NORC at the University of Chicago
The Industrial Mobility of Displaced Workers
Author(s): Bruce Chelimsky Fallick
Source: Journal of Labor Economics, Vol. 11, No. 2 (Apr., 1993), pp. 302-323
Published by: The University of Chicago Press on behalf of the Society of Labor
Economists and the NORC at the University of Chicago
Stable URL: http://www.jstor.org/stable/2535283
Accessed: 21-06-2017 11:20 UTC
REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/2535283?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
NORC at the University of Chicago, Society of Labor Economists, The University of
Chicago Press are collaborating with JSTOR to digitize, preserve and extend access to Journal of Labor
Economics
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
The Industrial Mobility
of Displaced Workers
Bruce Chelimsky Fallick, University of California,
Los Angeles
This article uses a two-industry model of unemployment duration
and job search to estimate rates of transition of displaced workers
from unemployment to employment, distinguishing between employment
in a worker's previous industry and in other industries. The
competing-risks model allows inferences about search strategies to be
drawn from data concerning employment outcomes and allows tests
of some fundamental implications of search theory. There is evidence
that improvements in the prospects for employment in their previous
industry induce displaced workers to reduce search intensity or increase
reservation wages in other industries.
I. Introduction
Between 1981 and 1985, approximately 11 million U.S. workers permanently
lost their jobs (see Horvath 1987). Many such job losses stem
from structural changes in the demand for labor. These changes may include
general and permanent contractions of employment in specific industries,
and such contractions may affect a worker's ability to find another job
similar to the one that he lost. The ability and willingness of displaced
workers to obtain employment and to move between industries in response
to the economic incentives engendered by structural change play a major
role in determining the economy's ability to adjust smoothly to new circumstances.
Changes in market conditions create dissimilar employment
Financial support for this work was provided by the National Science Foundation.
I am grateful to Michael Wachter, Paul Taubman, Janice Madden, and Scott Mayfield,
among others, for helpful comments and advice.
[Journal of Labor Economics, 1993, vol. 11, no. 2]
? 1993 by The University of Chicago. All rights reserved.
0734-306X/93/1 102-0003$01.50
302
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 303
opportunities across industries. We expect displaced workers to adapt by
concentrating their efforts to find jobs in those industries that offer the
best prospects, even if this means switching industries. Industry job prospects
include both the likelihood of finding a job and the wage levels to
be expected in the industry.
Macroeconomic incentives to change industries are, however, tempered
by displaced workers' long-term attachments to their industries. Frequently,
they will have acquired skills specific to the industry, and the value of
those skills at their next job will be crucial to their economic well-being.
They may have established credentials in a particular industry or have
information that facilitates productive and efficient job search in that industry.'
In sum, a worker's old industry may offer very different prospects
for finding new jobs than other industries. Together with more general
economic conditions, a displaced worker may find, for example, that his
old industry offers high wages but few jobs, while other industries offer
many jobs at lower rates of pay. Taking into account all of these factors,
a displaced worker should optimally allocate his search efforts among
industries and thereby affect both the duration of his unemployment and
the likelihood of changing industries.2
As we shall see, a model of job search over multiple labor markets can
incorporate both the macroeconomic incentives that may favor changing
industries and the more person-specific incentives that make reemployment
in the same industry more attractive. In such a model, an unemployed
worker's search strategy adjusts to favor finding and taking a job in those
industries where more promising conditions prevail. This prediction is a
generalization of the implication of single-sector analytical search theory
that an improvement in labor market conditions should induce a worker
to search more intensively while becoming more particular about which
job to take. This is a fundamental prediction of search theory and as such
is widely used.3 Much research depends on the validity of this hypothesis,
but it has not been extensively tested. The two-industry model used here
greatly facilitates empirical testing of this theoretical prediction. By exploiting
a competing-risks structure of unemployment duration, the effect
of a change in employment prospects on search behavior can be distin'
Workers displaced from high-wage industries may be especially inclined to
allow their past experience to unduly influence their expectations. Of course, "wait
unemployment," in which workers accept unemployment in the hope of obtaining
a job in a high-wage industry despite the relative scarcity of such jobs, may be a
rational response to market conditions. If workers displaced from high-wage industries
can expect a better position in the queue for such jobs, then this "displaced
worker syndrome" may be reasonable.
2 The same can obviously be said of other sectoral distinctions, notably, occupations
and geographic regions. The current study is confined to industrial shifts.
3 For examples, see Mortensen (1986, pp. 864-65).
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
304 Fallick
guished from
ences about s
outcomes without data on search behavior itself.
This article also investigates more generally the unemployment and job
search experience of a large sample of displaced workers in the United
States. I estimate the determinants of the rates of transition of displaced
workers from unemployment (i) to new employment in the industry from
which they were displaced and (ii) to reemployment in another industry.
In addition to being a natural population in which to study issues of multisector
search, a sample consisting of only displaced workers avoids
econometric problems arising from the endogeneity of voluntary separations.4
It should be noted, however, that sample selection bias may arise
from other sources, such as the exclusion from the survey of those recalled
from (ex post) temporary layoffs.5
Katz (1986) employed a competing-risks framework to great advantage
in studying the effects on job search of the possibility of recall from temporary
layoff. He found that several variables that increase the likelihood
of recall also increase the hazard rate for taking a new job, implying that
workers adjust their search behavior according to their prospects for recall
to their old jobs. Another way to infer changes in search behavior without
direct measures is by examining the results of variation in an aspect of a
worker's economic environment that has no direct effect on duration. The
most prominent examples of this are the many studies that examine the
relationship between unemployment insurance and unemployment
duration.
The studies just mentioned all confirm the predictions of received theory
concerning searchers' responses in the context of variations in their individual
economic environments. The present work extends this line of research
to include searchers' responses to variation in sectoral economic
conditions.
In this study I find evidence that displaced workers adjust their search
behavior in response to their prospects for new employment in each in'
Voluntary separations are obviously endogenous to the worker. By construction,
the Displaced Worker Surveys exclude anyone who quit her job, except that it
includes those who quit explicitly in anticipation of an involuntary separation.
This article proceeds under the assumption that the advance notice (or other information)
that makes such quits possible is provided or withheld exogenously.
Similarly, the advance notice variable is treated as uncorrelated with unobserved
heterogeneity in duration.
5 Both theoretical search models and empirical studies of the duration of unemployment
have tended to model search in a single market. Thomas and Bernhardt
(1991) is an exception. Several studies of the wage losses of displaced workers
(Jacobson 1978; Podgursky and Swaim 1987; Madden 1987, 1988; Addison and
Portugal 1989; and Kletzer 1991 ) recognize that changing industries or occupations
matters but do not examine the implications of this fact for search behavior.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 305
dustry and reemployment at their o
from industries with smaller aggreg
strategies that contribute to greater
only are more likely to end up in a job in a different industry but will
more quickly find acceptable jobs there. An additional finding is that the
determinants of the rate of reemployment in a worker's previous industry
differ from those for reemployment in other industries. Variables representing
the cause of displacement and measuring regional economic conditions
are more important determinants of the rate of reentry into the
worker's previous industry, while education, job tenure, gender, advance
notice, and wage premia are more important for entering other industries.
Personal characteristics such as the level of family income and heading a
household have large effects on both reemployment rates.
II. Theoretical Framework
Consider an individual with characteristics Xi who has been displa
from a job and is unemployed at the beginning of period t. The worker
has available an endowment of search "intensity" that he may devote to
seeking a job during this period or to other activities. "Intensity" may
refer to time, resources, or effort. For simplicity, think of it as the available
time that may be devoted to search. The worker may allocate the time he
devotes to search between two sectors of the economy: the industry from
which he was displaced (which I shall call the "old" industry) and the set
of all other industries (which I shall designate the "new" industry).6
Let the proportion of available time during period t that the worker
devotes to search in industry J be denoted by sj(t, Xi), where j = old, new.
Let the length of the period be sufficiently small that the probability of
receiving more than one job offer in a single period can be assumed to be
zero. If the worker were to devote all of his available time in period t to
searching in industry j, then the (marginal) probability that he would
obtain a job offer in period t from industry j is aj(t, Xi), which may loosely
be called the offer-arrival rate. Assume further that in general the (marginal)
probability that the worker will receive an offer in period t from industry
j is aj(t, Xi)G(sj(t, Xi)), where c6 is an increasing concave function with
6(0) = 0 and c6(1) = 1. The rate of current compensation associated with
a job offer (which I will denote as w and refer to simply as the wage) is
randomly drawn from a (marginal) distribution Fj(w; Xi).
Assume that for each industry j in each period t, the worker has a
reservation wage w(t, Xi) such that the worker will accept a job offer in
industry j with wage w if and only if w 2 w>(t Xi). Reservation wages
6 Compare the model of Thomas and Bernhardt (1991), wherein a person may
search in only one sector at a time.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
306 Fallick
and search
influenced by the a and F in both industries in addition to t and X.. In
this structure, the probability that a worker who was unemployed at the
beginning of period t will make a transition from unemployment to employment
in industry j during period t (i.e., the hazard function) is
hi., (t) = aj(t, X.) c(sj(t, X.)) [I -Fj(wjr(t, Xi.); Xi.)].(1
Equation (1) makes clear how changes in employment prospects (specifically,
the offer-arrival rates and wage-offer distributions) would affect
the hazard rates, holding search strategies constant, and how changes in
the elements of the search strategies would affect the hazard rates, holding
the employment prospects constant. If the search strategies were not to
change, that is, if sj and w were held constant, an increase in the offerarrival
rate aj or an improvement in the wage-offer distribution Fj would
increase the hazard rate h. , and reduce the duration of unemployment. An
increase in oaj or an improvement in Fj can also be expected to increase sj
and w, but such reactions are not necessary in order to induce a positive
relationship between, say, aj and hij. For this reason, a one-sector model
does not yield testable hypotheses about the influence of industry conditions
on search strategies in the absence of information on search behavior itself.
In such a model, any observed change in hi,, associated with a change in
industry conditions (aj or Fj) must be attributed jointly to the conditions
themselves and to their influence on search behavior, not separately to one
or the other.
In contrast, changes in the prospects for employment in the "other"
industry, Ctk or Fk, where k # /, would not affect h.,j at all if the search
strategies did not change. For example, any relationship between, ctk and
hij must be due to the way in which search in industry j reacts. I exploit
this consideration in order to make inferences about search behavior from
data on reemployment outcomes and to disentangle the indirect effect of
a change in employment prospects on hazard rates (i.e., the induced change
in the search strategy) from the more obvious direct effects. In other words,
the two-sector model yields testable hypotheses that its one-sector counterpart
cannot.7
It is intuitively sensible that any improvement in the prospects for employment
in industry k should lead to a reduction in search intensity de-
7The class of models that exhibit the property that conditions in one sector
affect only behavior (strategies) in the other sector includes many which are not
conventionally called search models. They all, however, can be represented by
something similar to eq. (1) and have the same essential character. In a rational
queuing model, for instance, the offer-arrival rates might be exogenous functions
of previous tenure, etc., but the reservation wage for one industry would still react
to the prospects in the other.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 307
voted to search in industry j and an
jobs in that industry. Fallick (1992) d
of job search in two sectors. The una
fall in response to an improvement in conditions in industry k (i.e., an
improvement in either atk or Fk), due to the induced changes in search
behavior. In contrast, while an increase in aj or an improvement in Fj
would probably lead to an increase in hi,1, the increase in wj that either
one would induce precludes a definitive prediction.8
A reduced form of the hazard function (1) may be written
hij(t) = g(t, Xi, common components of labor market (2)
conditions in the old and in the new industries),
where t is the number of periods of unemployment thus far, and by "common
components" I mean that part of the labor market conditions in each
industry (such as a and F) that are common to all individuals. Which
specific industries make up the "old" and "new" industries depends on
the individual's history. I estimate a version of the reduced-form hazard
function (2) for each industry.
III. Data
My sample is drawn from the Bureau of Labor Statistics (BLS) Disp
Worker Survey attached to the Current Population Surveys (CPS) of
uary 1984 and 1986. The survey was administered only to those indiv
in the CPS who were at least 20 years old and lost a job sometime du
the 5 years prior to the survey due to a plant closing, a layoff from
8 For some classes of improvements, a definitive prediction can be m
Fallick ( 1992). In brief, in each period, the unemployed worker chooses the elem
of his search strategy. Each of the reservation wages is set so as to equate the
expected value of remaining unemployed next period with the value of becoming
employed in that industry at that wage next period. Search intensity is allocated
so that the marginal value of additional search in each industry is equal to the
marginal cost of searching. Any change that raises the value of being unemployed
more than it raises the value of being employed in industry j at the reservation
wage will induce the worker to raise his reservation wage for jobs in that industry.
Any change that increases the marginal "productivity" of search in either industry
relative to the marginal cost of searching will induce a worker to increase his search
intensity overall, and any change that increases the marginal value of search in
industry j relative to that of search in industry k will induce the worker to search
more intensively in industry j relative to industry k. Therefore, an improvement
in the wage-offer distribution or offer-arrival rate in the old industry should cause
reservation wages to rise and search intensity to be reallocated away from the new
industry and toward the old industry.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
308 Fallick
Table 1
Distribution of Independent Variables
Variable Mean SD Variable Mean SD
EMPOLD .0188 .056 SUPPORT .22 ...
EARNOLD 329.08 65.09 CRAFT .20 ...
UNEMPOLD 9.64 3.52 LABORER .34 ...
EMPNEW .0151 .0263 FARMING .015 ...
EARNNEW 312.73 14.92 SERVICE .067
UNEMPNEW 8.40 1.25 AGE 36.4 11.4
SIZERAT 13.2 15.6 FEMALE .31 ..
NOTIFIED .53 ... FAMINC 5.91 3.15
SLACK .47 ... HEAD .68 ...
ABOLISH .15 ... STUNEMP 8.5 2.4
EDUC 12.2 2.6 WHITE .87 ...
PREMIUM .11 .65 UI .62 ...
TENURE 5.7 6.1 YEAR86 .48
New England .68 . .. MidAtlantic .13 ...
EN. Central .15 ... W.N. Central .10 ...
S. Atlantic .13 ... E.S. Central .06 ...
Mountain .10 ... Pacific .14 ...
NOTE.-N = 2,090. SD = standard deviation.
they were not recalled, or a similar reason.9 Call that job the worker's old
job. I include in the sample only workers who were employed full-time
at their old jobs for at least 1 year, were in the labor force at the time of
the survey or reported that they wanted a job, and last worked at their
old jobs (i.e., lost their old jobs) in the year prior to the survey. In addition,
only workers who reported that their old jobs ended due to a plant closing
or relocation, slack work, or the abolition of the position or shift were
included.'0 The characteristics of the members of the sample are summarized
in table 1.
The duration measure used is the number of weeks in which the individual
was without a job during the (at most) 1 year between displacement
and the survey. This may include time spent officially out of the labor
force rather than unemployed. Therefore, "nonemployment" may be a
more accurate term than "unemployment." Moreover, the survey asks only
See Flaim and Sehgal (1985) and Horvath (1987) for descriptions of the data.
The involuntary nature of the separations rules out the problems of selectivity that
concern much of the literature on voluntary turnover. Ideally, the sample would
include only clearly permanent separations, which would also eliminate the possibility
of recall to one's old job from influencing search behavior. As noted below,
however, while the Displaced Worker Surveys excluded anyone who had been, in
fact, recalled, the data do not live up to this ideal.
'0 I also included only those workers who were not missing relevant data, did
not report impossible values for important variables, and were not in the armed
forces at their old jobs. On these points, and for a further discussion of the sample
and problems with the data, see Fallick ( 1988).
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 309
for the number of weeks that an individual has been without a job since
becoming displaced, but the current job is not necessarily the first job since
displacement. The more time there has been for other jobs to intervene,
the less accurate the comparisons of people's inferred nonemployment
experiences will be. I have restricted the sample to those displaced no more
than 1 year prior to the survey in order to reduce inaccuracies in the
measured durations of nonemployment spells. A shorter period prior to
the survey also reduces the amount of simple recall error and bias in
the data. "
For this study, I have defined industries according to the 22 major industry
groups defined in the January CPS. They are listed in table 1. The
worker's "old" industry is the industry group of his old job. The "new"
industry is defined as the other industry groups taken together. Thus, the
identity of the old and new industries depends on the individual's history.
In addition to the data provided in the Displaced Worker Survey, I
constructed two variables meant to reflect the labor market conditions in
each industry: the rate of employment growth and median weekly earnings
in each industry during the relevant year for each worker.'2 They were
constructed using the full CPS for March 1983-86, each of which comprised
approximately 60,000 households. These variables are described in more
detail below. Industry unemployment rates are also used.
For the worker's old industry, constructing these variables was straightforward.
A worker's new industry, however, comprises several industry
groups, each of which may be relevant to his search for a job to a different
degree. In particular, the conditions prevailing in those industries that are
"close" to his old industry, in the sense that he is likely to find attractive
work there or can transfer there much of the human capital that he has
acquired in his old industry, are more likely to influence his search behavior
and the outcome of his efforts than are conditions in more "distant" industries.
Therefore, each of the variables reflecting the market conditions
in the new industry is a weighted average of the conditions in each of the
major industry groups of which it is composed.
The weights were constructed from data on workers' transitions between
industries from the March CPS Work History data from 1983 to 1988,
" Studies that use the same data sources but concentrate on wage changes often
include workers displaced up to 5 years prior to the survey.
12 For workers displaced during 1983, data from the March 1983 and 1984 CPS
were used; for workers displaced during 1985, the March 1985 and 1986 CPS were
used. Since search strategies depend on expected industry conditions (both current
and future), the question arises of the period to use in constructing industry variables.
It may be appropriate to include an industry trend, for example. My thinking was
that following a deep recession, and given the tendency for industrial declines to
manifest themselves primarily during recessions, current industry conditions would
be best.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
310 Fallick
using the procedure outlined in Shaw (1984, 1987). Essentially, two industries
are deemed to be close to each other if workers leaving jobs in
those two industries tend to move to the same other industries.13 Direct
flows between the two industries are not emphasized since they are more
likely to reflect temporal shifts in demand. Thus, the weights reflect the
relevance of an industry to the workers' search for jobs, whether due to
some intrinsic similarity between the industries, such as the transferability
of knowledge, or a circumstantial connection, such as geographic
proximity.
IV. Estimation Procedure
I assume that transitions from nonemployment to employment occur
as part of a search process in continuous time. The hazard functions in
(1) and (2) should be reinterpreted accordingly. I further assume that the
rates of transition follow the proportional hazards model. The hazard rate
of individual i for transitions from nonemployment to employment in
industry j at duration t is
hi.,(t) = ho,1(t)exp(X' 13,), (3)
where ho0j is a baseline hazard rate that may vary with t
of individual characteristics and conditions in the old and
and f,3 is a vector of coefficients on the Xi for transition
Since the baseline hazard function is not of interest in t
a specification that does not require assumptions about th
acter of the baseline hazard is best. This avoids the biases that such assumptions
can impart to the estimates of the rest of the function, at the
cost of discarding some of the information in the data. Therefore, I estimated
the coefficients f30d and fnew using a competing-risks version of
Cox's partial likelihood model (see Cox and Oakes 1984), which uses
information on the order of events only, not the times at which they occur.
Instead of the likelihood that each observation experiences the exit (from
unemployment) that we observe at the time we observe, the Cox model
deals with the likelihood that the observations exit in the order that we
observe, conditional on the observed timing of the exits. For each observed
"failure" time t, the likelihood that observation m should exit to industry
j given that someone exited to industry j is
ho, -(t) exp (X' Pj3)
I ho, -(t) exp (X'. Pj)
IER,
13 Details are available from the author.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 311
where R, is the set of observations s
baseline hazard rates cancel out, leaving us free to estimate the coefficients
,3.
The data report the duration of nonemployment in weeks, rather than
more specific actual durations. Consequently, there are numerous ties
(workers reporting the same number of weeks without a job) in the data,
which were handled using the approximation suggested by Breslow (1974).
The estimation was performed using BMDP software. The analysis was
repeated using both fully parametric and semiparametric specifications.14
The results reinforced those presented below and are not shown.
A comment is in order about the assumed independence of the hazard
rates in the Cox model. The estimation procedure just described does not
explicitly take account of possible stochastic dependence between the two
hazard rates, hold and hnew. Consequently, the estimated hazard rates should
be interpreted as the transitions rates conditional on the worker not yet
having exited into either industry."5 The alternative interpretation is that
each hazard rate is the transition rate conditional only on not having yet
exited into that particular industry, as if no other exits were possible."6
(The two interpretations are equivalent only under the assumption that
the latent failure times are independently distributed. See Cox and Oakes
1984, pp. 144-45.)
The search model does not rule out the possibility of stochastic dependence
between the two hazard functions given the information available
to the worker, so it would seem appropriate to estimate the hazard rates
without explicitly accounting for possible correlation and adopt the first
interpretation. However, there may be variables known to the worker but
unobserved by the researcher that affect both hazard rates and introduce
to the empirical model a source of correlation that is not present in the
search model. In order to account for such correlation, I have also estimated
a sequential multinomial probit model that allows for unrestricted cor'"The
semiparametric model was similar to that proposed by Prentice and
Gloeckler (1978).
15 These are sometimes called the "cause-specific" hazard rates and may be expressed
as
h ,1(t) = lim pr(t c Ti, < t + x I Tiold 2 t, Tinew 2 t)
X-O ~~~X
16 These are sometimes called the "counterf
as
hi~j(t) =limpr(t?Ti ?t+xlTi, 2 t)x-1so X
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
312 Fallick
Table 2
Estimates from the Cox Model
Variable hold Variable hold new
EMPOLD 5.4*** -4.7** SUPPORT .40*** -.12*
(2.3) (-2.0) (2.5) (-1.0)
EARNOLD -.0029* .0025* CRAFT .50*** .27*
(- 1.7) (1.8) (3.0) (-1.9)
UNEMPOLD .014 -.029 LABORER .32* -.23*
(.4) (-1.1) (1.9) (-1.7)
EMPNEW 32 -44 FARMING .76* -.031
(1.0) -(1.6) (1.7) (.1)
EARNNEW .14** .16*** SERVICE .14 .14
(-2.2) (3.3) (.5) (.8)
UNEMPNEW -.49 -.15 AGE -.016*** -.013***
(-.7) (-.3) (-3.1) (-2.9)
SIZERAT -.0093** .0012 FEMALE .11 .27***
(-2.0) (.4) (.8) (2.6)
NOTIFIED .04 .18** FAMINC .11*** .066***
(-4) (2.2) (6.6) (4.5)
SLACK -.44*** -.28*** HEAD -.63*** -.40***
(-4.2) (-3.0) (4.9) (3.8)
ABOLISH -.27** .055 STUNEMP -.090*** -.021
(-1.9) (.5) (-2.8) (-.8)
EDUC .014 .052*** WHITE -.55*** .70***
(.6) (2.8) (-3.2) (4.7)
PREMIUM .066 -.12* UI -1.2***
(.9) (-1.7) (-10.9) (-13.6)
TENURE -.0014 -.023*** YEAR86 -.14 .010
(-. 1) (-2.5) (-1.2) (.1)
Log likelihood -3,167 -4,318 _
NOTE.-Dummy variables for nine geographic regions were included but are not shown. tthe
coefficients appear in parentheses.
* Significantly different front zero at the 10% level.
** Significantly different from zero at the 5% level.
*** Significantly different from zero at the 1O/o level.
relation across risks. (See Sueyoshi [1991] for a discussion of binary
models as an alternative to the proportional hazards model.) The es
were qualitatively similar to those presented here."7
V. Results
In this section, I describe the explanatory variables and present their
estimated effects. The Appendix briefly defines the explanatory variables.
Table 2 presents the estimates of the Cox model for the sample.
A. Information and Market Conditions
One of the major themes of the theory of job search is that workers
should adjust their search behavior in response to information about their
17The estimated correlation coefficient was approximately 0.1. One could not
reject the null hypothesis of no correlation using a likelihood ratio test at standard
levels of significance. For this reason, and in order to retain greater comparability
with the literature, I report the proportional hazard estimates.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 313
prospects for employment in the re
cation of this theme is the concern
from "declining" industries move in
compared to workers displaced from industries that are doing better. To
reiterate, conditions in the old industry will affect the outcomes of the
search process by affecting workers' search strategies as well as by affecting
the distribution of outcomes which results from any given search strategy.
Within a single labor market, these two effects are confounded. When two
labor markets are examined, however, hnew(t) is unaffected by the conditions
in the old industry except through their effect on search strategies. Therefore,
for example, if an increase in cold is associated with any change in
hnee (holding other conditions constant), then we can infer that workers
change their search strategies in response to cold.
The industry variables are as follows: The rates of growth over the year
of full-time employment in each of the two industries are measured by
EMPOLD and EMPNEW. The variable EMPOLD is meant to serve as a
proxy for that part of aold which is specific to the old industry rather tha
to the individual. It is intended to answer questions like the following:
Holding the wage-offer distribution constant, do workers displaced from
industries characterized by lower offer-arrival rates adjust their search behavior
so as to increase hnew relative to workers displaced from more forthcoming
industries? According to the analysis in Section II, the answer
should be yes.
The median weekly earnings of full-time employees in each industry,
averaged between the two relevant surveys, are EARNOLD and EARNNEW.'8
EARNOLD is meant to capture that part of the expected wage
offer that is common to all workers from an industry, while EARNNEW
reflects the expected wage that workers from that industry can expect
elsewhere.'9 A better wage-offer distribution in one industry ought to reduce
the hazard rate for reemployment in the other industry by making jobs
there, and therefore search as well as acceptance of an offer, less attractive.20
18 Using mean weekly earnings rather than the median does not appreciably alter
the results. The rate of increase of median full-time earnings in the two industries
over the year was also tried, without effect.
19 Variables meant to reflect the spread of the wage-offer distributions, the variance
of weekly earnings, and the quartile range were too highly correlated with mean
or median earnings to separately estimate their effects. Only the median was included
in the reported results. Since a worker's search strategy truncatesf(w) from below
at Wr so that only the upper part of the distribution matters, an increase in the
standard deviation off (w) holding the mean constant is in most cases an improvement
from the point of view of the worker. Therefore, the median earnings can
still be interpreted as a measure of the "goodness" of the distribution.
20 Warner, Poindexter, and Fearn (1980) found that the predicted expected wage
for a worker (based on an estimated earnings function) was positively related to
the reservation wage and negatively related to the duration of unemployment spell
given the reservation wage.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
314 Fallick
The variables OLDUNEMP and NEWUNEMP are the industry unemployment
rates, as reported by the BLS. They are meant to capture slacknes
in the industries' labor markets.
The estimates in table 2 imply that EMPOLD is significantly and positively
related to hold, while it is significantly and negatively related to hnew.
It is no surprise that the rate of growth of employment in the worker's
old industry is estimated to increase the rate at which he goes to work
there, but this need not be due to any change in search behavior on his
part. At the same time, a larger rate of employment growth in his old
industry also decreases the rate at which he becomes employed in another
(the new) industry, and this does imply something about behavior. It appears
that workers do respond to lower offer-arrival rates in the old industry
by increasing their search intensity or decreasing their reservation wages
in the new industry. The order of magnitude of these effects may be indicated
by the predicted average effect of a change in EMPOLD on the
expected latent durations of unemployment. For becoming reemployed in
the old industry, a 1 percentage point increase in EMPOLD is estimated
to decrease the latent duration by approximately 1 week. For becoming
reemployed in the new industry, a 1 percentage point increase in EMPOLD
is estimated to increase the latent duration by approximately 1 week.21
The estimated coefficients on EMPNEW run contrary to expectations,
indicating that greater employment growth in the new industry increases
21 The latent durations (or latent failure times) may not be behaviorally meaningful
in this context, but they do provide an easily compared indication of the magnitudes.
For grouped duration data, the expected latent duration of unemployment for
reemployment in industry j is
Et] = Ad t[Si(t - 1)- V
t=1
where
Sj(t) = exp[ - hj(s)dsl
is the survival probability for transitions into industryj. For a proportional hazards
model,
dEt 0
d= = D E: t[Sj(t - 1 )ln Sj(t - 1 ) - Sj(t)lin Sj(t)] .
dX t=
The summation was calculated using life-table hazard rates. This
must be multiplied by the coefficient to yield the derivative, is -19
-22.32 for Etnew.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 315
search in the old industry and reduc
ployed in the new industry. However
have correspondingly large standard
to changes in the specification.22 The estimates for EMPOLD, in contrast,
were robust to the same changes.
The estimates for EARNOLD are similarly unexpected in sign, while
the estimates for EARNNEW have the expected signs but are suspiciously
large.23 For reemployment in the old industry, a $10 increase in weekly
earnings in the old industry (EARNOLD) is predicted to increase the
expected latent duration by .57 weeks, while a $10 increase in EARNNEW
increased the expected latent duration by 27.5 weeks. For reemployment
in the new industry, a $10 increase in EARNOLD is predicted to decrease
the expected latent duration by .56 weeks, while a $100 increase in EARNNEW
decreased expected latent duration by 35.7 weeks. The coefficients
on the industry unemployment rates are insignificantly different from zero
for both hazard rates.
One possible cause of the counterintuitive and nonrobust estimates for
several of the industry conditions variables is that the new industry is
defined as the complement of the old industry. For a given year, for example,
aggregate employment (the sum for both the old and the new
industries) is constant across old industries. By construction, then, there
is a strong negative relation (in fact, a one-to-one mapping) between each
of the variables for the old industry and its counterpart for the new industry
within each year. The estimates may confound their effects. To supplement
the results in table 2, I regressed each of the constructed industry variables
on its counterpart and a dummy for the year and used the residuals from
these equations to reestimate the model. The results appear in table 3. The
estimates for the other variables are unchanged and not included in the
table. The coefficients on the industry variables' residuals conform to expectations
about workers' responses, but the coefficients on the new industry
variables remain very large in magnitude. Only the estimates for
the growth rate in the old industry (EMPOLD) are robust to this trans-
22 Conditional on becoming reemployed in the old industry, an increase in EMPNEW
of 1 percentage point is predicted to decrease expected duration by 6.3
weeks. Conditional on becoming reemployed in the new industry, the same change
in EMPNEW would increase expected duration by 9.8 weeks.
23 It has been suggested that EARNOLD may reflect the degree of unionization
in the industry and therefore be correlated with the worker's union status, for
which I cannot control directly. Union status may matter because, in the words
of one unemployed worker, "nonunion employers . . . were reluctant to hire a
union man, believing he would be unhappy with the lower wage they offer" (Serrin
1986). Controlling for the rate of unionization in the old industry group did not
change the results significantly.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
316 Fallick
Table 3
Residuals for Industry Variables, Cox Model
Variable hoId hnew
EMPOLD 8.8* _4.8*
(2.7) (- 1.9)
EARNOLD -.0006 -.0025*
(-.4) (-1.8)
UNEMPOLD .042 -.040
(1.0) (-1.3)
EMPNEW -119*** 38
(-2.4) (1.0)
EARNNEW .15**.21***
(-2.1) (3.9)
UNEMPNEW -1.2 .48
(-1.4) (.8)
Log likelihood -3,167 -4,318
NOTE.-t-statistics for the coefficients appear in p
* Significantly different from zero at the 10% le
** Significantly different fromn zero at the 5% l
Significantly different from zero at the 1% lev
formation. This was also the one variable that was robust to changes in
specification under the original formulation.
These results offer some support for the claim that workers adjust their
job-search behavior in response to conditions in the labor markets in which
they may search. The one set of coefficients that appears to be robust to
changes in the specification and to the transformation, EMPOLD, indicates
that workers from relatively declining industries direct their search efforts
toward other industries.
Closer to home, the prospects for continuing at the old job may affect
search behavior before the job ends. The variable NOTIFIED in table 2
indicates that workers who knew or expected in advance that their jobs
would end had significantly higher hazard rates for reemployment in the
new industry than those who did not know, all else being equal.24
Similarly, a worker whose job was lost due to a plant closing or moving
(the omitted job loss category) will have little or no hope of being recalled
or rehired to his old job. A person whose job was lost due to the abolition
of his position or shift (ABOLISH = 1) may have some hopes, while a
person's whose reason was "slack work" (SLACK = 1) may have some
24 Note, however, that work by Ehrenberg and Jakubson (1989) and Addiso
and Portugal (1987) indicate that the primary effect of advance notice is to increase
the probability of zero unemployment, with little effect on the hazard rate once
unemployment is experienced. Moreover, Fallick (1991) and Ruhrm (1992) find
that advance notice is endogenously determined.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 317
reason for optimism (although no one in this sample had in fact been
recalled), and his prospective employers may fear the same possibility.25
Alternatively, workers may be more likely to know in advance that the
plant is scheduled to close than that a position is scheduled to be abolished
and more likely to expect either of those events than a layoff due to slack
work. The coefficients on SLACK and ABOLISH are consistent with either
of these views. The value hold is significantly lower for those laid off due
to slack work than for those whose position or shift was abolished, which
in turn is lower than hold for those whose plant closed or moved. The value
hnew is similar, but there is no significant difference indicated between plant
closing and abolition of shift.
B. Human Capital
Human capital considerations clearly have the potential to advance or
retard industrial mobility, depending on whether or not it is specific to
the old industry.26 The variable EDUC measures the number of grades of
schooling completed. It is significantly positive for hnew and insignificantly
different from zero for hold.27 The difference between these coefficients is
not statistically significant, but it suggests that education may do more to
improve prospects in the new industry than in the old. Education may be
a less informative indicator of a prospective worker's productivity in the
old industry, where superior indicators such as the worker's work history,
recommendations, and so forth from that industry are available. Alternatively,
education may increase or signal the worker's ability to learn,
which would be more important in the new industry than in the old.
The variable PREMIUM is the percentage by which the worker's earnings
in his or her old job exceeded the industry average, which is intended
to capture industry-specific human capital or match quality. The coefficient
on PREMIUM is significantly negative for hnew while insignificantly positive
for hold. This suggests that the effect of the wage premium, possibly the
25 Barron and Mellow ( 1979) found that better prospects for recall reduced the
amount of time devoted to search.
26 For reemployment in general, Mincer and Ofek (1982), among others, have
also argued that at higher levels of education human capital depreciates more rapidly,
so that individuals with more education have more incentive to keep the duration
of nonemployment short. Kiefer and Neumann (1979) found that more education
increases the reservation wage, while Barron and Mellow (1979) found that it
increases the amount of available time devoted to search.
27 The coefficients on the education variables were sensitive to specification of
the hazard function, especially to the inclusion or exclusion of an AGE. I report
what I consider to be the most reliable estimates. The results are similar if dummy
variables for less than 12, 12, 13-15, and at least 16 years of schooling are used
instead of the continuous variable EDUC.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
318 Fallick
premium itself, will not carry over into a new industry, although it may
carry over into a new job in the old industry.28
Job tenure is another proxy for firm-specific human capital, and in the
absence of data on industry tenure may represent industry-specific human
capital as well. If so, then we would then expect tenure to increase hold
and decrease hnew 29 The variable TENURE is significantly negative in the
equation for hnew, while insignificantly different from zero in the equation
for hold. For reemployment in the new industry, 1 more year of tenure is
estimated to increase the expected latent duration by about 1/2 week.
The discussion of this section relies, of course, on the assumption that
different jobs within an industry are sufficiently similar that the human
capital acquired at one job will likely carry over to another job in the same
industry to a greater extent than it will carry over to a job in another
industry.30 While a finer level of aggregation of industries than the 22
industry groups used here would lend more credence to this assumption,
considerations of sample size led me to use the broader definition. If,
however, the old industry is defined by the two-digit Standard Industrial
Classification (SIC) code of the old job, the coefficient on tenure at the
old job and on the intraindustrial wage premium do increase in magnitude
in the hazard function for employment in the old industry. They exhibit
no noticeable change in the other hazard function, and the coefficients on
the education variable change little in either function.
C. Occupations
To the extent that occupation-specific human capital is important and
that particular occupations tend to be employed in particular industries,
a worker's occupation may affect the two hazard rates differently. A set
of dummy variables controlled for the occupation of the worker. The
coefficients on a set of dummy variables for the worker's occupation at
his old job (by census major occupation code) indicate that the omitted
category of managerial and professional occupations have lower hold than
28 Together with EARNOLD, PREMOLD makes the inclusion of the worker's
weekly earnings at the old job redundant. Using those earnings instead of PREMOLD
only complicates interpretation.
29 Also, permanent job loss may come as more of a surprise to workers with
more seniority at a firm, and they may be more optimistic about the possibility of
being recalled. Therefore, they may begin to search later and with lesser intensity
than their less tenured co-workers. See Leighton and Mincer (1982). Tenure and
age are highly correlated, so one may wish to interpret the coefficients cautiously.
30 The discussion of industry conditions, in contrast, requires only that the industries
define identifiable sectors of the economy that a worker may see as differing
in job prospects and among which one may allocate search effort. We can confidently
assume that a worker coming from a job in, say, durable manufacturing will see
that sector as distinct from, say, retail trade.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 319
other occupations (excepting service occupations) and higher hnew than
production occupations. More detailed analysis (not shown in the tables)
indicated that within that group it is the executive, administrative, and
managerial occupations, as opposed to professional specialties, that differ
when it comes to reemployment in the old industry. For the new industry,
salespeople and production occupations appear to have lower hazard rates
than other occupations. These occupations may involve a significant amount
of industry-specific human capital. Managers and executives, however, appear
to have less industry-specific or more general skills and knowledge.
D. Demographic Characteristics
The demographic characteristics included do not appear to greatly influence
industrial mobility, although they do have large effects on the
overall rate of reemployment. While the coefficient on FEMALE indicates
greater mobility for women, on further analysis this turns out to be entirely
attributable to differences in the occupational distributions of men and
women. An exception may be "usual" family income (represented by FAMINC),
which has a positive effect on both hazard rates but a significantly
larger effect on hold than on hnew. Considering that the worker's earnings
on the old job are controlled for by EARNOLD and PREMIUM (the
result remains if the worker's earnings are entered directly), I interpret
FAMINC as representing income from other family members or assets.
The availability of these resources makes any direct costs of search less
onerous and so may increase search intensity. At the same time, such "nonlabor"
income may reduce time spent in search activities by increasing the
marginal utility of leisure. 3' The variable FAMINC may also represent the
household's social status. If so, then higher FAMINC implies more contacts
with people likely to be of assistance in finding a new job and a greater
stigma on remaining unemployed.
E. A Higher-tenured Subsample
The definition of a displaced worker has varied in the literature. The
present study focuses on the distinction between old and new industries
because of an interest in displaced workers who are likely to have formed
some attachment to the industries of their old jobs. Therefore, the sample
was restricted to those whose tenure at that job was at least 1 year. A more
stringent definition is sometimes used. Accordingly, I restricted a subsample
to those whose tenure at their old jobs was at least 3 years. The cost of
this restriction was to reduce the size of the sample from 2,090 to 1,266,
31 Barron and Mellow (1979) found that higher nonwage income leads to less
time devoted to search but for some individuals leads to larger financial expenditures
on search.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
320 Fallick
and the num
to 573 (45% of the subsample). Other than the consequent increase in the
standard errors of estimates, the results (not shown, in the interests of
space) are consistent with the above discussion.
VI. Conclusion
Search-theoretic models and empirical studies of the duration of u
ployment generally treat all jobs as belonging to a single homogeneous
labor market. Recognizing that a worker may search simultaneously in
more than one market allows one to test a fundamental class of implications
of job-search models and investigate their importance to the mobility of
labor across sectors of the economy. In particular, judging from the results
presented here, distinguishing between industries when analyzing the
reemployment of displaced workers is important because labor market
conditions vary across industries. Coming from an industry with a greater
rate of employment growth appears to induce workers to reduce their
search intensity and increase their reservation wages in other industries,
in accordance with models of job search. Other measures of industry conditions
are more problematic.
The distinction between the industry in which a person used to work
and all other industries is useful also in that factors such as education, job
tenure, and intraindustrial wage premia affect the hazard rates associated
with these two sectors differently. For example, at least through graduating
from high school, education increases the rate of reemployment into jobs
in a new industry but not the rate into jobs in the old industry. Studies of
displaced workers should take account of the distinctions between the
different industries in which a worker may find a job.
The evidence that displaced workers respond to macroeconomic incentives
concerning industrial mobility highlights society's interest in providing
job seekers with good and timely information about labor market conditions
in the various sectors of the economy. The effect of job tenure and
wage premia on mobility suggests that programs to assist displaced workers
in finding new jobs might profitably emphasize, in addition to prompt
and accurate information, locating matches that allow the greatest amount
of human capital to be transferred to the new job.
Appendix
Definitions of Explanatory Variables
ABOLISH = 1 if left job due to the elimination of shift or
position,
= 0 otherwise;
AGE = age in years;
CRAFT = 1 if occupation at the old job was precision
production, craft, and repair,
= 0 otherwise;
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 321
EARNNEW = the analog of EARNOLD for the weighted aggregate
of all major industries except the old industry;
EARNOLD = median weekly earnings of full-time employees in the
old major industry, averaged between the March 1983
and 1984 CPS, or 1985 and 1986 (as appropriate);
EDUC = the number of grades of schooling completed;
EMPNEW = the analog of EMPOLD for the weighted aggregate
of all major industries except the old industry;
EMPOLD = fraction increase in the number of full-time
employees in the old major industry in the January
Current Population Survey between 1983 and 1984,
or 1985 and 1986 (as appropriate);
FAMINC = usual family annual income, by category;
FARMING = I if occupation at the old job was farming, forestry,
or fisheries,
= 0 otherwise;
FEMALE = 1 if female,
= 0 if male;
HEAD = I if head of household,
= 0 otherwise;
LABORER = 1 if occupation at the old job was operator,
fabricator, or laborer,
= 0 otherwise;
NOTIFIED = 1 if expected in advance the loss of one's job,
= 0 if did not expect;
PREMIUM = the proportional difference between weekly earnings
on the old job and median weekly earnings for the
industry;
SERVICE = 1 if occupation at the old job was services,
= 0 otherwise;
SIZERAT = ratio of employment in the new industry to
employment in the old;
SLACK= 1 if left job due to slack work,
= 0 otherwise;
STUNEMP = the state unemployment rate for 1983 or 1985;
SUPPORT = 1 if occupation at the old job was technical, sales, or
administrative support,
= 0 otherwise;
TENURE = number of years working at the old job;
UI = 1 if unemployment insurance benefits were received,
= 0 otherwise;
UNEMPNEW = unemployment rate for the new industry;
UNEMPOLD = unemployment rate for the old industry;
WHITE= 1 if white,
= 0 if nonwhite;
YEAR86 = 1 if the observation came from the 1986 survey,
= 0 otherwise.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
322 Fallick
Dummy variables for geographic divisions are also used.
If SUPPORT = SERVICE = CRAFT = LABORER = FARMING = 0,
then the occupation at the old job was managerial or professional.
References
Addison, John, and Portugal, Pedro. "The Effect of Advance Notification
of Plant Closings on Unemployment." Industrial and Labor Relations
Review 41 (October 1987): 3-26.
. "Job Displacement, Relative Wage Changes, and Duration of
Unemployment." Journal of Labor Economics 7 (July 1989): 281-302.
Barron, John K., and Mellow, Wesley. "Search Efforts in the Labor Market."
Journal of Human Resources 14 (Summer 1979): 389-404.
Breslow, N. "Covariance Analysis of Censored Survival Data." Biometrics
30 (1974): 89-99.
Cox, D. R., and Oakes, D. Analysis of Survival Data. New York: Chapman
& Hall, 1984.
Ehrenberg, Ronald G., and Jakubson, George H. "Advance Notice of Plant
Closings: Does It Matter?" Industrial Relations 28 (Winter 1989): 60-
71.
Fallick, Bruce. "The Re-employment and Industrial Mobility of Displaced
Workers." Ph.D. dissertation, University of Pennsylvania, August 1988.
"Advance Notice and Destructive Attrition," Occasional Paper
no. 91-7. Los Angeles: UCLA Center for American Politics and Public
Policy, May 1991.
. "Job Security and Job Search in More than One Labor Market."
Economic Inquiry 3 (October 1992): 742-45.
Flaim, Paul O., and Sehgal, Ellen. "Displaced Workers of 1979-83: How
Well Have They Fared?" Monthly Labor Review 108 (June 1985): 3-16.
Horvath, Francis W. "The Pulse of Economic Change: Displaced Workers
of 1981-85." Monthly Labor Review 1 10 (June 1987): 3-12.
Jacobson, Louis S. "Earnings Losses of Workers Displaced from Manufacturing
Industries." In The Impact of International Trade and Investment
on Employment, edited by U.S. Department of Labor, Bureau of International
Affairs, pp. 87-98. Washington, D.C.: U.S. Government Printing
Office, 1978.
Katz, Lawrence F. "Layoffs, Recall, and the Duration of Unemployment."
Working Paper no. 1825. Cambridge, Mass.: National Bureau of Economic
Research, January 1986.
Kiefer, Nicholas, and Neumann, George. "An Empirical Job-Search Model
with a Test of the Constant Reservation Wage Hypothesis." Journal of
Political Economy 87 (February 1979): 89-107.
Kletzer, Lori G. "Earnings after Job Displacement: Job Tenure, Industry,
and Occupation." In Job Displacement: Consequences and Implications
for Policy, edited by John T. Addison. Detroit: Wayne State University
Press, 1991.
Leighton, Linda, and Mincer, Jacob. "Labor Turnover and Youth Unemployment."
In The Youth Labor Market Problem, edited by R. B. Freeman
and D. A. Wise. Chicago: University of Chicago Press, 1982.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms
Mobility of Displaced Workers 323
Madden, Janice Fanning. "Gender Differences in the Cost of Displacement:
An Empirical Test of Discrimination in the Labor Market." American
Economic Review 77 (May 1987): 246-51.
"The Distribution of Economic Losses among Displaced Workers:
Measurement Methods Matter." Journal of Human Resources 23 (Winter
1988): 93-107.
Mincer, Jacob, and Ofek, Haim. "Interrupted Work Careers: Depreciation
and Restoration of Human Capital." Journal of Human Resources 17
(Winter 1982): 3-24.
Mortensen, Dale T. "Job Search and Labor Market Analysis." In Handbook
of Labor Economics, vol. 2, edited by 0. Ashenfelter and R. Layard, pp.
849-920. Amsterdam: North Holland, 1986.
Podgursky, Michael, and Swaim, Paul. "Job Displacement and Earnings
Loss: Evidence from the Displaced Worker Survey." Industrial and Labor
Relations Review 4 (October 1987): 17-29.
Prentice, R. L., and Gloeckler, L. A. "Regression Analysis of Grouped
Survival Data." Biometrics 34 (March 1978): 57-67.
Ruhm, Christopher J. "Advance Notice and Postdisplacement Joblessness."
Journal of Labor Economics 10 (January 1992): 1-32.
Serrin, William. "For Texas Oil Workers, Economy Has Gone Dry." New
York Times (Wednesday, May 7, 1986), p. A20.
Shaw, Kathryn. "A Formulation of the Earnings Function Using the Concept
of Occupational Investment." Journal of Human Resources 19
(Summer 1984): 319-40.
. "Occupational Change, Employer Change, and the Transferability
of Skills." Southern EconomicJournal 53 (January 1987): 702-19.
Sueyoshi, Glenn T. "A Class of Binary Response Models for Grouped
Duration Data." Unpublished manuscript. San Diego: University of
California, September 1991.
Thomas, Jonathan, and Bernhardt, Dan. "Job Search Strategy, Expected
Wages, and Sectoral Movers and Stayers." Discussion Paper no. 810.
Cambridge: University of Cambridge, April 1991.
Warner, John T.; Poindexter, Carl, Jr.; and Fearn, Robert M. "EmployerEmployee
Interaction and the Duration of Unemployment." Quarterly
Journal of Economics 94 (March 1980): 211-33.
This content downloaded from 147.251.185.127 on Wed, 21 Jun 2017 11:20:14 UTC
All use subject to http://about.jstor.org/terms