Bonuses to Workers and Employers to Reduce Unemployment: Randomized Trials
in Illinois
Stephen A. Woodbury; Robert G. Spiegelman
The American Economic Review, Vol. 77, No. 4. (Sep., 1987), pp. 513-530.
Stable URL:
http://links.jstor.org/sici?sici=0002-8282%28198709%2977%3A4%3C513%3ABTWAET%3E2.0.CO%3B2-R
The American Economic Review is currently published by American Economic Association.
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/journals/aea.html.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact support@jstor.org.
http://www.jstor.org
Mon Dec 3 13:00:40 2007
Bonuses to Workers and Employers to Reduce Unemployment: 

Randomized Trials in Illinois
By STEPHENA. WOODBURY G. SPIEGELMAN*AND ROBERT
New claimants for Unemployment Insurance were randomly assigned to one of
two experiments that were designed to speed up the return to work. In the jirst
experiment, a $500 bonus was ofered to eligible claimants who obtained employment
within 11 weeks. This experiment reduced the number of weeks of insured
unemployment, averaged over all assigned claimants whether or not they participated,
by more than one week. In the second experiment, the $500 bonus was
ofered to the subsequent employer of the eligible claimant. This experiment
reduced the weeks of insured unemployment for only one important group -white
women -by about one week.
Between mid-1984 and mid-1985, the 11linois
Department of Employment Security
conducted two controlled social experiments
designed to test the effectiveness of cash
bonuses in reducing the duration of insured
unemployment. In one experiment, called
the Claimant Bonus Experiment (or simply
Claimant Experiment), a random sample of
new claimants for Unemployment Insurance
(UI) were instructed that they would qualify
for a cash bonus of $500 if they found a job
*Department of Economics, Michigan State University,
East Lansing, MI 48824, and W. E. Upjohn Institute
for Employment Research; and W. E. Upjohn
Institute for Employment Research, 300 South Westnedge
Avenue, Kalamazoo, MI 49007, respectively.The
experiments reported here were conducted by the State
of Illinois, Department of Employment Security, using
funds allocated by Illinois Governor James R. Thompson
under the Wagner-Peyser Act. The support of Sally
Ward, Director of the Department of Employment
Security, was central to the conduct of the experiments.
The research was performed by the W. E. Upjohn
Institute, jointly funded by the Institute and the State
of Illinois. Spiegelman was project leader and, with
Woodbury, co-principal investigator. The data underlying
the results presented here, with all identifiers suppressed,
are available on magnetic tape at cost. Inquiries
should be sent to the authors at the W. E.
Upjohn Institute. We are particularly grateful to Azman
Abdullah for excellent research assistance, and Susan
Pozo, Peter Schmidt, and an anonymous referee for
helpful discussions and comments.
(of 30 hours or more per week) within 11
weeks of filing the UI claim, and if they held
that job for four months. The intent was to
create an incentive for claimants to search
more intensely for work and to become reemployed
more rapidly than they would
otherwise.
In the other experiment, called the Employer
Bonus Experiment (or Employer Experiment),
a second random sample of new
UI claimants were told that their next employer
would qualify for a cash bonus of
$500 if they, the claimants, found a job
within eleven weeks of filing the UI claim,
and if they retained that job for four months.
The intent here was to provide a marginal
wage-bill subsidy, or training subsidy, that
might reduce the duration of insured unem-
ployment.
The impetus for the Illinois Unemployment
Insurance experimentswas a decade of
criticism of the UI system and the inability
of economists to establish clearly how the
behavior of UI recipients differs from what
it would be in absence of the UI system, if it
differs at all.
The most benign view of the UI system
would be that it provides unemployment
benefits to carry a worker through a spell of
unemployment resulting from involuntary
layoff.In such a benign view, UI benefits are
taken to be a nondistortionary transfer the
514 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1987
size of which cannot be affected by an individual's
behavior. But there are at least two
reasons why the presence of unemployment
benefits may prolong a jobless spell beyond
what it would be in the absence of unemployment
benefits. First, as job search models
such as those pioneered by Dale Mortensen
(1970) and J. J. McCall (1970) suggest,
UI benfits may act as a subsidy to additional
job search. Indeed, nearly all empirical work
to date on the relation between UI and the
duration of unemployment has been interpreted
in the context of one or another job
search model. Second, as Orley Ashenfelter
(1978a) has observed, UI benefits are also a
subsidy to the consumption of nonmarket
time (or leisure). If labor is not supplied
perfectly inelastically (i.e., if workers place a
positive value on their nonmarket time), then
the availability of UI benefits may increase
unemployment duration to the point where
the margnal utility of nonmarket time equals
the difference between the wage and the UI
benefit level. Recently, Jerry Kingston et al.
(1986) and Robert St. Louis et al. (1986)
have suggested that many UI beneficiaries
search for work less than the law requires,
and that UI benefits subsidize leisure rather
than job search.
The Illinois experiments provide the first
opportunity to explore, within a controlled
experimental setting, whether bonuses paid
to UI beneficiaries or their employers would
reduce the unemployment of beneficiaries
relative to a randomly selected control group.
The experiments yield strong evidence on
whether the UI system is simply a benign
income transfer that changes no worker or
employer behavior. However, these experiments
alone do not resolve the question of
how best to characterize the behavior of
unemployed workers. In Section I, we describe
the design and operation of the
experiments, and sketch our data sources.
Section I1 discusses the incentives facing
experimental participants. Section I11 is a
presentation of the responses to the experimental
treatments, and Section IV a development
of experimental benefits and costs.
Section V is a discussion of the implications
of the experiments.
I. Experimental Design and Operations
A. Treatment Design
Both treatments consisted of a $500 bonus
payment. In the Claimant Experiment, $500
was paid to a claimant who had found a job
within 11 weeks and held that job for four
months. In the Employer Experiment, $500
was paid to an employer who hired an eligible
claimant within 11 weeks of the initial
claim and employed that claimant for four
months. The size of the bonus, $500, reflected
a balancing of the experiments'
budget constraint (a maximum of $750,000
in bonus payments) against a somewhat arbitrary
judgment about how small a bonus
could still be expected to generate a response.
For the average UI claimant, $500
was on the order of 5 percent of annual wage
and salary payments, and represented about
four weeks of UI benefit payments. Such an
amount was believed large enough to provide
an incentive to at least some claimants
to accept employment more quickly (or
search more intensely), and to at least some
employers to hire a claimant.'
To qualify for the $500 bonus (or to make
their employer qualify), claimants had to
find a job of 30 hours or more per week
within 11 weeks of filing their initial claim,
and to hold that job for at least four months.
The period of 11 weeks was chosen arbitrarily
to be about 40 percent of the potential
duration of benefits in Illinois, which is
26 weeks. It was also chosen to be less than
the median duration of insured unemployment
experienced by Illinois UI beneficiaries
in the months preceding the e~periment.~
'The $500 bonus was low compared with the marginal
wage-bill subsidy provided under the Targeted Jobs
Tax Credit, which was up to $3,000 in the first year
after hiring an eligible worker. Because UI claimants
would seem to be more attractive prospective employees
than the disadvantaged workers who were eligible for
the Targeted Jobs Tax Credit, it seemed likely that a
smaller incentive might evoke a response from em-
ployers.
*Note that the 11-week period implies 10 weeks of
benefit payments, because of the 1-weekwaiting period
in Illinois.
515VOL. 77 NO. 4 WOODBURY AND SPIEGELMAN: UNEMPLOYMENT
Employment for four months was required
to avoid the possibility of fraudulent hire,
undertaken solely to obtain a bonus. Also,
four months was regarded as the shortest
period that would avoid payment of bonuses
to seasonal workers and employers.
B. Sample Design
To be eligible for either the Claimant
Experiment or the Employer Experiment, an
individual had to 1)file an initial claim for
UI between July 29,1984 and November 17,
1984; 2) be eligible for 26 weeks of UI
benefits; 3) register with one of 22 Job Service
offices in northern and central Illinois; and
4) be at least 20 years old, but less than 55.
Imposing these eligibility criteria increased
the homogeneity of the sample and
excluded claimants whose behavior might be
influenced by complicating factors. For example,
enrolling claimants filing claims other
than initial claims (i.e., additional, reopened,
or transitional clairns) would have meant
admitting claimants eligible for anywhere
from 1 to 26 weeks of UI benefits, and
would have complicated the evaluation unnecessarily.
Requiring enrollees to be Job
Service registrants restricted the experiments
to claimants who could be expected to obtain
a job through usual market channek3
Excluding younger and older claimants was
an attempt to reduce the number and kinds
of complicating factors-special programs
for young people, incentives to retire early
for older workers-that might influence the
job-finding behavior of those enrolled in the
experiments.
Each claimant was assigned to one of three
groups-the control group, the Claimant
Experiment treatment group, or the Employer
Experiment treatment group-by simple
random assignment, based on the last
two digits of his or her Social Security number.
Hence, claimants had an equal probability
of assignment to each of the three groups.
'Hence, workers on layoff with a definite recall date
and union members who find jobs through a hiring hall
were excluded from the design. In addition, recent
veterans and federal employees were excluded.
C. Site Selection and Sample Size
Three variables could have been used to
control the size of the sample: the number
(and type) of sites, the length of the enrollment
period, and the proportion of claimants
selected at any given site. Of the 22 Job
Service offices chosen as sites, 11 were in
metropolitan Chicago, 2 were in the outlying
metropolitan area, and 9 were in outlying
northern and central Illinois. Limiting the
experiments to these 22 offices, rather than
enrolling claimants at every Job Service office
in the state, lowered the costs of monitoring
the experiments and of communicating with
Job Service personnel. On the other hand,
having a fairly large number of sites helped
assure that experimental results would not
be specific to one or few offices, and that the
results would represent the response of the
diverse demographc and industrial mix of
Illinois. Also, having more sites permitted a
shorter enrollment period, which was viewed
as desirable in order to obtain results in a
reasonable period of time.
The duration of the experiments (originally
designed to be 13weeks, ultimately 16)
was selected with an eye to 1) the size of the
bonus budget, and 2) achieving a sample
large enough to allow detection of experimental
responses that would be relevant to
policy. As it turns out, the sample was large
enough to detect, at the 5-percent significance
level, a 0.017 change in the proportion
of claimants who found a job within 11
weeks (for example, a change from 25.6 to
27.3 percent). Such a small response is at the
lower bound of what could be regarded as
economically significant to UI reform (see
our report, 1987a, ch. 2).
D. Operational Issues and Data
Although the experimental design was unusually
simple, implementing it within an
existing agency posed a variety of complexities
(our report, 1987a, chs. 2 and 3). In
particular, several special instruments had to
be created to monitor and track the experience
of enrolled claimants. When a claimant
registered at the Job Serviceoffice, a base-line
516 THE AMERICAN ECONOMIC REVIEW SEPTEMBER I987
survey was administered to monetarily eligible
claimants aged 20 through 54, and a
special log was started on each claimant to
record the claimant's treatment assignment
and experience in the experiment. Claimants
who were assigned to the Claimant Experiment
or Employer Experiment groups were
asked to sign an "agreement to participate,"
and those who agreed received a packet of
administrative and instructional materials by
mail once their nonmonetary eligibility for
benefits had been determined. Participants
in the Claimant Experiment who found a job
within 11weeks submitted a Notice of Hire
(countersigned by the employer) to the 11linois
Department of Employment Security
(the Department), and those in the Employer
Experiment who found a job within
11 weeks gave a Notice of Hire to their
employer for submission to the Department.
The Department then returned a voucher to
the claimant or the employer, and the
voucher was in turn submitted for $500 cash
after the claimant had been employed continuously
for four months. All of these transactions
were recorded in the Job Service
office logs.
To analyze treatment responses, we had
access not only to the base-line survey and
office logs, but also to the administrative
data bases of the Department. Most important
were 1) the Benefits Information
System, which records the dates of claims
filed and the amount and timing of benefits
received, among other items, and 2) the Wage
Records data base, from which we have
drawn each claimant's quarterly earnings in
UI-covered employment for the third quarter
of 1983 through the third quarter of 1985
inclusive. Access to these administrative data
frees us from the problem of selective attrition
and associated nonresponse bias. That
is, we have complete data on each claimant's
experience in the experiment, on benefits
received, and on earnings in covered employment
before, during, and after the ex~
e r i m e n t . ~
4The data do impose an important limitation: We do
not know the labor force status of claimants who
terminated their benefits (particularly those who exhausted
benefits) but did not reenter UI-covered emTable
1 displays basic data on experimental
enrollment and use. Roughly 4,000 claimants
who were eligible for UI benefits and
eligible to participate in the experiments were
assigned to each of the three experimental
categories. Because the procedures used to
construct these three samples were identical,
each can be treated as a random sample
from the population of fully eligible initial
claimants for UI benefits who were aged 20
through 54.
The second row of Table 1 shows that
there were important differences between the
Claimant Experiment and Employer Experiment
in claimants' willingness to participate
in each. Whereas 84 percent of the eligible
claimants who were offered the chance to
participate in the Claimant Experiment
agreed to participate, only 65 percent of
those offered the chance to participate in the
Employer Experiment agreed. Table 1 further
indicates that actual use of the programs-as
shown by return of Notices of
Hire and actual cashing of bonuses-differed
greatly between the Claimant Experiment
and Employer Experiment. Whereas 14
percent of those assigned to the Claimant
Experiment received a bonus, only 3 percent
of those assigned to the Employer Experiment
were responsible for a bonus payment
to their employer.
We have explored participation in and use
of the experiments in detail elsewhere (1987a,
ch. 7), and for now note only two points
about the Employer Experiment. First, the
Employer Experiment was a more complicated
treatment than the Claimant Experiment
because it required the understanding
and participation of both the worker and
the employer. Second, the limited use of the
Employer Experiment suggests that it had
limited scope for reducing UI benefits paid
or weeks of insured unemployment.
Table 2 displays some descriptive statistics
of each of the three subsamples. Although
these data on sex, race, and other variables
may be of interest in their own right, their
ployment. These claimants could have found a job in
the uncovered sector, left the labor force, or remained
unemployed (i.e., continued to seek work).
VOL. 77 NO. 4 WOODBURY AND SPIEGELMAN: UNEMPLOYMENT 517
Claimant Employer
Control Experiment Experiment
Propor- Propor- ProporN
tion N tion N tion
~ l i g i b l e ~ . ~ 3,952 1.00 4,186 1.00 3,963 1.00
Agreed to
ParticipateC - - 3,527 0.84 2,586 0.65
Submitted
Notice of Hired - - 765 0.18 199 0.05
Bonus Paid - - 570 0.14 112 0.03
Sources: Eligibility from Illinois Department of Employment Security, Benefits Information System; other data from
office logs kept during the experiments.
aEligible for UI benefits by both monetary and nonmonetary criteria, met the age and initial claim restrictions of
the experiments, and were located in the Benefits Information System.
b~ total of 17,306 claimants completed the base-line survey and were assigned to one of the three groups; 1,857 of
these were monetarily ineligible, were not initial claimants (that is, were filing additional, reopened, or transitional
claims), or could not be located in the Benefits Information System. An additional 3,348 (1,171 controls, 1,104in the
Claimant Experiment, and 1,073 in the Employer Experiment) were nonmonetarily ineligible (i.e., failed to meet
separation and availability requirements, as determined by our constructed nonmonetary eligibility code), or failed to
meet the age restrictions of the experiments.
'Agreed to participate according to Job Service office records. 

Includes participants who ultimately received a bonus but never submitted a Notice of Hire. 

TABLE2-CHARACTERISTICS ASSIGNED GROUPSOF CLAIMANTS TO EXPERIMENTAL
Claimant Employer
Control Experiment Experiment
Propor- Propor- ProporN
tion N tion N tion
Total 3,952 1.000 4,186 1.000 3,963 1.000
Male 2,162 0.547 2,357 0.563 2,131 0.538
White 2,497 0.632 2,723 0.651 2,565 0.647
Black 1,072 0.271 1,050 0.251 1,014 0.256
Hispanic, Native
American,
Other 383 0.097 413 0.099 384 0.097
Age 20-29 1,680 0.425 1,827 0.436 1,679 0.424
Age 30-39 1,315 0.333 1,357 0.324 1,292 0.326
Age 40-49 708 0.179 776 0.185 740 0.187
Age 50-54 248 0.063 226 0.054 252 0.064
Weekly Benefit Amount:
$51 347 0.088 355 0.085 333 0.084
$52-$90 794 0.201 887 0.212 861 0.217
$91-$120 666 0.169 738 0.176 711 0.179
$121-$160 749 0.190 822 0.196 716 0.181
$161 1,396 0.353 1,384 0.331 1,342 0.339
Dependents'
Allowance 1,834 0.323 1,955 0.345 1,883 0.332
Source: Illinois Department of Employment Security, Benefits Information System.
Notes: "Weekly Benefit Amount" refers to weekly payment for which each claimant was eligible at the time of filing
the initial claim. The sample excludes claimants who were ineligible for UI benefits for monetary and nonmonetary
reasons (as determined by our constructed nonmonetary eligibilitycode), and who failed to meet the initial claim and
age restrictions of the experiments. Hence, all initial claimants who met the program criteria and were eligible for UI
benefits are included in the sample.
518 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1987
main import lies in the support they lend to
the randomness of the three subsamples. Indeed,
none of the difference~in proportions
between any pair of groups is statistically
different from zero at conventional significance
levels5 The randomness of the three
subsamples along the lines of observable
characteristics suggests that each of the three
subsamples was indeed randomly drawn
from the same population. It follows that
comparisons between the Claimant Experiment
group and the control group (or between
the Employer Experiment group and
the control group) implicitly control for all
observed and unobserved variables that may
have contributed to the outcomes that are of
interest-duration of insured unemployment
and post-reemployment earnings. Thus, a
simple comparison of the mean weeks of
insured unemployment for members of either
experimental group with the mean weeks of
unemployment for members of the control
group will show the impact of the treatment
in question on the duration of insured unem-
ployment.
11. IncentivesFacing Participants
The incentives facing UI claimants assigned
to the Claimant Experiment may be
viewed within the context of either a job
search model or an income-leisure model of
labor supply. In a job search context, the
$500 bonus creates an incentive for unemployed
workers either to lower their reservation
wage during the 11-week qualification
period and accept a job sooner than they
would otherwise, or to search more intensely
for a job in order to find a job sooner than
they would otherwise. In a labor supply context,
the $500 bonus raises the opportunity
cost of leisure consumed during the period
of time immediately following the initial
claim, hence creating an incentive to substitute
income for leisure. The basic results
of the Claimant Experiment cannot dis'We
have explored differences between the control
group and the experimental groups in several other
observable variables, and have found no statistically
significant differences.
tinguish whether one of these approaches(or
some other approach) is the appropriate way
to view how the Claimant Experiment
worked; such an enterprise requires imposition
of a structural model on the data and
we do not propose to do that here.
To the extent that leisure or job search
can be substituted easily from one time
period to another, some workers niay alter
their behavior both during the period of the
experiment and during the period after reemployment.
We shall test for this possibility
in what follows by examining the behavior
of the treatment group relative to the
control group before, during, and after the
experiment.
Marginal wage-bill subsidies resembling
the Employer Experiment have been implemented
in the United States before. As
Ashenfelter (1978b) has observed, the evaluation
of such subsidies is complicated by
the fact that the employers who would add
workers to their payrolls even without the
subsidy have the greatest incentive to participate
in such programs6 If an employer
pays the same wage rate to all workers of a
given skill class, then an employer who hires
an Employer Experiment enrollee will simply
receive a subsidy for hiring the enrollee
rather than someone else. Clearly, then, a
worker assigned to the Employer Experiment
has an increased probability of being
hired, but it does not follow that the Employer
Experiment must increase total employment.
The higher the turnover rate of
workers in a labor market, given the ratio of
Employer Experiment enrollees to total job
seekers in that market, the less likely it is
that the Employer Experiment will raise total
employment. Nevertheless, if the demand for
labor is sufficientlyelastic, the Employer Exhere
is of course a similar possibility in the Claimant
Experiment. Some workers assigned to the Claimant
Experiment-i.e., those whose planned unemployment
durations were shorter than 11weeks-could receive a
lump sum bonus without altering their behavior at all.
Nevertheless, if there is uncertainty about whether one
can find a job to fulfill precisely one's plan, even these
workers (if they are risk averters) may be induced by the
$500 bonus to become reemployed sooner than they
otherwise would.
519VOL. 77 ili0.4 WOODBURY AND SPIEGELMAN: UNEMPLOYMENT
periment bonus may induce an employer
who was not planning to hire additional
workers to increase total employment.7
A final consideration is that a worker assigned
to the Employer Experiment may
bargain with an employer to receive a wage
payment that is (in expected present value
terms) hgher than would otherwise be the
case by precisely the $500 bonus that the
employer will receive. If workers could strike
such bargains, which could be thought of as
"Coasean deals," then the Employer Experiment
and the Claimant Experiment would
establish precisely the same incentives for
workers, and the two experimental treatments
should have precisely the same effect
on workers' unemployment and earnings (net
of the bonus).'
111. Responses to the Treatments
A. 	Effects on Benejt Receipt and Duration
of Insured Unemployment
Table 3 displays the means of several program
variables by experimental group. These
means are based on the sample of all fully
eligible claimants-the same sample that underlies
Tables 1 and 2. We stress that this
sample includes eligible claimants who refused
to participate in one of the experiments,
so that the Claimant Experiment and
Employer Experiment groups are fully comparable
with the control group. (Examining
only Claimant Experiment and Employer
Experiment group members who agreed to
participate would involve a comparison of
self-selected groups-Claimant Experiment
and Employer Experiment agreers-with all
controls, some of whom would have refused
participation had they been offered the opportunity.
The result could be biased estimates
of the experimental effects.)
'For example, John Bishop (1987, ch. 4) has recently
concluded that the Targeted Jobs Tax Credit did result
in a net increase in employment in participating firms.
'see Ronald Coase (1960). Of course, it is also
possible for the Claimant Experiment and the Employer
Experiment to have identical effects without such
Coasean deals occurring.
Row 1under "Benefits Paid" ("State Regular,
First Spell") shows that the mean dollar
amount of state regular benefits received
by members of the control group during the
spell of unemployment immediately following
the initial claim was $2,267. For eligible
claimants assigned to the Claimant Experiment
and Employer Experiment, the comparable
figures are $2,074 and $2,159. Row 2
under "Benefits Paid" shows the mean of the
sum of state regular benefits and Federal
Supplemental Compensation received during
the spell of unemployment immediately following
the initial claim. Rows 3 and 4 show
state regular benefits received, and the sum
of state regular and Federal Supplemental
Compensation received, but this time for the
entire benefit year.
The full benefit year is the appropriate
time period to examine in determining the
impact of the experiments on benefit receipt,
rather than just the spell of insured unemployment
following the initial claim. As
noted in Section 11, it is possible that the
experimental treatments created incentives
to redistribute insured unemployment over
the benefit year, with insured unemployment
dropping immediately following the initial
claim, but increasing in the latter part of the
benefit year to compensate. We can capture
t h s effect, if it exists, by examining benefit
receipt and weeks of insured unemployment
over the full benefit year.
The two rows under "Weeks of Insured
Unemployment" show the means-for the
spell of unemployment immediately following
the initial claim, and for the full benefit
year-of the number of weeks of insured
unemployment for each group. It is worth
emphasizing that statistical tests performed
on these means are valid if interpreted as
tests on the number of weeks of insured
unemployment, because weeks of insured
unemployment are a censored measure of
total unemployment.
The entries at the bottom of Table 3 show
the proportion of claimants who exhausted
their state regular benefits, and the proportion
of claimants who terminated benefits
within 11weeks of filing their initial claim.
Table 4 displays differences between the
mean values of the control group and the
520 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1987
Control
Claimant
Experiment
Employer
Experiment
SE SE SE
Mean of Mean Mean of Mean Mean of Mean
Benefits Paid ($):
1) State Regular,
First Spell
2) Total, First
Spell
3) State Regular,
Benefit Year
4) Total, Benefit
Year
Weeks of Insured
2,267
2,558
2,487
2,786
27.5
33.8
27.0
33.1
2,074
2,329
2,328
2,592
26.7
32.9
26.3
32.2
2,159
2,446
2,426
2,725
27.4
33.8
27.0
33.8
Unemployment:
1) First Spell
2) Benefit Year
18.3
20.1
0.205
0.194
17.0
18.9
0.199
0.188
17.7
19.7
0.205
0.194
SE SE SE
Propor-
tion
of Pro-
portion
Propor-
tion
of Pro-
portion
Propor-
tion
of Pro-
portion
Proportion of
Claimants Who:
1) Exhausted
Benefits 0.478 0.008 0.446 0.008 0.464 0.008
2) Ended Benefits
within 11weeks 0.353 0.008 0.408 0.008 0.384 0.008
N 3,952 4,186 3,963
Sources: Tabulations from Illinois Department of Employment Security, Benefits Information System, and office logs.
Notes: "First Spell" refers to the spell of unemployment immediately following the initial claim for UI. "Total
Benefits Paid" refers to the sum of state regular benefits and Federal Supplemental Compensation. "Benefit Year"
refers to benefits paid or weeks of benefits paid during the full benefit year for each claimant. "Ended Benefits within
11Weeks" refers to termination of benefits within 11weeks of filing the initial claim (equivalently, 10 or fewer weeks
of benefit payments, because of the 1-weekwaiting period). Sample is the same as that underlying Table 2.
aSE denotes standard error.
Claimant Experiment group, and between trol group by somewhat over a week, again
the control group and the Employer Experi- measured over the full benefit year.
ment group. The differences are calculated The Claimant Experiment results are quite
from Table 3, and the standard error of each strong in that the $158 to $194 benefit redifference
is shown. The most striking results duction, and the 1.15-week reduction in the
shown by the table pertain to the Claimant duration of unemployment, were attained
Experiment: Average benefit receipt was on average over all eligible workers who
lower in the Claimant Experiment group than were assigned to the Claimant Experiment,
in the control group by $158 to $194 over whether or not they agreed to participate,
the full benefit year (depending on whether and whether or not they actually cashed a
Federal Supplemental Compensation benefits voucher for $500. Further, compared with
are included in benefits received). The dif- the control group, 5.5 percent more of those
ferences are statistically different from zero assigned to the Claimant Experiment ended
at the 1-percent level (using a two-tailed their spell of insured unemployment within
test). Further, the average number of weeks 11 weeks of filing, and 3.2 percent fewer
of insured unemployment was lower in the exhausted their UI benefits. These last findClaimant
Experiment group than in the con- ings suggest that the Claimant Experiment
521VOL. 77 NO. 4 WOODBURY AND SPIEGELMAN: UXEMPLOYMENT
TABLE4-DIFFERENCES BETWEENCONTROLGROUPAND EXPERIMENTALGROUPMEANS
Claimant Experiment Employer Experiment
minus Control minus Control
Difference Difference
of Means SE of Means SE
Benefits Paid ($):
1) State Regular,
First Spell
2) Total, First
Spell
3) State Regular,
Benefit Year
4) Total, Benefit
Year
Weeks of Insured
Unemployment:
1) First Spell
2) Benefit Year
Difference Difference
of of
Proportions SE Proportions SE
Proportion of
Claimants Who:
1) Exhausted
Benefits
2) Ended Benefits
within 11 Weeks
Source: Calculations based on Table 3. See Notes and fn. a, Table 3.
"Rejection of the hypothesis that the difference of means is zero using a two-tailed 1-percent significance level.
ejection of the hypothesis that the difference of means is zero using a two-tailed 5-percent significance level.
reduced the duration of insured unemploy- even in the first spell of unemployment. The
ment of workers throughout the distribution results suggest that, to the extent the Emof
unemployment spell lengths. ployer Experiment reduced the length of the
The results of the Employer Experiment initial spell of unemployment, this effect did
are quite different. Although there was an not persist over the full benefit year. Howinitial
reduction in benefits received by the ever, evidence presented in Section 111, Part
Employer Experiment group in the spell of C, shows that the Employer Experiment did
unemployment immediately following the reduce the benefits paid and the weeks of
initial claim, the reduction in benefits paid insured unemployment of white women over
to the Employer Experiment group over the the full benefit year. Hence, these overall
full benefit year is statistically insignificantly results mask an effect of the Employer Exdifferent
from zero. The evidence of an im- periment on at least one major group of UI
pact of the Employer Experiment on the claimants.
number of weeks of insured unemployment
is similar. Although there was a reduction in B. Efects on Earnings after
unemployment during the first spell, no sta- Reemployment
tistically significant difference between the
control group and the Employer Experiment The above results suggest strongly that the
group exists over the full benefit year. Claimant Experiment reduced the duration
In view of the comparatively low rate of of job search for claimants who participated
use of the Employer Experiment, it may in it. It is possible-indeed, Mortensen's and
seem surprising that there was an impact McCall's early models of job search suggest
522 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1987
TABLE5-MEAN PRE-AND POSTPROGRAM CLAIMANTSEARNINGS OF ELIGIBLE WITH EARNINGS
IN QUARTER TERMINATION, GROUPAFTER BENEFIT BY EXPERIMENTAL
Mean Earnings in: Control
Claimant
Experiment
Employer
Experiment
Base Period (Average
of Four Quarters)
Quarter before
Initial Claim
Quarter after Benefit
Termination
Notes: Standard error of each mean is shown in parentheses. The sample is constructed as follows: Starting with fully
eligible claimants who met the initial claim and age restrictions of the experiments, samples of those who showed
positive earnings in the quarter after they terminated benefits (2,531 controls, 2,786 Claimant Experiment enrollees,
and 2,550 Employer Experiment enrollees)were used to compute mean earnings in the base period, the quarter before
the initial claim, and the quarter after benefit termination. Note that all means are computed conditional on positive
earnings; thus, N used to compute mean earnings in the quarter before the initial claim is lower than elsewhere
because not all claimants in the sample showed earnings in the preclaim quarter.
-that the shorter search time induced by
the $500 bonus may result in a less-favorable
match between worker and job, which would
manifest itself in lower earnings in the subsequent
job. If a Claimant Experiment participant
who submitted a Notice of Hire (or
received a bonus) effectively lowered his or
her reservation wage and simply accepted
the first job that presented itself, then the
claimant's earnings after reemployment and
the efficiency of the labor market would both
be r e d ~ c e d . ~
Table 5 addresses the concern that Claimant
Experiment participants may have
sacrificed earnings in their postprogram job
in order to obtain the $500 bonus. The table
displays data on the pre- and postprogram
earnings of claimants in each of the three
groups. All figures are based on the subsample
of claimants who terminated benefits (at
some point following the initial claim that
brought them into the experiment), and had
positive earnings in the first full quarter following
benefit termination. That is, claimants
who exhausted benefits and failed to
find a new job, and claimants who dropped
'Kathleen Classen (1979) has shown that, in search
models that relax some of the assumptions of the early
Mortensen and McCall models, the relation between
unemployment duration and earnings after reemployment
is ambiguous. See also Kenneth Burdett (1979).
out of the labor force, are excluded from
consideration here. Since our concern focuses
on the earnings of those who found a new
job, and whether these earnings are lower for
the Claimant Experiment group, this is
clearly the appropriate group to examine.
The first row of Table 5 shows average
base period earnings of claimants in each of
the three groups, and the second row shows
earnings in the quarter before the initial
claim was filed. Note that there is no statistically
(or otherwise) significant difference
across groups in either of these preprogram
earnings measures.1° (The sample on whlch
earnings in the quarter before the initial
claim is based is smaller than the sample
used to calculate the other figures in the
table because not all claimants had earnings
in the quarter before they filed for benefits.)
The third row of Table 5 shows, for each
of the three groups, earnings in the first full
quarter after benefit termination (for those
claimants who had earnings after benefit
termination). The figures suggest strongly
1 ° ~ o rmean earnings in the base period, the standard
error of the difference between either experimental group
and the control group is about $63, which overwhelms
either of the mean differences ($17 and $23). The standard
error of the differences is about $72 for mean
earnings in the quarter before the initial claim, and
about $67 for mean earnings in the quarter after benefit
termination.
523VOL. 77 NO. 4 WOODBURY AND SPIEGELMAN: UNEMPLOYMENT
CONTROL AND EXPERIMENTALTABLE6-DIFFERENCEBETWEEN GROUP GROUPMEANSBY RACE AND SEX
Difference in 

Benefits Paid ($): 

Control vs. 

Claimant Employer
Race/Sex Experiment Experiment
White Women - 262.0a - 164.1b
(1,113; 1,170;1,166) (70.7) (70.7)
White Men - 185.1" - 9.8
(1,384; 1,553;1,399) (62.4) (63.6)
Black Women -65.9 -49.3
(527;512; 504) (104.7) (105.1)
Black Men -72.7 -33.2
(545;538;510) (115.6) (104.3)
Difference in 

Weeks of Benefits: 

Control vs. 

Claimant Employer 

Experiment Ewperiment 

- 1.623a - 1.008~
(0.503) (0.503)
-1.125~ +0.192
(0.444) (0.455)
-0.978 -0.539
(0.745) (0.748)
- 0.518 - 0.211
(0.730) (0.742)
Notes: A negative difference implies that the experimental mean is less than the control mean. Standard error of each
difference is in parentheses under the difference. Number of observations in the control, Claimant Experiment, and
Employer Experiment groups for each race and sex category is shown in parentheses in the first column. Sample sizes
for Hispanics, Native Americans, and others are too small to allow detection of experimental effects of the magnitude
found in the overall sample; hence, these groups are not considered separately.
a.b~eeTable 4.
that there is no difference between the postprogram
earnings of controls and of Claimant
Experiment workers-the average for
the controls is $3,121, whereas the average
for the Claimant Experiment group is $3,129.
The difference, $8, is swamped by the standard
error of that difference, which is $67.
We conclude that the relatively rapid reemployment
of Claimant Experiment participants
did not come at the expense of lower
earnings. Rather, the data are consistent with
the idea that the faster reemployment of
Claimant Experiment workers resulted from
more-intense job search efforts by Claimant
Experiment workers, and not from overly
rapid acceptance of job offers."
C. ESfects by Race and Sex
Those enrolled in the experiments compose
a diverse group, and it is possible that
certain groups of workers responded more
strongly than others to the experimental in"
~ ninterpretation within the framework of labor
supply is also possible. If a single wage offer (constant
over time) faced a worker, then increased search intensity
would be unnecessary in order to obtain a job at a
given wage. Participating claimants could then cut short
their spell of unemployment by substituting income for
leisure in the current period.
centives. We have explored elsewhere(1987a,
ch. 6) the effects of the experiments on several
different categories of workers and we believe
one of these breakdowns to be particularly
significant.
Table 6 displays the effects of the experiments
on 1) state regular benefits paid to
claimants during their full benefit year, and
2) the number of weeks of benefits paid to
claimants during the full benefit year, broken
down by four race and sexcategories.12These
results are dramatic because they show that.
for white women, the Employer Experiment
unambiguously reduced UI benefit payments
and weeks of insured unemployment. Moreover,
we have found that no loss of earnings
accompanied the response of whlte women
and their employers to the Employer Experiment
(1987a, ch. 6, Table 6-9). White
women stand out sharply from the other
three race and sex categories, each of whom
showed a response to the Employer Experiment
that is not statistically significantly different
from zero.
Table 6 also shows that the effects of the
Claimant Experiment appear to have varied
12Raceand sex subgroups other than the four shown
are too small to allow detection of experimental effects
of the magnitude found in the overall sample.
524 THEAMERICAN ECONOMIC REVIEW SEPTEMBER 1987
by race. The Claimant Experiment brought
about a larger reduction in the UI benefits
and weeks of insured unemployment of both
white women and white men than of black
women and black men.13 Indeed, the effect
of the Claimant Experiment on the insured
unemployment of blacks is not statistically
significantlydifferent from zero.
Some insight into the differences by race
and sex in the effects of the experiments can
be obtained from Table 7. The first panel
shows the proportion of each race and sex
group in each experiment that received a
bonus. Here we see that whites were nearly
three times more likely to receive a Claimant
Experiment bonus than were blacks. It is
also shown that whites' employers were
roughly ten times more likely to receive
an Employer Experiment bonus than were
blacks' employers.
We can decompose the differencebetween
any two groups in their propensities to receive
a bonus into two components. One
component reflects the difference between
the groups in their probabilitiesof qualifying
for a bonus (or in their probabilities of
qualifying an employer for a bonus), and the
other reflects the difference in the probabilities
that qualifying workers (or their employers)
will actuallycash a voucher for $500.
Define b,, as the proportion of race and sex
group i (i = white women, white men, black
women, and black men) in treatment group t
( t = control, Claimant Experiment, and Employer
Experiment) who actually receive a
bonus. Also, define e,, as the proportion of
all claimants in group i, t who qualify for a
bonus (because they find a job within eleven
weeks and hold it for four months), and r,,
as the proportion of qualifying claimants in
group i, t who actually receive a bonus (or
whose employers do). (We will call this conditional
probability r,, the "take-up rate"
because it reflects the proportion of qualifying
claimants who use the program.) We can
then express the proportion of a group who
"A case could be made that black women showed a
stronger response to the Claimant Experiment treatment
than did black men.
receive a bonus (b,,) as the product of the
proportion qualifying and the take-up rate:
bit= eitril.
Since we are interested in differences between
groups in their propensities to receive
a bonus, it is natural to write the proportional
difference between two groups as
Taking the natural logarithm of both sides
and rearranging gives
This is the desired decomposition. It attributes
differences between two groups' propensities
to receive a bonus to differences
between the two groups in (a) their probabilities
of qualifying for a bonus, and (b) their
probabilities of cashing a voucher if they
qualify.
The data required to compute any such
decomposition are displayed in Table 7. The
second panel shows the proportion of each
race and sex group in each experimental
group that (by finding a job within 11weeks
and holding the job for four months)
qualified for a bonus. The third panel shows
the take-up rate of each group. If we were
interested in decomposing the difference between
white and black women's probabilities
of receiving a Claimant Experiment bonus,
we would compute:
(Subscripts wf and bf denotewhite and black
females, and subscript a denotes the Claimant
Experiment.)The decomposition suggests
that 62 percent of the difference between
white and black women in their propensities
to receive Claimant Experiment bonuses can
be attributed to the higher probability that
white women qualified for a bonus, and that
38 percent of the difference can be attributed
525VOL. 77 NO. 4 WOODBURYAND SPIEGELMAN: UNEMPLOYMENT
TABLE7-PROPORTION OF CLAIMANTS FOR A BONUS,WHO RECEIVEDA BONUS,QUALIFIED
TAKE-UPRATE,AND EXPERIMENTAL AND BENEFITS,COSTS BY RACEAND SEX
Claimant Employer Claimant Employer
Control Experiment Experiment Control Experiment Experiment
1) Proportion Receiving a Bonus: 4) Bonus Cost per Assigned
Claimant ($):
Total - 0.1362
(0.0053) Total White
Women - 0.1761
(0.0111) White Women White
Men - 0.1700
(0.0095) White Men Black
Women - 0.0625
(0.0107) Black Women Black
Men - 0.0632
(0.0105) Black Men -
2) Proportion Qualifying for a Bonus:
5) Ratio of Benefit Payment
Total 0.207 Reduction to Bonus Cost
(0.006) per Assigned Claimant: 

White Women 0.237 

(0.013) Total -

White Men 0.244 

(0.012) White Women -

Black Women 0.139 

(0.015) White Men -

Black Men 0.127 

(0.014) Black Women -
3) Take-up Rate: Black Men Total
- 0.5448 

(0.0077) 6) Size of Group ( N): 

White Women - 0.5870 

(0.0263) ~ o t a l ~ 3,952 

White Men - 0.6139 White Women 1,113 

(0.0235) White Men 1,384 

Black Women - 0.3951 Black Women 527 

(0.0543) Black Men 545 

Black Men - 0.3821 

(0.0515) 

Note: Standard errors in parentheses. The proportion of each group qualifying for a bonus (second panel) is the
proportion of the group that found a job within 11 weeks of filing the initial claim and held the job for four months.
A group's take-up rate (third panel) is the proportion of the group receiving a bonus divided by the proportion
qualifying for a bonus. A group's bonus cost per assigned claimant (fourth panel) equals total bonus payments to the
group divided by the size of the group (N). The ratio of benefit payment reduction to bonus cost per assigned
claimant (fifth panel) states the reduction in UI benefit payments for each dollar of bonus paid. The standard error of
each ratio in the fifth panel is approximated by taking a Taylor expansion for the ratio.
aRejection of the hypothesis that the differenceof proportions between the experimental and control groups is zero
using a two-tailed 5-percent significance level.
he sum of white women, white men, black women, and black men is less than the total because the total
includes Hispanics, Native Americans, and others.
- - - - - ---
THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1987
Percentage of Difference 

Attributable to: 

Difference in Difference
Proportions in Proportions Difference in
Comparison Receiving Bonus Qualifying Take-up Rates
Women in Claimant
Experiment with
Men in Claimant
Experiment
Whites in Claimant
Experiment with
Blacks in Claimant
Experiment
Women in Employer
Experiment with
Men in Employer
Experiment
Whites in Employer
Experiment with
Blacks in Employer
Experiment
Women in Claimant
Experiment with
Women in Employer
Experiment
Men in Claimant
Experiment with
Men in Employer
Experiment
Whites in Claimant
Experiment with
Whites in Employer
Experiment
Blacks in Claimant
Experiment with
Blacks in Employer
Experiment
Note: Standard error of the difference in proportions receiving a bonus is shown in
parentheses below the difference. Equation (2) in the text defines the decomposition
displayed in the table.
"The difference in proportions receiving a bonus is statistically insignificant; therefore,
no decomposition is shown.
to the higher probability that qualifying play comparisons of the same group between
white women actually cashed a $500voucher. experiments.
Some further decompositions are shown in The differences between women and men
Table 8. To keep Table 8 of reasonable in bonus receipt within each experiment are
length only four groups-all women, all men, statistically insignificant and hence no deall
whites, and all blacks-are used in the compositions are shown (first and third rows
comparisons (rather than the less aggregate of the table). But the differences between
race and sex categories shown in Table 7). whites and blacks within experiment are
The top four rows of Table 8 display com- more revealing (second and fourth rows).
parisons of different groups within each ex- These decompositions show that differences
periment, whereas the bottom four rows dis- between whltes and blacks in bonus receipt
VOL. 77 NO. 4 WOODBURY AND SPIEGELMAN: UNEMPLOYMENT 527
can be attributed to white-black differences
both in the probabilities of qualifying for a
bonus and in take-up rates. In the Claimant
Experiment, 58 percent of the higher propensity
of whites to receive a bonus can be
attributed to whites' higher probability of
qualifying, and 42 percent to whites' higher
take-up rate. In the Employer Experiment,
only 29 percent of the higher propensity of
whites to receive a bonus can be attributed
to whltes' higher probability of qualifying,
and 71 percent to the higher take-up rate of
whites and their employers.
Differences between the Claimant Experiment
and the Employer Experiment in bonus
receipt are decomposed in the bottom four
rows of Table 8. The decompositions show
clearly that the difference between the experiments
in bonus receipt must be attributed
mainly to higher take-up rates in the
Claimant Experiment.
IV. Benefit-CostRatios and Take-upRates
The fourth and fifth panels of Table 7
display information on the costs and benefits
of the experiments to the state. The fourth
panel shows the bonus cost per assigned
claimant for each experimental and race-sex
group. For example, the figure $68.10 for
bonus cost per assigned claimant in the
Claimant Experiment equals the dollar
amount paid in Claimant Experiment
bonuses ($500 times 570 bonuses) divided by
the number of claimants assigned to the
experiment (4,186). Since the Claimant Experiment
was more heavily used than the
Employer Experiment, its costs per assigned
claimant were higher.
The cost figures displayed in the fourth
panel of Table 7 can be used to derive a
benefit-cost ratio for each experimental and
race-sex group. These ratios are shown in the
fifth panel. They are obtained by dividing
the reduction in UI benefit payments per
claimant (that is, the average treatment responses
shown in Tables 4 and 6) by the
bonus cost per assigned claimant from the
fourth panel of Table 7. For example, the
benefit-cost ratio for the Claimant Experiment,
2.32, equals the average treatment response
($158) divided by the bonus cost per
assigned claimant ($68). Ths benefit-cost
ratio is statistically significantly different
from zero. A straightforward interpretation
of this benefit-cost ratio is that, for the
Claimant Experiment overall, state regular
benefit payments were reduced by $2.32 for
each $1.00 of bonus payments made. It follows
that a program modeled on the Claimant
Experiment would be extremely attractive
from the state's point of view if the
presence of the program did not increase
unemployment among workers who were not
participants in the program.
The fifth panel also shows that the overall
benefit-cost ratio for the Employer Experiment
is 4.29, but it is not statistically different
from zero. The benefit-cost ratio for whte
women in the Employer Experiment, however,
is 7.07, and is statistically different
from zero. Hence, a program modeled on the
Employer Experiment also might be attractive
from the state's point of view if the
program did not increase unemployment
among nonparticipants. Since, however, the
Employer Experiment affected only white
women, it would be essential to understand
the reasons for the uneven effects of the
treatment on different groups of workers before
drawing conclusions about the efficacy
of such a program.
It is important to consider these benefitcost
ratios in relation to the take-up rates
developed in the previous section and displayed
in the third panel of Table 7. Those
figures show that only 54 percent of the
Claimant Experiment workers who qualified
for a $500 bonus actually took the steps to
claim the bonus.. It is also striking that only
12 percent of the employers who could have
received a $500 bonus simply by claiming it
actually did so.
If a program modeled on either the Claimant
or Employer Experiment were implemented,
it is possible that claimants' or employers'
take-up rates would be substantially
higher than those observed in the experiments.
Futher, if increased take-up rates were
unaccompanied by additional reductions in
UI benefits and unemployment, then the
benefit-cost ratio would decline. For example,
in the Claimant Experiment, if 100 percent
of the qualifying claimants had claimed
528 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1987
a bonus, then the benefit-cost ratio would
fall from 2.32 to 1.26. In the Employer Experiment,
if 100 percent of qualifying employers
had claimed a bonus, then the benefit-cost
ratio would be only 0.53. (For
employers of white women in the Employer
Experiment, a 100percent take-up rate would
lead to a decrease in the benefit-cost ratio
from 4.29 to 1.17.) Again, these calculations
assume an increase in the take-up rate
without any accompanying behavioral response.
How take-up rates would change in
an actual program, and whether there would
be further reductions in UI benefit payments
and unemployment, are important topics for
further research and experimentation.
V. Conclusion
The results of the Claimant Experiment
are unequivocal and strong. The incentive
created by the $500 bonus, which was actually
paid to only 570 out of 4,186 UI claimants
assigned to the experiment, reduced
state regular benefits paid to the randomly
selected treatment group by an average of
$158, and reduced average weeks of insured
unemployment by more than one week (over
the full benefit year), compared with the
randomly selected control group. We reemphasize
that these reductions in average benefit
payments and weeks of unemployment
were achieved over all 4,186 eligible claimants
in the Claimant Experiment sample,
whether or not they agreed to participate or
acted on the incentive.
Some of the results of the Employer Experiment
are also unequivocal and strong.
White women who were randomly assigned
to the Employer Experiment received $164
less in UI benefits, and experienced one week
less of insured employment (over the full
benefit year), than did white women who
were randomly assigned to the control group.
These reductions were achieved on average
over all 1,166 white women who were assigned
to the Employer Experiment. Only 54
of these women were responsible for a $500
bonus being received by the employer who
hired them. White women are the only race
and sex group who experienced a statistically
significant effect of the Employer Ex-
periment.
It is clear that the members of the experimental
treatment groups experienced less
unemployment than the randomly selected
control group. Thus, we may unambiguously
reject the hypothesis that the unemployment
insurance benefit system is only a benign or
nondistortionary income transfer. The results
reported here are the first experimental
demonstration of this proposition, and we
believe them to be quite convincing.
It is also clear that the members of the
experimental groups were, on average, made
no worse off by their assignment to the
experiments. On the other hand, it is not
possible to be certain whether the control
group's experience was identical to what it
would have been in the absence of any experimental
treatments. This is a problem
shared by all controlled social experiments.
Moreover, we have not addressed whether a
full-scale program modeled on either the
Claimant or Employer Experiment would
have significant displacement effects-that
is, would result in improvements for program
participants at the expense of nonparticipants.
Hence, we cannot conclude unambiguously
what the net social benefits of
such a program might be.
Another limitation of our results stems
from our finding that many claimants, and
most employers, failed to take the steps to
obtain a $500bonus even when they qualified
to do so. If the take-up rate in the Claimant
Experiment had been 100 percent (that is, if
all claimants who qualified for a bonus had
claimed one), and UI benefit reductions had
remained unchanged, then the benefit-cost
ratio of the experiment would have fallen to
1.26 from 2.32. If the take-up rate in the
Employer Experiment had been 100 percent,
then the benefit-cost ratio would have fallen
to 0.53 from 4.29. (For white women in the
Employer Experiment, the benefit-cost ratio
would have fallen to 1.17 from 4.29.) Thus,
even with a 100 percent take-up rate, the
Claimant Experiment would have reduced
UI benefit costs by considerably more than
the bonus costs incurred; however, the
Employer Experiment (overall) would not.
Because the take-up rates observed in the
experiments could understate those that
would occur in a program modeled on one
of the experiments, these findings pose im-
VOL. 77 NO. 4 WOODBURY AND SPIEGELMAN: UNEMPLOYMENT 529
portant questions for further research
and experimentation: Would take-up rates
change in an actual program, and if so, how?
To what degree would changes in take-up
rates be accompanied by further reductions
in UI benefit payments and unemployment?
Our analysis of the experimental data does
not allow us to determine conclusively why
the experimental treatments changed the
participants' behavior. We do have three
clear findings, however, that may be of considerable
significance for evaluating alternative
models of worker behavior: 1) Offering
unemployed workers the option of assigning
bonuses to employers who hired them had a
far smaller effect on employment than offering
unemployed workers the same bonus
directly. 2) The lower unemployment of
members of the Claimant Experiment group
occurred for workers at all parts of the distribution
of unemployment spell lengths. 3)
The earnings of members of the treatment
groups after they became reemployed did
not differ from the earnings of members of
the control group. It seems clear that additional
analysis of these data will be needed
before further conclusions can be reached,
particularly conclusions about whlch of various
models of worker and firm behavior are
supported.
Because such strong inferences can be derived
from experimental work, the desirability
of further experimentation seems unambiguous.
A significant improvement in the
design of the Illinois experiments over the
design of the income maintenance experiments
of the 1960's and 1970's is the reliance
on administrative records rather than surveys
for the collection of information about the
treatment and control groups.14 Use of administrative
records is an important development
in field experimentation in economics.
Experiments that rely solely on survey records
are vastly more expensive than those
that rely on administrative data. Moreover,
141n our earlier article (1987b, Sec. LC), we discuss
the advantages of administrative data, particularly in
avoiding the problems associated with selective attrition.
For analyses of attrition in the income maintenance
experiments, see Harold Watts et al. (1977), and
Philip Robins and Richard West (1980).
experiments that rely on survey data to measure
responses run the risk of confounding
measures of the experimental treatment with
measures of attrition from the survey, as
Ashenfelter and Mark Plant (1987) have recently
noted.
Although the Claimant and Employer Experiments
in Illinois have yielded strong results
about the effects of experimental
bonuses to workers and employers, additional
research and experimentation along
the lines of these experiments could usefully
address three different and important issues.
First, it would be desirable to measure the
response to variation in (a) the size of the
bonus provided to successfuljob seekers and
to employers, and (b) the length of the period
within which a UI claimant must become
reemployed in order to qualify for a bonus.
Second, it is important to learn how a longer
and more widespread experimental bonus
program would influence employee and employer
take-up rates as well as UI benefit
reductions. Third, analysis and experimentation
that would shed light on the extent to
whlch differences between treatment and
control groups in unemployment experience
represent costs to the control group (or other
nonparticipants) would be of special importance.
Further experiments hold out the possibility
of designing new programs that may
effectively reduce unemployment at low or
even negligible cost to unemployed workers
and to society.
REFERENCES
Ashenfelter, Orley, (1978a) "Unemployment
as a Constraint on Labour Market Behaviour,"
in M. J. Artis and A. R. Nobay,
eds., Contemporary Economic Analysis,
London: Croom Helm, 1978,149-81.
-, (1978b) "Evaluating the Effects of
the Employment Tax Credit," in U. S.
Department of Labor, Office of the Assistant
Secretary for Policy, Evaluation and
Research, ConferenceReport on Evaluating
the 1977 Economic Stimulus Package,
Washington: USGPO, 1978.
and Plant, Mark W., "Non-Parametric
Estimates of the Labor Supply Effects of
Negative Income Tax Programs," unpub-
530 THE AMERICAN EC:ONOMIC REVIEW SEPTEMBER 1987
lished paper, March 1987.
Bishop, John H., Subsidizing the Hiring and
Training of the Disadvantaged, Kalamazoo:
W . E. Upjohn Institute for Employment
Research, 1987.
Burdett, Kenneth, "Unemployment Insurance
Payments as a Search Subsidy: A Theoretical
Analysis," Economic Inquiry, July
1979, 17, 333-43.
Classen, Kathleen P., "Unemployment Insurance
and Job Search," in S. A. Lippman
and J. J. McCall, eds., Studies in the
Economics of Search, Amsterdam: NorthHolland,
1979, 191-219.
Coase, Ronald H., "The Problem of Social
Cost," Journal of Law and Economics, October
1960, 3, 1-44.
Kingston, Jeny L., Burgess, Paul L. and St.
Louis, Robert D., "Unemployment Insurance
Overpayments: Evidence and Implications,"
Industrial and Labor Relations
Review, April 1986, 39, 323-36.
McCall, J. J., "Economics of Information and
Job Search," Quarterly Journal of Economics,
February 1970, 84, 113-26.
Mortensen, Dale T., "Job Search, the Duration
of Unemployment, and the Phllips
Curve," American Economic Review, December
1970, 60, 847-62.
Robins, Philip K. and West, Richard W., "Program
Participation and Labor-Supply Response,"
Journal of Human Resources, Fall
1980, 15, 499-523.
St. Louis, Robert D., Burgess, Paul L. and
Kingston, Jerry L., "Reported vs. Actual
Job Search by Unemployment Insurance
Claimants," Journal of Human Resources,
Winter 1986, 21, 92-117.
Spiegelman, Robert G. and Woodbury, Stephen
A., (1987a) The Illinois Unemployment Insurance
Incentive Experiments, Final Report
to the Illinois Department of Employment
Security, Kalamazoo: W. E.
Upjohn Institute, February 1987.
and -, (1987b) "Controlled Experiments
and the Unemployment Insurance
System," in W. Lee Hansen and
James F. Byers, eds., Unemployment Insurance:
The Second Half-Century, Madison:
University of Wisconsin Press, forthcoming
1987.
Watts, Harold W., Peck, Jon K. and Taussig,
Michael, "Site Selection, Representativeness
of the Sample, and Possible Attrition
Bias," in Harold W. Watts and Albert
Rees, eds., The New Jersey Income-Maintenance
Experiment, Volume 111: Expenditures,
Health, and Social Behavior; and
the Quality of the Evidence, New York:
Academic Press, 1977, 441-66.
You have printed the following article:
Bonuses to Workers and Employers to Reduce Unemployment: Randomized Trials in
Illinois
Stephen A. Woodbury; Robert G. Spiegelman
The American Economic Review, Vol. 77, No. 4. (Sep., 1987), pp. 513-530.
Stable URL:
http://links.jstor.org/sici?sici=0002-8282%28198709%2977%3A4%3C513%3ABTWAET%3E2.0.CO%3B2-R
This article references the following linked citations. If you are trying to access articles from an
off-campus location, you may be required to first logon via your library web site to access JSTOR. Please
visit your library's website or contact a librarian to learn about options for remote access to JSTOR.
[Footnotes]
8
The Problem of Social Cost
R. H. Coase
Journal of Law and Economics, Vol. 3. (Oct., 1960), pp. 1-44.
Stable URL:
http://links.jstor.org/sici?sici=0022-2186%28196010%293%3C1%3ATPOSC%3E2.0.CO%3B2-F
14
Program Participation and Labor-Supply Response
Philip K. Robins; Richard W. West
The Journal of Human Resources, Vol. 15, No. 4, The Seattle and Denver Income Maintenance
Experiments. (Autumn, 1980), pp. 499-523.
Stable URL:
http://links.jstor.org/sici?sici=0022-166X%28198023%2915%3A4%3C499%3APPALR%3E2.0.CO%3B2-L
References
The Problem of Social Cost
R. H. Coase
Journal of Law and Economics, Vol. 3. (Oct., 1960), pp. 1-44.
Stable URL:
http://links.jstor.org/sici?sici=0022-2186%28196010%293%3C1%3ATPOSC%3E2.0.CO%3B2-F
http://www.jstor.org
LINKED CITATIONS
- Page 1 of 2 NOTE:
The reference numbering from the original has been maintained in this citation list.
Unemployment Insurance Overpayments: Evidence and Implications
Jerry L. Kingston; Paul L. Burgess; Robert D. St. Louis
Industrial and Labor Relations Review, Vol. 39, No. 3. (Apr., 1986), pp. 323-336.
Stable URL:
http://links.jstor.org/sici?sici=0019-7939%28198604%2939%3A3%3C323%3AUIOEAI%3E2.0.CO%3B2-A
Economics of Information and Job Search
J. J. McCall
The Quarterly Journal of Economics, Vol. 84, No. 1. (Feb., 1970), pp. 113-126.
Stable URL:
http://links.jstor.org/sici?sici=0033-5533%28197002%2984%3A1%3C113%3AEOIAJS%3E2.0.CO%3B2-4
Job Search, the Duration of Unemployment, and the Phillips Curve
Dale T. Mortensen
The American Economic Review, Vol. 60, No. 5. (Dec., 1970), pp. 847-862.
Stable URL:
http://links.jstor.org/sici?sici=0002-8282%28197012%2960%3A5%3C847%3AJSTDOU%3E2.0.CO%3B2-1
Program Participation and Labor-Supply Response
Philip K. Robins; Richard W. West
The Journal of Human Resources, Vol. 15, No. 4, The Seattle and Denver Income Maintenance
Experiments. (Autumn, 1980), pp. 499-523.
Stable URL:
http://links.jstor.org/sici?sici=0022-166X%28198023%2915%3A4%3C499%3APPALR%3E2.0.CO%3B2-L
Reported VS. Actual Job Search by Unemployment Insurance Claimants
Robert D. St. Louis; Paul L. Burgess; Jerry L. Kingston
The Journal of Human Resources, Vol. 21, No. 1. (Winter, 1986), pp. 92-117.
Stable URL:
http://links.jstor.org/sici?sici=0022-166X%28198624%2921%3A1%3C92%3ARVAJSB%3E2.0.CO%3B2-2
http://www.jstor.org
LINKED CITATIONS
- Page 2 of 2 NOTE:
The reference numbering from the original has been maintained in this citation list.