.
and
qa
and
qd<Pd
including damages
Rejected nondisabled applicants can also sue, falsely claiming to be disabled,
fees.
abled worker
v^-,
for
every job, Dp- of
and Lp as given.
Finally,
whom
are disabled,
we assume that
example by purchasing special equipment
We
firms can
at cost
C
per
worker, and that this expenditure increases the marginal productivity of disabled workers by
a fixed amoimt
B
per worker.
The
ADA
requires employers to
make such accommodations.
C<
B, employers would make these adjustments voluntarily, even in the absence of the
The
fact that
government regulation
The maximization problem
required suggests that typically
is
of the firm at time
t
=
C>
can be written
ADA.
B.
as:
max n = ^P'[F{LuDt)-Wa,tLt-Wd,tDt-cDt-fasLt-,-fdsDt_,
{Dt,Lt}
c
—C—B
and
= D_i =
is
costs of
(1)
^^Q
-ha {Lp where L_i
If
0, Wa^t
and
[L,
Wd^t
-
(1
-
5)L<_i])
-
hd
{Dp - [A -
(1
-
5) A-l])]
denote nondisabled and disabled wages at time
the net cost of accommodation after the
accommodation are assumed
to
be
zero.
ADA.
The
Pre-
ADA
first line
firing
of (1)
is
t,
and
and hiring costs
output minus the
wage
accommodation
bill,
costs
and
Firms discharge a fraction
firing costs.
s of their
employees
who receive an adverse match-specific shock,
incurring a firing cost of fj for each disabled layoff,
and
The second
fa for every nondisabled termination.
incurs as a function of the
When
—
Lt
number workers not
=
Lt-\ and Dt
line gives the "hiring costs"
hired out of the applicant pools,
Dt-\ so that employment
nondisabled and sDt-i workers to replace those
who
is
Since adjustment costs are linear and there
is
As noted above, ha and
off.
=
by hiring more workers.
Wa, Wd,t
These equihbrium employment and wage
=
"^d,
=
Lt
L, and Dt
= D
= Wa + (5sfa-[l + P{l-s)]ha
+ eD"]^ = Wd + psfd-[l + ^{l - s)]hd +
(2)
c.
Both equations equate the relevant marginal product to the flow marginal
firing costs, hiring subsidies,
also
impose market clearing
n~^ [mL)
r?"' is
of firms
m
the inverse supply function and
is
which holds with complementary slackness.
the
is
=
for
nondisabled workers,
Wa,
(3)
the equilibrium
number
of firms.
The number
determined by
is
n < r and
profits
cost, inclusive of
and the net costs of accommodation..
To determine equilibrium, we
where
\n
levels satisfy:
LP-'[U + eD''p'
eD"-' [L"
hd.
no aggregate imcertainty, firms immediately
adjust to steady state employment levels, and Wa^t
every period.
Lp and Dp-
not changing, the firm hires sLt-\
are laid
act as hiring subsidies, because the firm reduces its costs
the firm
equal to entry costs, or that there
is
m
> M,
(4)
This means that either the maximized value of
no entry and the number of
firms,
m,
is
equal to
minimum, M.
Finally, the
wages received by disabled workers are given by
where
7]
is
(mD)
In the absence of the
ADA,
a parameter.
disabled wages.
The
= max
[n^^
Wd
,T]Wa
t]
=
(5)
,
and there are no restrictions on
equal pay provision can be interpreted as setting
77
=
1.
Inspection of the equilibrium conditions yields the following conclusions:
1)
From
since hd
is
effectively
a hiring subsidy.
be best described as hd
case the
ADA to reduce the costs of employing the disabled,
The scenario envisaged by ADA proponents can probably
(2), it is clearly possible for
>
0,
fa
=
ha
=
ADA can increase employment
the
0,
and minimal
firing costs for
and wages of the disabled.
the disabled, in which
It is also
worth noting that
as long as
p> 0,
this effect
is
2)
than
ADA will
the
be small.
likely to
As we noted
in Section
hd. Similarly, costs of
the
II,
ADA
appears to have increased fd considerably more
employing the disabled are increased by the accommodation
Therefore, in practice, the
c.
reduce nondisabled employment eind wages in this case, though
ADA may
be more
costs,
reduce disabled employment and
likely to
wages.
3)
The equal-pay
creating involuntary
provision of the
unemployment
4)
0)
may have
increased disabled wages,
The equal-pay
the disabled supply curve.
Although the partial-equilibrium
n = r
>
r/
to start with, then
employment
effect of hiring costs is to increase
is
an increase
by reducing
effectively financed
in hd will
profits.
More
fall.
ADA
the free-entry and fixed-number cases suggests that the
generally, the contrast
will
and Ross, 1990, Chapter
5) Finalh', the
tlic
nondisabled.
The
/
ADA
and
might also
affect
ha
=
should have no
=
hd
firing
like taxes
is
and /d >
when / and h
effect
0.
We know
are pure transfers
on employment.
To
that workers anticipate they
Wd can
and keep Wd
fall
+
may be
fired,
fd
is
may
and hiring costs
for
from the work of Mortensen
and side payments are allowed,
a pure transfer
+
13s fd)-
0s fd constant, changes
ADA
It is
The reason
for
in fd
straightforward to see that
have no
on disabled
effect
limits this possibility.
include payments to third parties and because firms
risk-averse, characterizing these costs as a tax
and both parties
and therefore include the discounted
employment income.
employment. However, the equal-pay provision of the
because / and h
firm size
see this in our context, suppose that
Under the assumption that
fiow value of firing costs, /3s/d, in their
as long as
we use
on the firm-worker relationship
are risk-neutral, labor supply of disabled workers changes to nd[wd
the change
between
profitable (see, e.g., Schmalensee,
employment by changing
presumes that / and h act
(1978) and Lazear (1990) that
=
M
ha-
discu.ssion so far
firing costs
>
11).
rather than a transfer from the firm to the worker.
fa
more
for excess profitability, since large firms are
1989; Scherer
m
reduce employment most in
firms or uidustries where profits are close to entry costs. In the empirical work,
proxy
If
of the
cause some firms to shut down, causing both
disabled and nondisabled employment and wages to
as a
provision also
exacerbating the dechne in disabled employment.
disabled, the implicit hiring subsidy hd
and
(i.e.
and the costs of accommodation by preventing wages from faUing to
interacts with firing costs
offset these costs,
off
ADA
Moreover,
and workers are
on the employment contract seems
like
a better
stylized description of reahty.
Finally,
we note
that the analysis so far ignores the impact of firing costs on separations
In a previous version of the paper (details available
because separations are exogenous.
we
request),
allow for time-varying productivity and endogenous separations.
shows that ADA-related
The
firing costs are likely to
theoretical discussion therefore
provisions are most important.
provision
is
shows that the net
The accommodation and
not binding, equiUbrium
is
This analysis
reduce both hiring and separations.
ADA depends on which
imposed by the ADA axe
effect of the
firing costs
reduce employment, while hiring costs have the opposite
likely to
upon
effect.
If
the equal-pay
on the labor supply curves of disabled and nondisabled
workers, and employment declines are accompanied by
wage
declines.
More
generally, however,
the equal-pay provision can create "involuntary imemployment" off the supply curve.
We
therefore estimate reduced-form equations of the form:
y^t
where
Xii is
i
denotes individuals and
t
=
x[^(3
+
time; yu
6t
+
is
weeks worked or average weekly wages; and
a set of controls including disability status.
changes in aggregate conditions. Finally,
dummy
and a
for
post-ADA
years.
The
a^t
is
+ E^t,
-ita,t
The term
(6)
5t
is
a year effect to control for
the interaction of a disability status indicator
coefficient 7^
the disabled using the nondisabled as a control group.
measures the impact of the
Since the
ADA
ADA
on
potentially affects the
emplo}'mcnt of nondisabled as well as disabled workers, we also explore empirical specifications
that use \'ariation by firm size and state to separately identify effects on the disabled and
nondisabled.
IV.
The sample
women aged
is
Data and Descriptive Statistics
drawn from the March CPS's from 1988 through 1997, and limited
21-58, since this age group has strong labor-force attachment.
to
men and
Disabled workers
are identified in the
March CPS Income Supplement by the question: "Do you have a health
problem or
which prevents you from working or which limits the kind and the amoTmt
of
disability
work you can do?"
issues (e.g.,
SIPP
Kr\ise, 1995),
and
of the
ADA
on employment
diiring the previous calendar year
is
is
similar to disabiUty questions in the
PSID and
Burkhauser and Daly 1996; DeLeire, 1997).^
The impact
measure
This question has been used by other researchers working on disability
Krueger and
(see, e.g.,
worked
.
levels
is
evaluated by looking at data on weeks
from the March income supplement.
The wage
average weekly earnings, computed using annual earnings data from the supplement.
^Using data from the Retirement History Survey and the National Longitudinal Survey of Older men, Bound
(1989) compares objective measures of health status with self-reported measures like the one used here. His
results suggest there is no clear basis for preferring one type of measure over the other.
.
Although the
1994,
CPS
changed from paper questionnaires to computer-assisted interviewing
and the main labor force status
<piestions
in
were also revised at that time, the content of
the income supplement was not changed. Appendix
A
discusses an analysis of
data for March 1993 and 1994, and provides more information on the
CPS
matched CPS
redesign and
its
possible consequences for our analysis.''
The
variables in the income supplement refer to the previous calendar year, so the
has data
for
weeks worked and wages
supplement appears to
The
in 1987-1996.
refer to respondents' status at the
sample
disabihty status question in the
time of the survey (March of the survey
year), but actually serves as a lead-in question prefacing additional
and Figure
supplement questions about
disability
income the previous year. Except
statistics
dated by survey year, the tables and figures label estimates according to the year of
observation, which
CPS
is
for
Table
I
which present descriptive
I,
the survey year minus one.
supplement respondents provide information on the
Responses to
for longest in the past year.
this question are
size of the
employer they worked
grouped into three brackets;
24 employees (small), 25-99 employees (medium), and 100 or more employees (large).
analysis of hiring
CPS
and
firing
rotation groups from
brought
in
outcomes
March
is
The
based on changes in employment status in a match of
to April. Finally,
each state since 1992 to see
if
EEOC
we use
charge activity
is
data on ADA-related charges
related to employment.^
Descriptive statistics organized by survey year, age group, and sex, are reported in Table
The
table shows no major change in average weeks
currently employed, although there
and Figure
rates for
1
show an increase
men aged
is
worked by men or
men aged
between 1990 and 1994, which
it
proportion of
in work-related disability rates for
men and women.
I.
men
I
Disability
40-58 start increasing in the 1991 survey, and remain high thereafter, though
starting in 1994. For
lower than
in the
a modest increase in weeks worked by women. Table
with a slight decline in 1996 and 1997. Disability rates
is
1-
was
in 1990.
21-39, there
is
is
for
women aged
40-58 increase sharply
a small increase in self-reported disability rates
later reversed,
and
in
1996 the disabihty rate for this group
These patterns suggest the
ADA may
have had an
probability that people describe themselves as disabled, especially for
effect
women. ^ This
on the
raises the
The estimates we report are weighted bj' CPS sample weights. The weights were updated in 1994 to reflect
the 1990 Census. We use newly released updated weights for the 1990-1993 surveys as well. The 1988 CPS data
come from the so-called March "rewrite" file. This file includes firm size and other variables not on the original
1988 release, and reflects a revised imputation procedure (Bureau of the Census, 1991). The extract excludes
the hispanic oversample for each year. A few dozen households with duphcate household identifiers in the 1994
survey were also excluded because they could not be included in the matched samples.
^The analysis variables are state/year aggregates from our tabulations of EEOC microdata. These are similar
to less recent statistics published in EEOC reports.
^The fact that the disabled may be an elastic population has been noted by 01 (1991) and Kubik (1997).
On the other hand, Dwyer and Mitchell (1998) argue that disability status does not appear to be endogenous
in models of retirement behavior.
10
we return
possibility of composition effects, a point
Results
V.
A.
to in the discussion of results.
Employment and Wage Effects
Figure 2 plots average weeks worked by sex and disability status. While
on average over 45 weeks a
noteworthy
men work
year, disabled
less
first full
year in which the
disabled
women aged
ADA
was
There
in effect.
is
21-58 work
than 20 weeks on average. Especially
the 2 week decline in weeks worked by disabled
is
men aged
men between
no sharp decline
1992 and 1993, the
in
weeks worked by
21-58.
Weeks worked
FigTues 3a and 3b plot average weeks worked separately by age group.
by disabled men aged 21-39 dropped sharply between 1992 and 1993, while those by disabled
women aged
21-39 started falling in 1992.
Weeks worked by men aged 40-58
was an increase
decline between 1992 and 1993. In contrast, there
women aged
women when
Table
II
40-58 between 1992 and 1993, which
the two age groups for
women
is
why we do
in
not see a decline for disabled
are pooled.
(6).
are from regressions of weeks worked and log weekly wages on age and race
and year x disability interactions
coefficients of interest.
Because the
as post-treatment years.
1992
is
ADA
show a marked
weeks worked by disabled
reports ordinary least squares (OLS) estimates of equation
disability effects,
also
came
for 1991-1996.
The
These estimates
dummies, year and
interaction terms are the
into effect in July 1992,
we think
of 1993-96
possibly a transition year, while the effects for 1991 can be
seen as '"pre-treatment" specification
tests.
The
table also reports estimates from specifications
including a linear time trend interacted with disability status, and a single-effect
model that
includes a trend interaction with disability status, 1991 and 1992 interactions with disability
and a single post-ADA
status,
models with trends allow
dummy
for
for the possibility that
be explained by extrapolating different trends
The
results in
Table
II
suggest that the
changes in outcomes by disability status can
for the disabled
ADA
and nondisabled.
had a substantial and
negative effect on the employment of disabled people under 40.
Panel
by
A
1.8
statistically significant
For example, column
(1) of
shows that weeks worked by disabled men aged 21-39 were stable until 1992, but
weeks
in 1993,
and declined by an additional week
in 1995.
Column
(2)
controlling for disabihty-specific trends does not change these results. Results for
21-39 are similar (columns 4-6), but the decline starts in 1992.
in 1992
The
1993-1996 interacted with disability status.
and 1993 suggest that the
the law became effective. This
is
costs of the
ADA
shows that
women aged
The sharp employment dechnes
were not anticipated by employers before
not too surprising since the
11
fell
first
years of
ADA
enforcement
were characterized by some confusion regarding exactly what the act required
and Sims,
B
weeks
1.8
in 1993, but this effect
weeks after controlling
employment
into effect.
The
reports estimates for the 40-58 age group.
is
women aged
Moreover, columns
and
(5)
40-58, but this
show that these
(6)
is
column
(4)
evidence that the
ADA
had an
effect
the other hand,
shows a decline
in 1991, before the
is
effects
change
on the employment of disabled
we
came
control
So there
sign.
women
in the
ADA
disappear once
for disability-specific trends, and, in fact, the coefficients of interest
little
On
added.
men
generates a statistically significant, decline of 2.1
(3)
for disability-specific trends. Finally,
of disabled
disabiHty-year interaction for
halved when a linear trend
the single-effect model reported in column
relative
Veres
1995).
Panel
is
(see, e.g.,
is
in the 40-58
age group.
Columns
women aged
A
(7)-(12) in Panel
There
21-39.
effects disappears in
is
an
effect
Moreover,
inclusion of trends.
decline in wages for older
suggesting that the
on
men
models with a trend.
aged 21-39. The wage estimates
for older
in 1994,
There
is
but not in other years, and the 1994
no evidence of a wage
We
starts too soon.
ADA had little effect on
women
in
therefore interpret the
Panel B, the
wage
the relative wages of disabled workers.
the paper focuses on further analyses of employment effects only, and the analysis
demographic groups where the evidence
in
women
effect for
men, reported in Panel B, are also sensitive to the
the employment effects for older
like
women
men and
report estimates for the log weekly earnings of
for effects is strongest
results as
The
is
rest of
limited to
—women aged 21-39,
and men
both age groups.
Composition Bias
A
Figure
possible explanation for the
shows an increase
1
employment
results in
Table
11
is
in self-reported disability rates after 1991.
were disproportionately more
likely to identify
We
1993 respondents.
matched sample
those
who do
worked
in
is
effect.
Tmemployed people
II
ADA,
perhaps
could overestimate
ADA.
investigate the possibility of composition bias using a
and 1994 CPS's. In
If
themselves as disabled after the
because disability became more socially acceptable, the results in Table
the disemployment effects of the
a composition
principle, the
matched sample from the 1993
matched sample provides two observations
(In practice the
match rate
is
used here to compare individuals
lower; see
who
Appendix
A
for half of the
for details).
The
report a disability in both surveys to
not report a disability in either year. Since these surveys report data on weeks
1992 and 1993, the matched data set provides a short panel that straddles the
implementation date and
is
ADA's
imaffected by composition bias arising from changes in reporting
12
behavior.
Results using the matched sample are reported in Appendix Table Al.
the basic cross-sectional specification for 1993 and 1994 data only.
sample of individuals
for the
for
whom
Column
Cohrmn
the mat.ch was successful.
Al
are similar to those in Table
II.
(1)
reports results for
The
Comparing cohimns
cross-sectional
and
(1)
(3), it is
apparent that the restriction to the same disabihty status in both years has almost no
men aged
the results for
40-58 and
women aged
21-39. This
is
repeats
(2) reports results
(3)
the sample of individuals reporting the same disability status in both years.
estimates in Table
Column
effect
on
important evidence against the
presence of composition effects since disabihty rates increased more for these two groups than
younger men.
for the
men aged
For
21-39, the results using a sample
smaller than in the 93-94 cross-section.
in
weeks worked, as compared to a
matched
decline in the
to a
1
week
decline,
cross-section.
which
is
1.2
The
Adding controls
closer to the 1.5
allocated by the Census
effects either. In fact, the
since disability rates for
composition story
men aged
for personal characteristics
a
3.1
week
changes this
in the corresponding full
where supplement and weeks worked data were
of -1.3 (reported in
men aged
is
unchanged are
in the full cross-section, or
week decHne observed
Bureau generates an estimate
these estimates suggest that the results for
disability status is
disabihty-constant sample shows a 0.3 decline
week decline
Finally, excluding observations
cross-section.
where
column
(7)). Overall,
21-39 are not generated by composition
not plausible for this demographic group anyway,
21-39 are actually lower at the end of the sample period than
beginning, and the increase in disability rates could account for at most half of the decline
at
tlie
in
weeks worked.
Changes
in the
SSI and DI Programs
Another factor
affecting the interpretation of Table II
is
an increase
in the
number
of people
receiving disability payments from the Disability Insurance (DI)
and Supplemental Security
Income
al,
(SSI)
who worked
programs
in the early 1990s (see, e.g., Stapleton, et
long enough are entitled to receive DI payments
gainful activity.
1994). Disabled workers
when not engaged
in substantial
Disabled people without a work history can receive SSI, which
tested federal benefit that
adverse labor supply
is
effects,
supplemented by some
Appendix B using CPS data on
II
social security
Weidenbaum,
program
use.
are not explained by trends in SSI
the estimates).
13
account for the dechne in
1994).
The
a means-
SSI and DI payments create
may
increased use of these programs
disabled employment (a possibility suggested by
that the results in Table
states. Since
is
We
address this issue in
results of this analysis suggest
and DI
receipt (see Table
Bl
for
Magnitudes
The
estimates in Table
II
can be compared to estimates of the
effect of the
ADA
on
the costs of employing disabled workers. Unfortunately, there are no good estimates of these
costs, so
The average
our calculations are really just educated guesses.
was about 12 per 1000 disabled employees a
percent of
all
ADA
charges, employers
made
fraction of
charge rate
EEOC, about
In cases resolved by the
year.
14
average payments of over $15,000 per case. In the
remaining cases, either the charge was dropped,
know what
EEOC
pending, or there was a
is
ADA charges end up in court.
suit.
We
do not
However, between 1995-97 there were over
56,000 employment discrimination cases brought in federal court (Administrative Offices of U.S.
Courts, 1997).
The
total
number
employment discrimination charges
of
during this period was 245,000, which implies that 23 percent of
We
apply this fraction to
ADA
charges,
charge does not go to court or get settled by the
This
is
50,
ADA
400 x 0.012
charge
=
is
with the
EEOC
charges went to court.
and use the cost estimate of $210,000 per case offered
by Condon and Zolna (1997). To be on the conservative
average cost of an
all
filed
EEOC,
side,
we
also
assume that
+
0.14
x
ADA
The estimated
there are no other costs.
therefore equal to 0.23 x 210, 000
an
if
000
15,
=
50, 400.
$605 a year per disabled employee, or $12 for each week of exposm-e
to the risk of a suit.
Assessing the cost of reasonable accommodation
Network (1997) reports a monetary cost
cost.'
Our estimates
a $23 increase in weekly
12+23=35
in
even harder.
The Job Accommodation
accommodation, which we take as the net
of $930 per
of separation rates in the next section suggest that the average duration
of a job lield by a disabled
about $365
is
employee
employment
1991 and 1992.
dollars, our
is
10 months, which implies that
accommodation leads
The average weekly earnings
costs.
of the disabled were
Since the total weekly cost increeise due to the
rough estimate
is
that the
ADA
to
ADA
is
about
led to a 10 percent increase in the cost
of employing disabled workers.
In the theoretical
and are always on
model
their labor
employers take the total cost of labor as given
in Section III,
demand
curve. Since the results in Table II
show
ADA
falls
of a change in disabled wages, the cost increase generated
The
10 to 15 percent decline in weeks worked
about
-1 to -1.5 for disabled workers.
Hammermesh
This
is
is
by the
therefore consistent with
in the
little
evidence
on employers.
demand
elasticities of
range of elasticity estimates reported by
(1986) for workers in different demographic groups.
^This number excludes any productivity increases due to
or sub-optimal reorganization of the work environment.
14
accommodation and any
losses
from job changes
The Impact of the
B.
We
iised
ADA
on Hiring and Separations
matched CPS rotation groups from March
on hiring and separation
April match).
is
ADA
employed
An
individual
March
in
is
rates (see
of that year
March while
for details relating to the
and not
in
in April. Similarly,
March is employed
an accession
March-to-
t if
he or she
(hire) is
recorded
in April. Separations are defined for
accessions are defined for those not working in March. Disability
status always refers to March.
those used by Poterba and
A
coded as having experienced a separation in year
when someone who was not employed
those working in
Appendix
to April to investigate the effect of the
These measures of accession and separations are the same as
Summers
(1986),
and the resulting average accession and separation
rates are close to those they report.®
The estimates
by the
affected
of effects
CPS
on separations and accessions are imprecise and also potentially
come from the main CPS survey and
redesign (since the underlying data
not the supplement). Moreover, Poterba and
Summers
data are plagued by considerable measurement
error.
(1986)
We
show that labor-market
transition
therefore hmit the discTission in this
section to a brief graphical analysis.
Figure 4a-d plot log accession and separation rates by disabihty status, sex and age group.^
Figure 4a shows a post- ADA decline in separation rates for disabled workers aged 21-39, though
there
is
also a decline for the nondisabled.
On
the other hand. Figure 4b shows a post- AD A
decline in the luring of disabled workers aged 21-39 that
nondisabled, especially for women.
is
not mirrored in the data for the
Figure 4c, which plots separation rates for the 40-58 age
On
group, again shows no evidence of changes unique to the disabled.
as clear as the
change
in disabled hiring for the
of a relati\-e decline in hiring for disabled
liiring for
men and younger women
men
the other hand, while not
yoimg group. Figure 4d shows some evidence
in the older group.
are not surprising since
The apparent reductions
employment
groups
for these
fell.
in
It
important to note, however, that the lack of a clear reduction in disabled separations weighs
is
against a pure "firing costs" model of the
ADA.
Results By Firm Size
C.
Next,
less
we
look at employment patterns by firm
than 15 employees are not covered by the
size.
This
ADA, and
is
of interest because firms with
those with 16-25 employees were
^The average separation rate for 1988-1997 is 0.026 for men and 0.036 for women. The average accession
is 0.18 for men and 0.08 for women. The disabled have higher separation rates and lower accession rates
rate
than the nondisabled.
^For instantanous accession and separation
e
= ;^- We
plot log accession
rates,
and separation
77
and
(,
rates because
15
the steady-state
^
=
employment rate
(^^ - ^^)
.
(e(l
-
is
e)).
approximately
exempt. The
initially
since, as
ADA
noted in Section
might also have had a larger
III,
in small firms
small firms are probably less able to absorb ADA-related costs.
we might expect the
Together, these considerations suggest
on employment
on employment
effect
in firms that are sufficiently large to
enough to be vulnerable to an increase
ADA
to have
be covered by
ADA
had the
largest effect
provisions but small
in costs.
Figures 5a-c plot the log of the probabihty of working in a particular firm size category
As noted
divided by the probability of not working.
earlier,
the size category refers to the
The
worker's longest job last year as recorded in the supplement.
and 100 or more
(small), 25-99 (mediimi),
(large).
The
size categories are 1-24
figures give a visual representation of
the coefficients in a multinomial logit model where the dependent variable
size-category,
and non-workers are the reference group.
The log-odds
effects.
in
The independent
is
employment-by-
variables are year
each figure were computed separately for disabled and non-disabled
workers.
The log-odds
of working in a
than log-odds of working
groups. For
medium
size firm
in a small firm after
women, there
is
also a relative
appear to have fallen somewhat more steeply
1992 for disabled workers in
small contrast for
men aged
if
ADA
nondisabled workers, suggesting the
there were effects on the nondisabled,
effects
(e.g.,
t=1.4
21-39), but they are negative for all three
employment by firm
In contrast with this pattern, the log-odds of
for
it
three demographic
decUne in the probability of working
Estimates of these differing trends by firm size are not very precise
vs.
all
had no
seems
effect
in a large firm.
for the
mediimi
demographic groups.
size are essentially parallel
on the nondisabled. Of course, even
likely that
they would be
much
smaller than
on the disabled, and therefore harder to detect.
D.
Cross-state Variation
The
last strategy
we use
state-level variation in
in
ADA Charge
to estimate the
ADA
impact of the
charge rates.
to separately identify the impact of the
the purposes of this analysis, the
ADA
Rates
ADA
relates
changes in employment to
Like the firm-size analysis, this strategy allows us
ADA
on disabled and nondisabled employment. For
charge rate
is
defined as the
number
of ADA-related
discrimination charges, in a state, per thousand population disabled in the state in 1992 (the
latter figiu-e estimated using the 1992
1993, 1994
CPS). Charge rates are calculated separately
and 1995, but 1992 population data are always used
Charge rates vary considerably by
state.
The average cumulative
per 1000 disabled persons aged 21-58, varying from a
40/1000. Variation in charge rates
is
for the
minimum
for 1992,
denominator.
rate for 1993-95
of 6/1000 to a
was 13
maximum
of
generated by idiosyncratic differences in state labor force
16
composition, local awareness of ADA provisions, cross-state differences in employers' compliance
ADA, and whether
with the
Some
workers.
a state previously had an
had weak
states
laws, while others
penalties in cases where discrimination
is
FEP
statTite that
had laws that
covered disabled
set criminal as well as civil
proved (Advisory Commission on Intergovernmental
Relations, 1989).
The estimating equation
for this strategy
is:
96
Vidst
=
x'iP
+
+
4>ds
5dt+ Yl adAfhrTs^T-l)
+ Vidst,
(7)
r=93
where
ijidst is
and
t,
X, is
weeks worked by individual
i
with disabihty status d hving in state
The
a vector of individual characteristics (age and race dummies).
interest are interactions
between year dummies
(e.g.,
/i.ga)
includes separate state effects for each disability group, 0ds,
of interest, 0^,93, 0^,94, 0^,95
and
0:^,96,
regressors of
(e.g., rs,92).
Equation
and year x disability
are the coefficients
now
full set
of state x year x disability interactions in the models.
We
play the role of control variables.
also report estimates
trends for each state x disability group to equation
The
with a large number of
for
women aged
The
show that
results in Table III
effects in
ADA
21-39 and
in 1993-94,
on the state/year
we do not
The year x disability
from a model that adds
weeks worked by the disabled
Some
men
of the estimates
40-58 are negative and significant in other years as well.
1993 are implausibly large, probably because the charge rate in 1992 was very
annual charge rate of 4/1000
men aged
is
ADA
mean
predicted to reduce employment by 2.6 weeks based on the 1994
21-39.
In contrast with the estimates for the disabled, there
impact of
linear
in states
low, but estimates for other years are of a plausible magnitude. For example, hah-ing the
estimate for
effects
(7).
charges declined relative to other states.
men aged
(7) also
effects, 8dt-
charge rate interactions for each disability group. These effects are identified because
have a
year
and the charge rate per disabled
person in the previous year in the individual's state of residence
The parameters
s in
charges on the non-disabled.
nondisabled appear in models that control
The only
is
no consistent evidence of a negative
significant negative estimates for the
for linear trends.
For the most part, these estimates
are considerably smaller than the corresponding estimates for the disabled.
As a
nous.
final strategy,
we experimented with IV estimation
The instnmient was a dimamy
for
of (7) treating diTTs,T-\ as endoge-
whether a state previously had an
discrimination against the disabled, interacted with disabihty status and
mies.
The presence
of a preexisting
probably because the
ADA
was
FEP
less of
discrimination against the disabled.
law
is
FEP
law restricting
post-ADA year dum-
negatively correlated with
ADA
charge rates,
an innovation in states with their own laws restricting
For example, the states with preexisting
17
FEP
laws had
15 percent lower
suggest that the
ADA charge rates. The resulting IV estimates are imprecise but they also
ADA had larger disemployment effects on the disabled in states with more
ADA-related charges.
Conclusion
VI.
Some
social critics see the
ADA as part
of a process eroding the traditional employment-at-will
more
doctrine and making the U.S. labor market
ADA proponents see the ADA
as
like labor
markets
making the labor market more
without appreciably
inclusive,
Economic theory
increasing employer costs or reducing employment.
Europe. In contrast,
in
useful for evaluating
is
these alternative views and suggests avenues for inquiry, but does not
make imambiguous
predictions.
In 1993, the first full year in which the
the employment of disabled
employment
seem
men aged
ADA
in effect, there
21-58 and disabled
trends, composition effects,
and changes
to account for this decline, leaving the
ADA
women aged
were marked drops in
21-39.
Extrapolating
DI and SSI participation
in
do not
rates
as a likely cause. This interpretation
supported by evidence that employment of disabled
ADA-related charge activity and by
was
men
fell
more sharply
relative declines in disabled
in states
employment
in
is
also
with more
medium
size
firms.
Since the
ADA
and accession rates
provides a form of employment protection,
for
the disabled. There
is little
it
can lead to lower separation
evidence of an effect of the
separations, however. This and the fact that costs of reasonable
accommodation are probably
larger than the costs of litigation for wrongful termination suggests that
ha\'e
been more important
for
we found no evidence
its fiercest
opponents
(e.g.
accommodation
costs
employers than lawsuits. The absence of an offsetting decline in
wages suggests the equal pay provision has also played a
Finally,
ADA on disabled
role in the
of effects on nondisabled workers.
Olson, 1997),
it
seems
likely that
the
ADA's employment
effects.
Contrary to the concerns of
ADA
predominantly affected
the costs of employing the disabled, and has not led to a fear-of-litigation climate that reduced
the overall level of employment.
18
Appendix A: Data Issues
1.
March 1993
to
March 1994 Match
In principle, households in rotation groups 1-4 in 1993 are interviewed in
they are in rotation groups
move
or are lost for other
For the piirposes of the 93/94 match, we selected individuals in the relevant rotation
reasons.
groiips,
In practice, some of these households
5-8.
March 1994 when
with valid interview status (CPS item
H-HHTYPE=1), and
in the age range of interest
(CPS
(21-58 for 1993 and 21-60 for 1994). Records were matched using the household identifier
H-IDNUM), person
item
fine
H-MIS equaled 1993 H-MIS
number (CPS item A-LINENO), and rotation group
Of the March 1993 records ehgible
plus four).
for
(i.e.,
1994
matching, 76.2
percent were matched to a March 1994 record.
We
defined a successful
match
as a person with the
same sex and race
both years. 70.9
in
percent of eligible March 1993 records were matched according to this definition, a rate similar
to that reported in Peracchi
match
basic
on the 1994
restrictions
ICPSR
and Welch
93 percent of the records that satisfied the
(1995).
were also matched on sex and race. Because the household
incorrect, the
file is
matched data use a corrected household
identifier
identifier for
1994 provided by the Census Bureau.
2.
Consequences of the
As noted
CPS
Redesign
in the text, the redesign
computer-assisted interviewing
changed the main questioimaire and instituted universal
(see, e.g.,
Pohvka
(1996)).
The supplement questionnaire was
unchanged but these other changes may have affected the nature or likelihood of supplement
sponses.
A
conducted
parallel survey
and paper-and-pencil interview
that
some
in 1993
showed few differences between computer-assisted
Later, however,
results for the supplement.
of the annual earnings data collected in the 1994
lected for subannual
amounts (Bureau of the Census,
fixed in later surveys.
re-
To minimize the consequences
1994).
it
was discovered
supplement were mistakenly
The problem appears
col-
to have been
of errors in annual earnings,
we exclude
observations on weekly wages below $25 or above $2000 (in 1988 dollars).
One
rates,
possible consequence of the redesign
which were low .in the 1994
CPS and
high in the 1995
from G. Weyland, Bureau of the Census) This
.
people
who do not respond
of the redesign
is
affects the
in
CPS
supplement non-interview
(personal communication
data because Supplement variables for
are "allocated" by the Census Bureau. Another possible consequence
a change in the type of people
the impact of these phenomena,
data were collected using the old
these two years to people
was fiuctuation
who
report themselves as disabled.
we used the March 1993-94 matched
CPS
data.
To
assess
The March 1993
methodology. This allows us to restrict the sample for
who responded
to the
19
CPS supplement
in 1993
(i.e.,
their responses
were not allocated), and/or who identified themselves as disabled in both years, and/or who
answered the supplement questions on weeks worked in both years.
The
1994
results of this analysis are reported in Table Al. Again, the coefficient of interest
dummy
dimimy
interacted with a
being disabled. Except for
for
men aged
is
a
21-39, results
under a variety of sample restrictions are remarkably consistent with the main set of crosssectional results, though less precisely estimated.
The
considerably across columns, and in some cases are
column
sectional results in
lower for this age group.
This variability
(1).
results for
much
men aged
21-39 fluctuate
larger or smaller than the cross-
may be due
to the fact that
match
rates are
Controlling for schooling, marital status, and veteran status reduces
the fluctuation in estimates across samples.
3.
March
to
April Matches for 1988-1997
March households
in rotation
groups 1-3 and 5-7 are eligible for interviewing in April.
Observations to be matched were selected on the basis of rotation group, interview status,
and
age, as described for the
March 1993-94 match. Revised Census Bureau
rules necessitated additional criteria for
matching records
in
some
years.
We
confidentiality
therefore
matched
households using the household identifier (H-IDNUM), person line number (A-LINENO), rotation group status
(i.e.,
code (HG-STCO). Of the
April
H-MIS equaled March H-MIS
eligible
March
April record using these basic criteria
phis one), and 1960 Census state
records, 92.8 percent were successfully
matched
to an
(H-IDNUM, A-LINENO, H-MIS, and HG-ST60);
percent were successfully matched on both the basic criterion and sex and race, which
is
88.6
95.4
percent of tliose matched using only basic criteria.
Appendix
The CPS
B:
Effects of SSI and DI on Employment of the Disabled
SSI and social security income (OASDI). For
identifies individuals receiving
workers below retirement age, the latter consists of disability benefits and survivor benefits.
We coded
everyone receiving more than $75
(in
1988 dollars) a week in social security income as
receiving DI since survivors benefits are typically smaller than
(Department of Health- and
of
men and women
and SSI
rolls
The
still
ADA,
for pre-retirement
Services, 1992, Tables 5A.5-6). Figure
of different age groups receiving SSI and/or DI.
predate the
programs might
Human
DI
the timing
is
close
Bl
age groups
plots the fraction
Although increases
enough that disemployment
in the
DI
effects of these
provide an alternative explanation for the decline in disabled employment.
investigation of this possibility
is
compHcated by the
cause and a consequence of employment status.
20
fact that disabihty
income
is
both a
We iise the following simultaneous equation model to describe causal relationships
and employment:
Social Security benefits
where yu
weeks worked
is
disability status;
of disability status
Vit
=
x[i0o
+
as,t
su
=
x[iPi
+
6iyu
for individual
that are orthogonal to
have
(5i
<
x^t
and
a^-
for
and
Sit
But
who works more
even after controlling
also
time
at
£,(
in fact this increase
Table
II
One
Table
xu
is
(8b)
a vector of controls, time dummies, and
transfer payments,
and
a,^
The parameters
years.
and
t^jj
is
the interaction
of interest are the
are almost certainly negatively correlated,
will generally
be
observable characteristics.
income
less likely to receive disability
Since DI and SSI are means-tested,
we
captures the possible labor supply consequences of any transfers and
in
Table
II,
may
is
II
amount
to omitting
greater for those reporting a work-related disability.
therefore be biased
downwards
(i.e.,
strategy to correct this omitted variable bias
ignoring the fact that benefit receipt
dummy
for
income from SSI or DI
resulting coefficient estimate
men aged
21-39
receipt has
still
+ Vit,
Sjj
is
from equation
Figure Bl shows that the proportion receiving SSI and DI increased over the sample, and
(8a).
The
(8a)
are error terms that capture unobserved effects
rj^t
presumably negative. The regressions reported
a
+ itdit + Sit
0.
The parameter a
in
t,
dummy indicating federal
and a dummy for post-ADA
interaction coefficients 7j as before.
since an individual
i
a
s^t is
between
is
reduced, but
is
is
is still
is
add
to
endogenous.
and negative. As a
an even larger impact on weeks worked, so the
group disappear
s^t
in later years.
As
to the regressions reported in
We do
this in
Table Bl by adding
men and women aged
result, the
statistically significant.
negative and precisely estimated in 1993 and 1994.
for this
of 7t
too negative).
to our basic specification for
large
The estimates
For
impact of the
men aged
effects of the
ADA
21-39.^°
ADA
40-58, SSI or DI
are smaller, but
However, the disemployment
before, the effects for
women aged
on
effects
40-58 appear to
start too early.
Next,
we expand
imphcit tax rate
the model to recognize that federal benefits can have both income and
effects.
Instead of the constant coefficient a, the effect of benefit receipt in
^°These models also include the regressors used to predict wages in other specifications described below. The
regressors are age, age squared, a dummy for age 30-39 or age 40-49, two race dummies, dummies for high school
graduates, some college and college graduates, dummies for married, separated, widowed, and having served in
Vietnam or other mihtary service, state dummies and finally a dummy for answering yes to the question "Did
you ever retire or leave a job for health reasons" Adding these covariates to the regressions in Table II does
not change the results since the post-ADA/disabihty interactions are essentially uncorrelated with individual
characteristics.
.
21
equation (8a)
now assumed
is
to be:
=
Oiit
where wn
is
Oio
+ aiTiWit,
the predicted wage of the individual (as a function of the observables Xu), and
Ti is
the tax rate induced by DI or SSI means-testing, qq captiu-es the effect of additional unearned
income, and qiTi captures the fact that after-tax wages are
are presumably negative.
model with tax
Both
and
Sit
Columns
and
(2)
(8) in
—
(1
TiSit)Wit-
Both qq and ajTi
Table Bl report the results of estimating the
effects:
have negative
u),fSj(
the income effect
is
13 weeks,
For example, in column
coefficients.
and the
coefficient
worker earning $500 a week, an increase
employment by one week. Estimates
men aged
21-39,
on the interaction term suggests that
in the tax rate
for
a
from 50 percent to 60 percent reduces
impact of the
of the
(2), for
ADA
in these
models are similar to
those estimated in regressions without WnSit.
The
estimates so far do not account for the likelihood that benefit receipt
determined.
(i.e.,
It is
straightforward to show that the resulting estimates of
too small in absolute value).
strategy.
We
endogenoiisly
are biased upwards
therefore experimented with an instrumental variables
This approach begins by assuming that Su has two components, means-tested and
non-means-tested benefits. In particular,
like
jt
is
=
let s^t
5,t
-I-
where s^
Sit,
is
a means-tested benefit
SSI or DI. Only means-tested benefits should have a tax-rate effect so the estimating
equation becomes
Vrt
To implement
=
this strategy
ing veterans benefits, SSI, any
payment
(military, railroad,
eral stipend that
^uPo
is
we
+
set
Ck'oS,(
s^t
+
aiTiWuSu
federal
'jau
+
(9a')
Sit,
equal to an indicator for any federal stipend, includ-
OASDI payment,
and
+
or another federal pension or disability-related
employee disability pensions).
subject to means-testing,
i.e.,
SSI,
s,t
DI and any veterans benefits that are
means-tested. Means-testing of veterans benefits can be determined from a
asks whether respondents are required to
of Veterans Affairs.
benefits
who
The instrument
fill
indicates any fed-
CPS
question that
out an income questionnaire for the Department
for s^ is
an indicator
for individuals receiving veterans
are not subject to means-testing, on the presumption that these benefits are not
endogenous.^^
The
interaction term,
iy,«s,t, is
instrumented using the interaction of predicted
'^Means-tested veterans' benefits include some veterans pension and survivor benefits. Non-means-tested
veterans benefits include veterans compensation and schooling benefits (see, e.g., Department of Veterans AiTairs
1990)
22
wages with an indicator
The reasoning
9).
is
here
having had a previous disability that led to job loss (see footnote
for
is
that a worker
potentially eligible for DI.
The
who
left
a job because of a work-related disabihty
regressions include a
main
effect for
having had a previous
work-related disability, so the identification comes from the interaction with Wn.
Cohimns
(3),
(4),
(9),
and
two-stage least squares (2SLS).
for
(10) report
OLS
This generates results similar to those using only an indicator
SSI or DI as a control. 2SLS estimates of models without an interaction term, reported in
columns
and
(5)
(11),
show much smaller
effects of disability benefits
consequently larger disemployment effects of the
Finally,
These
columns
results
(6)
and
show smaller
of disabled
effect is
men aged
The
e\-idence of
in the
OLS
men aged
ADA
models.
both Su and
WitSit as
endogenous.
on young workers, but there are
21-39 by 2.3 weeks in 1993. For yoimg disabled
ADA
on weeks worked, and
ADA still appears to have reduced the employment
still
women, the
implicit tax
has a statistically significant negative effects in
40-58, the estimates are
week decline over the whole sample period. The
little
than
effects of federal disability benefits
considerably larger, but the
1993 and 1994. For
ADA
(12) report the results of treating
precisely estimated implicit tax effects.
2
estimates of the models to be estimated by
effects after controlling for
23
more
precise,
results for
and show an approximately
women aged
endogenous benefit
40-58 again show
receipt.
References
Abram, Thomas
Effect
(1993),
"The Law,
its
Interpretation, Levels of Enforcement Activity
and
on Employer Activity," American Economic Review Papers and Proceedings 83, 62-66.
Addison, John and Jean-Luc Grosso (1996), "Job Security Provisions and Employment:
Revised Estimates," Industrial Relations, 35, 585-603.
Administrative Offices of U.S. Courts (1997), Judicial Business of the United States Courts,
http://www.uscourts.gov/judiciaLbusiness/contents.html.
Advisory Commission on Intergovernmental Relations (1989), Disability Rights Mandates:
Federal and State Compliance with
Report A-111, Washington DC:
Gary
Becker,
Employment Protections and
US GPO,
Architectural
Earner Removal,
April.
The Economics of Discrimination, Second Edition, Chicago: The
(1971),
University of Chicago Press.
Bentohla, Samuel, and Giuseppe Bertola (1990), "Firing Costs and Labor
Bad
Is
Eurosclerosis?," Review of
Economic Studies
Demand: How
57, 381-402.
Bound, John (1989), "Self-reported Versus Objective Measures of Health Status
ment Models,", Journal of Human Resources
Bureau
of the
in Retire-
26, 106-138.
CD-ROM,
Census (1991), Current Population Survey, March 1988-91 on
Technical Documentation CPS-91-CD, Washington:
US Department
of
Commerce
(issued Oc-
tober 1991).
Bureau
of the
Census (1994), Income, Poverty, and Valuation of Noncash Benefits: 1993,
Current. Population Reports,
Consumer Income
Series,
Report P60-188, Washington:
US GPO.
Burkhauser, Richard and Mary Daly (1996), "Employment and Economic Well-Being Following the Onset of a Disability," in Jerry Marshaw
Benefits,
Kalamzoo: The Upjohn
Clarke, Richard (1996),
Underwriters Property
&
DeLeire,
Thomas
&
"Employment Practices
Liability Risks
on the Rise," National
On
the Rise of
EPL
Insurance," National
Casualty-Risk &Benefits Management, 09/15/97.
(1997),
of Health
Social Security Bulletin,
Work and Cash
Casualty-Risk &Benefits Management, 11/11/96.
Disabilities Act," University of
Department
Disability,
Institute.
Condon, David and Stan Zolna (1997), "Stakes
Underwriters Property
et al, eds..
and
"The Wage and Employment
Effects of the
Americans With
Chicago Department of Economics, mimeo, November.
Human
Services (1992),
Annual
Statistical
Washington DC: US Government Printing
24
Supplement
Office, 1992.
to the
Department of Veterans
USGPO,
Washington, DC:
Affairs (1990),
January 1990.
Dertouzos, James N. (1988), The
RAND
End
Report P-7441, Santa Monica: the
of Employment- at- Will: Legal and
RAND
Diamond, Peter and Eytan Sheshinski
Benefits" Journal of Public
Federal benefits for Veterans and Dependents,
Economics
Economic
Costs,
Corporation.
"Economic Aspects of Optimal Disability
(1995),
57, 1-23.
Donahue, John and Peter Siegelman (1991), "The Changing Nature of Employment Discrimination Litigation" Stanford
Law Review
43, 983-1033.
Donahue, John and Peter Siegelman (1993), "Law and Macroeconomics:
Discrimination Litigation Over the Business Cycle" Southern California
Employment
Law Review
66, 709-
765.
Dw'yer, D.
S.,
and 0. Mitchell
problems as Determinants of Retirement:
(1998), "Health
Are Self-Rated Measures Endogenous?,"
NBER
Working Paper 6503,
April.
Epstein, Richard A. (1992), Forbidden Grounds: The Case Against
Employment Discrim-
ination Laws, Cambridge: Harvard University Press.
EEOC
EEOC
(various years).
Annual Reports, Washington DC: US GPO.
(1995), Technical Assistance Program, Disability Discrimination:
Employment
crimination Prohibited by the Americans with Disabilities Act of 1990, Washington
Dis-
DC: US
GPO, May
Freeman, Richard B. (1981), "Black Economic Progress After 1964:
Why?,"
in
Sherwin Rosen,
ed., Studies in
Who
has Gained and
Labor Markets, Chicago: University of Chicago Press,
pp. 247-294.
Gruber, Jonathan (1994), "The Incidence of Mandated Maternity Benefits," American Eco-
nomic Review
84, 622-641.
Hammermesh, Daniel
S. (1986),
"The Demand
for
Labor
in the
Long Run," Chapter 8
O. Ashenfelter and R. Layard, eds.. The Handbook of Labor Economics, Volume
/,
in
Amsterdam:
North-Holland.
Heckman, James and Brook Payner
(1989), "Determining the
crimination Pohcy on the Economic Status of Blacks:
Economic Review
A
Impact of Federal Antidis-
Study of South Carohna" American
79, 138-77.
Hopenhayn, Hugo and Richard Rogerson (1993), "Job Turnover and Pohcy Evaluation:
A
General Equihbrium Analysis" Journal of Political Economy 101, 915-38.
Job Accommodation Network (1997), Accommodation Benefit/Cost Data, mimeo, West
Virginia University.
25
Jovanovic,
Economy,
Boyan
(1979), "Job
Matching and the Theory of Turnover" Journal of Political
87, 972-990.
Kemp, Evan
and Work:
Jr (1991),
"DisabiUty in Our Society," in Carolyn Weaver, ed.. Disability
Incentives, Rights
and Opportunities, Washington DC: The American Enterprise
Institute Press.
Krueger, Alan, and Douglas Krnse (1995), "Labor Market Effects of Spinal Cord Injuries
in the
Dawn
of the
Computer Age,"
NBER
Kubik, Jeffrey (1997), "Incentives
Disabilities:
ment
The Growth
of Economics,
Lazear,
Edward
Working Paper No. 5302, October.
for the Identification
of the SSI Child Disability
and Treatment of Children with
Program" University of Syracuse Depart-
mimeo, September.
(1990), "Job Security Provisions
and Employment," Quarterly Journal of
Economics, 105, 699-726.
Leonard, Jonathan (1989),
"Women and
Affirmative Action," Journal of
Economic Per-
spectives 3, 61-76.
Leonard, Jonathan (1990), "The Impact of Affirmative Action Regulation and Equal
ployment Law on Black Employment," Journal of Economic Perspectives
4,
Em-
47-63.
Leonard, Jonathan (1991), "Disabihty Pohcy and the Retm-n to Work," in Weaver op.
cit.
Mortenseu, Dale (1978), "Specific Capital and Labor Turnover," Bell Journal of Economics
9,
572-586.
Nickell,
Steven (1997), "Unemployment and Labor Market Rigidities: Europe versus North
America," Journal of Economic Perspectives
Oi, Walter (1991), "Disability
and
A
11, 55-74.
Workfare- Welfare Dilemma," in Weaver op.
Olson, Walter (1997), The Excuse Factory:
ican Workplace,
The
How Employment Law
is
cit.
Paralyzing the Amer-
Free Press.
Oyer, Paul and Scott Schaeffer (1998), "Layoffs and Litigation," Northwestern University,
Department of Economics, mimeo.
Peracchi, France and Finis Welch (1995),
"How Representative
Evidence from the CPS," Journal of Econometrics,
Pohvka, Anne E. (1996), "Data Watch:
are
Matched Cross-sections?
68, 153-179.
The Redesigned Current Population
Survey,"
Journal of Economic Perspectives 10(3), 169-180.
Poterba, James and Lawrence
Dynamics" Econometrica,
Summers
(1986),
"Reporting Errors and Labor Market
54, 1319-115.
Rosen, Sherwin (1991), "Disabihty Accommodation and the Labor Market," in Weaver
op.
cit.
26
Riihm, Christopher
Lessons
J.
(1998),
"The Economic Consequences of
From Europe," Quarterly Journal of Economics
Scherer, Frederick
Pcirental
Leave Mandates:
113, 285-318.
and David Ross (1990), Industrial Market Structure and Economic Per-
formance, Third Edition, Boston, Houghton
Mifflin.
Schmalensee, Richard (1989), "Intra-Industry Profitabihty Differences in U.S. Manufacturing:"
Journal of Industrial Economics 37, 337-358.
Stapleton, David, et al (1994), Labor Market Conditions, Socioeconomic Factors and the
Growth of Applications and Awards for SSDI and SSI Disability
VHI,
Benefits,
Arhngton: Lewin-
Inc.
Summers, Lawrence
(1989),
"Some Simple Economics
Economic Review Papers and Proceedings,
Veres, John,
and Ronald Sims (1995),
xvith Disabilities Act,
of
Mandated
Benefits,"
American
79, 177-183.
Human
Westport, Connecticut:
Resource Management and the Americans
Quorum
Books.
Waldfogel, Jane (1996), "The Impact of the Family and Medical Leave Act on Coverage,
Leave- Taking,
Employment and Earnings," Columbia University School
of Social
Work, mimeo.
May.
Weaver, Carolyn (1991),
Weaver
op.
"Incentives versus Controls in Federal Disability Policy,"
in
cit.
and Alec Levenson,
eds.,
"How Government Reduces Employment"
in
Lewis Solomon
Labor Employment Policy and Job Creation, Boulder:
The Westview
Weidenbaum, Murray
(1994),
Press.
27
o
m
so
00
d
o
00
d
ON
00
oo
OS
0\
en
00
O
O
o
o
d
NO
rf
00
iri
NO
O
d
00
00
o
vO
oo
ON
ON
On
d
<u
o
o
00
00
00
ON
o
NO
o
NO
^3-
oo
ON
ON
o
o
ON
00
NO
O
00
00
ON
m
O
o
d
NO
U-1
NO
00
00
ON
to
oo
p
oo
o
(N
NO
00
m
O
00
o
>
NO
rON
^
^
On
O
o
-a
00
00
On
o
«
3
NO
-a
a>
BO
>
o
ON
or)
r^
-*
(N
in
00
CO
00
a
o
O
d
O
NO
r-
00
0-
U
o
oo
r^i
d
CN
00
00
d
o
o
C^J
CN
CN
00
o
(N
f-
nD
On
ON
^
—
d
00
o
oo
d
o
o
CN
in
o
o
C3
o
CN
E
o
00
ON
(N
On
OO
c-~i
o
o
Q
CN
aj
<
On
ON
00
00
ON
ON
C-Nl
oo
U~)
On
O
o
tT
m
ON
On
ON
CN
00
o
o
CN
NO
CO
o
o
«
3
o
o
00
2
1)
DO
<
00
C3
o
o
00
oo
o
d
o
tu
wo
^
1
-^
o
1)
"o
u
r<-,
m
^
00
o
00
On
d
u
r3
§
«
^
o
d
<r~>
^
^r
1=
O
x>
CO
t/3
C/5
•5
t/5
•a
U
>
C3
t/5
ra
D
<n
X
Si
On
Tf
OO
CN
CN
On
>0
CN
CN
C3
>
lyj
C
>^
C«
a.
Qi
t-^
Q
o
Z
NO
00
00
CJ
d
CN
d
d
o
NO
CI
_
u-i
VO
CO
00
r<0
d
d
d
r-
d
«/o
iri
NO
>n
o
NO
m
r-
00
ON
'a-
ON
On
o
d
o
d
o
ON
o
d
O
On
o
NO
ro
00
On
On
NO
OJ
m
d
in
CO
en
o
>iO
00
ON
0\
d
<N
d
d
On
d
<
O
o
ON
00
ON
in
oo
^^
oo
00
On
f3
ro
NO
00
00
^
^-^
d
o
d
00
NO
r~
NO
vO
d
d
(N
On
p
in
ON
(N
d
d
O
d
00
OO
NO
OO
NO
O
O
O
d
00
00
ON
(U
•a
T3
t)
3
C
O
Ov
NO
On
On
00
o
E-
00
O
m
^->
o
00
ON
On
o
ON
-*
m
On
NO
00
On
ON
o
ro
O
O
O
C:
(^J
00
O
(^J
o
o
NO
o
^_^
^
r4
ON
NO
rr-,
NO
ON
u
CO
O
o
ON
00
0-
U
NO
in
rtN
O
d
o
o
o
'^
o
d
d
CO
o
OO
O
fN
<
o
>n
O
d
ON
On
00
OO
ON
ON
nD
00
d
""^
O
fN
ON
NO
0^
^
o
r^,
fN
ON
ON
CO
d
d
-sT
On
NO
ON
ri
Q
ON
o
o
o
OO
*
ON
m
d
ON
ON
OO
>
-^
OJ
C3
=1
o
o
JZ
u
c
HJ
E
o
^
OQ
00
<
CO
X o
T3
5
to
CO
"i
O
(U
SSI
C3
o
payments
<u
1)
00
<u
c
"o
U
>.
_o
a.
C3
-a
U
E
Receives
u
1
W
^f
DI
o
2
—
CN
in
O
od
c
u
E
o
'-
OS '^
CN
— r~
— o
o o
OS o\
OJ
—
m o
d d
O
o
9 o
00
o
d d
(N
O
d d
E
•^ ON
00 r-
v~i
o
o o
o o
vo
OS
r~~
o
— o
d d
o
o
in Tt
o o
o o
o o
o o
o
o
o o
00
(S -^
>n
o
-J
— t~
O
— o
d d
o o
c5 d
m
o o
d d
oo
rf oo
— ^
o o
o o
lO 0\
00 OS
o o
d d
— —
o o
(N
Tl-
d> di
r^
Tj-
o o
— OS
o o
d d>
o
o —
d d
—
—
o o
d cS
0\
o o
d d
00
00
VO
in
u
u
00
00
so
r~
IT)
o
u
00
00
so
rin
— o
o o
d d
vO
o
o o
o
O o
o o
so
OS
rso
>
d.
Q
oT
c
m
o
O o
(U
Os
o
o o
CM
u-1
O o
O o
o
o o
3 o
o 3
o
^^ o
o o
o o
OS
O o
o B
o
fN
o O
o O
o 00
o
^^ o
00
so
o o
o o
^
r~-
o o
9 o
O
o o
Os
rsO
-§
o
00
o o
o o
o
o m
o o
ros
m
o Co
o
o o
r-~
so
>^
00 oo
p^
d
OS
rj
S ~ —
rj
OS
r-
T3
5^
ON
in 00
o
O
^
(N
o
s.
r*-,
>
s 9
2.
00
OS sO
o
P-
fN
O
CO
9 o
so >n
so
m
O o
—
o
— o
00 00
o
^ —
o
c
o 00
Tf o
2;
"^
Zi
so so
r^,
Os
oc
r~,
^^
OS
ri
o)
'f
oT
(^1
o
c,
o in
O o
"3-
Os
00
o
ir-,
Os
ri "~
r^.
00
OS
in
ir,
— O
m
— o
in
^~
oo
m
o m
m o
CO
C3
OS
^
3
5
u
E
c
OS
OS
— o
tN ""
>
1-
O
—
m
rj TT
C
u u
OS
r)
O
n
o o
o
—
Os
~
o
O
r-
OJ in
OS
^^
^
— m
(N
d d
00 OS
r^,
~~
E
Ti-
OS rt^
r^, "~
=0
ii
_
00 \0
— —
o o
in (N
(N
O
d d
OS
m
CO
u
OS
OS
in
tN
o
o
OS
C 00
U C
o
so
Os
OS
QJ
uLU
Os
P^
1
CM
UJ
6JJ
<
c
a
S
OS
OS
OS
OS
OS
OS
OS
OS
m
OS
OS
*
SO
OS
Os
Uri
H
m
OS
c
U.
c3
*
o
J3
X3
K3
C3
C3
Xl
a
Xl
J3
C3
X
c
o
X a
O
>>
—
E
3
QJ
^-
Os
O.
c
-5
X5
C3
o
o o
S
Z
30
»
00
— m
o
oo
-"a-
fn
CO
c
c
u
E
o
rn ON
^ o
d d
o
o o
d d
\o
o
— o
d d
<^
so vO
VO OS
Oo
Oo
O o
o o
o
OS
O
00
oo
1
rsi
oo -^
— o
d d
^
00 rr
o o
d d
o o
o
d d
O
m
o
o o
m
o o
o o
m
IT)
mo
d d
oo
ON r-
oo
d d
-"I-
mo
OS <N
Tt oo
— O
d d
r-~
o o
n —
o o
o o
d d
o
— m
o
dd
ON
o o
d d
-^ OS
>0
00 ro
t-~
O o
o o
oo
00
o
OS
o
r^ so
ro
— m
o
d d
o
O
so
On
-J
•^ -rf
(N SO
OS
o o
d d
o
o o
>
OS
OS
d.
1)
Q
f7
i
o o
o
00
O o
o o
so
u-1
o
o o
o O
o O
so
o o
o o
SO
o
o
o o
SO
o
o o
O m
o
O o
so 00
oo
m
— O
d d
—
— o
—
d> d
ON
oo
SO >n
>/^
(N
d
o
ON
o
so
SO
so
cN in
so
sD
SO
__
O
o
d
—
m
o o
o o
—
—
d>
m
O
—
o
d
SO
m
X)
a
a
>
c
r
Q.
P.
ON
53-
o ro
o
d d
n-)
00
— CO
o o
d d
-Tt
m
so
— m
m
o o
d d
so
SO
SO
D
QJ
-o
J=
H
<1>
n
•a
>-,
— o
—
o o
u
so
ON
O
5
o
00
O O
cp "^
""
rn
~
o
in so
r4 ^~
m
o
00
00 r-
^~ (N
d d
m m
t>
<NI
in
^
C/5
SO
00
"^
cu
OS
00
csi
O
in in
00
a
u
d d
d d
—
—
— oo
r~ —
I
c
'^r
1:3
OS
00
in
(^
in
r^ so
r-
I^ ON
c
o
l~
>-.
oo ON
[^
E
-a
"2
c
c
•0
-a
aj
^
H
01
r^ ro
O
sD sD
SO
in
so
— O
o
m
Cp
so
O
m S
o O
o
O
ON
OS so
(N
ON
00
in
SO
m
fN
o
m
c
t/1
C/5
aj
o;i
»^
(ij
?
<
in
1
0/)
:3
— d
>
'
O m
d d
OS
so ON
O
t^
so CN
d d
CN
O
o o
o
m
so
c C
c
F
1)
•n
d.
3
^
-:<:
OJ
O
O
c
Tt in
o
OS so
o o
in
so
""
O
oo OS
O o
o o
9 -"
s
o
•"~
'^
•~
oo
n
(N —
d d
Tl-
~
OS
O
m
so
t
<u
S1
D.
n
n
(1)
v:
o O
o
m ?
^
O o
r-1
OS
OS
o O
^3-
— O
m ^
— o
t^
m ^
O
ON
1
«)
o J=
H
o
O o
ro
01)
f^
>
a
on
r<-l
SO
1
T3
o
u
so
OS
OS
Lb
UJ
00
vn
o
'^
y
60
<
oa
c
C3
S
™
-D
^^
ON
ON
*
OS
ON
OS
OS
*
ON
so
ON
OS
*
*
>>
X5
C3
Q
OS
ON
m
OS
eg
C8
a
X)
(5
ro
OS
OS
*
c
lU
H
£2
Urn
a
c
£
(1>
(A
T5
*
C3
XI
O
Xi
=
P
XJ
XJ
ca
C3
•0
CO
..
-a
ID
^
o o
S
2
?/
'
T3
C
o
vO
00
\0
vo
O
—
—
.'
00 ON
\0 CN
vo o\
t~~
(^
o o
C\ CO
—
o o
Tt
CO
—
O
(««
lA
O u
o
<u
trt
ca
T3
:
-1
U
g>
ou
<
c
C
o
r~ 00
00 OS
— m
00
m
ON CN
VO C^
(N 00
— o
ON r—I
o
'
o o
Ul
< o
Qi
<
a>
o
"u
c
> n
JLJ
E
u
U
'5
C3
U
O
00
(/:
O
DC
u
ct
u,
r3
U
U
O
OS
O
ri O
m
m
r>r)
—
Tt
(^ VD
—
vo
en
o—
r~-
O—
m
O
CM —
-o
T3
<
C3
r^,
in
\D 00
(N
CN
O
-5
ON
VO
C
o
O
ON
m
(N O
-*
o
Z
<
<
Ml
U
—^
il
CQ
4J
3
00 en
o o
—
rn
CN
r<^
1
h-
(N ON
<N
O
—
CM
OS
" c
—
c/j
Ov
O
O
a.
o
»y-i
—
<
CN
to
XI
n
u
-a
H
\0 r4
00 rj
CN ro
l/~l
m
m
IT)
ON
CN
m
O VD
O
ro
o
r-~
r^
o o
00
o
u c
Q. -
U-,
2 i ^
=
o B
t s-g
"E
m
C3
U
ro
ON
Ov
^
Ov
ON
m
ON
ON
VO
ON
as
f»^
Ov
ON
^
ON
ON
U-1
ON
ON
VO
ON
ON
"-
"
^
oj
—
2
C5
S
O U
J3
Z
J2L
"^ ><
D. >,
1
A
Table
Robustness of Basic Results
to
Sample Restrictions
Matched March 1993-March 1994 Data
Full
All
Same
Valid
Valid
Valid
Sample
Matched
Disability
Supplement
Supplement
Supplement
Records
Status in
Response
Response
Response
1993&1994
(1)
(2)
1993
(4)
(3)
in
in
1993&1994
(5)
Previous
&
Column
Excludmg
Same
Allocated
Disability in
1993&1994
Weeks
Worked
(6)
(7)
Men Aged 2 J -39
Without
-1.23
-3.11
-0.30
-3.47
-3.57
-0.04
-0.87
Covariates
(0.98)
(1.10)
(1.35)
(1.13)
(1.18)
(1.41)
(1.43)
With Covariates
-1.54
-3.16
-0.98
-3.42
-3.55
-0.67
-1.27
(0.95)
(1.05)
(1.31)
(1.09)
(1.13)
(1.37)
(1.38)
19271
13251
12669
12072
11247
10811
10656
N
Men Aged 40-58
W:thoul
-3.75
-3.62
-3.40
-4.18
-3.68
-2.76
-2.65
Covariates
(0.77)
(0.83)
(0.91)
(0.86)
(0.88)
(0.93)
(0.93)
With Covariates
N
-4.15
-3.99
-3.58
-4.52
-4.06
-2.99
-2.94
(0.74)
(0.79)
(0.89)
(0.82)
(0.85)
(0.91)
(0.91)
15105
12719
1914
1488
10637
10060
9951
Women Aged 21-39
Without
-2.69
-1.13
-2.50
-1.75
-1.51
-2.46
-2.54
Covariates
(1.39)
(1.69)
(2.28)
(1.75)
(1.81)
(2.33)
(2.35)
With Covariates
N
-2.29
-1.24
-2.79
-1.90
-1.78
-2.76
-2.85
(1.35)
(1.63)
(2.21)
(1.70)
(1.75)
(2.26)
(2.27)
21372
15130
14467
13878
12958
12469
12388
Women Aged 40-58
Without
1.09
1.55
-0.42
2.32
2.21
-0.17
-0.19
Covariates
(1.12)
(1.24)
(1.55)
(1.30)
(1.35)
(1.61)
(1.62)
With Covariates
N
Notes
1.01
1.38
-0.59
2.27
2.18
-0.36
-0.38
(1.08)
(1.20)
(1.51)
(1.27)
(1.31)
(1.57)
(1.57)
16407
13925
12907
12523
11522
10807
10699
Standard errors are shown
in parentheses. All entries are OLS estimates of coefficients on the disabled*
1994
equations for weeks worked. Models without covariates include year, age, race, and disabled main effects.
Models with covariates also contain education, marital status, and veteran status controls.
:
dummy
in
33
rj
00 io
«t On
>
O
^
V
V
VO in
ro 00
(N
o
1
o
o o
\o
00 00
^
^^
VO 00
ra
o
^ oo
tT
(N
\o 00
'
1
ON
o
o
^
>n
o
so so
d —
o—
*^ C-
— o
d d
oo
00
On
"Z
-
- >
c
u
E
o
^
2
c^ on
o
~ J
O
'='
— o
d d
— d
—
— o
00
so
00 00
m
o d
00
o
ON
rn
rt-;
d
0\ vo
SO r~
mo
r- so
o
cs
00 so
— o
rn
d
rn
t
d
>n t^
^ d
—
o
ro
^^
*
--0
00 Tf
c-^
d
m
O
so 00
00 r~
^ d
r^
os
so
d
11
o O
o Bo -Eu
d d
(N
1
•
y—»v
Os
ro
o O
O o
o o
0\ 00
^^ rso r—
o
OS so
ro r~r<->
'
m
o
SO
r-
o
rn
r~
1
m
o
r~-
E I
00 00
04 t^
(N
«N
oi
o
rsi
d
u
Q ^
o ^
u
I
T3
aj
o
i-
^
^—
on
oo -J
su On
ro
O O
^
m o
00
— o
o
o o
on
^.—
,,<-*v
sD
OS CO
On r~
p~- i^
— o
OS rr- p-
— o
O
in
so
1^
r<",
^
f. so
^
r^
^-v
,,^-s
r--
o
OS so
r~
O
o
^.—^
m
— o
Tf
so t~
0\ 00
t~- r-
— o
1
OS 00
so 00
o
<1-
00 r-
— o
o
OO
<N
o
OS
r<-i
3>
sO f^
P-^P-'
oo
o
"J
O
T
oo
—
<^
2 T 2
<n.
S
'^
2-
ro
oo <n
IT-,
o o
T 3
O
—
O
O r— d
OS
SO sD
(N
Jd
u
E :=
o "
CO
O "
c
.u
-- 00
00 «
m
o o
o
o o
1
0\ so
o
m o
^—
„«—
>
-a
><
a
o\
1
ON -J
^
o o
d d
C/5
^
>
"a
(J
tLt
« I
3 O
> —
Q
on
on
n
^
-3-
F
_
r-,
•*
>
^
OS
Ov
""
oo
in
—
on
-1
2
1^
Ov
on
-J
9.
o
o
^
—
OS
^
2
tN
m
O O
in
so
in
in
>n
—
OS
in m,
O
rs)
in
in
in
o
m o
in
so
OS in
r<-i
in
—
in
no
— o
o o
— o
O
sD
M
O
in
in
O
ro
in
d
M
in
— d
d
(N
d
—
oo tN
in sD
^- in
ri
CQ
S
in
o m
(N o
m,
r«-,
(N
CO
— d
o O
o
d d
r^.
On
sd
r^,
d
r«-i
oo
fN
rs|
d
'
^
on
•""
ri -J
in
rrN
o
o r^
n
mi o
—o
)
n
o o
1^
so in
00 r4
so in
o o
1
oo
r^
^
—
li^i
£•
00 ^r-- in
o
"^
i^
00 in
^~
2-
O m
"" o
^p- mj
rj
On in
^
— O
o o
m
ro
so in
^"
o
1
^
C o
o m
in
m
— o
m
— o
tT
r- in
o
in
o in
m £
rs<
(-)
rs
O
O
o o
E
E
3
—
rr- CN
r~-
o
-o
Q.T3
OS
(N
On
OS
OS
OS
On
On
On
On
*
*
in
ON
On
SO
OS
OS
!0
on
C
u
Q
J3
Q
T5
*
S"
<
a
X)
XI
J3
XI
o
on
on
(5
?7
00
-o -d
Si
E
u
.S-
.E
S
«
2
.9-
^
"s
a
>.
on
-;?
T3
»O
on
on
<U
00
5
5
b.
c
<
2
(>n
I
o
—
1
fN
>
>o
m o
O o
m
O O
U1
On VO
CTn
ra
00
o
o
O o
OS
O
VO
O
o o
Ov
00 {N
VD
— O
—
OV
O
— O
o
d d
00
OO
Ov
^
X
a
- >
f^ in
t^
o
m vD
— O
so
VO
s
1
sS
9 o
m ON
>n
o o
r~-
n 00
m
o
m
<N
1
O
u
u u
u
OO
<
«
« 3
S
>n •o
r--
t/v
>
-n
^ 00J
o
c
u
E
o
VD (N
30
<n
o
in CO
00
o O
in
Ov
00
VO
vO
— o
o o
m
O o
VO
ov
— o
o
1
1
^^
— (N (N
o
00 m
o
o O
Ov O
o o
Ov
1
-C
j=
o o
E 1)
q>
^
(N
--o
in
—
rn
(N
(^
vO ~>
00 -J
-O
— vo
— d
m
O
00 vD
d d
m
^ vO
— d
00
^
00 -"t
(N vO
— d
—
in vo
d d
VO
vD 0\
vD in
d d
ON ov
ro
m
d d
VO r-
r<-)
O m
oi d
r-~
vD
in
Ov (N
00 r-
—
^
d
OO rCO
d
c*
03
-n
9>
^ d
— d
— rr- m
— d
E
o
u
^.^
Q
O
u
F
o
m
m
o
o O
d d
o
•o
ii
Q.
4;
.^ OJl
00 a
00 :S
—
^
I
m
o
00 —
O VO
d d
n
o
00
O
vO
vO
rn
o
in
r<-,
O vO
— O
fN VD
— O
—
°
—
vO
S
o
^ a^
m
o o
m
— O
Ov
in
VO OO
— o
> •a
u
m O
h)
_^
HJ
P
3 o
a
—
>
c
r4
^ 3 O
vO ri
~
-T
00
-1
^
00 OJ
^
^ o
(^j
:r^
Ov
^ —
r~-
o
OO
GO
ll:
r^,
*
9 o
O;
(-.J
B O
^
2.
vO
(^)
— -*
— O
ov
00
^
o o
in
^
vO
^~
o^
m
£,
3
S-
o
(N
M
"*
o
^
s
o
ov
Ov"
""
O
00
00
""
s
t^
in
rJ
—
^
O
vD ov
vD
OJ
Ov Ov
00
m
^
""
vO OO
in
>n
"~
O
O
in
O
O^
r<-.
r<^,
ri
S
"~
r-i
m
o
^
OO rn
00 -^
o
o
OO
ov
ov ov
OO (N
— O
'
" O
2-
OO
O
vD
^ O
OO
vO OO
00
r^,
— m
o
o O
u u
o o
—
in
vO rj
m
o O
O
O
vO
t^ —
£ >
E "5
2 22
5
C3
^
c/;
00
-J
—o
(-4
—
—
Q
00 (N
—
£
OO
^
O o
in Ov
"^
O
in o^
"
2-
OO
00
"^
2-
9 O
OO
OO
o
o
^"
OO
O
o
^ O
vO
o o
o
o o
is
.E
u
*^
C
OJ
-§
>
>1
^ 00
—O
r- CM
t^
]
o
00
in
^
o o
ov"
00
— o
o^
o
— O
— O
vO
O o
o 00
vO
O o
tN
Ov
O
(N
o
ffl
—
D.T3
T3
T3
^.^
OO
l/'l
o
>^
jz
60
ov
ov
fN
Ov
OV
*
*
'jz:
.^
Xi
C3
DQ
Q
OV
Ov
Tt
ov
m
Ov
Ov
Ov
*
vO
OV
OV
x>
XI
C3
1/5
t/5
t/5
03
V5
Q
Q
Q
i5
?S'
c
Q.
O
(U
n
.
00
-O
2i
*
*
0)
o
T3
Q
X3
C3
Q
r^.
t=
C
00
C3^
>
>
00
o
u
1)
#
l-H
Urn
Q.
o
Q
*
I
X
C3
o
o
(U
H^
60
Q
OD
00
00
00
T3
C C
« aj
o.
CO
1)
<
OO
<
0.U
^
0.10
e
a
1
^^/
0.09
/
/
/
t
h
0.08
A
f
f
0.07
e
c
t
0.06-
s
W
0.05
I
0.04
0.03
-
-
88
89
90
92
91
93
95
94
96
Survey Year
Sex/Age Group
Men age 23-39
Women age 21-39
Figure
1.
Work - related
disability rates for
Men
Women
men and women aged 21-58
J6
age 40-58
age 40-58
in the
1988-97 March CPS.
97
—
r
50
45
W
-
40 4
e
e
k
s
35 ^
W
25
-f
r
k
e
d
20
15
10
—
—
I
87
88
89
90
92
91
94
93
Year
nondisabled
nondisabled
Figure
2.
Weeks worked
last
year by sex and
men
women
disabled
disabled
disability status,
31
CPS
men
women
respondents aged 21 — 58.
95
96
nondisabled
nondisabled
Figure 3a.
Weeks worked
last
men
women
disabled
disabled
year by sex and disability status,
3S
CPS
men
women
respondents aged 21 — 39.
50
45-
w
40
e
e
k
s
35
W
25
r
k
e
d
20
15
10
-
-
87
89
90
92
91
&4
93
Year
nondisabled
nondisabled
Figure 3b. Weeks worked
last
men
women
disabled
disabled
year by sex and disability status,
J*?
CPS
men
women
respondents aged 40 — 58.
95
96
Separation Rates
-1.0 4
-1.5
y.
-2.0
-2.5
3.0
-
-3.5
-
-4.0
88
I
I
89
90
93
92
91
94
95
Survey Year
nondisabled
nondisabled
Figure 4a. (Log) Separation rates by
•
•""•
""•
men
women
disability status
and
sex.
io
disabled
disabled
CPS
men
women
respondents aged 21-39.
96
97
Accession Rates
-1.0
-f
-1.5-
Z.0
-2.5
-3.0-
-3.5
-
-4.0
1
88
89
1
90
i
91
1
1
92
93
1
94
1
1
95
96
Survey Year
nondisabled
nondisabled
men
women
• • •
*--*--*
Figure 4b. (Log) Accession rates by disability status and sex.
Vf
CPS
disabled
disabled
men
women
respondents aged 21 — 39.
97
Separation Rates
-1.0
-f
-1.5-
-2.0-
-2.5
-
-3.0
-
-3.5
-
-4.0
-
-4.5
-5.0
I
88
89
90
91
92
93
94
95
Survey Year
nondisabled
nondisabled
Figure
4c.
men
women
disabled
disabled
(Log) Separation rates by disability status and sex.
V^
CPS
men
women
respondents aged 40-58.
—r96
97
r
Accession Rates
-1.0 4
-1.5-
-2.0
-
-2.5
-3.0-
-3.5
-4.0
-
88
—
89
90
91
92
93
94
95
Survey Year
nondisabled
nondisabled
men
women
——
• •
•""•-"
"•
Figure 4d. (Log) Accession rates by disability status and sex.
H3
CPS
disabled
disabled
men
women
respondents aged 40 — 58.
96
97
r
Log odds
1-
-1
—
87
88
89
90
92
91
93
96
95
94
Year
Firm
— —•
•
Figure
•
5a.
MNL
size
< 25, nondisabled
< 25, disabled
size
plot of log
employment
size
^ -•--•
and
size
probabilities
disability status
25-99, nondisabled
25-99, disabled
size
by firm
Reference group consists of nonworkers.
yv
size
size 100
size
+
,
nondisabled
100+
,
and disabOity status for men aged
disabled
21
— 39.
—
Log odds
1-
-1-
2-
3
-
—
I
1
89
87
90
92
91
93
96
95
94
Year
Firm
size
< 25, nondisabled
< 25, disabled
size
Figure 5b.
MNL
plot of log
employment
size
and
size
*--•--•
probabilities
disability status
25-99, nondisabled
25-99, disabled
size 100
size
by firm
Reference group consists of nonworkers.
V^
size
and
disability status for
+
,
size 100
nondisabled
+
,
disabled
women aged 21-39.
Log odds
3
1
-
0-
-1
87
88
89
90
92
91
93
94
96
95
Year
Firm
size
Figure
5c.
MNL
< 25, nondisabled
< 25, disabled
size
and
size
disability status
25-99, nondisabled
25-99, disabled
size
^-•--•-
plot of log
employment probabOities by firm
size 100
size
size
Reference group consists of nonworkers.
y^
and
disability status for
+
,
size 100
nondisabled
+
,
disabled
men aged 40—58.
0.06-
0.05-
/
s
s
I
/
0.04
r
0.03
D
I
0.02
0.01
87
88
89
90
92
91
94
93
Year
Sex/Age Group
Men
Women
Figure Bl. SSI or DI recipiency for
Men
Women
age 21-39
age 21-39
men and women aged 21-58
yf
in
the
age
40-58
age 40-58
1988-97 March CPS.
95
96
707
D
J
Lib-26-67
MIT LIBRARIES
3 9080 01972 0819