Full text of Working Papers (Federal Reserve Bank of Chicago) : Estimating the Returns to Community College Schooling for Displaced Workers, Working Paper 2002-31

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Federal Reserve Bank of Chicago

Estimating the Returns to Community
College Schooling for Displaced
Workers
Louis Jacobson, Robert LaLonde and
Daniel G. Sullivan

WP 2002-31

Estimating the Returns to Community College
Schooling for Displaced Workers

Louis Jacobson
Westat Inc.
Robert LaLonde
University of Chicago and NBER
Daniel G. Sullivan
Federal Reserve Bank of Chicago

December 2002

The authors thank Jeff Jaksich and the Washington State Training and Education Coordinating Board for providing
the data for this project and answering many questions about its construction. We have benefited from comments by
John Ham, James Heckman, Guido Imbens, Alona Lerman, Kevin Hollenbeck and participants at the HSFR
Conference on Unemployment and Labor Market Policy, in Sweden, and workshops at Chicago, Harvard, Princeton,
Northwestern, Michigan, Michigan State, the National Bureau of Economic Research, UCLA, Maryland, Syracuse,
Bureau of Labor Statistics, the Upjohn Institute for Employment Research, CIDE, and the Federal Reserve Bank of
Chicago. Ken Housinger, Jim Sullivan, and Abigail Waggoner provided technical assistance. Some of this research
was funded under U.S. Department of Labor ETA contract 99-0-0584-75-055-01 and by the Washington State
Workforce Training and Education Coordinating Board. The views expressed in this paper are solely those of the
authors and do not necessarily represent the views of the state of Washington, the U.S. Department of Labor, the
Federal Reserve Bank of Chicago or the Federal Reserve System.

Abstract
Studies show that high-tenure displaced workers typically incur substantial long-term earnings
losses. As these losses have become increasingly apparent, policy makers have significantly
expanded resources for retraining, much of which takes place in regular community college
classes. To analyze the effectiveness of such training, we link administrative earnings records
with the community college transcript records of workers displaced from jobs during the first half
of the 1990s in Washington State. We explore several issues of statistical specification for regression models quantifying the impact of community college credits on earnings. These include (i)
the need to allow for a transition period immediately after the end of workers’ schooling when
their earnings may be temporarily depressed, (ii) whether earnings gains are strictly proportional
to credits earned, and (iii) how to model worker-specific unobserved heterogeneity. In our preferred specification, we find that the equivalent of an academic year of community college
schooling raises the long-term earnings of displaced workers by an average of about 9 percent for
men and about 13 percent for women. However, these average returns mask substantial variation
in the returns associated with different types of courses. On the one hand, we estimate that an academic year of more technically oriented vocational and academic math and science courses raise
earnings by about 14 percent for men and 29 percent for women. On the other hand, we estimate
that less technically oriented courses yield very low and possibly zero returns. About one third of
the increase in earnings associated with more technically oriented vocational and academic math
and science courses is estimated to be due to increases in wage rates, with the remainder attributable to increased hours of work.
Louis S. Jacobson
Westat Inc.
1650 Research Road
Rockville, MD 20850-3129
Robert J. LaLonde
Irving B. Harris Graduate School of Public Policy Studies
The University of Chicago
1155 East 60th Street
Chicago, IL 60637
Daniel G. Sullivan
Economic Research Department
Federal Reserve Bank of Chicago
230 S. LaSalle St.
Chicago, IL 60604

0

I. Introduction
Layoffs of experienced workers are an enduring aspect of U.S. labor markets (Farber, 1993, 1997;
Aaronson and Sullivan, 1998; Helwig, 2001). Studies using longitudinal data indicate that
affected workers tend to incur substantial long-term earnings losses long after they return to work
(Ruhm, 1991; Jacobson, LaLonde, and Sullivan, 1993a, 1993b, 1993c; Fallick, 1996; Kletzer
1998). Given this prospect of significantly lower earnings, it is natural for displaced workers to
consider returning to school to obtain new skills that would allow them to find better jobs. Indeed,
policy makers often propose subsidizing retraining as a way to cushion the blow associated with
job displacement. As the costs of worker displacement became more apparent in the 1990s, and
the incidence of displacement became more wide spread among industrial sectors, policy makers
substantially expanded expenditures on employment and training services for displaced workers,
even as they scaled back expenditures on similar services for youths (LaLonde 2002). Much of
this training takes place in regular courses taught in the nation’s community colleges.
Policy makers’ heightened support for worker retraining has occurred even though there is relatively little evidence on its likely impacts. The many studies of public sector-sponsored training
for economically disadvantaged persons are not directly relevant because most displaced workers
have substantial work experience and thus are likely to be more employable without retraining
than the workers in those studies. Neither does the voluminous literature on the returns to schooling necessarily apply because displaced workers usually have been out of school for many years,
often take only a few courses, tend to be work more hours for pay while attending school, and are
usually much older than the typical student. Such factors may make it more difficult to learn new
skills.
Much of the knowledge we do have of the effects of training displaced workers comes from several demonstration programs implemented during the 1980s (Bloom, 1990; Leigh, 1990). A more
recent study examines the impact of training on a relatively small subset of displaced workers
who lost their jobs as a result of increased import competition (Corson, Decker, Gleason, and
Nicholson, 1993). These studies indicate that displaced workers benefit from job search assistance; that access to classroom vocational instruction or on-the-job training usually has little
effect on subsequent earnings; and that female displaced workers likely benefit more from
employment and training services than do their male counterparts (LaLonde, 1995).
This paper presents new evidence on the benefits of retraining prime-aged adults by analyzing the
impact of community college schooling on displaced workers’ earnings in Washington State during the 1990s. We focus on this group of workers because of our ability to construct a large longi-

1

tudinal database well suited to the study of training impacts. We constructed this database by
merging administrative earnings and unemployment insurance records with community college
transcript records. The administrative earnings records cover long periods both before workers’
displacements and after their schooling, while the transcript records provide unusually detailed
information about the type, intensity, and duration of classroom training.
We explore several issues in the specification of longitudinal regression models that quantify the
impacts of community college credits on earnings and other labor market outcomes. First,
because it apparently takes time to realize the full benefits of training, we find that it is important
to allow for the possibility that workers’ earnings may be temporarily depressed immediately
after they leave community college. Accounting for such a transition substantially increases estimates of long-term earnings gains.
Second, we find that it is important to allow for departures from strict proportionality between
earnings increases and credits earned, an assumption that has been imposed, for example, by Kane
and Rouse (1995) in a study of the returns to community college. Indeed, in some specifications
that relax proportionality, earning even a single credit is estimated to significantly increase expectations of subsequent earnings. The earnings increase associated with this first credit is many
times greater than that associated with additional credits. It also is probably higher than can be
plausibly attributed to factors such as access to wider networks for job search that may be provided by the community college environment. We, therefore, conjecture that the higher earnings
associated with merely “showing up” at community college are due mainly to the effects of nonrandom selection into training. Thus we focus our attention on the per-credit earnings increases
associated with completing additional credits. In most specifications, relaxing proportionality
lowers somewhat the estimated impact of community college credits on earnings.
Finally, because we have longitudinal data, we can estimate the gains from schooling in models
that allow for forms of unobserved heterogeneity that may be correlated with training decisions.
We find that the standard fixed-effect specification, which has been employed to estimate the
returns to schooling by Angrist and Newey (1991), is inadequate for our purposes because workers who earn many credits, especially many credits in nontechnical subjects, tend to have more
slowly growing earnings in the period before training than one would expect given their observable characteristics. Thus, we also estimate models that include worker-specific time trends in
addition to fixed effects. These models effectively control for any unobserved worker characteristics that are changing at a constant rate over time. Allowing for this richer form of worker-specific
heterogeneity raises estimates of the effects of training, especially the effects of nontechnical
training.

2

In our preferred specification, which allows for (i) post-schooling transition effects, (ii) deviations
from strict proportionality, and (iii) individual-specific time trends, we find that the equivalent of
a year of community college schooling increases displaced workers’ earnings by about 9 percent
for men and 13 percent for women.1 These gains from schooling compare favorably to those
reported for younger persons in the returns to schooling literature. We also find that returns differ
substantially by type of course. Credits completed in courses teaching quantitative or more technically oriented vocational subject matter generate earnings gains that average 14 percent for men
and 29 percent for women. By contrast, courses teaching non-quantitative or non-technical vocational skills are associated with small or possibly zero earnings gains.
The remainder of the paper proceeds as follows: Section II provides institutional background on
public sector-sponsored retraining programs for displaced workers and describes the data. Section
III presents our econometric model for estimating the effects of community college schooling. In
section IV, we present our estimates and some concluding remarks follow in Section V.

II. Training for Displaced Workers
The Role of Community Colleges
Community colleges have played an increasingly prominent role in worker retraining policy during the last 30 years. This change coincides with the greater emphasis that these institutions have
given to vocational training. Although community colleges continue to offer academically
inclined students traditional college-level courses, they now also offer vocationally oriented
courses that in the past were offered mainly by proprietary schools and vocational institutes (Freeman, 1974; Grubb, 1993; Kane and Rouse, 1999.) Typical course offerings cover areas as diverse
as computer information systems, food preparation and management, real estate, word processing, respiratory therapy, the construction trades, and automobile repair. Moreover, students who
complete these kinds of courses can often obtain certification in a particular trade or take state
licensing exams.
Given the existence of these vocational programs, community colleges are natural providers of
public sector-sponsored retraining. Indeed, several federal programs have provided funding to
support displaced workers’ community college attendance. These programs include those funded
1. Relatively few displaced workers attend as much as an academic year of community college. Nevertheless, computing the value of a year’s worth of credits provides a convenient benchmark to compare our
results to standard estimates from the returns to schooling literature.

3

under the Trade Adjustment Assistance Act (TAA), the 1972 Higher Education Amendments
(Pell Grants), and Title I of the Workforce Investment Act (WIA), which in 1998 replaced the
Economically Displaced Worker Adjustment Act (EDWAA). Under WIA, the federal government allocates funds to state and local authorities. They, in turn, provide displaced workers with
Individual Training Accounts, which workers can use to obtain training from certified providers.
In practice these funds are frequently spent on regular two-year community colleges courses. In
these mainstream programs, displaced workers take classes with non-displaced workers and fulltime students.2 It is the impact of these regular community college courses on displaced workers’
earnings that we study in this paper.
Data Sources
We analyze the earnings histories of workers who had three or more years of job tenure when they
were permanently laid off from their jobs between 1990 and 1994 in Washington State and who
remained attached to the state’s labor force from 1987 through 1995. Of the approximately 97,000
workers in our sample, about 16,000 earned some community college credits by 1996. We
excluded “low-tenure” displaced workers, because public policy has been most concerned about
the long-term consequences of the job losses of experienced workers. We exclude workers with
low labor market attachment because for such workers it is difficult to distinguish between true
unemployment and other reasons why administrative data show no earnings. The latter include
the possibilities that workers have moved out of state, that they are working in the uncovered or
illegal sectors, and that they are working under an incorrect social security number.
The administrative records of the state’s unemployment insurance system provide information on
individuals’ quarterly earnings and hours worked from 1987 to 1995, their birth year, race and
sex, their quarter of separation, job tenure and level of education at separation, and, for each calendar year, their primary employer's 4-digit SIC code, geographic location, and number of
employees. Community college records provide information on students’ demographic characteristics, listings of the credit and noncredit courses in which they enrolled, and the grades that they
received in courses taken for credit. Together they provide a rich data base for assessing the
effects of retraining on displaced workers’ earnings. The Appendix provides further details about
the construction of the sample.

2. About one-third of community college students in the United States are over 30, and the vast majority
work at least part-time (Kane and Rouse, 1999).

4

Characteristics of Displaced Workers
Like much of the nation, Washington State’s economy worsened in 1990. The unemployment rate
rose, with several sectors especially hard hit, including the aerospace and wood products industries. In response, the state legislature established a program in 1993 that provided community
colleges with additional funding aimed at boosting displaced worker enrollment. But because the
substantial increase in federal subsidies for retraining displaced workers had not yet taken place,
most of the state’s unemployed workers attended community college at their own expense. Thus,
the workers that we study should be reasonably typical of those displaced in a relatively weak
labor market and who decide on their own to return to school for more training.
Displaced workers who enroll in community college classes typically are prime-aged adults,
many of whom have attended college before, and who have been displaced from jobs from a wide
array of industries. As shown by Panel A of Table 1, the average age of displaced workers in the
Washington State who enrolled and completed community college courses was approximately 36
for males and 37 for females. Approximately one-half of both the male and female community
college participants had previously attended college at some point during their lives. The table
also indicates that many were displaced from outside the durable goods industries (c.f., Farber,
1997; Aaronson and Sullivan, 1998; Helwig, 2001). Indeed, approximately 70 percent of the
female displaced workers lost jobs held in the non-goods producing sector of the economy.
Community college participants differ in some important ways from other displaced workers in
the sample. As shown by Panel A of Table 1, participants tend to be younger, have slightly less
prior job tenure, and are more likely to be female. They also are slightly more likely to be located
in one the state’s rural counties. Most striking, however, is that those displaced workers who completed community college courses are substantially more likely to have previously attended college than their counterparts who did not complete community college courses.
As shown by Panel B of Table 1, participants completed about 30 credits on average. In Washington State community colleges, a full-time academic year consists of 45 credits. Thus community
college participants who earned at least one credit typically completed about two-thirds of a year
of schooling. Panel B of Table 1 also shows the breakdown of credits into two broad categories.
“Group 1” courses include academic courses in math and science as well as courses related to the
health professions, technical trades (e.g. air conditioner repair), and technical professions (e.g.,
software development). “Group 2” consist of all other courses and includes academic social sciences and humanities courses as well as courses in business, sales and less technical vocational
tracts, personal health, physical education, consumer education, basic skills, and English as a sec-

5

ond language courses.3 Approximately one-third of the credits completed by female displaced
workers were in courses teaching the more technically oriented academic or vocational skills.
Male displaced workers completed more than one-half of their credits in these Group 1 courses.
This difference between the content of courses completed by males and females suggests that
community college schooling may have different impacts for men and women.
Figure 1 plots mean earnings over time for workers displaced in 1991 who earned community college credits and a comparison group composed of those who did not. Among those who earned
credits, we distinguish between those who earned more Group 1 credits and those who earned
more Group 2 credits. Consistent with the evidence in Jacobson, LaLonde, and Sullivan (1993a),
Figure 1 shows that displaced workers’ real quarterly earnings (1995 dollars) declined during the
two-year period prior to their displacements. Moreover, their earnings fell sharply after their displacements. Further, displaced workers’ quarterly earnings remain roughly $2,000 below their
peak pre-displacement levels even after four years. In addition, we observe that while the earnings
of Group 1 participants are similar to the comparisons, the earnings of the Group 2 participants
are consistently lower than the earnings of the other groups, both prior to and after their job
losses.
Though informative about the general characteristics of the data, simple comparisons between
earnings of the groups shown in Figure 1, do not necessarily indicate the effectiveness of community college schooling. For instance, although Group 1 concentrators seem to have lost ground relative to nonparticipants, this does not necessarily mean that training was ineffective. These
participants differ from the nonparticipants in gender, race, age, prior education, and the industries from which they were displaced, factors which may have an impact on their subsequent earnings. The statistical model we have developed and the empirical work presented below account
for such possibilities.

III. The Econometric Model
We take three different approaches to parameterizing the impact of community college participation on earnings. The first two approaches have been widely implemented in the program evaluation and returns to schooling literature. Our third approach is a hybrid of the first two that we
believe does a more credible job accounting for selection and focuses on the variation in training
that is most likely to give an unbiased estimate of the impact of community college schooling.
3. Table A1 shows the categories of courses included in the two groups of courses.

6

In our first approach, we treat participation as a single discrete or “binary” event, as in much of
the literature on “program evaluation” (Ashenfelter, 1978; Ashenfelter and Card, 1985; Heckman,
LaLonde, and Smith, 1999). This approach compares regression-adjusted post-program earnings
of workers who did, and did not, complete community college schooling. In our regression analysis, we account for unobserved characteristics that are either fixed or changing at a constant rate
over time by exploiting information on individuals’ earnings over several years prior to their
schooling.
In our second approach, we assume the effects of community college are proportional to the number of credits earned. Unlike the first approach, this approach identifies the effects of training both
from differences in the earnings growth of participants and non-participants and from differences
in earnings growth of participants who obtain differing numbers of credits. Thus, in a sense, the
“trainees” serve as their own comparison group. Indeed, given the assumption that earnings gains
are proportional to credits, the effects of training could, in principle, be estimated without using
any data on non-participants. In practice, of course, there would still be important efficiency gains
from including data on displaced workers who did not attend community college.
Our second approach is analogous to specifications employed in the “returns-to-schooling” literure. Where that literature measures the return to a year of schooling, our approach measures the
impact of a community college credit. This is of value because the policy choice is not simply
whether displaced workers should attend community college, but also how long and how intensively they should attend. In the literature on the returns to community college schooling, this
approach has been employed by Kane and Rouse (1995).
Our third approach is a hybrid of the first two approaches. We allow for an impact of community
college that is proportional to credits as in the “returns to schooling” approach and, in addition,
for a discrete effect of training participation, as in the program-evaluation approach. There are
two advantages of this third approach. First, it may capture actual effects of community college
attendance that are not proportional to the number of credits earned. For instance, attending community college may provide “networking” opportunities that facilitate job search. Much of such
benefits are likely to be felt soon after students start school. Thus, allowing for a discrete effect of
having enrolled in school, which does not depend on credits earned, may identify such effects.
A second advantage of our hybrid approach is that it allows us to take into account the possibility
that displaced workers who do, and do not, participate in training may differ in some time-varying
manner that is not fully accounted for by regression adjustment. In such cases, estimates of the
impact of credits employing the pure “returns-to-schooling” approach are likely to be biased.

7

However, if given their decisions to attend community college, workers acquiring different numbers of credits differ systematically only in respects fully accounted for by regression adjustment,
the resulting estimates of the per-credit effect in the hybrid approach will be unbiased. In previous
work (Jacobson, LaLonde, and Sullivan, 1999, 2002), we have found that among those earning a
positive number of credits, the relationships between the number of credits earned and various
observable variables are significantly weaker than the corresponding relationships between these
variables and whether a displaced worker participates in community college schooling. It seems
reasonable that a similar pattern should hold for unobservable variables. Thus, we believe estimates of the impact of a community college credit obtained from the hybrid model are more reliable than estimates obtained by assuming strict proportionality between credits and schooling
outcomes.
To implement our three approaches for estimating the impact of community college training on
workers’ earnings, we estimated several linear regression models of the form,
(1)

y it = α i + ω i t + γ t + x it β + δ it ( s i, z i ) + τit ( c i, f i, l i, z i ) + ε it .

According to model (1), workers’ quarterly earnings, y it , depend on (a) unobserved worker-specific fixed effects and time trends as well as quarter-specific fixed-effects, αi + ω i t + γ t ; (b)
observed time-varying demographic characteristics, x it β ; (c) the effects of their displacement,
δ it ( s i, z i ) , which may depend on the date of their job loss, s i , and worker characteristics, z i ; (d)
the effects of training, τ it ( c i, f i, l i, z i ) , which may depend on the workers’ characteristics as well
as the number of credits earned, c i , and the first and last quarters of community college attendance, f i and l i ; and (e) time varying unobserved characteristics, ε it . We discuss the detailed
specification of factors (a) to (d) below.
Controls for worker and time period heterogeneity
In all of our models, we include time period fixed effects and worker-specific fixed effects. The
former controls for forces that effect earnings for all workers equally in a particular quarter. The
latter controls for any characteristic that is fixed over time for an individual worker. In addition, in
some models we include worker-specific time trends. These controls account for individual-specific heterogeneity that change at a constant rate over time. Previous program evaluations indicate
that models with such controls may be more likely to yield unbiased estimates of the impact of
training (Bloom, 1984; Ashenfelter and Card, 1985; Heckman and Hotz, 1989).

8

Observed time-varying demographic characteristics
In each of our statistical models, we include separate second order polynomials in age for each of
the four combinations of sex and minority status. It is important to adjust for age, gender and
minority status, because community college participants are typically younger, are disproportionately female and differ from the norm in terms of race, all factors that are associated with differences in earnings growth independent of any effects of schooling.
Effects of displacement
Previous research has documented the temporal pattern of the impact of displacement on workers’
earnings (e.g., Jacobson, LaLonde, and Sullivan, 1993). Displaced workers’ earnings typically
decline during the period prior to displacement; drop sharply following the quarter of their job
loss; and then rise relatively rapidly during the next few quarters before beginning to increase at a
slower rate in subsequent periods. Figure 1 indicates that this pattern holds in the Washington
State sample analyzed in this paper.
To capture the baseline pattern of displacement effects, we introduce dummy variables for the
number of quarters relative to displacement. Specifically, we assume that the effect of displacement on the earnings of a worker in quarter t who is displaced in quarter s i is a parameter δ t – si if
t ≥ s i – 8 and is zero otherwise.4 Thus the baseline displacement effect can be written as
δ it ( s i, z i ) =

∑

k ≥ –8

D itk δ k , where D itk is an indictor variable equal to one when t – s i = k and zero

otherwise.
We allow the temporal pattern of displacement effects to differ from the baseline according to
such factors as a workers’ sex, age, race, location, previous industry, and job tenure. Since
whether and to what extent a worker seeks retraining also may depend on these characteristics, it
is important to control for differences in the associated displacement effects. We do this by introducing additional interactions between such characteristics and a set of five factors that capture
different aspects of the temporal pattern of displacement effects. Specifically, we define,
F it1 = t – s i + 8 if s i – 8 ≤ t < s i and F it1 = 0 otherwise,

4. In preliminary work, we found that displacement effects were very limited prior to two years before separation.

9

F it2 = 1 if t = s i and F it2 = 0 otherwise,
F it3 = 1 if t > s i and F it3 = 0 otherwise,
F it4 = t – s i if t > s i and F it4 = 0 otherwise, and
F it5 = ( t – s i ) 2 if t > s i and F it5 = 0 otherwise.
Workers are assumed to differ in their “loadings” on these factors in a manner that is linear in
their characteristics, z i . Thus, the full displacement effect in our statistical model becomes:

(2)

δ it ( s i, z i ) =

∑

k ≥ –8

5

D itk δ k +

∑ Fitj zi φj ,
j=1

where the φj are additional parameter vectors to be estimated. Variation in the loading, z i φ 1 , on
the first factor allows for variation in the rate at which earnings dip in the period before displacement. Similarly, variation in the loading, z i φ2 , on the second factor allows for the earnings of
workers with certain characteristics to drop more sharply in the quarter of their jobs loss. Finally,
variation in the loadings, z i φ3 , z i φ 4 , and z i φ 5 , the third through the fifth factors, allow for a
quadratic departure of workers’ earnings from the baseline pattern of displacement effects during
the period after displacement.
In our empirical work, the vector of characteristics, z i , consists of indicator variables for gender,
minority status, four levels of previous education, three regions, for having a previous job in aerospace, a previous job in another manufacturing industry, a previous job in other parts of the goodproducing sector, previous job tenure of less than six years, and age at job loss in the 20s, 30s,
40s, or 50s. Altogether, our specification of the effects of displacement on earnings is quite flexible, containing approximately 150 parameters.
Effects of community college attendance
We allow the form of the “schooling effect,” τ it ( c i, f i, l i, z i ) , to depend on whether the worker is
in school or finished with classes. In the former case, we expect that schooling may lead to some
foregone earnings and specify the impact of schooling on earnings as

10

(3a)

ci
τ it ( c i, f i, l i, z i ) = ψ 0 + ------------------------- ψ if f i ≤ t ≤ l i ,
( l i – fi + 1 ) 1

ci
where c i is the total number of credits earned, and thus ------------------------- is the average number of
( li – fi + 1 )
credits earned per quarter, during the period workers were in school and where ψ 0 and ψ 1 are
additional regression parameters.
According to (3a), attending community college may be associated with lower earnings and the
effect may increase with the number of credits taken. To the extent that school attendance truly
causes earnings to decline, the parameters ψ 0 and ψ 1 are of significant policy importance as they
would represent a large part of the costs associated with retraining that need to be balanced with
its benefits.
It is possible, however, that in some cases, workers who earned more credits simply had difficulty
finding a full-time job following their job loss and therefore elected to fill what would otherwise
have been idle time by taking classes. In this case, the in-school effects on earnings is at least partially attributable to the lack of success of displaced workers’ job searches rather than a true cost
of school attendance. Such an interpretation may have most relevance in the quarter or two immediately after job loss when lack of luck in job search may plausibly explain unemployment or
underemployment. By contrast, it seems likely that by the time workers have spent several quarters in school, they could have found a full time job, but instead chose to forego some earnings in
the expectation of increased future wages.
Our primary interest is in the effects of schooling in the period after it is complete. In this case, we
have found it important to allow for a transition period after workers leave school. That is, immediately after leaving community college classes, worker’s earnings tend to be lower than they
eventually become in the long-term. Indeed, immediately after leaving school we almost always
have found that workers appear to have been worse off than they would have been without obtaining additional retraining. However, earnings tend to rise quickly in the year following the end of
training, before leveling out to a new, higher level. This finding underscores the importance of
having access to data for a long follow-up period when evaluating education and training programs for adults. Without the ability to observe at least some individuals for several quarters after
leaving school, it would be very difficult to gauge the value of training.
To capture this temporal pattern in schooling effects, we specify the post-schooling effects as

11

(3b)

1 τ + c  1 τ , t > l .
τ it ( c i, f i, l i, z i ) = τ 0 + c i τ 1 +  ---------i  ---------- 3
i
t – l i 2
t – li

In (3b), the parameters τ 0 and τ 2 capture discrete earnings differences between displaced workers who complete at least one community college credit and their counterparts who either do not
enroll or enroll but do not complete any courses, as in the program-evaluation approach. The
parameters τ 1 and τ 3 capture the effects of schooling that are proportional to the number of credits earned, as in the returns-to-education approach.
The long-run effects of community college attendance are given by the parameters τ 0 and τ 1 in
1
(3b). The term  ----------  , which is equal to one in the period immediately after schooling, converges
t – li
to zero with the passage of time. Thus, τ 2 and τ 3 capture short-run deviations from these longrun impacts. In particular, the discrete impact of community college attendance is τ 0 + τ 2 in the
period just after training, while the long-run effect is simply τ 0 . Similarly, the per-credit impact
of community college schooling is τ 1 + τ 3 in the period just after training, while the long-run
effect is simply τ 1 . This specification allows for the possibility that the impact of community college credits on earnings is small or negative during the first few quarters after leaving school, but
may grow (or decline) over time before leveling off in the long-term.
As noted, (3a) and (3b) allow for a discrete jump in earnings between zero and positive credits.
However, the impact of additional credits is assumed constant. To test the robustness of our
results to the latter assumption, we estimate models in which earnings are a step function of credits with several possible steps. This fits into the above framework if c i is reinterpreted to be a vector of indicator variables for worker i earning credits in various ranges and the τ k parameters in
(3a) and (3b) are conformable vectors.
Finally, our analysis of the data revealed that it is important to allow the effects of Group 1 and
Group 2 courses to differ. In this case, c i is a vector whose components are the number of each
type of credits earned and the τ k parameters in (3a) and (3b) are conformable vectors.
While somewhat complicated, the model consisting of (1) - (3) is simply a large linear regression
model. We estimate it by ordinary least squares using somewhat over 3.2 million worker-quarter

12

observations on between 160 and 300 time varying independent variables plus 97,000 worker
fixed effects and, in some cases, worker-specific time trends. This is accomplished using the usual
Frisch-Waugh two step approach to eliminate worker-specific intercepts and/or trends. That is,
writing the model as y it = α i + ω i t + w it λ + ε it where w it includes all the terms whose coeffi˜ it are deviations
˜ it λ + ε̃ , where y˜ it and w
cients are not worker-specific, we estimate y˜ it = w
it
from worker-specific means in the case the case of simple worker-specific fixed effects ( ω i are
all zero) and deviations from worker-specific linear time trends in the case in which worker-specific time trends are also present.5 Stacking all the observations for worker i to form ỹ i and w̃ i ,
˜ i ′w̃ i )
the estimator can be written as λ̂ = ( ∑ w

–1

∑ w˜ i ′ỹi . Letting ei

= y˜ i – w˜ i λ̂ be the vector of

residuals, we report standard errors based on
Cov ( λ̂ ) = ( ∑ w̃ i ′w̃ i ) ( ∑ w̃ i ′e i e i ′w̃ i ) ( ∑ w̃ i ′w̃ i ) , the panel data analog of White’s (1980)
–1

–1

covariance matrix estimator, which is consistent in the presence of arbitrary forms of worker-specific heteroscedasticity and serial correlation.

IV. The Impact of Community College Schooling
Fixed Effects Estimates
Table 2 reports estimates of the post-schooling training parameters from several variants of the
fixed-effect version of our statistical model. In addition to the variables shown in the table, all
models include the controls discussed in the previous section of the paper.6 The model shown in
column (1) includes only post schooling dummies for males and females (the τ 0 terms of model
5. The computation of the deviations from worker-specific time trends is sped up considerably by noting that
the deviation from trend of a variable that is a person-specific constant times a time varying variable is that
same person-specific constant times the deviation from trend of the time varying variable. For instance, the
j

j

deviation from individual specific trend of a variable such as F itj z i is F̂ it z i where F̂ it is the deviation from
trend of F itj . Thus the regression on time can be done just for the Fitj variables and the result multiplied by
the appropriate person-specific constants to get the full set of variables expressed as deviations from workerspecific time trends.
6. These include worker-specific and quarter-specific fixed effects, the age, gender, and minority status controls, the baseline and heterogeneous displacement terms, and separate dummies for male and female workers being in their schooling period and the interaction of these dummies with the number of courses taken in
the quarter.

13

(3b) or (3b’)). The results indicate that during their post-schooling period, male workers who
completed at least one credit earned on average about $140 per quarter more than would otherwise have been expected on the basis of their other characteristics. The comparable figure for
females is about $55 per quarter. For men, such a figure represents about 3.3 percent of post-displacement earnings, while for women it is about 1.9 percent of post-displacement earnings. Given
that workers complete about two thirds of a year of school on average, these earnings increases
are lower than would be expected on the basis of standard estimates of the return to schooling.
As noted in the previous section, we find that the earnings of schooling participants are temporarily depressed in the quarters immediately after their training. Thus, their average earnings over
the relatively short follow-up period of our study may not reflect the average value of the training
over the much longer horizon that is relevant for judging the value of the training. The model represented in the second column of Table 2 accounts for this phenomenon by adding the interaction
1
between the post-schooling indicator and ---------- . As discussed in the previous section, the coeffit – li
cient on the post-schooling dummy, τ 0 , gives the estimate of the long-run effect and the sum of
1
the coefficients on the post-schooling dummy and its interaction with  ----------  , τ 0 + τ 2 , gives the
t – l 
i
effect in the quarter immediately after training.
As shown by the second column of Table 2, allowing for a transition period increases the estimated long-run effect of schooling to about $385 per quarter for men and about $220 for women.
These figures represent about 9.3 percent of post-displacement earnings for men and about 7.6
percent for women. Given the number of credits completed by the typical participant, these estimates are in line with, or perhaps somewhat above, conventional estimates of the returns to
schooling. The estimates in the second column also indicate that immediately after training, earnings are substantially below their expected levels. In the case of men, the estimate is - $453 ($385
- $839), while for women it is - $334. The estimates also indicate that the effect on earnings
becomes positive in between two and three quarters after leaving school.
The models shown in the first two column of Table 2 use the traditional “program evaluation”
approach that treats training as a binary event. In contrast, the models shown in the third and
fourth columns are in the spirit of the “returns to schooling” literature which assumes the effect of
training is proportional to the amount of schooling.7 The third column simply includes the number
of credits earned, c i , interacted with the post schooling indicator, while the fourth column also

14

1  . As with the first two colincludes the interaction of the post schooling indicator with  ---------t – l i
1  , substantially raises the estimate of the long-run
umns, adding the adjustment variable,  ---------t – l 
i
effect of schooling. With the adjustment term included, the long-run estimate of the impact of an
additional community college credit is about $11 per quarter for men and about $6.50 per quarter
for women. Thus, a year of school (45 credits) would yield increases in earnings of about $485
per quarter or 11.7 percent of post-displacement earnings for men and about $300 per quarter or
10.4 percent of post-displacement earnings for women. Again, the estimates of the effects immediately after schooling ends are substantially negative, taking three quarters to turn positive.
The last two columns of Table 2 show the results of including both terms that are binary indicators and terms that are proportional to the number of credits. Focusing on the long-run effects
shown in the last column, we see that male participants’ quarterly earnings would be expected to
eventually increase by about $285 plus $5.50 for each credit earned. The comparable figures for
women are about $160 plus $3.75 per credit earned. Taken at face value, the estimated intercept
terms ($285 and $160) suggest quite large returns to “networking” or other activities associated
with minimal accumulation of credits. The figures suggest that “just showing up” raises men’s
earnings by 6.8 percent and women’s earnings by 5.5 percent. However, as was noted above, it
seems possible that much of these estimates are due to selection of more able or more motivated
people into training. That is, even after controlling for individual fixed effects and observable
variation as described above, those workers who earn a positive number of credits may have been
destined to have somewhat higher earnings even without receiving any training.
The implausibly high estimates of the effects of “just showing up” suggest that the long-run percredit impacts shown in column (6) better represent the true impact of training. The values of
about $5.50 per credit for men and of about $3.75 per credit for women imply that a year of
schooling raises post-displacement earnings by about 6.0 percent for men and 5.8 percent for
women. The positive intercept terms and the lower per-credit coefficients estimated in the sixth
column suggest to us that the estimates of the per-credit coefficients shown in the fourth column,
7. Of course, that literature usually works with the log of wage rates rather than the level of earnings as the
dependent variable. We report results for log hourly wages below. Another difference between our results
and the returns to schooling literature is that we control for workers’ age while the returns-to-schooling literature controls for workers’ potential labor market experience. Replacing potential experience with age in the
typical returns-to-schooling specification tends to reduce the estimates returns to schooling by a percentage
point or two, an effect which should be kept in mind in comparing our results to those in the returns-toschooling literature.

15

which are in the spirit of “returns to schooling” studies like Kane and Rouse (1993; 1995)), overstate the true effects of community college schooling on earnings.
Worker-specific Time Trend Estimates
Table 3 has the same format as Table 2, but reports estimates from models that allow for workerspecific time trends as well as fixed effects. The pattern of results across models is similar to that
in Table 2. In particular, models that allow for a transition process after workers leave school continue to indicate much larger long-run impacts of attendance and credits than those that do not.
Moreover, there continues to be a positive earnings differential associated with minimal credit
accumulation. Finally, the per-credit impacts from column (6), which allows for effects of just
showing up, are smaller than those from column (4), which do not.
The estimated per-credit impacts in the most general models (the last column in each of the
tables) are considerably larger when we include individual-specific time trends. As shown in the
sixth column of Table 3, the quarterly earnings gain for men is estimated to be $8.72 per credit,
while for women it is about $8.41. These estimates imply that an academic year of credits
increases men’s earnings by 9.4 percent of post-displacement earnings, whereas the comparable
figure for women is 13.1 percent. Both these figures are quite respectable by the standards of the
returns-to-schooling literature.
The fact that estimated impacts increase with the inclusion of worker-specific time trends indicates that, controlling for observable factors including age, workers who earn more credits tend to
have had earnings that were rising less rapidly than average in the period before they enrolled in
community college courses. Thus the accumulation of credits seems, at least to some extent, to be
a means of compensating for certain negative factors that effect those workers’ earnings prospects. Once we allow for differences among workers in such earnings trends, our estimates of the
per-credit impact of training increase.
Nonlinear specifications
The estimates in Tables 2 and 3 are based on model (3b), which assumes that the relationship
between credits and earnings is linear. To explore how sensitive our results are to this assumption,
Table 4 presents estimates of models that allow the relationship between credits and earnings to
follow a step function, with separate estimates of the effects of earning 1 - 5 credits, 6 - 10 credits;
11 - 20 credits; 21 - 40 credits; 41 - 75 credits, and more than 75 credits, a level near that necessary to obtain an AA degree. There are approximately equal numbers of participants in each of
these categories, except the first, which contains more workers. This specification allows us to

16

evaluate whether the relationship between credits and earnings is nonlinear, or even non-monotonic.
The step function estimates that control for worker-specific time trends are generally consistent
with the estimates shown in Table 3 in indicating that individuals who complete more credits have
higher subsequent earnings. The estimated magnitude of the effects do, however, depart somewhat from strict linearity in credits. In particular, for males, post-training earnings rise sharply
with completed credits until an individual has obtained about 40 credits. After this point, the estimates suggest little additional benefit of more schooling. For females, the estimated gains associated with completing up to 40 credits are more modest than for males. But, in contrast to estimates
for males, female participants appear to benefit substantially from completing enough credits to
obtain a degree. As shown by the third column of Table 4, women who completed more than 75
credits earned on average more than $700 per quarter more than their counterparts who completed
41 - 75 credits.
The fixed effect estimates in Table 4 depart more significantly from what is predicted by the
affine specification in Table 2. As shown by the first column of Table 4, we estimate that the earnings of male participants increase with credits only up to the 41-75 credit level. Men who completed more than 75 credits are estimated to earn about $425 per quarter less than observationally
similar men who completed 41 to 75 credits. It seems implausible that completing additional credits could have this adverse of an impact on long-term earnings. Given that the worker-specific
time trends estimates suggest earnings increase with additional credits, it is likely that the drop in
the fixed effect estimates at high levels of credits are due to such workers having lower earnings
trends, rather than to any true negative effects of acquiring additional community college schooling.
Estimates of effects by type of credit
Table 5 shows separate estimates of the earnings impacts of Group 1 and Group 2 courses. Fixed
effects estimates are in the left two columns and individual trends estimates are in the right two
columns.
We find that the estimated long-run impacts of the more technically-oriented Group 1 courses are
much larger than the impacts of Group 2 courses. The fixed effect estimates suggest that each
Group 1 credit increases quarterly earnings by $15.72 for men and $17.13 for women. This
implies that a year of Group 1 credits ultimately would raise male post-displacement earnings by
17.0. For women, the long-run earnings increase associated with a year of Group 1 credits is 26.7

17

percent. The estimates of the long-run impact of Group 1 credits obtained from the model incorporating worker-specific time trends also are large and similar to the fixed effect estimates. These
results indicate that the earnings gain associate with an academic year of Group 1 credits is 13.6
percent for men and 29.3 percent for women.
In contrast to the positive results for Group 1 credits, the results in Table 5 suggest that obtaining
Group 2 credits does not raise earnings in a meaningful way. Indeed, the long-run fixed-effects
estimates are negative for both men and women. Negative estimates should not be rejected out of
hand. Work experience tends to raise earnings. Thus, if attending a course imparts no useful skills
and reduces workers’ time in the labor market, it would have a negative impact on subsequent
earnings. However, the fixed effect estimate of -$9.98 per credit for men is probably too negative
to be reasonable. It implies that obtaining a year of Group 2 credits lowers men’s earnings by the
equivalent of 10.7 percent of post-displacement earnings, a figure that would be much larger than
the likely effect of the loss of a year of labor market experience.
In work not reported here in detail, we examined the characteristics of workers who earned large
numbers of Group 2 credits, finding that such workers tended to have both lower earnings levels
and lower earnings growth in the periods before they enrolled in school. These unfavorable trends
in their earnings suggest that they would have continued to have lower earnings had they not
earned community college credits. The estimates shown in the last two columns of Table 5, which
control for worker-specific time trends, suggest that the impacts of Group 2 courses are positive,
but relatively close to zero. These results seems more reasonable than the corresponding fixed
effect estimates. Of course, they still suggest that completing Group 2 courses does not, on average, increase displaced workers’ earnings as substantially as Group 1 courses. Indeed, given the
estimated standard errors, we cannot reject the hypothesis that the true effect of Group 2 credits is
zero.
As with Tables 2 and 3, the estimates in Table 5 are based on a specification that assumes a linear
relationship between earnings and Group1 and Group 2 credits. To explore how sensitive our
results are to this assumption, we again examine a more flexible specification. In this case, earnings depend on discrete levels of both Group 1 and Group 2 credits. To avoid extremely small cell
sizes, we limited the categories of each type of credit to 0 credits, 1 -5 credits, 6 - 20 credits and
21 or more credits. Estimates of the long-run earnings impact of each combination of Group 1 and
Group 2 credits relative to earning no credits of either kind are shown in Table 6. For each of the
four blocks of the table, the number of Group 1 credits increases as one reads across rows and the
number of Group 2 credits increases as one reads down columns of the table.

18

With a few exceptions, the results of Table 6 are consistent with the simpler specifications shown
in Table 5. For both the fixed effect and worker-specific time trends models, there is almost
always a pronounced increase in earnings as the number of Group 1 credits increases for a fixed
number of Group 2 credits. One exception is the case of the worker-specific time trends estimates
for males with 6 to 20 Group 2 credits. In that case, the maximum earnings increase is when the
number of Group 1 credits is between 1 and 5. Whether increasing numbers of Group 2 credits are
associated with higher or lower earnings depends on the form of the controls for unobserved heterogeneity. As one reads down the columns of the fixed effect results, earnings tend to decrease,
except in the case of women earning one to five Group 1 credits, for which the pattern is relatively
flat. On the other hand, as one reads down the columns of the worker-specific time trends results,
earnings tend to increase at least weakly. Both of these results are consistent with those in Table
5. Thus, the additive linear specification appears to provide a reasonable description of the data.
A Specification Test
As a means of testing the specifications underlying the estimates shown in Table 5, we added
additional variables to model (3) that allow for an effect of schooling on earnings in periods after
workers are displaced, but before they actually start school. Such effects must be zero in reality.
However, as shown by Table 7, our estimates suggest that workers who will eventually earn credits tend to have especially depressed earnings in the period before they start school.
In the case of the fixed effect specification, male workers who will eventually earn a positive
number of credits have earnings in the period between job loss and the start of school that are
lower by about $225 plus $5 for each Group 1 credit they will eventually earn plus $8 for each
Group 2 credit they will eventually earn. In the case of women, the negative association with preschooling period earnings is lower for minimal numbers of credits, but increases more sharply as
the number of Group 1 and Group 2 credits increases. These results suggest that the fixed effects
specification fails to account for all factors that determine earnings and community college participation. The results shown in Table 7 for the worker-specific time trends specification also reveal
a negative association between pre-training earnings and community college attendance and
course completions. However, the magnitudes of the coefficients are somewhat smaller, except in
the case of the discrete effect of participation for males.
The unobserved factors that cause workers who will eventually participate in community college
retraining to have lower earnings in the period before that retraining are likely to be at least somewhat persistent. Thus, in the absence of retraining, such workers likely would have continued to
have somewhat lower than expected earnings. This likely negative correlation between unob-

19

served factors influencing post-training earnings and the extent of retraining participation implies
that the departures from the model’s assumptions exhibited in Table 7 likely cause some downward bias in the estimates of the effects of community college attendance and credits on postschooling earnings.
Effects of Training on Hours Worked and Wage Rates
So far, we have examined the impact of community college schooling on displaced workers’
overall earnings. However, it is also of interest to know what portion of these impacts are due to
increased hourly wage rates and what portion to increases in the number of hours worked. In particular, a finding that community college credits raise hourly wage rates is strong evidence that
these programs increase the productivity of displaced workers. By contrast, increases in employment probabilities or the number of hours worked among those who are employed may to some
extent come at the expense of other workers who do not receive training.
Washington State administrative records report both quarterly earnings and quarterly “hours
paid” for most workers. Accordingly, it is possible to separately estimate the impacts of schooling
on wages and hours worked. To do this we estimate a version of model (1) in which the dependant
variable is individuals’ log hourly wage as well as one in which the dependant variable is log
quarterly hours worked. In both cases the sample is limited to the quarters in which hours paid
were positive.
We report the fixed effect estimates of the impact of community college schooling on log quarterly hours worked and log hourly wages in Table 8 and the corresponding worker-specific time
trends estimates in Table 9. In each table, the last column reports estimates of the impact of community college schooling on log earnings during quarters in which individuals worked. These
estimates differ from those reported in Tables 5 and 6 because worker-quarter observations with
zero reported hours or earnings are excluded.8
As shown by Table 8, fixed effect estimates suggest that completing at least one community college credit is associated with increases in both log hours worked and log hourly wage rates. For
males, completing minimal credits is associated with about a 2.5 percent increase in quarterly
hours worked and a 3.0 increase in average wage rates. For females, completing at least some
community college schooling is associated with about a 3.0 percent increase in hours, but a trivial
increase in wages. As explained above, these impacts could be the effects of improved job search
8. In Washington State, employers are instructed to report “hours paid,” but in some case they do not.

20

through chances to “network” at community college, but a significant portion of the estimated
effect may be due to complex forms of selection into community college programs following the
loss of a job.
The fixed effect results in Table 8 indicate that for men, Group 1 community college credits raise
both hours worked and wage rates, while for women, the effect is limited to higher wages rates. In
particular, increasing the number of Group 1 credits by 45, which corresponds to one academic
year of schooling, raises male hours worked by about 6.5 percent (0.00144*45) and their wage
rates by about 5.6 percent. The same increase in credits increases female wage rates by about 10.5
percent, but has a negligible impact on their hours worked. These effects of Group 1 credits on
wages are thus at least comparable to those typically obtained in the returns to schooling literature, and in the case of women somewhat higher.
As shown by the bottom two rows of Table 8, fixed effects estimates suggest that for both men
and women, completing Group 2 credits is associated with increased hours worked, but also with
significantly lower hourly wages. An academic year of Group 2 credits raises quarterly hours
among those who work by about 7.2 percent for men and 6.5 percent for women. By contrast,
such schooling is associated with a 5.9 percent decline in hourly wages for males and a 4.2 percent decline in hourly wages for females. As discussed in conjunction with the results in Tables 5
and 6, such wage estimates likely reflect specification error and not the true impact of Group 2
courses on worker productivity.
To address the foregoing specification problem, we also estimated the impact of community college schooling on hourly wages and hours worked controlling for worker-specific time trends. As
shown by Table 9, the discrete effect of earning a positive number of credits on hours increases to
5.0 percent for men and 4.5 percent for women when we allow for the more general form of unobserved heterogeneity. By contrast, the estimate of the discrete effect of training participation on
wages becomes negative for both groups. In the case of men, the discrete effect is -3.3 percent,
while for women it is -3.5 percent.
The worker-specific time trend results in Table 9 suggest that increases in Group 1 credits lead to
somewhat larger increases in hours worked than the corresponding results controlling only for
worker-specific fixed effects. For men, the increase in hours associated with an academic year of
Group 1 credits is 9.0 percent, while for women it is 4.9 percent. The increase in wage rates associated with a year of Group 1 credits is somewhat smaller after controlling for worker-specific
time trends, but remains quite significant at 4.7 percent for men and 9.9 percent for women.

21

Finally, the worker-specific time trend results in Table 9 suggest that Group 2 courses raise the
hours worked of women, but not men, and lead to small, but statistically insignificant, decreases
in wage rates for both sexes. The increase in hours associated with completing an academic year
of such credits is over 8 percent for women. The predicted drop in wage rates is 2.6 percent for
men and 1.6 percent for women. Neither of these latter two effects, nor the effect of Group 2 credits on hours worked for men is statistically significant.
The foregoing results indicate that while earning Group 1 credits increases workers’ wage rates, it
does not increase them by as much as it does their earnings. Indeed, for both men and women,
only about a third of the increase in earnings is attributable to increases in wage rates. The larger
impact on earnings comes from increases in the probability of working and from the greater hours
worked when employed. This conforms to the findings of the relatively small number of evaluations which have provided evidence on both earnings and wage rate impacts of training programs
(LaLonde,1995).

V. Conclusions
The foregoing results have several implications for the methodology of evaluating the impact of
retraining on mid-career workers’ labor market outcomes. First, when follow up periods are relatively short, allowance needs to be made for a transition period following the end of schooling.
Because it appears to take several quarters before something close to the full effects of education
are felt, a simple average of the effects during a short follow up period will tend to underestimate
the true, long-run effects of retraining. Our device of including an additional term that converges
to zero as time since schooling increases substantially raises the long-run estimates of retraining
impacts.
Second, we find that it is important to allow for a discrete effect of schooling that is associated
with just “showing up” and earning at least one credit. Previous work by Kane and Rouse (1995)
has assumed that the effect of schooling on workers’ labor market outcomes is proportional to the
number of credits they earn. However, when we relax that assumption, we find both a highly significant discrete effect of training and a lower estimate of the effect of credits. The exception is
our result for females in specifications that include worker-specific time trends.
There are at least two possible interpretations of our estimates of the effect of just “showing up”
and earning at least one credit. It is possible that even very limited participation in community
college training may aid workers’ job searches by expanding their networks of contacts. How-

22

ever, the estimates of these effects, at least for males, seem implausibly large. Thus it seems likely
that much of the estimated discrete effect of training is due to some form of non-random selection
of community college participants. We obtain large estimates of discrete training effects even
controlling for worker-specific fixed effects and, in the case of males, worker-specific time
trends. Thus, if these estimated discrete effects are due to selection, it must be a relatively complex form of selection in which workers who have recently experienced some change in circumstances associated with increased future earnings are more likely to enter training. But, such
forms of selection do not necessarily undermine our findings. If, conditional on entering retraining, the number of credits earned by workers is not related to other factors that cause workers’
earnings to deviate from trend, then our measures of the effects of credits are not biased by such
forms of selection. By contrast, in the pure returns-to-schooling specification that omits the discrete impact of retraining, estimates of the impact of credits are upwardly biased.
Third, allowing for worker-specific time trends in addition to worker-specific fixed effects is necessary to adequately control for the effects of unobserved heterogeneity on earnings. Displaced
workers who obtain community college credits, especially those that obtain large numbers of
Group 2 credits, tend to have more slowly growing earnings even before entering retraining.
Thus, simple fixed effects estimates of the effects of credits are biased downwards. Adding
worker-specific time trends removes this source of bias.
Unfortunately, while we have made significant methodological progress, our models for quantifying the effects of community college retraining on workers’ earnings still appear to have some
shortcomings. In particular, our specification test finds “impacts” of training before it occurs, suggesting that our framework fails to adequately account for all the factors that may simultaneously
effect earnings and retraining decisions. In addition, we view the large estimated impacts of minimal numbers of credits for males as further evidence of model misspecification.
Though they must be qualified in light of the evidence of the misspecification just mentioned, we
also believe we can draw some tentative substantive conclusions about the impacts of community
college attendance on displaced workers’ earnings. In our preferred specification, the equivalent
of a year of community college credits is estimated to increase displaced workers’ earnings by
about 9 percent for men and 13 percent for women.
These estimates seem more likely to be biased downward than upward. First, the above figures
assume the estimated discrete impact of attendance on earnings is entirely due to selection bias.
But, there may be actual positive effects of “just showing up” in community college. As we have
noted above, these effects could derive from more efficient job search due to an expanded net-

23

work of contacts. Second, our specification test suggests that workers who choose to return to
community college tend to have suffered larger than normal earnings declines in the period before
they begin retraining. Moreover, among those who return, the pre-training drop in earnings is
larger for those who subsequently earn more credits. Given that there is likely to be positive serial
correlation in the unobservable factors responsible for such outcomes, those workers with larger
pre-training drops in earnings would probably have continued to suffer larger than normal losses
in the future had they not entered training. This suggests that our estimates of the effects of community college training on earnings are biased downward due to our inability to fully control for
all the factors that effect the magnitude of earnings declines.
We also found that the effects of community college credits on earnings depend significantly on
the nature of the courses taken. A year of credits in Group 1 courses, which consist of more technically-oriented vocational and math and science courses, is estimated to increase earnings by 14
percent for men and 29 percent for women. We estimate that about one third of these increases are
due to increased hourly wage rates, with the rest attributable to higher hours of work. In contrast,
we estimate smaller effects of earning credits in Group 2 courses, which consist of less technically
oriented vocational and humanities and social science courses. Point estimates in models with
worker-specific time trends suggest that an academic year of Group 2 credits increases earnings
by about four percent. Unfortunately, the standard errors associated with these estimates are large
enough that we cannot be confident that such effects are nonzero.
Given our findings on community college schooling for displaced workers, we estimate that the
social benefit-to-cost ratios associated with the mix of community college schooling completed
by displaced workers in our sample to be roughly 1.07. However, this net benefit results because
of the very large returns to Group 1 courses and because about one-half of the credits completed
by the males and about one-third of the credits completed by the females in our sample were in
these subject areas. Our analysis suggests that the internal rate of return to such courses is roughly
10 percent for males and roughly 17 percent for females. Besides the direct costs of the schooling,
these rates of return estimates also take into account (1) the welfare cost associated with taxes
used to subsidized community college schooling and (2) estimates of the foregone earnings associated with completing such schooling.9 By contrast, for Group 2 courses we estimate the internal
rates of return to be negative for both groups. These figures indicate that, on average, community
college schooling is at least a marginally productive social investment for displaced workers stud9. We estimate the social cost of an academic year of community schooling to be approximately $19,000 for
displaced males and $18,000 for displaced females. See the Appendix for details on how we arrived at these
estimates.

24

ied in this sample. Whether it is socially productive more generally appears to depend on the type
of schooling that displaced workers receive.

25

References
Aaronson, Daniel and Daniel Sullivan (1998), “The decline of job security in the 1990s:
Displacement, anxiety, and their effect on wage growth,” Economic Perspectives, Federal
Reserve Bank of Chicago, Vol. 22, No. 1, First quarter, pp. 17-43.
Ashenfelter, Orley (1978). “Estimating the Effect of Training Programs on Earnings,” Review
of Economics and Statistics 60 (February): 47-57.
Ashenfelter, Orley, and Card, David (1985). “Using the Longitudinal Structure of Earnings to
Estimate the Effect of Training Programs.” Review of Economics and Statistics 67 (November):
648-60.
Ashenfelter, Orley, and Robert LaLonde, (1997), “The Economics of Training” in Handbook
of Human Resources, D. Lewin, D. Mitchell, and M. Zaidi, eds., JAI Press, 1997.
Angrist, Joshua, and Newey, Whitney (1991). “Over-Identification Tests in Earnings
Functions With Fixed Effects,” Journal of Business and Economic Statistics, 9(3): 317 - 323
(July).
Bednarzik, Robert; and Jacobson, Louis (1994). “Analysis of the Dislocated Workers'
Educational Training Program (DWETP): A Locally Funded Voucher-Like Program in
Pittsburgh, Pennsylvania,” mimeograph, Westat Inc.
Bloom, Howard (1984), “Estimating the Effect of Job-training Programs, Using Longitudinal
Data: Ashenfelter's Findings Reconsidered,” Journal of Human Resources, 19(4): 544-56.
Bloom, Howard (1990), Back to Work: Targeting reemployment Services for Displaced
Workers, Kalamazoo, Michigan: W.E. Upjohn Institute for Employment Research.
Browning, E., (1987). “On the Marginal Welfare Cost of Taxation,” American Economic
Review, 77(1): 11 - 23.
Corson, Walter; Decker, Paul; Gleason; and Nicholson, Walter (1993). International Trade
and Worker Dislocation: Evaluation of the Trade Adjustment Assistance Program, Final Report
DOL contract No: 99-9-0805-75-071-01, Princeton: Mathematica Policy Research Inc.
Fallick, Bruce (1996). “A Review of the Recent Empirical Literature on Displaced Workers,”
Industrial and Labor Relations Review, 50 (1): 5 - 16.
Farber, Henry (1993). “The Incidence and Costs of Job Loss: 1982 - 1991,” Brookings Papers
on Economic Activity: Microeconomics, 1993(1): 73 - 119.
Farber, Henry (1997). “The Changing Face of Job Loss in the United States, 1981-1995”
Brookings Papers on Economic Activity: Microeconomics, 1997(1): 55-128.
Flaim and Seghal (1985). “Displaced Workers of 1979 - 83: How well Have They Fared?”
Bulletin 2240, U.S. Department of Labor, Bureau of Labor Statistics.
Freeman, Richard B. (1974), “Occupational Training In Proprietary Schools and Technical
Institutes,” Review of Economics and Statistics, 63:310 - 318.
Grubb, Norton (1993). “The Long-Run Effects of Proprietary Schools on Wages and

26

Earnings: Implications For Federal Policy,” Education Evaluation and Policy Analysis, 15(1): 17
- 33 (Spring).
Heckman, James, and Robb, Richard (1985). “Alternative Methods for Evaluating the Impact
of Interventions.” In Longitudinal Analysis of the Labor Market Data, ed. J.J. Heckman and B.
Singer. Cambridge: Cambridge University Press.
Heckman James J., and Hotz, V. Joseph (1989). “Choosing Among Alternative
Nonexperimental Methods for Estimating the Impact of Social Programs: The Case of Manpower
Training.” Journal of the American Statistical Association 84 (December): 862-74.
Heckman, James, Robert LaLonde, and Jeffrey Smith, (1999) The Economics and
Econometrics of Active Labor Market Policies,'' in The Handbook of Labor Economics, eds.
Orley Ashenfelter and David Card, Amsterdam: North-Holland, 1999.
Jacobson, Louis S.; LaLonde, Robert J.; and Sullivan, Daniel G. (1993a). “Earnings Losses of
Displaced Workers,” American Economic Review, 83(4):685 - 709.
---- (1993b). The Costs of Worker Dislocation, Kalamazoo, Michigan: W.E. Upjohn Institute
for Employment Research.
---- (1993c). “Earnings Losses of High-Seniority Displaced Workers,” Economic
Perspectives, Federal Reserve Bank of Chicago, Vol. 17, No. 6, November/December, pp. 2-20.
---- (1999). “Participation in Community College Schooling and Its Effects on Displaced
Workers’ Earnings and Employment Prospects: A Study of Displace Workers From Washington
State,” Final Report for U.S Department of Labor, Employment and Training Administration, #K6307-7-00-80-30, (September).
--- (2002). “Measures of Program Performance and the Training Choices of Displaced
Workers,” Targeting Employment Services, Randall W. Ebberts, Christopher J. O’Leary, and
Stephen A. Wadner eds., Kalamazoo, Michigan: W.E. Upjohn Institute for Employment
Research.
Kane, Thomas; and Rouse, Cecila (1993). “Labor Market Returns to Two- and Four- year
College,” mimeograph, Princeton University.
Kane, Thomas; and Rouse, Cecila (1995). “Labor Market Returns to Two- and Four- year
College,” American Economic Review, 85(3, June).: 600 – 14.
Kane, Thomas; and Rouse, Cecila (1999). “The Community College: Educating Students at
the Margin Between College and Work,” Journal of Economic Perspectives, 13(1, Winter).: 63 –
84.
Kletzer, Lori (1998). “Job Displacement,” Journal of Economic Perspectives, 12(1): 115 – 36.
Leigh, Duane E. (1990). Does Training Work for Displaced Workers: A Survey of Existing
Evidence, Kalamazoo, Michigan: W.E. Upjohn Institute for Employment Research.
Ruhm, Christopher (1991). “Are Workers Permanently Scarred by Job Displacements?”
American Economic Review, 81:1, 319-323.

27

Saint-Paul Gilles (1994). “Out in the Cold,” European Economic Perspectives, London:
Centre for Economic Policy Research, 4: 5-6.
White, Halbert (1980). “A Heteroscedasticity-Consistent Covariance Matrix Estimator and a
Direct Test for Heteroscedasticity,” Econometrica 48:817-38.

28

Appendix
Notes on Displaced Workers, Administrative Earnings and Community College Records
There is no consensus as to the definition of a displaced worker. However, there are three common elements in most descriptions of such individuals:
(i) Displaced workers have not been discharged for cause;
(ii) Displaced workers have permanently separated from their former employer or have only a
very small likelihood of being recalled to their old jobs;
(iii) Displaced workers have had strong prior attachment to the industry of their pre-displacement
employer.
Studies of the consequences of job loss and of programs and policies to aid the victims of displacement, therefore, have usually limited their attention to unemployed workers with some or all
of these characteristics. Recognition that job loss is potentially more costly for high-tenured
workers has lead to U.S. Department of Labor in some of its publications to define displaced
workers as persons having at least 3 or more years of tenure when they permanently lost their jobs
(Flaim and Seghal, 1985).
Our Washington State sample is constructed from a data base on every individual who lost a job
in that state between 1990.II and 1994.IV, who filed a valid claim to receive unemployment insurance benefits and who had accumulated at least six quarters of job tenure with his employer. From
the state, we received two types of administrative records. First, we received quarterly earnings
and unemployment insurance benefits administrative records that started in 1987 and continued
through 1995. Second, we received machine-readable transcripts for all of these individuals who
had enrolled in one of the state’s 25 community colleges. These records cover the period starting
with the fall term of 1989 and continuing through the end of 1995.
We exclude from our sample individuals who transferred to a public four-year college or who
acquired more than 135 community college credits. We also exclude workers who were in school
at the beginning of the transcript records because of the possibility that they were actually in
school in earlier quarters as well.
We restrict our sample of displaced workers in several additional ways in order to focus on individuals for whom the effects of retraining would likely be of the greatest policy interest. First, we

29

limit our analysis to adults between the ages of 22 - 60 years old. We drop younger workers
because of the traditional policy interest in the displaced as a group of experienced workers who
require retraining assistance and also because studies of the economically disadvantaged population suggest that youths respond differently to retraining services than do adults (Heckman,
LaLonde, and Smith, 1999). We drop older workers because there is less policy interest in training workers close to retirement age.
We also limit our analysis to workers who had accumulated at least 12 quarters of service with
their pre-displacement employers. We exclude workers with fewer quarters of tenure with their
former employer because such individuals do not fall within common definitions of a displaced
worker. In addition, because a rationale for subsidizing displaced workers’ retraining is that the
earnings losses associated with their displacements may be substantial, we chose not to analyze a
group for whom on both theoretical and empirical grounds the adverse impacts of job loss are
likely to be small.
Finally, we also limit our sample to displaced workers who remained “continuously attached” to
the state’s work force. We define such workers to be those whose only periods of eight consecutive quarters without earnings were those containing their job loss or enrollment in community
college. This sample restriction ensures that each member of the sample was consistently attached
to Washington State's employed work force throughout the sample period.
The Washington State sample used in our analysis contains more than 3 million person/quarter
observations, on 97,262 individual workers. Of the latter number, 16,103 completed at least one
credit in a Washington State community college.
Computing the Benefits and Costs of Community College Schooling
In the text we report the social benefit-cost ratios and internal rates of return for community college schooling. In our computation of the social benefits we consider only gains in earnings. We
analyze the expected benefits for a representative participant who works for 25 years and acquires
the average number of Group 1 and Group 2 credits as reported in Table 1B. The 25 years figure
is a plausible, although probably upwardly biased, estimate of the remaining work life for the
average participant in our sample. We take account of the transition period after leaving community college schooling. When computing the social benefit-cost ratios, we discount future earnings gains at a rate of 4 percent.
We arrive at our estimates of the social cost of community college schooling as follows: First,

30

using Kane and Rouse (1999), we estimate that the average cost of a year of community college
schooling is about $8,000 per full-time equivalent student. This amount reflects the cost of tuition
and expenditures by the public sector other than transfers designed to cover participants’ living
expenses. Of this amount, tuition typically covers about 20%, with the rest paid for out of tax
receipts. Second, we estimated the welfare cost or deadweight loss associated with the taxes
raised to subsidized community colleges to be about $3,000 (Browning, 1987; Heckman,
LaLonde, and Smith, 1999). This amount is about 50 percent of 80% of $8,000 or 50 percent of
the amount of the direct community college costs paid for by the public sector.
Finally, we use the estimated parameters of (3a) in the text to estimate the foregone earnings associated with acquiring community college schooling. These estimates are about $8,000 per academic year for males and about $7,000 per academic year for females. The estimates from our
model do not suggest large differences between the average foregone earnings of male and female
community college participants. To the extent that our estimates of the foregone earnings associated with community college participation overstate lost earnings, the estimates of internal rates
of return as well the social benefit-cost ratios given in the text will be too small. When computing
the social benefit-cost ratios for the displaced workers in our sample, we take into account that the
average participant completed about two-thirds of an academic year of schooling.

31

Table A: Classifications of Washington State Community College Classes
Group1: Quantitative or Technically Oriented Vocational courses
Health related courses
Technical/professional courses
Technical trades
Science/math academic courses.
Group 2:Non-quantitative Courses
Sales/service courses
Vocational courses not in Group 1
Social Science/humanities courses
Health/PE/consumer-oriented courses
Basic skills education
Other courses.

32

Table 1: Characteristics and Community College Credits of Displaced Workers
Panel A: Characteristics of Individuals in Sample of Washington State Displaced Workers

Males

Mean age at separation

Females

Participants

Comparisons

Participants

Comparisons

(1)

(2)

(3)

(4)

35.9

39.0

37.2

40.3

Aged 20 - 29 at separation

0.306

0.190

0.257

0.158

Aged 40 - 59 at separation

0.307

0.426

0.373

0.497

Minority

0.113

0.140

0.100

0.136

Less than H.S. degree

0.081

0.151

0.051

0.122

More than H.S. degree

0.491

0.374

0.511

0.399

Mean job tenure at separation
(quarters)
Greater than 6 years tenure

18.8

18.8

19.1

19.4

0.195

0.201

0.231

0.241

Aerospace/Motor Vehicle

0.177

0.104

0.118

0.082

Other Manufacturing

0.288

0.232

0.149

0.142

Other Goods Producing

0.155

0.227

0.043

0.055

Non-Goods Producing

0.380

0.437

0.690

0.721

Rural

0.367

0.337

0.321

0.311

Seattle-Tacoma MSA

0.504

0.539

0.546

0.569

Other MSA

0.140

0.123

0.133

0.120

Industry of former employer

Region of state

Mean earnings in 4 quarters
prior to separation (1995$)

$29,486

$30,351

$22,790

$22,184

Mean earnings 1994Q3
to1995Q3 (1995$)

$16,643

$18,404

$11,550

$12,281

Number of Observations

8,041

49,532

8,062

31,627

Notes: Comparison groups are comprised of displaced workers who did not earn community college credits during
or after Fall quarter 1989. Workers earning more than three academic years of credits are excluded from the samples. See appendix for details of how the sample was constructed from administrative records. Numbers are proportions of the group with the given characteristic unless noted.

33

Table 1 (continued)
Panel B: Credits Completed by Displaced Workers Who Completed at Least One Credit

Mean

1-5

6-10

11-20

21-40

41-75

Males

31.9

0.279

Females

28.9

0.323

75+

0.150

0.147

0.139

0.133

0.153

0.153

0.153

0.125

0.126

0.136

1-5

6-20

20+

All Credits

Mean

0

Group 1 Credits

Males

16.9

0.312

0.229

0.219

0.241

9.6

0.420

0.247

0.194

0.140

Males

15.1

0.283

0.225

0.260

0.232

Females

19.3

0.164

0.266

0.275

0.295

Females
Group 2 Credits

Notes: The first column gives the average number of credits completed in Washington State community colleges
during and after the Fall quarter of 1989 by displaced workers who completed at least one such credit. Subsequent
columns give the proportion of such workers earning the indicated number of credits. Group 1 credits are from
courses teaching more technical academic and vocational skills. Group 2 credits are from other courses. See Table
A1 for a more detailed description of courses and the appendix for details on the transcript data from Washington
State’s 25 community colleges.

34

Table 2: Fixed Effects Estimates of Impact of Community College Credits on Displaced Workers’ Earningsa
Model

(1)

(2)

137.25
(45.62)

385.37
(57.42)

(3)

(4)

(5)

(6)

199.10
(56.22)

285.24
(71.15)

Males
Post-Collegeb
Post-College * 1/kc

-838.59
(74.33)

Credits*Post-College

-415.02
(95.74)
0.12
(1.16)

Credits*Post-College * 1/k

10.79
(1.57)

-3.17
(1.43)

-27.25
(1.79)

5.48
(1.95)
-19.31
(2.34)

Females
Post-College
Post-College * 1/k
Credits*Post-College
Credits*Post-College * 1/k

54.97
(35.11)

219.75
(43.62)

102.42
(20.24)

-553.53
(56.91)

157.87
(51.65)
-254.39
(71.20)

-1.04
(1.08)

6.66
(1.43)
-20.04
(1.59)

-2.76
(1.26)

3.69
(1.69)
-15.07
(2.00)

a. The dependent variable is quarterly earnings. All models include demographic, heterogeneous displacement, and in-school controls as well
as worker-specific fixed effects and period fixed effects. See section III of the text for detailed discussion of the specification. Robust standard
errors are in parentheses.
b. Indicator variable equal to one during quarters after training participant left school.
c. 1/k denotes the reciprocal of the number of quarters after worker left school.

Table 3: Worker-Specific Time Trends Estimates of Impact of Community College Credits on Displaced Workers’ Earningsa
Model

(1)

(2)

80.61
(51.74)

503.25
(71.12)

(3)

(4)

(5)

(6)

118.68
(61.26)

322.13
(85.80)

Males
Post-Collegeb
Post-College * 1/kc

-816.30
(70.74)

Credits*Post-College

-423.75
(91.22)
-0.18
(1.24)

Credits*Post-College * 1/k

13.84
(1.77)

-1.92
(1.47)

-25.00
(1.69)

8.72
(2.15)
-17.83
(2.22)

Females
Post-College
Post-College * 1/k
Credits*Post-College
Credits*Post-College * 1/k

-27.13
(40.98)

199.24
(56.24)

-12.63
(47.04)

-435.06
(55.10)

45.88
(66.12)
-120.71
(69.53)

-1.04
(1.15)

9.20
(1.74)
-18.30
(1.61)

-0.84
(1.32)

8.41
(2.05)
-16.17
(2.04)

a. The dependent variable is quarterly earnings. All models include demographic, heterogeneous displacement, and in-school controls as well
as worker-specific fixed effects and time trends and period fixed effects. See section III of the text for detailed discussion of the specification.
Robust standard errors are in parentheses.
b. Indicator variable equal to one during quarters after training participant left school.
c. 1/k denotes the reciprocal of the number of quarters after worker left school.

Table 4: Estimates of Nonlinear Impact of Community College Credits on Earningsa
Fixed Effects
Long-Run
Impactb

Worker-Specific Time
Trends

Transition
Effectc

Long-Run
Impactb

Transition
Effectc

Males
1 to 5 credits

102.98
(94.60)

-256.69
(132.49)

135.00
(114.30)

-205.48
(128.12)

6 to 10 credits

385.42
(122.69)

-657.81
(170.06)

398.45
(150.28)

-645.73
(165.31)

11 to 20 credits

572.66
(132.06)

-1000.76
(175.00)

743.20
(158.21)

-991.95
(168.58)

21 to 40 credits

794.20
(134.28)

-1354.89
(175.62)

989.38
(167.24)

-1378.02
(171.17)

41 to 75 credits

810.80
(166.26)

-1663.15
(207.74)

780.71
(194.73)

-1422.49
(213.35)

greater than 75 credits

387.73
(200.26)

-1878.46
(223.05)

992.44
(214.75)

-2000.42
(212.14)

1 to 5 credits

122.52
(61.81)

-200.26
(89.99)

68.48
(80.33)

-127.58
(89.14)

6 to 10 credits

169.88
(90.47)

-299.09
(134.82)

104.49
(116.08)

-180.84
(132.62)

11 to 20 credits

456.15
(119.44)

-806.11
(148.82)

341.82
(137.68)

-476.54
(141.53)

21 to 40 credits

216.56
(118.82)

-815.59
(150.51)

127.71
(144.88)

-747.27
(147.73)

41 to 75 credits

309.41
(151.84)

-1149.22
(167.93)

311.49
(187.00)

-937.86
(172.68)

greater than 75 credits

543.26
(178.61)

-1655.98
(204.00)

1049.51
(203.61)

-1730.48
(204.75)

Females

a. The dependent variable is quarterly earnings. All models include demographic, heterogeneous displacement, and in-school controls as well as individual and period fixed effects. The models shown in
columns 3 and 4 also include worker-specific time trends. See section III of the text for detailed discussion of the specification. Robust standard errors are in parentheses.
b. Coefficient of indicator variable equal to one during quarters after training participant left school and
credits in range indicated.
c. Coefficient of indicator variable equal to one during quarters after training participant left school and
credits in range indicated times reciprocal of number of quarters after schooling.

37

Table 5: Estimates of Impact of Group 1 and Group 2 Community College Credits on
Earningsa
Fixed Effects
Males
Post-Collegeb
Post-College * 1/kc
Group 1 Credits*Post-College
Group 1 Credits*Post-College * 1/k
Group 2 Credits*Post-College
Group 2 Credits*Post-College * 1/k

Females

Worker-Specific Time
Trends
Males

Females

317.81
(71.48)

171.79
(51.56)

329.17
(86.54)

58.04
(65.88)

-407.96
(96.44)

-255.53
(71.29)

-410.05
(92.20)

-120.66
(69.49)

15.72
(2.59)

17.13
(3.71)

12.61
(2.92)

18.69
(5.04)

-20.90
(3.07)

-19.48
(4.06)

-15.63
(2.88)

-16.90
(4.65)

-9.98
(3.17)

-4.31
(2.29)

3.68
(3.73)

2.36
(2.80)

-16.66
(3.93)

-12.34
(2.72)

-21.49
(3.88)

-15.69
(2.72)

a. The dependent variable is quarterly earnings. All models include demographic, heterogeneous displacement, and in-school controls as well as individual and period fixed effects. The models shown in columns 3
and 4 also include worker-specific time trends. See section III of the text for detailed discussion of the specification. Robust standard errors are in parentheses.
b. Indicator variable equal to one during quarters after training participant left school.
c. 1/k denotes the reciprocal of the number of quarters after worker left school.

38

Table 6: Estimates of Long-Run Nonlinear Impact of Group 1 and Group 2 Community College Credits on Earningsa

Group 2 credits

0

Fixed Effects

Worker-Specific Time Trends

Group 1 credits

Group 1 credits

1 to 5

6 to 20

20+

525.65
(149.46)

1052.37
(144.95)

1465.89
(206.69)

0

1 to 5

6 to 20

20+

273.00
(182.55)

608.07
(196.15)

494.38
(263.61)

Males
0

0.00

0.00

1 to 5

-201.12
(120.37)

25.66
(306.30)

906.78
(334.01)

990.31
(327.41)

22.64
(146.55)

539.07
(319.50)

903.83
(360.29)

1179.08
(457.57)

6 to 20

-133.47
(148.00)

550.75
(246.98)

824.24
(290.03)

606.29
(293.11)

-17.74
(175.00)

1781.12
(270.58)

1345.64
(324.50)

1042.49
(311.20)

20+

380.94
(225.11)

237.91
(240.92)

201.54
(277.11)

418.97
(259.37)

428.11
(266.28)

714.71
(291.98)

1309.55
(314.32)

1372.06
(269.62)

0.0

212.81
(109.54)

847.98
(192.59)

963.58
(403.88)

0.0

145.91
(141.41)

471.85
(233.00)

471.90
(659.99)

1 to 5

48.42
(74.72)

315.04
(196.11)

577.29
(299.91)

758.39
(509.57)

-11.16
(99.45)

230.34
(225.30)

619.24
(333.61)

643.51
(479.39)

6 to 20

108.74
(94.98)

510.37
(195.38)

222.08
(217.70)

1027.55
(327.37)

100.12
(118.18)

422.33
(223.27)

181.37
(281.74)

514.22
(351.89)

-172.38
(202.23)

376.22
(192.78)

158.46
(158.51)

337.61
(212.52)

84.39
(239.92)

435.89
(267.81)

268.69
(183.97)

971.36
(243.51)

Females
0

20+

a. The dependent variable is quarterly earnings. All models include demographic, heterogeneous displacement, and in-school controls as well as individual and period fixed effects. The models shown in columns 5 through 8 also include worker-specific time trends. Table entries are coefficients of indicator
variables equal to one during quarters after training participant left school and Group 1 and Group2 credits in ranges indicated. Models also include interactions of these indicator variables with the reciprocal of the number of quarters after workers left school. See section III of the text for detailed discussion of the specification. Robust standard errors are in parentheses.

Table 7: Estimates of Impact of Group 1 and Group 2 Community College Credits on
Earnings After Displacement But Prior To Community College Attendancea
Fixed Effects
Males
Pre-Collegeb

Females

-226.52
(99.32)

-132.94
(68.12)

Group 1 Credits*Pre-College

-5.09
(2.66)

Group 2 Credits*Pre-College

-8.05
(2.72)

Worker-Specific Time
Trends
Males

Females

-259.62
(88.68)

-116.75
(63.19)

-10.22
(3.26)

-3.85
(2.46)

-7.38
(2.79)

-8.56
(1.97)

-4.31
(2.43)

-6.75
(1.76)

a. The dependent variable is quarterly earnings. All models include demographic, heterogeneous displacement, and in-school controls as well as individual and period fixed effects. The models shown in columns 3
and 4 also include worker-specific time trends. See section III of the text for detailed discussion of the specification. Robust standard errors are in parentheses.
b. Indicator variable equal to one during quarters after training participant was displaced and before training
participant earned first community college credit.

40

Table 8: Fixed Effect Estimates of Impact of Group 1 and Group 2 Community College Credits on Hours Worked, Average
Wage Rate, and Earningsa
Log Hours
Males

Females

Log Wage
Males

Females

Log Earnings
Males

Females

Post-Collegeb

0.02568
(0.01087)

0.03126
(0.01150)

0.03003
(0.00807)

0.00714
(0.00716)

0.05570
(0.01261)

0.03840
(0.01328)

Post-College * 1/kc

-0.03723
(0.01765)

-0.06006
(0.01816)

-0.03088
(0.01142)

-0.01933
(0.01031)

-0.06811
(0.01892)

-0.07938
(0.01959)

Group 1 Credits*Post-College

0.00144
(0.00041)

0.00026
(0.00080)

0.00124
(0.00033)

0.00234
(0.00053)

0.00268
(0.00047)

0.00260
(0.00096)

Group 1 Credits*Post-College * 1/k

-0.00302
(0.00070)

-0.00250
(0.00110)

-0.00193
(0.00042)

-0.00214
(0.00069)

-0.00496
(0.00073)

-0.00464
(0.00122)

Group 2 Credits*Post-College

0.00161
(0.00061)

0.00144
(0.00060)

-0.00131
(0.00041)

-0.00094
(0.00035)

0.00030
(0.00072)

0.00051
(0.00067)

Group 2 Credits*Post-College * 1/k

-0.00528
(0.00100)

-0.00315
(0.00090)

-0.00175
(0.00054)

-0.00118
(0.00047)

-0.00703
(0.00110)

-0.00433
(0.00094)

a. Dependent variables are log quarterly hours, log average wage rate, and log earnings. All models include demographic, heterogeneous
displacement, and in-school controls as well as individual and period fixed effects. Samples are limited to those worker-quarters for which
the logarithms are defined. One academic year of community college schooling equals 45 credits. See section III of the text for detailed discussion of the specification. Robust standard errors are in parentheses.
b. Indicator variable equal to one during quarters after training participant left school.
c. 1/k denotes the reciprocal of the number of quarters after worker left school.

Table 9: Worker-Specific Time Trends Estimates of Impact of Group 1 and Group 2 Community College Credits on Hours
Worked, Average Wage Rate, and Earningsa
Log Hours
Males

Females

Log Wage
Males

Females

Log Earnings
Males

Females

Post-Collegeb

0.05034
(0.01620)

0.04538
(0.01698)

-0.03283
(0.01041)

-0.03502
(0.00990)

0.01751
(0.01726)

0.01036
(0.01805)

Post-College * 1/kc

-0.05571
(0.01970)

-0.06808
(0.01970)

0.01834
(0.01185)

0.01299
(0.01091)

-0.03737
(0.02039)

-0.05509
(0.02041)

Group 1 Credits*Post-College

0.00201
(0.00071)

0.00108
(0.00132)

0.00104
(0.00041)

0.00220
(0.00078)

0.00305
(0.00073)

0.00328
(0.00146)

Group 1 Credits*Post-College * 1/k

-0.00257
(0.00080)

-0.00164
(0.00143)

-0.00183
(0.00044)

-0.00142
(0.00078)

-0.00440
(0.00080)

-0.00306
(0.00153)

Group 2 Credits*Post-College

0.00001
(0.00093)

0.00186
(0.00084)

-0.00058
(0.00050)

-0.00035
(0.00048)

-0.00056
(0.00096)

0.00151
(0.00087)

Group 2 Credits*Post-College * 1/k

-0.00311
(0.00113)

-0.00286
(0.00099)

-0.00204
(0.00053)

-0.00144
(0.00051)

-0.00516
(0.00119)

-0.00430
(0.00101)

a. Dependent variables are log quarterly hours, log average wage rate, and log earnings. All models include demographic, heterogeneous
displacement, and in-school controls as well as worker-specific fixed effects and time trends and period fixed effects. Samples are limited to
those worker-quarters for which the logarithms are defined. One academic year of community college schooling equals 45 credits. See section III of the text for detailed discussion of the specification. Robust standard errors are in parentheses.
b. Indicator variable equal to one during quarters after training participant left school.
c. 1/k denotes the reciprocal of the number of quarters after worker left school.

Figure 1: Eamings of Trainees and Comparisons
Wot1c:ens Dleplaoec:l In 1881

..

!

8000
7000

•0'
:I

8000

l

15000

!

~

,

4000

0

i.

3000
2000

. - ......... ~''
..
..........

~

-----

••.
••

·v
•• ••••..
• ••

___________....

...··••····•·········

••

•

1000 -tr=:::;:::::::;=::;=r=;:::::;::::::;:::::;:::::::;=::;=r=;:::::;::::::;:::::;:::::::;=::;=r=;:::::;::::::;:::::;=::;::::::;l
' '
I '
' I
' ' I ' '
' I ' ' ' I
'
-8
-4
0
4
12
18
8

quarter relatlv8 to dlaplacement
PLOT

group 1 concentrators
oornpari110ne

43

•••••••· group 2 ooncentratore

Working Paper Series
A series of research studies on regional economic issues relating to the Seventh Federal
Reserve District, and on financial and economic topics.
Extracting Market Expectations from Option Prices:
Case Studies in Japanese Option Markets
Hisashi Nakamura and Shigenori Shiratsuka

WP-99-1

Measurement Errors in Japanese Consumer Price Index
Shigenori Shiratsuka

WP-99-2

Taylor Rules in a Limited Participation Model
Lawrence J. Christiano and Christopher J. Gust

WP-99-3

Maximum Likelihood in the Frequency Domain: A Time to Build Example
Lawrence J.Christiano and Robert J. Vigfusson

WP-99-4

Unskilled Workers in an Economy with Skill-Biased Technology
Shouyong Shi

WP-99-5

Product Mix and Earnings Volatility at Commercial Banks:
Evidence from a Degree of Leverage Model
Robert DeYoung and Karin P. Roland

WP-99-6

School Choice Through Relocation: Evidence from the Washington D.C. Area
Lisa Barrow

WP-99-7

Banking Market Structure, Financial Dependence and Growth:
International Evidence from Industry Data
Nicola Cetorelli and Michele Gambera

WP-99-8

Asset Price Fluctuation and Price Indices
Shigenori Shiratsuka

WP-99-9

Labor Market Policies in an Equilibrium Search Model
Fernando Alvarez and Marcelo Veracierto

WP-99-10

Hedging and Financial Fragility in Fixed Exchange Rate Regimes
Craig Burnside, Martin Eichenbaum and Sergio Rebelo

WP-99-11

Banking and Currency Crises and Systemic Risk: A Taxonomy and Review
George G. Kaufman

WP-99-12

Wealth Inequality, Intergenerational Links and Estate Taxation
Mariacristina De Nardi

WP-99-13

Habit Persistence, Asset Returns and the Business Cycle
Michele Boldrin, Lawrence J. Christiano, and Jonas D.M Fisher

WP-99-14

Does Commodity Money Eliminate the Indeterminacy of Equilibria?
Ruilin Zhou

WP-99-15

A Theory of Merchant Credit Card Acceptance
Sujit Chakravorti and Ted To

WP-99-16

1

Working Paper Series (continued)
Who’s Minding the Store? Motivating and Monitoring Hired Managers at
Small, Closely Held Firms: The Case of Commercial Banks
Robert DeYoung, Kenneth Spong and Richard J. Sullivan

WP-99-17

Assessing the Effects of Fiscal Shocks
Craig Burnside, Martin Eichenbaum and Jonas D.M. Fisher

WP-99-18

Fiscal Shocks in an Efficiency Wage Model
Craig Burnside, Martin Eichenbaum and Jonas D.M. Fisher

WP-99-19

Thoughts on Financial Derivatives, Systematic Risk, and Central
Banking: A Review of Some Recent Developments
William C. Hunter and David Marshall

WP-99-20

Testing the Stability of Implied Probability Density Functions
Robert R. Bliss and Nikolaos Panigirtzoglou

WP-99-21

Is There Evidence of the New Economy in the Data?
Michael A. Kouparitsas

WP-99-22

A Note on the Benefits of Homeownership
Daniel Aaronson

WP-99-23

The Earned Income Credit and Durable Goods Purchases
Lisa Barrow and Leslie McGranahan

WP-99-24

Globalization of Financial Institutions: Evidence from Cross-Border
Banking Performance
Allen N. Berger, Robert DeYoung, Hesna Genay and Gregory F. Udell

WP-99-25

Intrinsic Bubbles: The Case of Stock Prices A Comment
Lucy F. Ackert and William C. Hunter

WP-99-26

Deregulation and Efficiency: The Case of Private Korean Banks
Jonathan Hao, William C. Hunter and Won Keun Yang

WP-99-27

Measures of Program Performance and the Training Choices of Displaced Workers
Louis Jacobson, Robert LaLonde and Daniel Sullivan

WP-99-28

The Value of Relationships Between Small Firms and Their Lenders
Paula R. Worthington

WP-99-29

Worker Insecurity and Aggregate Wage Growth
Daniel Aaronson and Daniel G. Sullivan

WP-99-30

Does The Japanese Stock Market Price Bank Risk? Evidence from Financial
Firm Failures
Elijah Brewer III, Hesna Genay, William Curt Hunter and George G. Kaufman

WP-99-31

Bank Competition and Regulatory Reform: The Case of the Italian Banking Industry
Paolo Angelini and Nicola Cetorelli

WP-99-32

2

Working Paper Series (continued)
Dynamic Monetary Equilibrium in a Random-Matching Economy
Edward J. Green and Ruilin Zhou

WP-00-1

The Effects of Health, Wealth, and Wages on Labor Supply and Retirement Behavior
Eric French

WP-00-2

Market Discipline in the Governance of U.S. Bank Holding Companies:
Monitoring vs. Influencing
Robert R. Bliss and Mark J. Flannery

WP-00-3

Using Market Valuation to Assess the Importance and Efficiency
of Public School Spending
Lisa Barrow and Cecilia Elena Rouse
Employment Flows, Capital Mobility, and Policy Analysis
Marcelo Veracierto
Does the Community Reinvestment Act Influence Lending? An Analysis
of Changes in Bank Low-Income Mortgage Activity
Drew Dahl, Douglas D. Evanoff and Michael F. Spivey

WP-00-4

WP-00-5

WP-00-6

Subordinated Debt and Bank Capital Reform
Douglas D. Evanoff and Larry D. Wall

WP-00-7

The Labor Supply Response To (Mismeasured But) Predictable Wage Changes
Eric French

WP-00-8

For How Long Are Newly Chartered Banks Financially Fragile?
Robert DeYoung

WP-00-9

Bank Capital Regulation With and Without State-Contingent Penalties
David A. Marshall and Edward S. Prescott

WP-00-10

Why Is Productivity Procyclical? Why Do We Care?
Susanto Basu and John Fernald

WP-00-11

Oligopoly Banking and Capital Accumulation
Nicola Cetorelli and Pietro F. Peretto

WP-00-12

Puzzles in the Chinese Stock Market
John Fernald and John H. Rogers

WP-00-13

The Effects of Geographic Expansion on Bank Efficiency
Allen N. Berger and Robert DeYoung

WP-00-14

Idiosyncratic Risk and Aggregate Employment Dynamics
Jeffrey R. Campbell and Jonas D.M. Fisher

WP-00-15

Post-Resolution Treatment of Depositors at Failed Banks: Implications for the Severity
of Banking Crises, Systemic Risk, and Too-Big-To-Fail
George G. Kaufman and Steven A. Seelig

WP-00-16

3

Working Paper Series (continued)
The Double Play: Simultaneous Speculative Attacks on Currency and Equity Markets
Sujit Chakravorti and Subir Lall

WP-00-17

Capital Requirements and Competition in the Banking Industry
Peter J.G. Vlaar

WP-00-18

Financial-Intermediation Regime and Efficiency in a Boyd-Prescott Economy
Yeong-Yuh Chiang and Edward J. Green

WP-00-19

How Do Retail Prices React to Minimum Wage Increases?
James M. MacDonald and Daniel Aaronson

WP-00-20

Financial Signal Processing: A Self Calibrating Model
Robert J. Elliott, William C. Hunter and Barbara M. Jamieson

WP-00-21

An Empirical Examination of the Price-Dividend Relation with Dividend Management
Lucy F. Ackert and William C. Hunter

WP-00-22

Savings of Young Parents
Annamaria Lusardi, Ricardo Cossa, and Erin L. Krupka

WP-00-23

The Pitfalls in Inferring Risk from Financial Market Data
Robert R. Bliss

WP-00-24

What Can Account for Fluctuations in the Terms of Trade?
Marianne Baxter and Michael A. Kouparitsas

WP-00-25

Data Revisions and the Identification of Monetary Policy Shocks
Dean Croushore and Charles L. Evans

WP-00-26

Recent Evidence on the Relationship Between Unemployment and Wage Growth
Daniel Aaronson and Daniel Sullivan

WP-00-27

Supplier Relationships and Small Business Use of Trade Credit
Daniel Aaronson, Raphael Bostic, Paul Huck and Robert Townsend

WP-00-28

What are the Short-Run Effects of Increasing Labor Market Flexibility?
Marcelo Veracierto

WP-00-29

Equilibrium Lending Mechanism and Aggregate Activity
Cheng Wang and Ruilin Zhou

WP-00-30

Impact of Independent Directors and the Regulatory Environment on Bank Merger Prices:
Evidence from Takeover Activity in the 1990s
Elijah Brewer III, William E. Jackson III, and Julapa A. Jagtiani

WP-00-31

Does Bank Concentration Lead to Concentration in Industrial Sectors?
Nicola Cetorelli

WP-01-01

On the Fiscal Implications of Twin Crises
Craig Burnside, Martin Eichenbaum and Sergio Rebelo

WP-01-02

4

Working Paper Series (continued)
Sub-Debt Yield Spreads as Bank Risk Measures
Douglas D. Evanoff and Larry D. Wall

WP-01-03

Productivity Growth in the 1990s: Technology, Utilization, or Adjustment?
Susanto Basu, John G. Fernald and Matthew D. Shapiro

WP-01-04

Do Regulators Search for the Quiet Life? The Relationship Between Regulators and
The Regulated in Banking
Richard J. Rosen
Learning-by-Doing, Scale Efficiencies, and Financial Performance at Internet-Only Banks
Robert DeYoung
The Role of Real Wages, Productivity, and Fiscal Policy in Germany’s
Great Depression 1928-37
Jonas D. M. Fisher and Andreas Hornstein

WP-01-05

WP-01-06

WP-01-07

Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy
Lawrence J. Christiano, Martin Eichenbaum and Charles L. Evans

WP-01-08

Outsourcing Business Service and the Scope of Local Markets
Yukako Ono

WP-01-09

The Effect of Market Size Structure on Competition: The Case of Small Business Lending
Allen N. Berger, Richard J. Rosen and Gregory F. Udell

WP-01-10

Deregulation, the Internet, and the Competitive Viability of Large Banks and Community Banks WP-01-11
Robert DeYoung and William C. Hunter
Price Ceilings as Focal Points for Tacit Collusion: Evidence from Credit Cards
Christopher R. Knittel and Victor Stango

WP-01-12

Gaps and Triangles
Bernardino Adão, Isabel Correia and Pedro Teles

WP-01-13

A Real Explanation for Heterogeneous Investment Dynamics
Jonas D.M. Fisher

WP-01-14

Recovering Risk Aversion from Options
Robert R. Bliss and Nikolaos Panigirtzoglou

WP-01-15

Economic Determinants of the Nominal Treasury Yield Curve
Charles L. Evans and David Marshall

WP-01-16

Price Level Uniformity in a Random Matching Model with Perfectly Patient Traders
Edward J. Green and Ruilin Zhou

WP-01-17

Earnings Mobility in the US: A New Look at Intergenerational Inequality
Bhashkar Mazumder

WP-01-18

The Effects of Health Insurance and Self-Insurance on Retirement Behavior
Eric French and John Bailey Jones

WP-01-19

5

Working Paper Series (continued)
The Effect of Part-Time Work on Wages: Evidence from the Social Security Rules
Daniel Aaronson and Eric French

WP-01-20

Antidumping Policy Under Imperfect Competition
Meredith A. Crowley

WP-01-21

Is the United States an Optimum Currency Area?
An Empirical Analysis of Regional Business Cycles
Michael A. Kouparitsas

WP-01-22

A Note on the Estimation of Linear Regression Models with Heteroskedastic
Measurement Errors
Daniel G. Sullivan

WP-01-23

The Mis-Measurement of Permanent Earnings: New Evidence from Social
Security Earnings Data
Bhashkar Mazumder

WP-01-24

Pricing IPOs of Mutual Thrift Conversions: The Joint Effect of Regulation
and Market Discipline
Elijah Brewer III, Douglas D. Evanoff and Jacky So

WP-01-25

Opportunity Cost and Prudentiality: An Analysis of Collateral Decisions in
Bilateral and Multilateral Settings
Herbert L. Baer, Virginia G. France and James T. Moser

WP-01-26

Outsourcing Business Services and the Role of Central Administrative Offices
Yukako Ono

WP-02-01

Strategic Responses to Regulatory Threat in the Credit Card Market*
Victor Stango

WP-02-02

The Optimal Mix of Taxes on Money, Consumption and Income
Fiorella De Fiore and Pedro Teles

WP-02-03

Expectation Traps and Monetary Policy
Stefania Albanesi, V. V. Chari and Lawrence J. Christiano

WP-02-04

Monetary Policy in a Financial Crisis
Lawrence J. Christiano, Christopher Gust and Jorge Roldos

WP-02-05

Regulatory Incentives and Consolidation: The Case of Commercial Bank Mergers
and the Community Reinvestment Act
Raphael Bostic, Hamid Mehran, Anna Paulson and Marc Saidenberg

WP-02-06

Technological Progress and the Geographic Expansion of the Banking Industry
Allen N. Berger and Robert DeYoung

WP-02-07

Choosing the Right Parents: Changes in the Intergenerational Transmission
of Inequality  Between 1980 and the Early 1990s
David I. Levine and Bhashkar Mazumder

WP-02-08

6

Working Paper Series (continued)
The Immediacy Implications of Exchange Organization
James T. Moser

WP-02-09

Maternal Employment and Overweight Children
Patricia M. Anderson, Kristin F. Butcher and Phillip B. Levine

WP-02-10

The Costs and Benefits of Moral Suasion: Evidence from the Rescue of
Long-Term Capital Management
Craig Furfine

WP-02-11

On the Cyclical Behavior of Employment, Unemployment and Labor Force Participation
Marcelo Veracierto

WP-02-12

Do Safeguard Tariffs and Antidumping Duties Open or Close Technology Gaps?
Meredith A. Crowley

WP-02-13

Technology Shocks Matter
Jonas D. M. Fisher

WP-02-14

Money as a Mechanism in a Bewley Economy
Edward J. Green and Ruilin Zhou

WP-02-15

Optimal Fiscal and Monetary Policy: Equivalence Results
Isabel Correia, Juan Pablo Nicolini and Pedro Teles

WP-02-16

Real Exchange Rate Fluctuations and the Dynamics of Retail Trade Industries
on the U.S.-Canada Border
Jeffrey R. Campbell and Beverly Lapham

WP-02-17

Bank Procyclicality, Credit Crunches, and Asymmetric Monetary Policy Effects:
A Unifying Model
Robert R. Bliss and George G. Kaufman

WP-02-18

Location of Headquarter Growth During the 90s
Thomas H. Klier

WP-02-19

The Value of Banking Relationships During a Financial Crisis:
Evidence from Failures of Japanese Banks
Elijah Brewer III, Hesna Genay, William Curt Hunter and George G. Kaufman

WP-02-20

On the Distribution and Dynamics of Health Costs
Eric French and John Bailey Jones

WP-02-21

The Effects of Progressive Taxation on Labor Supply when Hours and Wages are
Jointly Determined
Daniel Aaronson and Eric French

WP-02-22

Inter-industry Contagion and the Competitive Effects of Financial Distress Announcements:
Evidence from Commercial Banks and Life Insurance Companies
Elijah Brewer III and William E. Jackson III

WP-02-23

7

Working Paper Series (continued)
State-Contingent Bank Regulation With Unobserved Action and
Unobserved Characteristics
David A. Marshall and Edward Simpson Prescott

WP-02-24

Local Market Consolidation and Bank Productive Efficiency
Douglas D. Evanoff and Evren Örs

WP-02-25

Life-Cycle Dynamics in Industrial Sectors. The Role of Banking Market Structure
Nicola Cetorelli

WP-02-26

Private School Location and Neighborhood Characteristics
Lisa Barrow

WP-02-27

Teachers and Student Achievement in the Chicago Public High Schools
Daniel Aaronson, Lisa Barrow and William Sander

WP-02-28

The Crime of 1873: Back to the Scene
François R. Velde

WP-02-29

Trade Structure, Industrial Structure, and International Business Cycles
Marianne Baxter and Michael A. Kouparitsas

WP-02-30

Estimating the Returns to Community College Schooling for Displaced Workers
Louis Jacobson, Robert LaLonde and Daniel G. Sullivan

WP-02-31

8