P.
L.
than

...»...

£55..
. f
35%
n .6 v! .
......ﬂm. #9:: !2..
:vQFth. sh. 5.7:... .2...
a...‘ .31.?
.tlmnnhnﬁ.

)
5x
I

it
y

.I
a I»
2.:-

{on}.

.3.

, i. n.
14‘
1.9.3
2".

Huh

Xvi!

. .Ly

. ‘

....sbyushuii
......33‘.....:...
x. “1......

ii.“
5.8%... u... :3: 9 . ..
.42.. :13: ..s. . ....xei
.. mmﬁﬂmgzm
52.135. . J
335i:
using.

5:.

..l
.\4!
.19

.h

L
)9

i231

!
3

11‘

l
1
u
I

. 3% v
. , . h. ‘
5x...

3

\‘1 .\

3
6

...! ...:
«27.3;
. .51

I
1

)1...-
o! .1!“ 2
I31? 31,;
790:...
5

E

. rat I,
3... ‘I ...-1.....1...‘
a)!!! 11.104111511~Jl;
......u. .... t y; . ..r.:9l\
{£373. x:
31.3.}!
...-la...

Iraq}: \
53.... l. n

. 2.3.).
\‘

u it’ll...-
zn...‘ :‘I-ld 1:“..0
33f... 5:1.w1t55. 7
1 vs? ....6\ at.li\-»!r\|)a« u . .
1.133.513: ‘19 :52. l is!!! .A\
2 . 3.2;"...9‘11 5‘ 1' t
‘ . ‘1. 3711.]! a.
I): \f‘n“.

3‘!

 

131:1;
. . I)! ...
\I

. ~ t;\.
he;

..1. 1 3
(an. . .n ant
(31.2.21;

 

.iﬂmﬁg «an

11in. ‘
I552... .\
a 10:1: Ill.- ‘1‘
I. I ‘wil
. |1 13.x?

 

 

 

 

 

 

 

 

 

 

 

 

I “may mI'll/lll’lliullilllﬁllilﬁl
Michigan State 3 1293 01390 2923
University

dissertation entitled

THE EFFECT OF GEOGRAPHIC INDUSTRY
CONCENTRATION ON THE RE-EMPLOYMENT
EXPERIENCES OF DISPLACED WORKERS

presented by

Wendy Alston Stock

has been accepted towards fulﬁllment
of the requirements for

Ph ; D . degree in Economics

 

 

Major professor

 

 

MSULI an Afﬁrmative Action/Equal Opportunity Institution O~12771

.—_ nﬁsﬁ_

 

 

PLACE Ii RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or More doto duo.

‘ DATE DUE DATE DUE DATE DUE

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU Is An Affirmative Action/Equal Opportunity lnotltmion
W

”3-9. 1

 

THE EFFECT OF GEOGRAPHIC INDUSTRY CONCENTRATION
ON THE RE-EMPLOYMENT EXPERIENCES OF DISPLACED WORKERS
By

Wendy Alston Stock

A DISSERTATION
Submitted to
Michigan State University

in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Economics

1996

ABSTRACT

THE EFFECT OF GEOGRAPHIC INDUSTRY CONCENTRATION ON THE
RE-EMPLOYMENT EXPERIENCES OF DISPLACED WORKERS

by

Wendy Alston Stock

Workers become displaced when they involuntarily lose their jobs because of
changes in product and labor demand which cause plant closures, mass layoffs, or slack
work. As a result of these changes, they often suffer extended periods of unemployment
with signiﬁcant earnings losses.

This dissertation examines how concentration of industries in local labor markets
affects re-employment outcomes for displaced workers. Particular attention is paid to the
effect of geographic industry concentration on the probability that a displaced worker will
switch industries upon re-employment. Employment shares for each detailed industry
within each metropolitan statistical area covered by the Current Population Survey are
calculated and included in reduced form models of re-employment, industry switch, and
earnings losses. Policy implications and methodological approaches to measuring displaced
worker earnings losses are also explored. A key ﬁnding is that workers displaced from
industries which account for a small Share of local employment are more likely to switch
industries upon re-employment. Switching industries is found to result in larger earnings

losses for displaced workers.

This dissertation is dedicated to Ken, who always inspires me to lead a balanced life.

ACKNOWLEDGMENTS

A project like this can never be undertaken alone, and many individuals have
given me their support, advice, and patience along the way.

David Neumark, my advisor and mentor, has been a positive inﬂuence throughout
my years at Michigan State. His advice has been incredibly helpful and important to this
work and has greatly improved my knowledge of economics and economic research. His
expertise in labor economics, econometrics, and research techniques has never ceased to
amaze me. His teachings provide a ﬁrm foundation upon which to continue building my
own skills as an economist.

Numerous other faculty members at Michigan State have provided guidance at
various stages in the process and I feel lucky to have had so many outstanding economists
at my disposal. Jeffrey Wooldridge was extremely inﬂuential in introducing me to the
tools of econometrics, as well as in helping me to choose the right tool for the task at
hand. Robert LaLonde and Harry Holzer provided numerous insightful suggestions and
comments along the way.

I also owe thanks to my fellow graduate students. Discussions shared with

Kathleen, Dan, and Amelia have been particularly valuable and inspirational.

iv

The inﬂuence of my father, Richard Alston, has shaped this work in numerous
ways, both obvious and subtle. His masterful teaching attracted me to this ﬁeld, and his
words of support, guidance and wisdom have kept me here. He has been a constant source
of pride for me and I hope to do as well in my career as he has in his. I also appreciate
the support and concern shown by my entire family. Thanks Mom, Aaron, Bette, Roy,
and Ruth.

Finally, I owe the largest debt to Ken Stock, my husband. The sacriﬁces he made
to come to Michigan and to put up with a largely absent (and absent minded) spouse
throughout our years here are incalculable. I will Spend a lifetime (I hope) repaying the

debt I owe him.

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................... viii
LIST OF FIGURES ............................................................................................. x
INTRODUCTION ............................................................................................... 1
CHAPTER 1
THE EFFECT OF GEOGRAPHIC CONCENTRATION ON THE
RE-EMPLOYMENT EXPERIENCES OF DISPLACED WORKERS .............. 6
1. Introduction ........................................................................................ 6
2. Data .................................................................................................... 8
2.1 The Displaced Worker Survey Data ..................................... 8
2.2 The County Business Patterns Data ...................................... 10
2.3 Employment and Earnings Data ........................................... 12
3. An Overview of the Displacement Problem ....................................... 12
4. Previous Research ............................................................................... 15
4.1 Earnings Losses .................................................................... 16
4.2 Local Labor Market Conditions: Unemployment Rates ....... 17
4.3 Local Labor Market Conditions: Industry Growth Rates ..... 18
4.4 Effects of Industry Switching ............................................... 20
4.5 Factors Inﬂuencing Industry Switching ................................ 21
5. Theoretical Motivation - Industry Switching and Earnings Losses....22
5.1 Industry Switching and Earnings Losses .............................. 22
5.2 Factors Inﬂuencing Industry Switching ................................ 23
6. Statistical Framework and Empirical Results ..................................... 27
6.1 Sample Characteristics .......................................................... 27
6.2 Industry/Area Characteristics ................................................ 27
6.3 Re-employment ..................................................................... 29
6.4 Industry Switching ................................................................ 30
6.4.1 Sample Selection Issues ......................................... 32
6.4.2 Endogeneity Issues ................................................. 33
6.5 Earnings Loss Model ............................................................ 35
6.5.1 Sample Selection Issues ......................................... 37
6.5.2 Endogeneity Issues ................................................. 37
7. Conclusion .......................................................................................... 39

vi

Appendix .................................................................................................. 52

References ................................................................................................ 56
CHAPTER 2
INTER-INDUSTRY MOBILITY AND DISPLACED WORKER
ASSISTANCE PROGRAMS .............................................................................. 60
1. Introduction ......................................................................................... 60
2. Programs for Displaced Workers ........................................................ 62
3. Evidence on Program Effectiveness ................................................... 66
3.1 Evidence on Job Search Assistance ...................................... 66
3.2 Evidence on Training Programs ............................................ 67
3.3 Evidence on Industry Switching ........................................... 69
4. Predicting Industry Switching ............................................................. 72
4.1 Data ....................................................................................... 74
4.2 Sample Characteristics .......................................................... 76
4.3 Estimations and Predictions .................................................. 77
5. Conclusion .......................................................................................... 81
References ................................................................................................ 89
CHAPTER 3
COMPARISON GROUP USE AND THE EARNINGS LOSSES
OF DISPLACED WORKERS ............................................................................. 91
1. Introduction ......................................................................................... 91
2. Previous Research and Use of Comparison Groups ........................... 93
2.1 Use of Comparison Groups: Non-DWS Data ....................... 96
2.2 Use of Comparison Groups: DWS Data ............................... 97
3. A Model of Earnings Losses ............................................................... 102
4. Comparison of Earnings Loss Measures Using
a CPS Comparison Group ................................................................... 108
4.1 Potential Problems and Biases .............................................. 109
4.2 Sample Characteristics .......................................................... 110
4.3 Estimation of Earnings Losses .............................................. 112
5. Conclusion .......................................................................................... 1 15
References ................................................................................................ 125

vii

CHAPTER 1

Table 1:

Table 2:

Table 3:

Table 4:

Table 5:

Table 6:

Table 7:

Table 8:

Table 9:

Table 10:

Table A1:

Table A2:

Table A3:

Table A4:

LIST OF TABLES

Characteristics of Displaced Workers 1987-1992 ....................... 41
Most and Least Concentrated Industries and Areas

excluding retail sales .................................................................... 43
Comparison of Switch Rates ........................................................ 44

Predicted Effects on Industry Switching and Earnings Losses....45

Mean Values of Variables ............................................................ 46
Re-Employment Effects ............................................................... 47
Switch Effects .............................................................................. 48
Switch Effects by Industry and level of Employment Share ....... 49
Earnings Change Effects .............................................................. 50
Earnings Change Effects by Gender ............................................ 51
Re-Employment Effects-~Probit Model ....................................... 52
Switch EffectS--Probit Model ...................................................... 53
Switch Effects-Two-Stage Least Squares Estimation .................. 54

Earnings Change Effects-Two-Stage Least
Squares Estimation ....................................................................... 55

viii

CHAPTER 2

Table 1: Federal Programs for Displaced Workers .................................... 83
Table 2: Variable Means ............................................................................ 84
Table 3: Switch Effects .............................................................................. 85
Table 4: Former Industry Effects on Switching ......................................... 86
CHAPTER 3
Table 1: Comparison of Earnings Loss Measurement Methods ................ 118
Table 2: Comparison Group Use and Findings from Studies

which measure the Distribution of Earnings Losses

for Displaced Workers ................................................................. 119
Table 3: Estimating the Differential Impact of outcome S ......................... 122
Table 4: Mean Values of Variables ............................................................ 123
Table 5: Comparison of Earnings Loss Estimates ..................................... 124

ix

LIST OF FIGURES

CHAPTER 2
Figure 1: Industry Switch Rates by Level of Employment Share ............... 87
Figure 2: Actual and Predicted Industry Switch Probabilities .................... 88

INTRODUCTION

Workers become displaced when they involuntarily lose their jobs because of
changes in product and labor demand which cause plant shutdowns, mass layoffs, or slack
work. As a result of these changes, displaced workers have little or no chance of being
recalled to their former job, or perhaps even to their former industry. They often suffer
extended periods of unemployment with signiﬁcant losses of earnings and health beneﬁts,
as well as readjustment difﬁculties. Their communities often sustain indirect employment
losses in service activities and intermediate production.

Although a substantial amount of research has been done on displaced workers,
ﬁndings in the displaced worker literature are extensive and varied, and problems in
estimating displaced‘worker earnings losses persist.

Chapter 1 is an attempt to overcome two shortcomings in the displacement
literature. First, it is often argued that local labor market conditions play a role in
determining the experiences of displaced workers. However, the exact channels through
which local labor market conditions affect displaced workers remain unclear. One local
labor market inﬂuence that has been neglected in previous research on displacement is the
geographic concentration of the displaced workers' industries in their local labor market.
Deﬁning geographic concentration as an employment share (i.e. the percentage of
employment in a local labor market attributable to each industry in that market), 1 construct

geographic concentration measures and estimate the impact of geographic concentration on

various outcomes. Second, although several previous researchers have shown that
switching industries upon re-employment increases earnings losses, very few have
estimated the impact of factors which affect whether a worker switches industries upon re-
employment. I estimate the inﬂuence of various factors, including geographic industry
concentration, upon whether a worker switches industries upon re-employment. My
ﬁndings indicate that workers displaced from geographically concentrated industries are
less likely to switch industries upon re-employment. I also ﬁnd that workers who do not
switch industries upon re-employment tend to suffer smaller earnings losses.

Worker displacement is an important problem. The back-to-back recessions in the
early 19805, the corporate down-sizing during the recession of the early 19903, and recent
legislation affecting both the environment and international trade have raised questions
about the extent of worker displacement, the earnings losses which result from
displacement, and the costs displacement imposes upon others. As evidenced by the large
expenditure on displaced worker assistance, policy makers (and their constituents) are
concerned with the effects of structural changes in the economy on displaced workers,
particularly the potentially large and persistent earnings and human capital losses they
endure and the time they spend unemployed. Policy makers who attempt to offset these
losses through government sponsored assistance and training programs face a difﬁcult and
important set of questions. A particular focus in this work is whether programs for displaced
workers should be altered to ﬁt different local labor market conditions. As one example, if
workers displaced from geographically concentrated industries are less likely to switch

industries upon re-employment, programs which focus upon job search assistance, as

opposed to re-training, may be more effective as a response to displacement in communities
with high levels of geographic concentration in displaced workers’ industries. Chapter 2 is
an attempt to test the feasibility of such a reallocation of the resources dedicated to
training programs. Despite existing evidence that ﬁnds that workers who switch
industries upon re-employment tend to suffer larger earnings loses and to have longer
unemployment spells than their non-switching counterparts, very little research has
explicitly estimated the factors that inﬂuence whether a displaced worker switches
industries upon re-employment. The results presented in Chapter 2 suggest that although
it is possible to identify factors which strongly inﬂuence switching, they only explain a
small fraction of the variation in switch behaviOr. Thus, it would be difﬁcult a priori to
identify workers who are more (or less) likely to switch industries, at least based solely
upon the types of information available in large, nationwide data sets like the Displaced
Worker Survey (discussed below). However, the information presented in here, in
combination with information currently available from local agencies, can be used to
assist policy makers and assistance service providers in designing effective programs for
displaced workers.

The most commonly used data on displaced workers comes from the Displaced
Worker Surveys (DWS), which have supplemented the January Current Population
Surveys (CPS) biennially since 1984. Individuals in the CPS who are identiﬁed as having
lost their jobs in the ﬁve years prior to the survey because of plant shutdowns, mass
layoffs, or slack work are asked supplemental questions about their current and former

earnings, job characteristics, and experiences since displacement. Thus, the DWS

provides detailed demographic and earnings information on a large, nationwide sample of
individuals and has been used extensively to study the experiences of displaced workers.

One of the major disadvantages of the DWS is its lack of comparable information
on the earnings of non-displaced workers, which has made it difﬁcult to use the DWS to
measure displaced workers’ earnings losses as the difference between their actual
earnings growth (or decline) and what that earnings growth would have been had
displacement not occurred. Put more simply, the DWS does not readily include the
information necessary to obtain a comparison group of non-displaced workers. Various
techniques have been invoked to obtain information on the earnings of non-displaced
workers, most notably the use of non-DWS/CPS data, as well as the use of earnings data
taken from the outgoing rotation groups of the CPS.

In Chapter 3, I ﬁrst review previous attempts to obtain comparison groups and to
estimate the eamings losses of displaced workers. Second, I evaluate the potential gains
and biases introduced by comparison groups and propose a method for obtaining a
comparison group by using the average earnings of workers in the CPS. Finally, using
data on workers in the 1992 DWS and workers in the 1986 through 1992 CPS, I compare
earnings loss estimates made while using the comparison group against loss estimates
made while simply comparing the pre- and post-displacement earnings of displaced
workers.

Although the ideas presented in Chapter 3 are applicable in more general contexts,
I focus attention throughout the paper on the problem of estimating the impact of industry

switching on displaced worker earnings losses. Many researchers have found that

displaced workers who switch industries upon re-employment experience larger earnings
losses, but my particular interest lies in determining how estimates of the impact of
industry switching on earnings losses vary depending upon ﬁrst, whether a comparison
group is used to measure earnings losses and second, how the earnings of the comparison

group change over time and with industry Switching.

Chapter 1
THE EFFECT OF GEOGRAPHIC INDUSTRY CONCENTRATION ON THE

RE-EMPLOYMENT EXPERIENCES OF DISPLACED WORKERS
1. Introduction

Workers become displaced when they involuntarily lose their jobs because of
changes in product and labor demand which cause plant closures, mass layoffs, or slack
work. As a result of these changes, displaced workers have little or no chance of being
recalled to their former jobs, or perhaps even to their former industries. They often suffer
extended periods of unemployment with signiﬁcant losses of earnings and health beneﬁts,
as well as readjustment difﬁculties.l

A substantial amount of research has been done to measure the number of displaced
workers, the earnings losses which result from displacement, and the impact of training
programs designed for displaced workers. Findings in the displaced worker literature are
extensive and varied, and problems in estimating displaced worker earnings losses persist.
This paper is an attempt to overcome two shortcomings in the displacement literature.

First, it is often argued that local labor market conditions play a role in determining
the experiences of displaced workers. However, the exact channels through which local

labor market conditions affect displaced workers remain unclear. One local labor market

 

I The readjustment difﬁculties could include psychic costs arising from relocation to a new area, as well as job
search costs. There is extensive research on the relationship between job search and mental distress. Gordus,
Jarley, and Fermen review some of this research (1981, ch. 6).

inﬂuence that has been neglected in previous research on displacement is the geographic
concentration of the displaced workers' industries in their local labor market. Deﬁning
geographic concentration as an employment share (i.e. the percentage of employment in a
local labor market attributable to each industry in that market), I construct geographic
concentration measures and estimate the impact of geographic concentration on various
outcomes.

Second, although several previous researchers have shown that switching industries
upon re-employment increases earnings losses, only one study (Fallick, 1993) estimates the
impact of factors which affect whether a worker switches industries upon re-employment. I
extend F allick’s analysis by estimating the inﬂuence of various factors, including
geographic industry concentration, upon whether a worker switches industries upon re-
employment. My ﬁndings indicate that workers displaced from geographically concentrated
industries are less likely to switch industries upon re-employment. I also ﬁnd that workers
who do not switch industries upon re-employment tend to suffer smaller earnings losses.

Worker displacement is an important problem. The back-to-back recessions in the
early 1980s, the corporate down-sizing during the recession of the early 19905, and recent
legislation affecting both the environment and international trade have raised questions
about the extent of worker displacement, the earnings losses which result ﬁ'om
displacement, and the costs displacement imposes upon others. As evidenced by the large

expenditure on displaced worker assistance,2 policy makers (and their constituents) are

‘D

 

2 For example, the Economically Displaced Worker Adjustment Assistance Act (EDWAA) had a budget
allocation of approximately $500 million in ﬁscal year 1993, while the Trade Adjustment Assistance

concerned with the effects of structural changes in the economy on displaced workers,
particularly the potentially large and persistent earnings and human capital losses they
endure and the time they spend unemployed. Policy makers who attempt to offset these
losses through government sponsored assistance and training programs face a difﬁcult set of
questions. Although this paper is not a policy analysis per se, it begins to address an
important policy question: should programs for displaced workers be altered to ﬁt different
local labor market conditions? AS one example, if workers displaced from geographically
concentrated industries are less likely to switch industries upon re-employment, programs
which focus upon job search assistance, as opposed to re-training, may be more effective as
a response to displacement in communities with high levels of geographic concentration in

displaced workers’ industries.

2. Data
I utilize three data sets in this study, the Displaced Worker Survey, County Business

Patterns, and Employment and Earnings data published by the BLS.

2.1 The Displaced Worker Survey Data. During every other wave of the January
CPS, workers who have identiﬁed themselves as having lost a job in the last ﬁve years due

to their plant closing, their employer going out of business, or a layoff from which they

 

(TAA) program had a budget allocation of approximately $80 million in ﬁscal year 1993 (Stock, 1996). In
addition, many displaced workers are eligible to receive unemployment insurance beneﬁts, Pell Grants, and
other federal assistance (Jacobson, LaLonde, and Sullivan, 1994).

were not recalled, are asked a series of supplementary questions about their unemployment
spell and the characteristics of their former and (if applicable) current jobs. This
supplement, known as the Displaced Worker Survey (DWS), has been conducted biennially
since 1984. The January 1992 DWS serves as the primary data set for the analysis in this
paper.

Following the federal deﬁnition of a displaced worker, I include in my sample
workers in the DWS who had at least three years of tenure prior to their displacement.3 In
order to control for local labor market conditions, I limit the sample to workers who live in
metropolitan statistical areas (MSAS) and who did not move after displacement.4 Finally, in
order to ensure that workers had enough time after displacement to conduct a job search, the

sample is further restricted to include only workers who were displaced between January

 

3 The ofﬁcial federal deﬁnition of a displaced worker is one who lost a job after three or more years of tenure,
owing to slack work, abolition of the job, mass layoffs, or plant closure (F lairn and Seghal, 1985). As noted by
Swairn and Podgursky (1991), most recent research on displaced workers follows the federal deﬁnition.
Researchers who utilize a standard which varies from the federal deﬁnition generally alter the years of tenure
standard. For example, Howland and Peterson (1988) include as displaced workers those with at least one year
of tenure. Seitchik and Zomitsky (1989) also include workers with low levels of tenure, arguing that not to do
so introduces bias due to non-random sample selection. They also argue that low tenure does not necessarily
imply weak labor force attachment, particularly among displaced workers who may ﬁnd themselves moving
from one low tenure job to another in the face of “hired last, ﬁred ﬁrst” seniority systems. Alternatively,
Jacobson, LaLonde, and Sullivan (1993) impose a six-year tenure standard.

’ Workers who moved to ﬁnd re-employment are excluded from the estimates since the labor market
conditions, including the geographic concentration of their former industry, cannot be calculated.
Unfortunately, the DWS does not ask workers about the geographic area in which they were employed prior to
displacement. However, the survey does ask "since displacement, has [the reference person] moved to a
different city or county to look for work or to take a different job?" As this question does not reﬂect whether
workers changed MSAs as a result of displacement, it would be erroneous to include workers who moved in
the analysis. I ran several tests to analyze the effect of excluding movers from the sample. I found a negative,
although insigniﬁcant effect of having moved on the probability that a displaced worker was re-employed.
There was no signiﬁcant effect of moving upon either the size of earnings losses or upon whether a worker
switched industries upon re-employment.

10

1987 and December 1990.5

The DWS provides detailed demographic and earnings information on a large,
nationwide sample of individuals. However, the retrospective nature of the DWS questions
will tend to lead to measurement error and biases in econometric analysis since workers are
less likely to report more remote instances of job loss, particularly if the job loss resulted in

only a small change in earnings.6

2.2 The County Business Patterns Data. I utilize the Country Business Patterns
(CBP) data to compute geographic industry concentration measures. The CBP is an annual
data set containing employment and payroll information, by county and industry, for
establishments covered by the Federal Insurance Contributions Act (FICA). The basic data
items in the CBP are extracted from the Standard Statistical Establishment List, a ﬁle of all
known single and multi-establishment companies maintained and updated by the Bureau of
the Census. The Annual Company Organization Survey is the source of data for multi-
location ﬁrms, while data on Single-location ﬁrms are obtained from various programs

conducted by the Census Bureau7 (County Business Patterns, 1993).

 

5 The DWS asks workers about the year but not the month of their displacement. It is unlikely that workers
displaced in late 1991 or January 1992 had enough time to conduct a job search before the January 1992
survey. Together, the three exclusion restrictions drop 2127 of the original 2880 observations from the sample.
The exclusion of workers with fewer than three years of tenure drops 1527 of the original 2880, the exclusion
of those displaced in 1991 or 1992 drops 442 of the remaining 1353 observations, and the exclusion of movers
drops 158 of the remaining 911 observations from the sample.

6 See Horvath (1982) for a discussion of bias in retrospective surveys.

7 These programs include the Annual Report of Organizations, the Annual Survey of Manufactures, Current
Business Surveys, and Internal Revenue Service (IRS) administrative records.

ll

Deﬁning the local labor market as the worker's MSA and the geographic industry
concentration of the worker’s industry as the employment share of his industry in his MSA
gives rise to the following formula, which I use to calculate the employment shares (ES) for
each of the 50 detailed industries and 203 MSAs deﬁned in the Current Population Survey

(and the DWS):

ES = (number of workers in industry j in MSA k) / (number of workers in MSA k)8, 9

For each of the years 1986 to 1992, the employment share ﬁgures are merged with

the DWS, attaching an ES value for the relevant industry and MSA to each individual in the

DWS.

 

8 Alternative measures of industry geographic concentration are available. In a diﬁ'erent context, Diamond
and Simon (1990), and Diamond (1988) use a Hidindahl Index (HI) of specialization in a local market,
deﬁned as: HI= 2‘. (ES)2 across all industries in the local labor market. The HI measure characterizes a local
labor market as perfectly specialized (Hl=l) if its employment is concentrated in a single industry and
perfectly diversiﬁed when its employment is evenly distributed across all industries. Note that although this
measure allows for comparison of the relative industrial diversity across local labor markets, it prohibits
comparison of the relative concentration of industries within local labor markets.

9 The ES variable can also be computed at the major industry level. There are 23 major (one-digit)
industries and 51 detailed (two-digit) industries in the CPS. They differ primarily in their degree of
specialization in both the durable and non-durable goods manufacturing industries. For example, “durable
goods manufacturing primary metals” and "durable goods manufacturing fabricated metals” are two
distinct detailed industries. However, they are grouped together (along with 12 other durable goods
manufacturing industries) into one major industry category, “durable goods manufacturing.” I brieﬂy
discuss the effects of ES ﬁgures which are calculated at the major industry level in the footnotes of Section
6.

12

2.3 Employment and Earnings Data. Unemployment ﬁgures obtained from the
CPS are reliable at the state level, but the CPS is insufﬁciently large to accurately estimate
MSA-level unemployment rates. For this reason, the unemployment data for this study
come from the Bureau of Labor Statistics (BLS) publication Employment and Earnings,

which contains unemployment rate statistics for most MSAs in the CPS.'0

3. An Overview of the Displacement Problem

Table l, which is reproduced from assorted tables in Gardner (1993) and
supplemented with information from Employment and Earnings, the DWS, and the CBP,
provides summary information on the total number of workers displaced between 1987 and
1992, as well as re-employment rates and real weekly earnings changes experienced by the
displaced workers covered in this study. Table 1 also summarizes the employment share

ﬁgures calculated from the CBP data.”

 

‘0 State-level unemployment rates were assigned to the 25 MSAs not covered in Employment and Earnings.
Although primarily from the same sources, the BLS data differ ﬁ'om CPS data at the sub-state level because
the BLS adjusts statewide CPS unemployment rates based on administrative statistics (compiled from reports
ﬁled by establishments covered under state unemployment insurance laws) in order to obtain sub-state
unemployment estimates. Comparisons of MSA-level unemployment rates calculated using the CPS outgoing
rotation groups for the years 1987 to 1992 and the BLS Employment and Earnings data reveal that the CPS
unemployment ﬁgures were consistently much smaller (by an average of about two percentage points) than
the BLS ﬁgures.

" The employment share ﬁgures in Table l were estimated by computing an employment share value for
each industry and MSA in the CBP and then averaging these values across the regions and industries
Shown in the table.

13

By the federal deﬁnition, about 5.6 million workers were displaced between January
1987 and January 1992 (Gardner, 1993).l2 About 2 million of the workers displaced from
1987 to 1992 lost jobs in manufacturing, accounting for approximately 35 percent of all
displacements during the period.'3 Although worker displacement was most strongly
concentrated in the South, which accounted for roughly 33 percent of all displacements, the
New England states also had a disproportionately large share of displacements (Gardner,
1992). '4

Although the workers in the South suffered the largest number and percentage of
displacements, the re-employment rate of displaced workers there was relatively high. In
this sample, roughly 94 percent of the southern displaced workers had found jobs at the
time of the 1992 DWS. Note that the re-employment ﬁgures for this sample, which
excludes more recently displaced workers, are much higher than if workers displaced in
1991 and 1992 are included in the sample. Including the recently displaced resulted in re-
employment ﬁgures which are approximately 20 percentage points lower than those in

Table 1.15

 

'2 In comparison, roughly 5.1 million workers were displaced from January 1981 to January 1986 (Horvath,
1987), and a similar number were displaced from January 1979 to January 1984 (Flaim and Seghal, 1985).

’3 Declines in manufacturing accounted for roughly one-half of all displacements during the period from
January 1979 to January 1986 (Flaim and Seghal, 1985 and Horvath, 1987).

'4 The share of displacements for each region was calculated as the percent of the region’s work force which
was displaced over the period as compared to the percent of the US work force which was employed in the
region at the beginning of the period (Gardner, 1992).

'5 If the year of displacement restriction is not imposed, the overall average re-employment rate falls to .79.
Re-employment rates in the four regions (Northeast, Midwest, South, and West) fall to .72, .82, .84, and .81,
respectively. Re-employment rates by industry fall to: construction (.72), manufacturing (.77), transportation
(.80), trade (.84), FIRE (.84), services (.77), and public administration (.73).

14

The real weekly earnings change ﬁgures reported in Table 1 are calculated as re-
employed workers’ real earnings in 1992 less their earnings at the time of displacementm’ '7
The earnings change ﬁgures indicate that workers displaced from the ﬁnance, insurance,
and real estate industry (FIRE) experienced the largest drop in their earnings (-161.86 on
average, representing real earnings losses of 28 percent). Workers in the personal and
business services industry had relatively smaller earnings changes. Workers in the
Northeast experienced the largest weekly earnings changes (-98.76 on average), while
workers in the West experienced the smallest earnings changes over the period (although
their percentage decrease in real earnings was still 11 percent).

The wholesale and retail trade industry had the largest employment share ﬁgures
across all regions (.18), while the manufacturing and transportation/public utilities
industries had the lowest levels of employment share across the regions (.03).

The breakdown of employment shares in Table l masks a wide dispersion. The
overall mean level of geographic concentration for the workers in this study is .07. The
maximum level of geographic concentration (.35) occurred in the retail sales industry in the

Lawton, Oklahoma MSA. The minimrun level of geographic concentration (.00003)

occurred in the forestry and ﬁsheries industry in the Los Angeles-Anaheirn-Riverside MSA.

 

'6 Real earnings ﬁgures were calculated using the Consumer Price Index (base year 1984).

'7 Note that the earnings changes in Table 1 do not include the earnings losses of those workers still
unemployed at the time of the survey. Since their current incomes (not including government transfer
payments such as unemployment insurance beneﬁts) are essentially zero, their inclusion in Table 1 would
cause the earnings loss ﬁgures to rise.

15

To illustrate the wide dispersion of employment share ﬁgures, Table 2 provides a
tabulation of ten highly concentrated and the ten least concentrated areas and industries in

18
For

this study. Table 2 also serves to illustrate the geographic concentration issue.
example, it is unlikely that a worker displaced from the furniture and ﬁxtures manufacturing
industry in Hickory, NC (which has an employment share value of 27 percent), will have
the same re-employment experience as a worker displaced from the motor vehicles and
equipment manufacturing industry in Las Vegas, NV, where his industry accounts for only
.01 percent of total MSA employment. Most importantly, the two workers would likely

have different probabilities of switching industries upon re-employment, which (as

discussed in Section 5) would affect their subsequent earnings.

4. Previous Research

Economic theory suggests that workers who become involuntarily unemployed will
suffer earnings losses even if they ﬁnd subsequent employment. These losses could come
from increases in earnings associated with speciﬁc human capital or a good job match,
losses of wage premiums due to unions or efﬁciency wages, or, for workers with long

tenure, from an earnings-tenure proﬁle which is steeper than their productivity proﬁle.19

 

’8 Note that retail sales accounted for the ten most concentrated industries. I list the ten most concentrated non-
retail sales industries in Table 2.

'9 See Willis (1986) and Becker (1975) for a discussion of the loss of speciﬁc human capital; Lewis (1986) for
a discussion of union relative wage effects; Shapiro and Stiglitz (1984) for a discussion of eﬁiciency wages;
Mortensen (1986) for a discussion of job search and match; and Lazear (1981) for a discussion of earnings-
tenure proﬁles.

16

This section reviews existing evidence on displaced worker earnings losses and the effects

of local unemployment rates and local industrial growth rates on those losses.

4.1 Earnings Losses. Many studies have estimated the earnings losses experienced
by displaced workers.20 In studies reviewed by Hamermesh (1989), the median re-employed
worker suffered a reduction in wages of between ﬁve and 15 percent and an unemployment
spell of as long as 40 weeks. High tenure workers, less educated workers, and workers who
changed occupations or industries upon re-employment tended to experience both greater
wage losses and longer periods of unemployment. Podgursky and Swaim (1987) estimate
that men and women displaced from blue-collar occupations suffered earnings losses of
11.7 and 26.2 percent, respectively, while men and women displaced from white-collar
occupations experienced losses of between 4.3 and 16.3 percent, respectively. They also
ﬁnd that displaced workers with substantial speciﬁc human capital investments tended to
suffer more enduring losses.

The above results generally refer to short term outcomes. Jacobson, LaLonde, and
Sullivan (1993) ﬁnd that high-tenure prime-age workers suffer substantial and persistent
average earnings losses. Even six years after displacement, the quarterly earnings of
displaced workers were $1,600 less than expected levels, representing a loss of 25 percent

of pre—displacement earnings. The authors conclude that because the estimated losses do

 

20 For example, Corson and Nicholson (1981) estimate an lifetime earnings loss of $9000 (1978 dollars).

Hamermesh (1987) estimates a loss of job speciﬁc human capital of $7000 (1980 dollars).

17

not signiﬁcantly decline even beyond three years after displacement, there is "little evidence
that displaced workers' earnings will ever return to their expected levels" (Jacobson,

LaLonde, and Sullivan, 1993: p.697).

4.2 Local Labor Market Conditions: Unemployment Rates. Previous studies
suggest that local unemployment rates play a signiﬁcant role in the re-employment
experiences and earnings losses of displaced workers. Jacobson, LaLonde, and Sullivan
(1993) estimate the effects of both long-term local labor market conditions as well as short-
term business cycle conditions on displaced worker earnings losses. Using employment
growth rates to measure long-term differences in local labor market conditions, they ﬁnd
that workers’ quarterly earnings losses are as much as $500 greater, even ﬁve years after
displacement, in areas with depressed rates of employment growth. Short-term effects are
measured using both variation in local employment deviations from their trends and
variation in local employment rates. Even ﬁve years after displacement, workers displaced
in areas which were experiencing cyclical downturns at the time of their displacement
suffered losses which were nearly $1500 larger than their counterparts displaced during the
best cyclical conditions. Howland and Peterson (1988) also conclude that displaced workers
tend to suffer larger economic losses in depressed local eoonornies. They ﬁnd that strong
overall growth in the local economy reduces the losses of displaced white-collar workers,

but not of blue-collar workers. Podgursky and Swaim (1987) ﬁnd that for each percentage-

18

point increase in the local unemployment rate, displaced workers’ re-employment earnings
fell by between one and two percent.

In addition to the studies surveyed here, many case studies have focused on the
experiences of displaced workers and their communities. Although they tend to be limited
in scope, the case studies generally provide support for the empirical results summarized
above. For example, researchers studying the Armour and Company meat packing plant
closures in the 19603 concluded that the condition of the local labor market had a strong
impact on the experiences of workers displaced as a result of Armour plant closures

(Ullman, 1969).21

4.3 Local Labor Market Conditions: Industry Growth Rates. While no research
on displaced worker earnings losses has estimated the impact of local industry geographic
concentration, a few researchers have controlled for local industry growth rates when
studying the experiences of displaced workers. The limited ﬁndings suggest that workers
displaced from industries which are growing locally tend to suffer smaller losses than
workers displaced from industries which are declining locally.

Using an establishment level data set which includes only manufacturing workers,
Howland (1988) ﬁnds that workers displaced from manufacturing industries which were

declining locally tended to suffer the longest stretches of unemployment and the largest

 

2' See Schultz (1964), Conant (1965), and Stern (1969) for more examples. More recent studies include one
done by the US General Accounting Ofﬁce (1993), which focuses on the impact of declines in timber
harvesting in the Paciﬁc Northwest.

l9

earnings losses. Howland concludes that the re-employment success of workers displaced
from manufacturing industries is more sensitive to local conditions in the manufacturing
industry than to conditions in the overall local economy. In a separate paper, Howland and
Peterson (1988) conclude that re-employment success for displaced workers depends more
strongly upon local employment growth in their 2—digit industry than on overall local
employment grth for blue-collar workers. For white-collar workers, overall local
employment growth was more inﬂuential. The authors argue that the result arises because
white-collar workers tend to have more general human capital, which is easily transferable
across industries.

Carrington (1993) computes state-level industry employment changes and estimates
their effects on displaced worker earnings losses using the DWS. He ﬁnds that losses for
high experience workers are sensitive to declines in employment in their state, industry, and
occupation, but losses for high tenure workers are not sensitive to these downturns. He
concludes that losses for high tenure workers are primarily due to losses in ﬁrm-speciﬁc or
at least narrowly deﬁned industry-speciﬁc skills. Although Carrington’s study begins to
assess the impact of local industry characteristics on earnings losses, he computes industry
growth rates at the state level, which is too broadly deﬁned to be applicable to local labor
markets. Furthermore, Carrington does not analyze the impact of industry geographic

concentration on OUICOITICS.

20

4.4 Effects of Industry Switching. Jacobson, LaLonde, and Sullivan (1993)
compare the regression adjusted mean earnings losses over time for workers who become
re-employed in the same industry against those who switch industries upon re-employment.
They ﬁnd that earnings losses are larger for workers who switch industries upon re-
employment. Displaced manufacturing workers who obtained new jobs in the
manufacturing sector suffered losses between 18 and 20 percent of their pre-displacement
earnings, while displaced manufacturing workers who left the manufacturing sector upon
re-employment suffered losses equal to 38 percent of their pro-displacement earnings.
Similarly, Seitchik and Zornitsky (1989) ﬁnd that displaced workers who changed
occupation or industry upon re-employment had re-employment earnings ﬁgures which
were 15 percent lower than those of displaced workers who had not changed occupations or
industries upon re-employment. Podgursky and Swaim (1987) ﬁnd that (i) becoming re-
employed in the same industry or the same occupation has a positive and highly signiﬁcant
effect on earnings for blue-collar men, (ii) becoming re-employed in the same industry has a
positive and highly signiﬁcant effect on earnings for white-collar men, but (iii) neither
becoming re-employed in the same industry nor becoming re-employed in the same

occupation has a signiﬁcant effect on the earnings of women.

4.5 Factors Inﬂuencing Industry Switching. Although each of the studies
described above estimates the impact of switching industries upon the earnings losses of

displaced workers, none analyze the factors which affect whether or not a worker switches

21

industries upon re-employment. Fallick (1993) provides some evidence in this area. Using
data ﬁ'om the 1984 and 1986 DWS, he estimates rates of transition (hazard rates) for two
groups of displaced workers: those moving from unemployment into employment in new
industries, and those moving from unemployment into employment in their former
industries. Although he primarily uses these estimates to test implications of job search
models, Fallick’s estimates of hazard rates for displaced workers moving from
unemployment into employment in new industries provide some evidence about possible
factors which inﬂuence industry switching.

Fallick ﬁnds that for workers who switched industry, an increase in employment
in their former industry increased their unemployment duration. Thus, even though the
workers eventually switched industries, they remained unemployed longer if employment
in their former industry was rising during the year of their displacement. Alternatively,
for workers who did not switch industries, an increase in employment in their former
industry decreased their unemployment duration. Fallick argues that these results arise
because workers change their search intensity and/or reservation wage in new industries
in response to information about employment prospects in their old industry.

To summarize, previous researchers have found that displaced workers tend to
suffer substantial earnings losses which vary in size according to workers’ industry,
occupation, tenure, human capital, and gender. Researchers have also found that
unemployment rates and industry growth rates affect the size of displaced worker earnings

losses. Workers displaced from depressed local economies and/or declining industries tend

22

to suffer larger losses than workers displaced ﬁom growing local economies and/or growing
industries. Furthermore, workers who switch industries upon re-employment tend to suffer
larger earnings losses than workers who do not switch industries upon re-employment.
Despite this evidence, based on the existing research, the exact channels through which
local labor market conditions (particularly geographic industry concentration) affect

displaced worker earnings losses, industry switching, and re-employment remain unclear.

5. Theoretical Motivation - Industry Switching and Earnings Losses

5.1 Industry Switching and Earnings Losses. There are numerous reasons why
switching industries upon re-employment should affect displaced worker earnings losses.
Both human capital models and search models suggest that workers who switch industries
upon re-employment will suffer larger losses than workers who do not switch industries.
Human capital models suggest that workers who lose any job will suffer losses in speciﬁc
human capital acquired during their tenure in that job. Assuming that human capital can
take forms which include job-speciﬁc, ﬁrm-speciﬁc, and industry-speciﬁc skills, human
capital losses will be greater when workers begin jobs which are substantially different from
their former jobs, or work in industries which are different from their former industries.

Because search methods are industry-speciﬁc,22 workers who switch industries will tend to

 

22 For example, searchers in the fast-food industry tend to ﬁle applications and wait for interviews, those
searching for jobs in the government may take civil service tests, and economists primarily use the annual
AEA meetings to search for positions. Holzer (1987) ﬁnds that ﬁrms recruiting employees use different
recruiting methods across industries. For example, the farm in manufacturing industries in Holzer’s study
most frequently used employment services, while ﬁrms in the ﬁnance and service industries tended to rely
most heavily on recruiting through newspaper want ads and least heavily on walk-in applicants.

23

have longer, less efﬁcient searches than their non-switching counterparts because
“switchers” must spend time leaming about the best search methods to use in different
industries.23 Workers who switch industries may also accept relatively lower wages if they
do not have adequate information about the equilibrium wage in the new industry and if the
marginal cost of obtaining information about the industry is high enough. Together, the
human capital and search models imply larger losses for workers who switch industries

upon re-employment.

5.2 Factors Affecting Industry Switching. Both the level of geographic
concentration of a worker's industry and the growth or decline in the geographic
concentration of that industry are potentially important inﬂuences upon whether a worker
switches industries upon re-employment. Holding all else constant, a displaced worker will
be more likely to switch industries upon re-employment if his former industry accounts for
only a small number of the jobs in the area (i.e. the industry has a low level of geographic
concentration). This effect will be intensiﬁed if the industry is declining in the local labor

market (i.e. the industry has a low and decreasing level of geographic concentration).

 

23 Although space limitations prohibit reporting of the results here, I ran simple regressions in order to test the
hypothesis that workers who switch industries upon re-employment tend to have longer searches. Regressions
of weeks unemployed after displacement on a dummy variable for switching industries, geographic industry
concentration, local unemployment rate, and other demographic and industrial controls (these variables are
described in Section 6 below) imply that workers who switched industries upon re-employment spent
approximately 4.5 more weeks unemployed than their non-switching counterparts. Results from these
regressions are available upon request. Fallick (1993) also ﬁnds that industry switchers tend to have longer
unemployment durations.

24

Other important factors which may inﬂuence industry switching (and subsequent
earnings losses) include the type of displacement and the turnover rates in the local labor
market after displacement occurs. If a worker is displaced from an industry which is highly
concentrated in his area (which would tend to decrease his probability of switching
industries), the probability that he switches industries upon re-employment will most likely
be higher if he is displaced as part of a mass layoff rather than as an isolated event that does
not affect many other workers in the industry. Similarly, even a worker displaced from a
highly concentrated industry may be more likely to switch industries upon re-employment
if the turnover rate in her former industry declines. Thus, in order to capture effects of
geographic concentration on displaced worker industry switching, it is important to control
for the nature of the worker’s layoff as well as the turnover rates in industries in the area.

An example which illustrates the possible inﬂuences of geographic industry
concentration, type of layoff, and changes in industry turnover rates is given in Table 3. For
hypothetical workers, Table 3 compares the probability of switching to a different industry
after displacement, while allowing geographic industry concentration, type of layoff, and
turnover rate to vary.24 There are two industries in the area. Industry A, the layoff industry,
is the larger of the two. Industry B, from which no workers are laid off, is the smaller

industry. Panel I shows outcomes for an idiosyncratic layoff of one worker from industry

 

2’ In order to simplify the example in Table 3, I assume that hiring is random. Thus (ignoring concentration,
turnover rates, and type of layoff), any worker would have the same probability as any other worker of being
hired at either industry. However, the idea that human capital can be industry speciﬁc implies that hiring
would not occur randomly, which in turn implies that workers would probably have lower probabilities of
switching industries than those proposed in Table 3.

25

A, while Panel 11 shows outcomes for a mass layoff of half of the workers from industry A.
Columns (1) and (3) Show outcomes for workers displaced when industry A is highly
concentrated (accounting for 90 of the 100 jobs in the area). Columns (2) and (4) show
outcomes for workers displaced when industry A is less concentrated (but still larger than
industry B, accounting for 60 of the 100 jobs in the area). The bottom three rows of Table 3
show the probability that workers who are laid off from industry A will switch to industry
B. This probability varies with type of displacement and with different turnover rates in
industry A following displacement.25

There are three conclusions which can be drawn from the example in Table 3. First,
as the turnover rate in the layoff industry falls, the probability of switching industries rises,
since when the number of positions being ﬁlled in their former industry declines, many
workers will simply move into other industries. Second, the probability of switching
industries is higher for a worker displaced as part of a mass layoff than it is for a worker
displaced as part of an idiosyncratic layoff. For a given level of concentration and a given
turnover rate in their former industry, workers displaced as part of a mass layoff would be
less likely to ﬁnd positions in their former industry than those displaced as part of an
idiosyncratic layoff because there are fewer positions remaining in the mass layoff industry
after displacement. Third, for a given turnover rate in industry A and either type of layoff,

workers displaced ﬁom highly concentrated industries are less likely to switch industries

 

2’ As industry B does not layoff any workers in the scenario presented in Table 3, its turnover rate remains
constant at 10 percent. If the turnover rate in industry B is lower than 10 percent (for example, if it falls as a
result of declining local economic conditions), the probabilities given in Table 3 will be smaller.

26

upon re-employment than their counterparts in less concentrated industries.26 This result
arises because, for a given turnover rate and after a similar type of layoff, the highly
concentrated industries have more remaining positions than the relatively less concentrated
industries.

The general conclusions reached from the example in Table 3 are summarized in
Panel A of Table 4. Workers should be more likely to switch industries upon re-
employment (as indicated by a (+) sign) if they are displaced from industries which are less
concentrated in their local area. Similarly, workers should be more likely to switch
industries upon re-employment if they are displaced as part of a mass layoff, since the
number of jobs in their former industry is decreasing in the local labor market. However, as
discussed above, if a worker is displaced as part of a mass layoff from a concentrated
industry, the industry switching outcome is inﬂuenced by two opposing factors, the
negative impact of employment share and the positive impact of having been displaced as
part of a mass layoff.

As stated above, switching industries is predicted to have a negative impact on
earnings regardless of the nature of the layoff. Combining this prediction with the effect of
geographic industry concentration on the probability of switching industries allows for
“reduced form” predictions of the effect of geographic industry concentration on earnings

losses, which are presented in Panel B of Table 4. Displaced workers should suffer

 

26 This conclusion is reached by comparing the probabilities in column 1 against the probabilities in column 2,
and the probabilities in column 3 against the probabilities in column 4.

27

relatively larger losses and be worse off (indicated by a (-) sign) if displaced from a less
concentrated industry, since it is unlikely that they will be re-employed in one of the
relatively few jobs in their former industry. Workers should also tend to suffer larger losses
if they are displaced as part of a mass layoff, since it is more likely that they will switch
industries upon re-employment. Thus, the strongest predictions are for highly
concentrated/idiosyncratic layoff and less concentrated/mass layoff situations. In the next
section, I formally test the effects of geographic industry concentration while controlling for

both the nature of layoffs and the economic conditions of the local labor market.

6. Statistical Framework and Empirical Results

6.1 Sample Characteristics. Table 5 presents the mean values of the variables
used in this study. Displaced workers who became re-employed experienced earnings losses
(measured as the difference between pre- and post-displacement earnings) of about $44 per
week. The sample is 60 percent male, 88 percent white, and on average 41 years old with
8.7 years of job tenure. In addition, 63 percent of the re-employed workers switched

industry after displacement.

6.2 Industry/Area Characteristics. The employment share (ES) variable measures
the level of concentration of the workers’ industries in their area at the time of
displacement. Change in employment share (year of displacement to year after

displacement) and change in employment share (1986-1992) measure both short- and long-

28

term changes in geographic concentration. The average worker was displaced from an
industry which accounted for approximately seven percent of local MSA employment at the
time of his displacement. From the year of displacement until the year afterward, the
average worker’s industry became less concentrated in the local area, although the change
in concentration was very small (-.002). Even over the longer period, 1986 to 1992,
industry concentration changed by only a small amount (-.007 on average).

The variable FELL] 0 is used to control for the nature of workers’ layoffs. F ELL] 0

is a dummy variable equal to one if the number of jobs in the worker’s industry within the

worker’s MSA fell by 10 percent or more from the year of displacement until the year after
displacement. In this sample, approximately 19 percent of the workers were displaced from
industry/MSA combinations which fell by 10 percent or more during the year after
displacement, suggesting that approximately one-ﬁfth of the workers in this sample were
displaced as part of mass layoffs.

I control for longer term changes in the labor market using the variables change in
industry employment (1986-1992) and change in MSA employment (1986-1992). Although
average MSA employment grew by about 178,000 jobs over the six year period ﬁ'om 1986
to 1992, workers were being displaced from industries which declined by about 8,500 jobs

on average over the period.

29

6.3 Re-employment. Table 6 summarizes the results of OLS regressions of
employment on the geographic industry concentration variables, the local labor market

controls, and demographic controls, as speciﬁed in equation (1).

(1) empl = bO + b. x + b2 z + b, 1),, + b, Es +b5FELL10 + u

Whether a worker is re-employed at the time of the survey is dependent upon both
local economic conditions at the time of displacement (Z) as well as individual
characteristics (X). The variables D,,,,27 are dummies for the year of displacement (1990
excluded). Inclusion of these dummies captures the idea that workers displaced in, 1987,
for example, have had more time since displacement and may therefore have different re-
employment outcomes than workers displaced closer to the survey date.28 The variables ES
and FELLlO are deﬁned as in Table 5.29

AS Table 6 shows, the level of geographic industry concentration has a positive and

marginally signiﬁcant impact upon the probability that a displaced worker will become re-

 

” The subscripts used throughout this section are deﬁned as follows: t==time of survey and s=years since
displacement.

28 Although not shown in the table, the coefﬁcients on the all of the year of displacement dummies were
positive and signiﬁcant (the dummy for displacement in 1990 was excluded).

29 Similar regressions could be run with the employment share variables deﬁned for the worker’s re-
employment (rather than displacement) industry. However, I focus on workers’ former industries in this paper
since they are the most relevant from a policy perspective. The geographic concentration rates of the workers’
re-employment industries are unknown at the time of displacement, and are therefore less relevant when
tailoring policy to suit different local labor market conditions.

30

employed.30 For a one percentage point increase in employment share, the probability of
re-employment increases by about .002, or about .2 percent for the average displaced
worker. Workers displaced from mass layoff industries were slightly more likely to be re-
employed than other workers. Other variables had generally small and insigniﬁcant impacts

on the probability of re-employment.31

6.4 Industry Switching. Equation (2) presents a model of industry switch. The
variable switch is equal to one if a worker switched detailed industries upon re-

employment. The other variables are deﬁned as in equation (1).

(2) switch = b0 + b, x + b2 2 + b, D,_, + b, Es +b, FELLIO + u32

 

30 For the re-employment regressions, the employment share of a worker’s major industry had only slightly
larger effects upon re-employment than employment share of a worker’s detailed industry.

3’ In addition to the regression speciﬁcations reported in Table 6, I ran speciﬁcations which included
change in employment share (year of displacement to year after displacement), change in employment
share (1986-1992), as well as an interaction between ES and FELLIO. The coefﬁcients on all three
variables were positive but insigniﬁcant in all speciﬁcations and the other coefﬁcients in the speciﬁcations
remained qualitatively unchanged.

’2 Although equation (2) models switch behavior, there may be unobserved individual effects which affect
whether a worker switches industries upon re-employment. For example, some workers may be more inclined
to search for work in different industries or may have greater ability to learn new, industry-speciﬁc skills.
These unobserved effects will inuoduce bias into estimates of equation (2) if they are correlated with the right
hand side variables. Estimates of equation (2) which control for these ﬁxed effects would require the use of
panel data over at least two periods on individuals in order to "difference out" the ﬁxed effect, a technique
which is unworkable using the DWS.

31

Estimates of equation (2) are provided in Table 7. The coefﬁcients on the
employment share variable show that the more concentrated a worker's displacement
industry, the less likely it is that the worker will switch industries upon re-employment. For
example, a one percentage point increase in employment share is estimated to decrease the
probability of switching industries by about .011, or about 1.75 percent for the average
displaced worker. A one standard deviation increase in employment share is estimated to
decrease the probability of switching industries by about 12 percent for the average
displaced worker.33

As shown by the coefﬁcient on F ELLIO, workers displaced as part of mass layoffs
were slightly more likely to switch industries upon re-employment. However, the
signiﬁcance of the coefﬁcient on FELLlO falls when former industry and occupation
controls are included in the model. Other signiﬁcant variables in the industry switch
regressions include the unemployment rate in workers’ MSAs at the time of their
displacements, which had a negative and signiﬁcant impact on industry switching,
suggesting that a one percentage point increase in the unemployment rate is associated with
approximately a three percent decrease in the probability of switching industries for the

average worker. In addition, the positive and signiﬁcant coefﬁcient on weeks unemployed

 

33 In addition to the speciﬁcations presented in Table 7, I also ran speciﬁcations while including change in
industry employment (1986-1992), which had a negative and insigniﬁcant coefﬁcient in all speciﬁcations. 1
also ran speciﬁcations that included change in employment share (year of displacement to year after
displacement) and change in employment share (1986—1992). Both variables were negative and
insigniﬁcant in all speciﬁcations. Finally, in order to control for possible non-linear effects of BS on
industry switching, I included ES-Squared in all speciﬁcations. Although the coefﬁcient on ES-squared was
positive, it was insigniﬁcant.

32

suggests that workers who remained unemployed for longer periods of time were slightly
more likely to switch industries upon re-employment.

Table 8 ﬁrrther examines factors which inﬂuence industry Switching. Column (1) of
Table 8 repeats the Speciﬁcation presented in column (5) of Table 7, while using more
aggregated industry controls in order to estimate impacts of former industry on the industry
switching outcome. As shown in column (1 ), relative to workers displaced from the services
industry, workers in the construction; transportation, communications and public utilities;
and ﬁnance, insurance, and real estate (FIRE) industries were less likely to switch industries
upon re-employment. Column (2) includes interactions between workers’ former industries
and ES, thus allowing the impact of E8 to vary across different industries. For example, for
the average worker displaced from the construction industry, a one percentage point
increase in ES is associated with a 4.90 percentage point decrease in the probability of
switching industries. The effect of BS on industry switching is larger for workers displaced
from the FIRE and transportation, communications, and public utilities industries, and

smaller for workers displaced from the manufacturing and trade industries.

6.4.1 Sample Selection Issues. It should be noted that switch is only observed for
displaced workers who become re-employed. If workers who become re-employed are not
necessarily representative of displaced workers in general, failure to control for the

difference between re-employed and unemployed displaced workers will lead to biased

33

estimates of the coefﬁcients in equation (2).34 To test and correct for possible sample
selection bias, I run a two-step sample selection correction model. In the ﬁrst step, re-
employment is regressed on all of the independent variables in the model, plus two factors
which affect re-employment (a dummy variable for the presence of children under 18 years
old in the household and a variable for the number of wage earners in the household).35 The
inverse Mills ratio is computed from this regression and included as a regressor in the
second-step switched industries regression. Sample selection tests reveal no evidence of
selection bias in the model and the coefﬁcients in the corrected speciﬁcations do not differ

substantially from those presented in Table 7.

6.4.2 Endogeneity Issues. Apart from the sample selection issues raised when
estimating equation (2), it is also possible that the variable weeks unemployed is

Simultaneously determined with switched industries upon re-employment. Factors which

 

3" For this sample, which includes only workers displaced at least one year prior to the survey date, re-
employment rates are extremely high (92 percent re-employed as of the survey date). As noted in Section 2,
this is not the case for more recently displaced workers, who had signiﬁcantly lower re-employrnent rates.
Sample selection bias is likely to be a more relevant issue when estimating industry switch and earnings loss
regressions for the recently displaced.

35 Those individuals with children at home may have different re-employment experiences than others because
of their family responsibility. Those who have other earners in the household may have more time to search
and may have less pressure to take the ﬁrst job available. In making sample selection corrections, I ﬁrst
employed the Heckman selection correction procedure, which resulted in little change in the size and
signiﬁcance of the results presented. However, the Heckman procedure is not well behaved when estimating
this model, and may be inappropriate for this model structure (dummy selection variable, dummy regression
variable). I also use three other speciﬁcations to test for sample selection bias: a linear probability model for
both the switch and re-employment equations, a probit model for both switch and re-employment, and a
linear probability model for re-employment and a probit model for switch. All four procedures gave no
evidence of selection bias and did not generate signiﬁcantly different results from the uncorrected estimates.

34

affect industry switching (employment share, for example) could also affect weeks
unemployed, causing bias in the estimates of equation (2). To correct for this possible
endogeneity problem, I use a two-stage least squares procedure. The presence of children
under 18 in the household and the number of wage earners in the household variables are
used as instruments for weeks unemployed in the ﬁrst stage.36 Estimates from this procedure
are presented in Appendix Table A3. Although the predicted effect of ES remains negative,
it becomes insigniﬁcant in the two-stage least squares estimation. The effect of weeks
unemployed remains positive but also becomes insigniﬁcant. The p-values from tests of the
endogeneity of weeks unemployed (shown in Table A3) suggest that the null hypothesis
that weeks unemployed is exogenous in the switched industry regressions can be rejected at
the two percent signiﬁcance level. In addition, Hausman tests could not reject the null
hypothesis that the industry switch model was overidentiﬁed when using the children and
number of earners in the household variables as instruments for weeks unemployed. The p-

values from this test are also reported in Table A3.37

 

36 Individuals with children in the household may have different values for weeks unemployed than other

workers if their family responsibilities cause them to engage in shorter job searches. Similarly, those with
fewer wage earners in the household may also engage in shorter searches. The use of the children and number
of earners in the household as identifying variables in both the sample selection correction and the two-stage
least squares procedures is justiﬁable since the two outcomes (weeks unemployed and re-employment) are
closely related and since I do not make sarnpie selection corrections when estimating the two-stage least
squares regressions.

37 For the Hausman test, the residuals from the second stage switched industry regression were regressed on
all of the regressors from the ﬁrst stage and the instruments. The LM statistic (number of observations
times the R-squared from the residual regression) is distributed as chi-squared with one degree of freedom.
For the endogeneity tests, the residuals from the ﬁrst stage weeks unemployed regression were included in
the switched industry regression, along with weeks unemployed and all other exogenous variables. A t-test
was then performed on the included residual’s coefﬁcient.

35

6.5 Earnings Loss Model. In order to estimate the effects of the employment
share variables on the earnings losses of displaced workers, I use an earnings loss model
similar to those used by Podgursky and Swaim (1987)38 and Howland and Peterson (1988).
In this model, earnings losses are measured as the difference between a displaced worker's

current (1992) earnings and his earnings at the time of displacement, as shown in equation

(3).

(3) (yt - yt_S ) = b0 + b, X + b2 switch + b3 Z + b4 D,_S + b5 ES +b6 FELLIO + u

The variables yt and yt,s represent the log of weekly earnings on the worker's current
and former jobs, respectively. The X, Z, and D,_s vectors and the variables switch, ES and
FELLlO are deﬁned as in equation (2). Estimates of equation (3) are presented in Table 9.39

Regressions which include only ES, DH, and demographic controls (column (1))
reveal no statistically signiﬁcant reduced form effects of employment share, although the
coefﬁcient implies that workers displaced from more concentrated industries tend to

experience smaller earnings losses than workers displaced from less concentrated industries.

 

38 Podgursky and Swaim (1987) use current earnings minus former earnings as a measure of earnings loss.
They inﬂate former earnings by a trend index of growth in earnings (by industry and occupation). They use
the Employment Cost Index to obtain the inﬂation coefﬁcients.

39 Because the regressions include both a constant and year of displacement dummies, the dependent variable
measures real earnings changes.

36

That no statistically Signiﬁcant reduced form effects of employment share on earnings
losses arise despite the statistically strong impacts of employment share on switched
industry and of switched industry on earnings losses is problematic. Recall, however, that
the employment share variables have only a small amount of explanatory power in the
switched industry regressions in Table 7 (the R-squared from the regressions of switched
industry on the employment share variables is only about .03). Thus, although employment
share affects industry switching, it does not explain enough of the variation in industry
switching to produce strong reduced form effects on earnings losses.40

Workers who switched industries upon re-employment experienced about 13
percent smaller earnings changes (13 percent larger losses) than those who remained in the
same detailed industry upon re-employment. Workers who spent more weeks unemployed,
workers in more populated areas, and workers whose MSA employment declined over the
period 1986-1992 also experienced larger earnings losses. Although the coefﬁcients are not
shown, workers with higher tenure and workers who changed from full-time to part-time

job status also experienced larger earnings losses.

 

4° The issue at hand can be described as attempting to infer the reduced form effect of employment share

(ES) on earnings losses (LOSS), where employment share affects earnings losses only through industry
switching (SWITCH):

a LOSS / 6 ES = a LOSS / 6 SWITCH “ a SWITCH / a BS

I ran simple inference tests of the net effect of employment share on earnings losses by jointly estimating the
two relationships above: earnings losses as a function of switched industry and switched industry as a function
of employment share (including demographic and year of displacement controls in both regressions). The
product of the coefﬁcient on employment share in the switched industry regression (-l.l 1) and the coefﬁcient
on switched industry in the earnings loss regression (-.13) is the predicted partial effect of employment share
on earnings losses (.14). The standard error for this predicted coefﬁcient is about .05, implying that the
predicted partial effect is highly statistically signiﬁcant.

37

Table 10 presents earnings change regressions, by gender, for the speciﬁcations in
columns (5) and (6) of Table 9. The regressions are separated in order to estimate possible
gender differences in the effects of industry Switching, weeks unemployed, and other
industry/area variables. Separating the regressions by gender reveals that although the effect
of switching industries is negative and signiﬁcant for both males and females, the effect of
switching industries is larger for females than for males. The negative effect of weeks
unemployed is also slightly larger for females than for males. Other effects are generally

similar for the two groups.

6.5.1 Sample Selection Issues. As with equation (2), tests for possible sample
selection bias were run using the standard Heckman procedure (the children and number of
earners in the household variables were used as identifying variables). The procedure gave
no evidence that sample selection causes bias in the earnings loss regression and the

corrected coefﬁcients did not change in any relevant way.

6.5.2 Endogeneity Issues. Both the switched industries and weeks unemployed
variables may be endogenous in the earnings loss model because factors that affect the size
of a worker's earnings loss may also affect whether the worker switches industries upon re-

employment.4| For example, a worker with high levels of general human capital may be

 

4' Recall that workers who moved after displacement are excluded from this analysis. The moving effect
parallels the industry switch effect. It is likely to be endogenous in an estimate of earnings losses, and its
exclusion introduces bias if one extends the results presented here across all displaced workers, instead of
applying them to only those who do not move as a result of displacement.

38

more likely to switch industries upon re-employment since her human capital is more easily
transferable across industries. If her general human capital Skills are highly rewarded in the
labor market, her subsequent earnings losses will be affected. Alternatively, a worker with
large amounts of industry-speciﬁc skill may want to remain in the same industry when
searching for re—employment. If that particular industry is declining, her industry-speciﬁc
skills will earn lower compensation upon re-employment.

Estimates which correct for the possible endogeneity of switched industries and
weeks unemployed using a two-stage least squares procedure are presented in Table A4. The
two employment share variables, the dummy for the presence of children in the household,
and the number of earners in the household variables are used as instruments for switched
industries and weeks unemployed in the ﬁrst stage. The Sign of the switched industries
coefﬁcient does not change in the two-stage estimation, although it becomes insigniﬁcant.
There were no other important differences from the uncorrected estimates, with the
exception that the coefﬁcient on weeks unemployed became insigniﬁcant. The p-values
from tests of endogeneity of the switched industry and weeks unemployed variables are
provided in Table A4, along with p-values from Hausman tests of overidentiﬁcation. The
endogeneity tests suggest that the null hypotheses that switched industry and weeks
unemployed are exogenous cannot be rejected at conventional signiﬁcance levels. In
addition, the Hausman tests could not reject the null hypotheses that the model was

overidentiﬁed.

39

7. Conclusion

This paper uses the Displaced Worker Survey to test the effects of geographic
industry concentration on displaced worker re-employment experiences and earnings losses.
I ﬁnd that geographic industry concentration affects the re-employment experiences of
displaced workers, primarily through its inﬂuence upon workers’ industry switch outcomes.

Workers displaced from industries concentrated in their local labor market were less
likely to have switched industries upon re-employment. For example, a one percentage
point increase in employment Share was found to decrease the probability of switching
industries upon re-employment by about 1.75 percent. Alternatively, a one standard
deviation increase in employment share was found to decrease the probability of switching
industries upon re-employment by about 12 percent for the average displaced worker. I also
ﬁnd that the impact of employment Share on industry switching varies across industries,
with effects being particularly large in the FIRE and transportation, communications, and
public utilities industries.

Like previous researchers, I ﬁnd that workers who switch industries upon re-
employment experience larger earnings losses than their counterparts who do not switch
industries. The negative impact of industry switching on earnings losses was slightly larger
for females than for males. In addition, reduced form estimates of earnings losses on
geographic concentration variables reveal positive but insigniﬁcant effects of geographic
concentration on displaced worker earnings losses. These results suggest that geographic

concentration may play an indirect role in affecting displaced worker earnings losses.

40

Although this paper is not a formal policy analysis, the results presented here have
implications for displaced worker policy. The relatively sparse literature on the
effectiveness of training for displaced workers suggests that while general job search
assistance training is an effective way to increase the earnings and employment of displaced
workers, both speciﬁc on—the-job training and more general classroom training for displaced
workers has little, if any, impact upon the earnings and employment of displaced workers.
Corson et. al. (1993) conclude that impacts of training programs for displaced workers may
be stronger for workers who switch industries. This conclusion, combined with the results
on industry switching presented here, suggests that policy makers designing programs may
more efﬁciently allocate resources for displaced workers by altering their assistance
programs based upon industry concentration in local labor markets. As workers displaced
from concentrated industries are less likely to switch industries upon re-employment,
resources designated for these workers may be better spent on relatively inexpensive job
search assistance programs rather than on training programs which focus upon skills
required for jobs in different industries. On the other hand, workers displaced from non-
concentrated industries are more likely to switch industries upon re-employment, and
assistance which focuses upon re-training workers to learn skills useful in different
industries may be more appropriate for them. Thus, the evidence presented here can be
combined with information currently available from local industry councils, from local
employment offices, and from interviews with individual workers and be used to assist

policy makers in assessing and designing appropriate policy responses to displacements.

41

 

 

8.- 2.- w _ .- 3.- 3.- E; 8.- aEeam E amaze 03520..

5:.- Rﬂé- it. 3.2 I. 3: _- 8.3- 3.8- amass awEEam £83 :3

8. 8. 2. 8. 8. so. 8. 22% E2835

ca. 2: 2: of w». as. 3. Sum EQEEEEm-om

9A R mom 2 as. 3 £2 :88 82..an .3832
$0322

3.. S- _ T an- R.- 3- £- 353 E owEEO cassava

31E- Nmim- 3.2.- 8.2:- m _ .8- 3N2- ©23- owEEO mmEssa £33 :3

8. 8. m _. S. 8. 3. 8. 225 E3335

3. a. 8. a. a. 8. a. cam a=o§o_aEm-om

cm x: an a E. m2 2: $88 REESE c3532
manor—:02

m _ .- 2.. m _ .- 8.- 8.- R.- 8- $55 E Esau S8528

8.”...- ew. E- 8.3- 3.8.- 3.3- $9- 8.3- mango awESE has? .8”.

8. so. w _. 8. 8. 3. S. 33% E3335

3. a. 3. so. 8. a». 8. mama “55835-01

...a N2. 32: Mam as. am NSC $88 BESS 33:2
.98

38mm :31 ova:-
85235 :59: 82:5 0:25
moatom 855m 282055 :ozmtoamcmF 3:38:52 canon-3:00 EEO-H-

 

ausa-$3 23.83 83.35.. assess-.26

— manta—-

42

<3 do .5 .5 .02 .5 .22 .>2 .5 .9 .8 .N< 083 an .08 .5 .5. .2: .02 51.2.2: .<> .8 .02 d2 do J: .0:
.5 H€50 ”am dz .m2 .0: .22 .3. .<_ .E .:o :2 .2: .-= 3:032: 2.: .>2 .2 .:-> .5 .22 <2 .52 .5 3805.2 032.0: a a. 8:02 58 05:58 3:90 0

Ace—"vma—V 08::— 0E£ 583300 05 .3 me50 BEEo: 55:0:
.3 359:8 0:03 mmE50 .00: .32 555.. ES. 20800256 :8 0E: 05 69-: 90%: m0_::m=:E :5 m:Ew0: 05 ha me50 50—003 E 0w5:0 50:: 05 5 :05502 m

.:Emm=om_: :o: :8: 00m .:Ew0: 05 E .O:m::E :50 8 038.535 :Ew0: :50 E ::0E.Ao_:E0 :22 :o 055080: 05 m: 35.0: m: 055 ::0E.€.:Em v

A:Ew0:

:8 .C:m::E 05 E 5532: :22 05 :0 $9520: a 5 32 E :0..Ao_:E0-0: 0:03 0:3 03:50- 0.: E 0:00:03 3025:: :o :0:E=: 0...: 5 3:23:00 53 05: 20:32:80-0: m

.038 05 ES: :0:=_0x0 05 3:00:03 8an 3055:0855 :8 3:385

553 ..:o5::m_EE:: 0:59. :5 3:00:95 ooodwm .A_05::_xo::5 :o: 3585 5023 10:6: 00:30:50 .933: 05 E $00.83 03000: 030: 08:05 E?- :o: 0: 0:09:32 a

.38 :5: :0 5630.2: :8 :8: 00m .82 :5 two. :00.$:0n 3025:: 0:03 0:3 :5 2080530.: :05 0.6:: 8: 2: 2:: 0:00:03 2 :0.:0: 3:33: me50
b.4003 :5 .055 60:32:80 .05: ::0Exo_:E0-0: 0:; .mao. :5 52 50.50: 08: x5 5 3025:: 0:03 0:3 :5 E08825: :05 m:Ew0: 30: 9 Biz: 2:: 0:00:03
0:20E 8:29.: 5035:: 5:25: 0.:- .~oo_ E 8:028: 9:00:03 05 o: 0:0:0: :Ew0: .59.: 3025:: 53 :0x:o.$ 05 .932: 05 9 0:0:0: .903:— .m¢Em :0 2028: E05 :0
E25235 05 :o 1:03 E5? .336 :o 3:50? :53 .«o 3500: no“ a :8. of» 05:0: :0 $50» 0:2: .8 m 5:: 38:0: 2 :30: 8:5 .3053: 0w5no 0w?» :5 20583600:
5: x0>::w 50:03 :05EEQ 05 :5 ”$053: 055 ::0E.Ao_:E0 :0: :5: 3:05: mm0Em=m bgoo 0.:- Am05wc 0:00:03 3025:: :0 59:3: 8.: $3: :0::5O M0053 H

meEam E 0w5:0 033020:

$.- 8. 8. _ T 8.- 3.- :.-
NS- :S: 03 :3...- 005- N23- 58- 0052.: meeﬁ 30.8.5 :8:
8. 8. 2. 8. 8. 8. S. 220 0:05:23:
5. 8. a. 8.: a. 8.: a. 2.: 2053350:
8: m: x: :0 3m 2: z: _ :88 88.3.5 5:52

amok,
5.- .2.- 0N- m _ .- 0m- 2; 5.- 5.5: E 00:20 005520:
8%- 5.5. 83- N _ .8- a3:- mﬂa- 8. a- 0052.: meEam 5003 .8:
3. S. a. 8. 8. 0o. 8. 0:25 20.52%:
0:. 3. 0o. 8.: 8. 8. g. 03: 22335-0:
05 a: 0mm 2.: 80 S: 3: :88 883:: .0552

.528: _ :33:-

Snow

Share of Industry

Employment

.270 Furniture and Fixtures
Manufacturing

.249 Personal Services

.183 Personal Services

.179 Business Services

.172 Textile Mill Products
Manufacturing

.145 Textile Mill Products
Manufacturing

.140 Other Transportation
Equipment
Manufacturing

.137 Machinery, Except
Electrical
Manufacturing

.136 Apparel and Other
Finished Textile
Manufacturing

.136 Lumber and Wood

TABLE 2

Most and Least Concentrated Industries and Areas
excluding retail sales1

 

Highest Ten

Products Manufacturing

Area

Hickory NC

Las Vegas NV

Reno NV

Washington DC

Anderson SC

Hickory NC

Wichita KS

Rockford IL

Anderson IN

Eugene-
Springﬁeld OR

 

Lowest Ten
Share of Industry
Employment
.00003 Forestry and Fisheries

.0001

.0003

.0003

.0003

.0004

.0005

.0006

.0006

.0007

 

Motor Vehicles and
Equipment
Manufacturing

Motor Vehicles and
Equipment
Manufacturing

Motor Vehicles and
Equipment

Manufacturing

Electrical Machinery and
Supplies Manufacturing

Petroleum and Coal
Manufacturing

Leather and Leather
Products Manufacturing

Mining

Mining

Motor Vehicles and
Equipment
Manufacturing

Area

Los Angeles—
Anaheim-
Riverside

Las Vegas NV

Pittsburgh PA

Worchester MA

Anchorage AK

New York-New
Jersey-Long
Island
Cincinatti-
Hamilton

Milwaukee-
Racine

Norfolk-Virginia
Beach- Newport
News VA

Boston-Laurence-
Salem MA

I The Retail Sales industry actually covered all ten of the highest concentration ﬁgures, ranging from .353 to .298. 1
list the top ten most concentrated non-retail sales industries in the table.

43

TABLE 3

Comparison of Switch Rates '

 

 

 

 

Panel I Panel II
Idiosyncratic Layoff - Mass Layoff -
One Worker from Industry A 50% of Workers from Industry A
(1) (2) (3) (4)
Industry A Industry A Less Industry A Industry A Less
Highly Concentrated Highly Concentrated
Concentrated Concentrated
Number of jobs
in A before
layoff 90 60 9O 60
Number of jobs
in A after layoff 89 59 45 30
Number of jobs
in B 10 40 10 40

Turnover Rate
in
A = 10%

Turnover Rate
in
A = 5 %

Turnover Rate
in
A = 1%

 

Probability of re-employment

Probability of re-employment

 

 

in industry B in industry B
(1/9.9) = .10 (4/9.9) = .40 (1/5.5) = .18 (4/7) = .57
(1/5.45) = .18 (4/6.95) = .58 (1/3.25) = .31 (4/5.5) = .73
(1/1.89) = .53 (4/4.59) = .87 (1/1.45) = .69 (4/4.3) = .93

 

 

 

I The table compares the probability of switching to another industry for hypothetical workers displaced in four
different displacement/market scenarios (idiosyncratic layoff from highly concentrated industry, idiosyncratic layoff
from less concentrated industry, mass layoff from highly concentrated industry, mass layoff from less concentrated
industry) and three different turnover rates in the industry which is laying off workers (industry A). The non-layoff

industry (B) has a turnover rate of ten percent in all cases.

44

 

TABLE 4

Predicted Effects on Industry Switchinﬂnd EamingLIJosses

 

 

 

 

 

Panel A' Panel B
Predicted Effect on Switching Industries Predicted Effect on Earnings Losses
Mass Layoff Idiosyncratic Mass Layoff Idiosyncratic
(+) Layoff (-) Layoff
(-) (+)
Displaced Displaced
from highly from highly
concentrated ( + /-) (-) concentrated (+ / -) ( +)
industry industry
(-) (+)
Displaced Displaced
from less from less
concentrated (+) ( +/—) concentrated (-) (+/-)
industry industry
( +) 1 (-)

 

 

 

 

1 A negative sign in panel A indicates that the worker is less likely to switch industries upon re-employment. A
negative sign in panel B indicates that the worker suffers a larger earnings loss (i.e. they are worse off).

45

 

TABLE 5

Mean Values of Variables1

 

Sample Characteristics

 

Current weekly earnings (1992)

Weekly earnings on former job

Change in weekly earnings

Re-employed

Weeks unemployed

Switched detailed industry upon re-employment ‘2
Years of tenure on former job

Full time (>35 hours/week) on former job *
Age

Male *

White *

Married *

Industry/Area Characteristics

 

ES - Employment share of former industry in year of displacement

Change in employment share of former industry
(year of displacement to year after displacement)

Change in employment share of former industry (1 986-1992)
Unemployment rate in MSA (year of displacement)

F ELLIO - Industry employment fell by 10 percent or more
(year of displacement to year after displacement)‘

Change in Industry Employment (0008) (1986-1992)
Change in MSA Employment (000s) (1986-1992)
Another MSA within 25 miles *

Number of observations

 

' Standard deviations are reported in parentheses.

459 (258)
505 (308)
-44 (219)
.92 (27)
17'(21)
.63 (48)
8.72 (6.55)
.96 (20)
41 (10)
.60 (49)
.88 (32)

.69 (46)

.07 (.08)

-.002 (.008)

4007 (018)
5.15 (1.45)

.19 (39)

-8.56 (76.33)
178 (279)
.37 (48)

753

2 Variables denoted with an asterisk are dummy variables equal to one if the condition holds and zero otherwise.

46

TABLE 6

Re-Employment Effects

 

Dependent Variable: Employed in January 19921

 

(1) (2) (3) (4)
Employment Share of Former .22*2 .22* .20* .17
Industry in year of displacement (ES) (.12) (.12) (.12) (.26)
FELLIO - .02 .02 .06**
(.03) (.03) (.03)
Unemployment rate in MSA - - .00 .01
(year of displacement) (.01) (.01)
Percent Change in MSA - - .09 .04
Employment (1986-1992) (. 10) (.1 1)
Another MSA within - - -.01 -.02
25 miles (.02) (.02)
Former Industry and no no no yes
Occupation dummies
Adjusted R-squared .03 .03 .03 .03

 

1 n = 753. All coefficients result from linear probability (OLS) regressions. White (heteroskedasticity robust) standard
errors are reported in parentheses. Probit speciﬁcations of these regressions are presented in Table A1. The qualitative
results are similar to the results presented here. All regressions also include demographic controls (tenure on former job,
full-time on former job, age, age-squared, white, married, male), four schooling dummies, and dummies for year of
displacement (l989-l987), with those displaced in 1990 as the excluded group. The coefﬁcients for these variables are not
reported in the table, but are available from the author.

2 " Signiﬁcant at the 5 percent level. * Signiﬁcant at the 10 percent level.

47

TABLE 7

Switch Effects

 

Dependent Variable: Switched Industries upon Re-EmploymentI

 

(1) (2) (3) (4) (5)
Employment Share of Former -1.11M2 -1.13** -l.13** -1.10** -1.49**
Industry in year of displacement (.25) (.25) (.25) (.24) (.65)
FELLIO - .09** .10** .10** .03
(.04) (.04) (.04) (.05)
Unemployment Rate in MSA - - -.02* —.02* -.02*
(year of displacement) (.01) (.01) (.01)
Percent Change in MSA - - .27 .27 .25
Employment (1986-1992) (.22) (.22) (.22)
Another MSA within - - .03 .02 .02
25 miles (.04) (1)4) (.04)
Weeks Unemployed3 - - - .002** .001*
(.001) (.001)
Former Industry and no no no no yes
Occupation dummies
Adjusted R-squared .03 .03 .04 .05 .14

 

l n = 693. All coefﬁcients result from linear probability (OLS) regressions. White (heteroskedasticity robust) standard
errors are reported in parentheses. Probit speciﬁcations of these regressions are presented in Table A2. The qualitative
results are similar to those presented here. All regressions also include demographic controls (tenure on former job, change
from full—time to part-time status. change from part-time to full-time status, age, age-squared, white, married, male), four
schooling dummies, and dummies for year of displacement (I989-l987), with those displaced in 1990 as the excluded
group. The coefﬁcients for these variables are not reported in the table, but are available from the author.

2 ** Signiﬁcant at the 5 percent level. " Signiﬁcant at the 10 percent level.

3 Regressions which correct for the possible endogeneity of Weeks Unemployed in this model are presented in Table
A3. The qualitative results are similar to those presented here. See text for discussion.

48

TABLE 8

Switch Effects by Industry and level of
Employment Share

 

Dependent Variable: Switched Industries upon Re-Employmentl

 

 

 

 

 

 

 

 

 

 

 

(l) (2)
industry effects industry * ES interactions
Employment Share of -l .25 -
Former Industry in year (.42)
of displacement (ES)
Agriculture and Mining .10 Agriculture and 3.09
(.1 l) Mining*ES (3.88)
Construction -.21"‘2 Construction*ES -4.90**
(.08) (1.22)
Manufacturing .01 Manufacturing*ES -2.54**
(.06) (.71)
Transportation, -.25** Transportation, -10.6l **
Communications, and (.09) Communications, and (2.15)
Public Utilities Public Utilities*ES
Trade .02 Trade*ES -1 .04**
(.08) (.27)
Finance, Insurance, and -.40** FIRE*ES -8.56**
Real Estate (FIRE) (.08) (2.31)
Services - Services*ES -.08
(.70)
Adjusted R-squared .13 .12

 

 

 

 

 

 

 

1 n = 693. All coefﬁcients result from linear probability (OLS) regressions. White (heteroskedasticity robust) standard
errors are reported in parentheses. All regressions also include all variables from column (5) of Table 7.

2 " Signiﬁcant at the 5 percent level. “' Signiﬁcant at the 10 percent level.

49

TABLE 9

Earnings Change Effects

 

Dependent Variable: log (current earnings) - log (former earnings)1

 

 

(I) (2) (3) (4) (5) (6)
Switched Industries - -.13M2 -.14** -.13** -.11** -.15**
upon re-employment3 (.04) (.04) (.04) (.04) (.04)
FELLlO - - .06 .07 .07 .05
(.05) (.05) (.05) (.05)
Unemployment Rate in MSA - - - -.02 -.02 -.02*
(year of displacement) (.01) (.01) (.01)
Percent Change in MSA - - - -.38* -.40* -.30
Employment (1986-I992) (.22) (.21) (.23)
Another MSA within - - - -.l4** -.13** -.14*"‘
25 Miles (.04) (.04) (.04)
Weeks Unemployed - - - - -.004** -.004"”"

(.001) (.001)

Employment Share of Former .10 - — - - _
Industry in year of displacement (ES) (.24)

Former Industry and no no no no no yes
Occupation Dummies

Adjusted R-Squared .17 .18 .19 .19 .21 .25

 

I n = 693. All regressions also include demographic controls (change in full-time or part-time status, tenure on fonner
job, age, age-squared, white, married, male), four schooling dummies, and dummies for year of displacement (19894987),
with those displaced in 1990 as the excluded group. The coefficients for these variables are not reported in the table, but
are available from the author.

2 ” Signiﬁcant at the 5 percent level. "‘ Signiﬁcant at the 10 percent level.

3 Regressions which use two-stage least squares to correct for the possible endogeneity of both switched industries
upon re-employment and weeks unemployed are presented in Table A4. See text for discussion.

50

TABLE 10

Earnings Chang; Effects by Gender

 

Dependent Variable: log (current earnings) - log (former earnings)l

 

 

 

 

(I) (2) (3) (4)

Males Females Males Females
Switched Industries -.08"‘2 -.l7" -.12*“' -.25"‘*
upon re-employment (.04) (.06) (.05) (.08)
FELLIO .03 .13 .03 .12

(.05) (.09) (.06) (.l l)
Unemployment Rate in MSA -.02 -.01 -.03"‘ -.01
(year of displacement) (.02) (.02) (.02) (.02)
Percent Change in MSA -. 19 -.61 -.07 -.67
Employment ( 1986-1992) (.26) (.3 7) (.29) (.45)
Another MSA within -. l 7" -.08 -.15** -. 16*
25 Miles (.05) (.07) (.06) (.09)
Weeks Unemployed -.003" -.005" -.003""‘ -.006“

(.OOl) (00]) (00]) (.001)
Former Industry and no no yes yes
Occupation Dummies
Adjusted R-Squared .18 .23 .21 .24
Number of Observations 409 284 409 284

 

1 All regressions also include demographic controls (change in full-time or part-time status, tenure on former job, age,
age-squared, white, married), four schooling dummies, and dummies for year of displacement (l989-l987), with those
displaced in I990 as the excluded group. The coefﬁcients for these variables are not reported in the table, but are available

from the author.

2 "' Signiﬁcant at the 5 percent level. "‘ Signiﬁcant at the 10 percent level.

51

APPENDIX

TABLE Al

Re-Employment Effects-Probit Model

Dependent Variable: Employed in January 1992I

 

(1) (2) (3) (4)

Employment Share of Former .19 .17 .16 .22
Industry in year of displacement (ES) (. 13) (.13) (.13) (.38
FELLlO - .02 .02 .05
(.02) (.02) (.02)

Unemployment Rate in MSA - - .00 .01
(year of displacement) (.01) (.01)
Percent Change in MSA - - .09 .05
Employment (1986-1992) (. 12) (. 12)
Another MSA within - - -.01 -.01
25 miles (.02) (.02)
Former Industry and no no no yes

Occupation dummies
Pseudo R-squared .09 .09 .09 .24

Number of Observations 2 753 753 753 581

 

1 All coefﬁcients represent partial effects on the dependent variable which are estimated using a probit model. Standard
errors are reported in parentheses. All regressions also include demographic controls (tenure on former job, full-time on
former job, age, age-squared, white, married, male), four schooling dummies, and dummies for year of displacement
(l989-1987), with those displaced in 1990 as the excluded group. The coefficients for these variables are not reported in
the table, but are available from the author.

2 The number of observations changes across columns because variables and observations which perfectly predicted
re-employment behavior were excluded from the regressions. The existence of perfect predictors also prohibits
reporting of the constant term.

52

TABLE A2

Switch Effects-Probit Model

 

Dependent Variable: Switched Industries upon Re-Employmentl

 

(1) (2) (3) (4) (5)
Employment Share of Former -1.10"""2 -1.l4** -1.l9** -1.13** -1.67**
Industry in year of displacement (ES) (.25) (.25) (.25) (.25) (.77)
FELLIO - .10** .11** .11** .04
(.05) (.05) (.05) (.07)
Unemployment Rate in MSA - - -.02* -.02* -.03
(year of displacement) (.01) (.01) (.02)
Percent Change in MSA - - .29 .30 .40
Employment (1986-1992) (.24) (.24) (.29)
Another MSA within - - .03 .02 .04
25 miles (.04) (.04) (.05)
Weeks Unemployed - - - .003“ .002*
(.001) (.001)
Former Industry and no no no no yes
Occupation dummies
Pseudo R-squared .05 .04 .05 .06 .18
Number of Observations3 693 693 693 693 635

 

1 All coefficients represent partial effects on the dependent variable estimated from a probit model. Standard errors are
reported in parentheses. All regressions also include demographic controls (tenure on former job, change from full-time to
part-time status, change from part-time to full-time status, age, age-squared, white, married, male), four schooling
dummies, and dummies for year of displacement (1989-1987), with those displaced in 1990 as the excluded group. The

coefﬁcients for these variables are not reported in the table, but are available from the author.

2 “ Signiﬁcant at the 5 percent level. " Signiﬁcant at the 10 percent level.

3 The number of observations changes across columns because variables and observations which perfectly predicted
switch behavior were excluded from the regressions. The existence of perfect predictors also prohibits reporting of the

constant term.

53

TABLE A3

Switch Effects-Two-Stirge Least Squares Estimation

 

Dependent Variable: Switched Industries upon Re-Employmentl

 

(1)2 (2)
Employment Share of Former -. 17 -.30
Industry in year of displacement (ES) (1.58) (2.09)
FELLIO .16 .21

(.12) (.17)
Unemployment Rate in MSA -.02 -.03
(year of displacement) (.03) (.03)
Percent Change in MSA .38 .43
Employment ( 1986-1992) (.60) (.49)
Another MSA within -. l2 -. 13
25 miles (.21) (.14)
Weeks Unemployed .04 .03

(-06) (.03)
Former Industry and no yes
Occupation dummies
P-value: test of endogeneity of weeks unemployed .02 .02

(null hypothesis: weeks exogenous)

P-value: overidentiﬁcation test .34 .31

(null hypothesis: model overidentiﬁed)

 

l n=572. White (heteroskedasticity robust) standard errors are reported in parentheses. Regressions also include
demographic controls (tenure on former job, change from full-time to part-time status, change from part-time to full-time
status, age, age-squared, white, married, male), four schooling dummies, and dummies for year of displacement (1987-
1989), with those displaced in 1990 as the excluded group. The coefﬁcients for these variables are not reported in the
table, but are available from the author. Two instruments were used in the ﬁrst stage Weeks Unemployed regression: a
dummy variable for presence of children under 18 years old in the household, and a variable for the number of wage
earners in the household. The number of observations differs from that in Table 7 because some individuals had missing

values for these instruments.

2 As the regressions presented in columns (1 )-(5) in Table 7 of the text do not include Weeks Unemployed, they do not
require the two-stage least squares estimation and are not reported here. Accordingly, column (1) above corresponds to

column (6) in Table 7, etc.

54

TABLE A4
Earnings Change Effects - Two-Stage Least Squares Estimation

Dependent Variable: log (current earnings) - log (former earnings)l

 

 

(I); (2) (3) (4) (5)
Switched Industries -.09 -.09 -.1 l -.79 -.56
upon re-employment (.20) (. 19) (.19) (1.00) (.58)
FELLIO - .06 .07 .20 .15
(.05) (.05) (.21) (.15)
Unemployment Rate in MSA - - -.02 -.03 -.03
(year of displacement) (.01) (.04) (.03)
Percent Change in MSA - - -.38"‘3 .03 .04
Employment (1986-1992) (.22) (.64) (.40)
Another MSA within - - -.14** -.22 -.18
25 Miles (.04) (.17) (.11)
Weeks Unemployed - - - .03 .01
(.05) (.02)
Former Industry and no no no no yes
Occupation Dummies
Adjusted R-Squared .18 .20 .19 .00 .00
Number of Observations“ 693 693 693 572 572
P-value: test of endogeneity of .84 .82 .91 .28 .77
switch and weeks unemployed
P-value: overidentiﬁcation test .37 .44 .97 .87 .67

 

1 White (heteroskedasticity robust) standard errors are reported in parentheses. All regressions also include demographic
controls (change in full-time or part-time status. tenure on former job, age, age-squared, white, married, male), four
schooling dummies, and dummies for year of displacement (1987-1989), with those displaced in 1990 as the excluded
group. The coefficients for these variables are not reported in the table, but are available from the author. Five instruments
were used in the ﬁrst-stage Switched Industries and Weeks Unemployed regressions: the three employment share variables,
the children under 18 in the household dummy and the number of wage earners in the household.

2 As the regressions presented in columns (1)-(2) in Table 8 do not include either the Switch or Weeks Unemployed
variables, they do not require the two-stage least squares estimation and are not reported here. Accordingly, column
(1) above corresponds to column (3) in Table 8, etc.

3 " Signiﬁcant at the 5 percent level. "' Signiﬁcant at the 10 percent level.

4 The number of observations differs from that in Table 8 because some individuals were missing the children under
I 8 and number of wage earners instruments.

55

BIBLIOGRAPHY

BIBLIOGRAPHY

Becker, Gary S. (1975) Human Capital, 2nd ed. New York: NBER.

Carrington, William J. (1993) “Wage Losses for Displaced Workers: Is It Really the Firm
that Matters?” Journal of Human Resources, 28(3):435-462.

Conant, Eaton H. (1965) "Report and Appraisal: The Armour Fund's Sioux City
Project," Monthly Labor Review, 88(1 1):1297-1301.

Corson, Walter, Decker, Paul, Gleason, Phillip, and Walter Nicholson (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.

Corson, Walter and Walter Nicholson (1981) "Trade Adjustment Assistance for
Workers," Research in Labor Economics, 4:417-469.

County Business Patterns (1993) County Business Patterns, 1 990-1991 on CD-ROM,
prepared by Data Access and Use Branch, Data User Services Division, Bureau of the
Census. Washington: The Bureau.

Diamond, Charles A. and Curtis J. Simon (1990) “Industrial Specialization and the
Returns to Labor,” Journal of Labor Economics, 8(2):]75-201.

Fallick, Bruce Chelimsky (1993) “The Industrial Mobility of Displaced Workers,”
Journal of Labor Economics, 1 1(2):302-323.

Flairn, Paul and Ellen Seghal (1985) "Displaced Workers of 1979-83: How Well Have
They F ared?" Monthly Labor Review, 108(6):3-16.

Gardner, Jennifer M. (1993) Displaced Workers 1987-1991. Bulletin 2427. Bureau of
Labor Statistics. July.

Gordus, Jeanne R, Paul Jarley, and Louis A. F erman (1981) Plant Closings and
Economic Dislocation. Kalamazoo, MI: W.E. Upjohn Institute.

Hamermesh, Daniel S. (1989) "What Do We Know About Worker Displacement in the
US?" Industrial Relations, 28(1): 51-9.

56

57

----------- (1987) "The Costs of Worker Displacement," Quarterly Journal of
Economics, 102:51-76.

Heckman, James J. (1979) "Sample Selection Bias as Speciﬁcation Error."
Econometrica, 47(1): 1 53-161 .

Holzer, Harry (1987) “Hiring Procedures in the F irrn: Their Economic Determinants and
Outcomes,” in Kleiner, et. al, eds. Human Resources and the Performance of the Firm.
Madison, WI: Industrial Relations Research Association.

Horvath, Francis W. (1987) "The Pulse of Economic Change: Displaced Workers of
1981-85," Monthly Labor Review, 110(6): 3-12.

---------- (1982) "Forgotten Unemployment: Recall Bias in Retrospective Data," Monthly
Labor Review, 105(3):40-44.

Howland, Marie (1988) Plant Closings and Worker Displacement: The Regional Issues.
Kalamazoo, Michigan: W. E. Upjohn Institute.

---------- and George E. Peterson (1988) "Labor Market Conditions and the Re-
employment of Displaced Workers." Industrial and Labor Relations Review, 42(1): 109-
122.

Jacobson, Louis S., LaLonde, Robert, and Daniel G. Sullivan (1993) "Earnings Losses of
Displaced Workers," American Economic Review, 83(4):685-707.

---------- (1994) “The Returns from Classroom Training for Displaced Workers,”
Working Paper 94-27, Federal Reserve Bank of Chicago.

Jacobson, Louis (1977) Earnings Losses of Worker Displacement When Unemployment
Declines in the Steel Industry, Ph.D. Dissertation, Northwestern University.

Lazear, Edward P. (1981) "Agency, Earnings Proﬁles, Productivity, and Hours
Restrictions," American Economic Review, 71(4):606-20.

Lewis, H. Gregg (1986) "Union Relative Wage Effects" in Ashenfelter, O. and Richard
Layard, eds. Handbook of Labor Economics, New York: Elsevier Science Publishers.

Madden, Janice F. (1988) “The Distribution of Earnings Losses Among Displaced
Workers,” Journal of Human Resources, 23(1):93-107.

58

Mortensen, Dale T. (1986) "Job Search and Labor Market Analysis," in Ashenfelter, O.
and Richard Layard, eds. Handbook of Labor Economics. New York: Elsevier Science
Publishers.

Podgursky, Michael and Paul Swaim (1987) "Job Displacement and Earnings Loss:
Evidence From the Displaced Worker Survey," Industrial and Labor Relations Review,
41(1):17-29.

Ruhm, Christopher J. (1994) "Advance Notice, Job Search, and Post-displacement
Earnings," Journal of Labor Economics, 12(1): 1-28.

---------- (1991) “Are Workers Permanently Scarred by Job Displacements?” American
Economic Review, 81(1):319-324.

---------- (1987) "The Extent and Persistence of Unemployment Following Permanent
Quits and Layoffs," mimeo, Boston University.

Seitchik, Adam and Jeffery Zornitsky (1989) From One Job to the Next: Worker
Adjustment in a Changing Labor Market. Kalamazoo, Michigan: W.E. Upjohn Institute.

Shapiro, Carl and Joseph Stiglitz (1984) "Equilibrium Unemployment as a Discipline
Device," American Economic Review, 74(3):433-44.

Shultz, George P. (1964) "The Fort Worth Project of the Armour Automation
Committee," Monthly Labor Review, 87(1):53-57.

Simon, Curtis (1988) “Frictional Unemployment and the Role of Industrial Diversity,”
Quarterly Journal of Economics, 103 :71 5-728.

Stern, James L. (1969) "Evolution of Private Manpower Planning in Armour's Plant
Closing," Monthly Labor Review, 92(12):21-28.

Stock, Wendy A. (1996) “Inter-industry Mobility and Displaced Worker Assistance
Programs,” Unpublished Manuscript, Michigan State University.

Swaim, Paul and Michael Podgursky (1991) " The Distribution of Economic Losses
Among Displaced Workers: A Replication," The Journal of Human Resources,
26(4):742-755.

Ullman, Joseph C. (1969) "Helping Workers Locate Jobs Following a Plant Shutdown,"
Monthly Labor Review, 92(4):35-40.

59

United States General Accounting Ofﬁce, Human Resources Division (1993) Dislocated
Workers: A Look Back at the Redwood Employment Training Programs (Brieﬁng Report
to Congressional Requesters) Washington DC

Willis, Robert J. (1986) "Wage Determinants: A Survey and Reinterpretation of Human
Capital Earnings Functions," in Ashenfelter, O. and Richard Layard, eds. Handbook of
Labor Economics. New York: Elsevier Science Publishers.

Chapter 2
INTER-INDUSTRY MOBILITY AND DISPLACED WORKER
ASSISTANCE PROGRAMS

1. Introduction

The official federal deﬁnition of a displaced worker is one who has lost his or her
job after three or more years of tenure as a result of a plant shutdown, mass layoff,
abolition of the job, or slack work (Flaim and Seghal, 1985). The large body of literature
on displaced workers suggests that they suffer substantial lifetime earnings losses,
readjustment difﬁculties, losses of speciﬁc human capital, and long spells of
unemployment. Certain groups of workers seem to be particularly hard hit by
displacements, including blue collar workers, workers with substantial speciﬁc human
capital investments, workers displaced during periods of high unemployment in their
local economy, and workers who switch industries upon re-employment.

In response to displacements, federal and local governments have instituted
assistance programs for displaced workers which provide services ranging from direct
cash payments to formal training programs. The Clinton administration is currently
considering a system providing training vouchers for workers which will allow them to
choose the type of training, if any, they wish to undertake.

Despite the evolution of displaced worker training programs over more than 30
years, a surprisingly small amount of research analyzes the impact, if any, of assistance

programs designed for displaced workers. Evidence suggests that many of the current

60

61

assistance programs have had only limited success and are ineffective for some groups of
workers (LaLonde, 1995; Leigh, 1990). This evidence, combined with both the high cost
of providing formal training to displaced workers and policy makers’ and researchers’
interest in improving the effectiveness and efﬁciency of programs for displaced workers,
has led some researchers to suggest that formal training only be offered to select groups
of workers (Leigh, 1990; National Commission for Employment Policy (N CEP), 1995).

One recent study (Corson et al., 1993) ﬁnds that the beneﬁts of training programs
may be greater for displaced workers who switch industries or occupations upon re-
employment. This result makes intuitive sense: workers who switch industries or
occupations upon re-employment will lose human capital speciﬁc to that industry or
occupation, and training may serve to aid the transition from one industry to another by
providing displaced workers with the skills required in other industries. In addition,
because displaced workers who change industry generally experience larger earnings
losses than other displaced workers, participation in training is relatively less costly for
them since the value of their foregone earnings is relatively lower. Accordingly, it may
be cost effective to design and offer training programs primarily for workers who are
likely to switch industries or occupations upon re-employment.

This paper is an attempt to test the feasibility of such a reallocation of the
resources dedicated to training programs. Despite existing evidence that ﬁnds that
workers who switch industries upon re-employment tend to suffer larger earnings loses
and to have longer unemployment spells than their non-switching counterparts, very little

research has explicitly estimated the factors that inﬂuence whether a displaced worker

62

switches industries upon re-employment.l The results presented here suggest that
although it is possible to identify factors which strongly inﬂuence switching, they only
explain a small fraction of the variation in switch behavior. Thus, it would be difﬁcult a
priori to identify workers who are more (or less) likely to switch industries, at least based
solely upon the types of information available in large, nationwide data sets like the
Displaced Worker Survey (discussed below). However, the information presented here, in
combination with information currently available from local agencies, can be used to
assist policy makers and assistance service providers in designing effective programs for
displaced workers.

The paper proceeds as follows: the next section presents an overview of the
largest federal programs for displaced workers. In section (III), I review existing evidence
on assistance program effectiveness. Section (IV) presents evidence on industry

switching outcomes for displaced workers. The paper concludes in section (V).

2. Programs for Displaced Workers
The primary federal programs for displaced workers are summarized in Table 1.2

These programs provide a mix of assistance services including job search and re-

 

’ Only one paper to date, Fallick (1993), which is primarily focused on testing implications of job search
theory, estimates displaced workers’ rates of transition from unemployment into a new industry.

2 In addition to the federal programs, many states and locales have specialized programs for displaced
workers (see Leigh (1990) for examples).

63

employment assistance (J SA), formal training, and cash payments to workers. Displaced
workers may also be eligible for student loans, Pell grants, and welfare payments.3

The Job Training and Partnership Act (JTPA) Title III was originally established
in 1982, but in 1988, JTPA Title III was amended by the Economic Dislocation and
Worker Adjustment Assistance Act (EDWAA).4 Within broad guidelines, EDWAA is
designed to focus upon individuals who have been terminated or laid off or who have
received notice of layoff, who are eligible for or have exhausted their unemployment
insurance beneﬁts, and who are unlikely to return to their previous industry or
occupation. Thus, EDWAA is intended to at least partially focus upon workers who are
likely to switch industries upon re-employment. In addition, EDWAA provides beneﬁts
to the long term unemployed who have limited opportunities for employment or re-
employment in the same or a similar occupation in the area where they live (NCEP,
1995).

EDWAA authorizes ﬁve different types of services for displaced workers and
their communities. First, it funds the establishment of rapid response teams which
quickly contact employers and employees and disseminate information about assistance

resources available. Second, EDWAA provides basic readjustment services including job

 

3 Approximately 75,000 displaced workers received Pell grants in 1990-1991 (Jacobson, LaLonde, and
Sullivan, 1994).

4 In addition to the EDWAA program described in the text, JTPA Title III also includes three other
programs for displaced workers. The Defense Conversion Adjustment Program and the Defense
Diversiﬁcation Program are designed to assist workers displaced by reductions in national defense
expenditures. The Clean Air Employment Transition Assistance program is designed to assist workers
displaced as a result of reduced employment by ﬁrms in order to comply with the Clean Air Act. The
eligibility requirements of all three programs differ slightly from the EDWAA program, but their beneﬁts
are generally similar.

64

search assistance (JSA), career counseling, skills assessment, and placement assistance.
Third, EDWAA provides funding for retraining services including classroom, on-the-job,
occupational skill, and entrepreneurial training. Fourth, EDWAA provides direct
payments to individuals who are participating in training and education programs and
who are either no longer, or never were, qualiﬁed for unemployment compensation.
Fifth, EDWAA calls for coordination of services among agencies providing services to
displaced workers.

Under the EDWAA program, states and localities play major roles in building
delivery systems and administering assistance ﬁinds, which primarily take the form of
formula block grants.5 At the state level, governors, Job Training Coordinating Councils,
private industry councils, and local elected ofﬁcials work together to plan assistance
activities for sub-state areas. At the local level, EDWAA grantees, which include private
industry councils, non-proﬁt organizations, local government agencies, community
colleges, and area vocational schools, receive state funds to develop training plans, decide
upon services to be offered, design budgets, and select training service providers.6 Service
providers generally include community based organizations, 'educational institutions,

private training establishments, businesses, and other organizations. The service

 

5 The federal ﬁrnding allocation formula is as follows: one-third of the grant is based on the number
unemployed in the state relative to the number unemployed in all states; one-third is based on the number
of individuals in the state who have been unemployed for more than 15 weeks, relative to the number of
individuals in all states who have been unemployed for more than 15 weeks; one-third is based on the
excess number of people unemployed in the state, as compared to the total number of excess people
unemployed in all states (excess is deﬁned as the number unemployed in excess of 4.5 percent of the
civilian labor force) (NCEP, 1995).

6 At the sub—state level, one-half of the funding is allocated based on a legislative formula which must
include information on: (1) the number of insured unemployed in the area, (2) unemployment
concentration in the area, (3) plant closing and mass layoff data, (4) declining industry data, (5) farmer-
rancher hardship data, and (6) long-term unemployment data (NCEP, 1995).

65

providers, in turn, screen and select participants for eligibility, assess skills and needs,
and provide training as appropriate. Thus, although EDWAA programs involve
considerable bureaucratic processes, services are generally designed and administered at
local levels.

The Trade Adjustment Assistance program (TAA), ﬁrst enacted in 1962 as part of
the Trade Expansion Act, is designed to assist workers when imports have contributed
signiﬁcantly to their displacement. Along with job search and relocation assistance, TAA
currently provides cash payments, called Trade Readjustrnent Allowances (TRAs). The
TRA cash beneﬁts are conditional upon training program participation and prior
exhaustion of unemployment insurance beneﬁts. TAA also provides workers with job
search and relocation assistance.

Unlike the EDWAA program, the TAA program is generally centralized at the
federal level. Firms representing groups of workers petition the Department of Labor in
order to gain TAA eligibility for their workers, while individual workers petition state
Employment Security Agencies for eligibility. The Department of Labor provides TAA
funds to states and the funds are then made available to local Employment Service
ofﬁces. The Employment Service ofﬁces either provide assistance services themselves,
or contract with other providers to offer services to TAA recipients (N CEP, 1995).

There are two other programs which, although they are not primarily designed for
displaced workers, provide a substantial portion of the assistance that displaced workers
receive. The US Unemployment Insurance (UI) system was established in 1935 to

provide weekly payments to job searchers. It is often the ﬁrst source of assistance for

66

displaced workers, particularly since some displaced worker assistance program beneﬁts
are contingent upon prior exhaustion of U1 beneﬁts. Recipients of U1 must demonstrate
that they are able and available to work (in some states they must also be actively seeking
work). Individuals generally receive beneﬁts for up to 26 weeks, although extended
beneﬁts are available.

The US Employment Service plays a large and increasing role in assisting
displaced workers (NCEP, 1995). In addition to serving as the UI beneﬁts liaison in
many states, the Employment Service is designed to match job seekers with job openings
by providing information about the local labor market and employer recruitment efforts.
The Employment Service also offers skills testing, job search assistance, and resume
writing instruction. All persons who are legally authorized to work in the US are eligible

for Employment Service programs.

3. Evidence on Program Effectiveness

3.1 Evidence on Job Search Assistance (JSA) Programs. During the 19808, an
array of policies, including re-employment bonus programs, on-the-job training, J SA, and
formal classroom training, were tested in a series of demonstration projects designed for
displaced workers. Leigh (1990) provides a comprehensive review of the projects. Three
projects, conducted in New Jersey, Buffalo, and Texas, allowed for estimation of the
impact of JSA. In all three projects, individuals who received JSA were found to have
higher employment rates and higher earnings than similar workers who received no

beneﬁts. In addition, J SA was found to decrease the net unemployment insurance beneﬁt

67

paid by $87 and $220 in the Texas and New Jersey projects, respectively. Because of its
strong impact, low cost per worker, and relative ease of implementation, Leigh argues
that J SA should be at the core of adjustment services for displaced workers.

Jacobson, LaLonde, and Sullivan (1994) estimate the effects of classroom training
for displaced workers who participated in the Displaced Workers Educational Training
Program (DWETP) in Pennsylvania during the 19808. The DWETP program provided
recruiting and counseling services, developed curricula for displaced worker training at
the Community College of Allegheny County (CCAC), and paid participants’ out-of-
pocket expenses for courses at CCAC. Although the program was primarily designed to
provide traditional classroom training, it also included an optional J SA non-credit course.
After controlling for the effects of schooling, grade point average, and enrollment in other
non-credit courses available from the DWETP, Jacobson, LaLonde, and Sullivan
conclude that enrollment in the J SA course had a “modest” impact (an increase in average
quarterly earnings of $150) on the earnings of males, which persisted throughout 28
quarters of the post-training period (Jacobson, LaLonde and Sullivan, 1994; pg. 30).

However, J SA did not appear to beneﬁt women in the DWETP study.

3.2 Evidence on Training Programs. Leigh (1990) argues that ﬁndings on the
effectiveness of classroom and on-the-job training programs for displaced workers are
generally inconclusive. In all four training demonstration projects evaluated in Leigh’s
study, classroom training failed to have a signiﬁcant and sizable effect on earnings and

employment beyond the effects of J SA. Classroom training also tended to be very

68

expensive to provide. Even on-the-job training programs, designed for both displaced
workers and workers at high risk of displacement, failed to consistently improve earnings
or employment rates.

In their evaluation of the DWETP (discussed above), Jacobson, LaLonde, and
Sullivan (1994) ﬁnd that participation in training lowered participants’ earnings during
the training period, which reﬂects the costs of foregone work time and experience while
in training. In addition, participants’ earnings tended to be lower for about nine quarters
after leaving the program. After the seventh post-training year, however, the relative
earnings differential of training participants leveled off to approximately $250.00 per
quarter, giving a crude estimated return to training of about 6.3 percent (250/4000
quarterly earnings). However, as is generally the case with displaced worker assistance,
this modest increase in earnings was not enough to overcome the initial losses due to
displacement. Furthermore, because the provision of training was so expensive and
because participants realized only modest gains in earnings, the authors conclude that the
net social beneﬁts of the program were much smaller.

Corson, Decker, Gleason, and Nicholson (1993) analyze the Trade Adjustment
Assistance (TAA) program. In their analysis, they found that TAA trainees remained
unemployed longer and earned less after their initial UI claim than other TAA recipients,
who were eligible, but did not participate in TAA training. As Corson et al. point out, part
of the difference in earnings of trainees and non-trainees reﬂects the self-selection of
workers with relatively bleak re-employment prospects into training. In fact, many of the

workers in the TAA training programs did not enter into training until after a long period

69

of unemployment. For workers who have re-employment opportunities, participation in
training generally entails reduced hours of work and lower earnings during training,
which reﬂects the investment decision to forego current employment and earnings in
anticipation of higher compensation in the future. Unfortunately, this investment was not
rewarded for the TAA trainees, at least not within three years following their initial UI
claim. After three years, the trainees’ earnings and employment outcomes were similar to
those of non-trainees once the authors controlled for observable differences between the
two groups.

To summarize, research to date has not found consistent evidence to support the
idea that formal training signiﬁcantly improves either the earnings or employment of
displaced workers. Leigh (1990) argues that due to both the lack of evidence to support
strong effects of training programs and their relatively high cost, skill training should be
offered sparingly and only for individuals with well-speciﬁed needs, and where adequate
training resources are present. Corson et al. (1993) provide evidence (discussed below)
which suggests that displaced workers who switch industries upon re-employment may

be good candidates for this type of “targeted” assistance.

3.3 Evidence on Industry Switching. Previous research on the earnings losses
of displaced workers has found that individuals who switch industries or occupations
upon re-employment suffer larger earnings losses than non-switchers (Stock, 1996;
Podgursky and Swaim, 1987; and Jacobson, LaLonde, and Sullivan, 1993). One

interesting ﬁnding in the Corson et al. (1993) study is that trainees were more likely than

70

non-trainees to have switched industries or occupations upon re-employment. Among the
switchers in the Corson et al. study, trainees had ﬁve to ten percent higher wages than
non-trainees, although the difference was not statistically signiﬁcant. The authors
conclude that training appears to be part of a transition as workers move from one
industry to another, and that their study provides some evidence that training may
actually have a positive impact for industry or occupation switchers.

The evidence on industry switching suggests that there are at least two reasons
why it may be beneﬁcial to focus retraining programs toward workers who are likely to
switch industries. First, the “ﬁxed” costs of providing training would be the same for
industry switchers as for other workers, but the beneﬁts of training may be higher for
industry switchers than for other workers. In addition, workers who switch industries
upon re-employment experience larger earnings losses than other workers, which makes
the earnings they forego during training (i.e. their opportunity cost of training) smaller
than for other workers.

Training programs may also be beneﬁcial in encouraging industry switching for
workers displaced from locally declining industries, since workers who would not have
otherwise switched industries may do so if the training offered to them is of relatively
higher value in new industries. For example, researchers have argued that workers
displaced from the timber industry in the Paciﬁc Northwest are so strongly attached to
their former industry that many of them would be more likely to remain unemployed in

hopes of ﬁnding jobs in the timber industry than to switch industries (U S Department of

71

Agriculture (USDA) and US Department of Interior (USDI), 1994).7 In an effort to
transit workers from the declining timber industry in the Paciﬁc Northwest, the Clinton
Administration has offered displaced timber workers adjustment incentives ranging from
small business development loans to formal training to re-location allowances (Yu,
1993). This type of “targeted training” may serve to both reduce unemployment duration
and reduce repeated spells of displacement.

Alternatively, for workers displaced as part of a small layoff from an otherwise
vibrant local industry, it may be unnecessary to offer re-training since many workers will
likely ﬁnd jobs with other ﬁrms in their former industry.

The use of training as an adjustment tool highlights the importance of deﬁning
clear policy goals. If the goal of policy makers is to reduce worker earnings losses due to
displacement, inducing workers to switch industries by offering industry-speciﬁc skills
training would be considered ineffective if it results in earnings which are lower than they
would have been had the worker not switched industries. However, if the goal of policy
makers is to assist workers in making adjustments to changing economic conditions, a
training could be considered a success even if it results in earnings which are lower than
what workers’ earnings would have been if the workers had found employment in their
former industries. Alternatively, it is possible that some workers will switch industries
regardless of having training offered to them. Because these workers already plan to

switch industries, they may place high value on learning skills required in other industries

 

7 In their analysis of proposed changes in timber harvesting levels in the Paciﬁc Northwest, ofﬁcials in the
USDA and USDI argue that the timber workers in the area have a high degree of attachment to their
lifestyle and location, which would lower the probability of industrial or locational mobility if incentives to
move are not relatively high (USDA and USDI, 1994).

72

and may therefore be more likely to participate in training. Since the workers would have
switched industries even without the training program, and since training generally
entails reduced employment, one could argue that the program is unnecessary or even
counterproductive if the goal of policy makers is to encourage the rapid re-employment
of displaced workers.

In sum, evidence suggests that training may have positive effects as transition
device for workers who switch industries and targeting retraining assistance toward
workers who are more (or less) likely to switch industries may be an effective tool for
promoting adjustment to changing local labor market conditions, particularly in the face

of local industrial shifts.

4. Predicting Industry Switching

Switching industries is an outcome which is unknown prior to re-employment.
Thus, policy makers and assistance providers who want to focus retraining toward
workers who are likely to switch industries must make estimates about which workers
would be more likely than others to switch industries. Since industry decline is one likely
inﬂuence on industry switching, information which will assist in estimating worker
industry switching is already being gathered because some sub-state funding for the
EDWAA program is allocated based upon a legislative formula which includes
information about declining industries (NCEP, 1995).

There are many other possible characteristics which may also inﬂuence workers’

industry switching. For example, since the remaining years of work for older workers are

73

generally fewer than for younger workers, an investment in learning the skills required in
new industries is worth relatively less for older workers. For obvious reasons, it is
probably not prudent (or legal) for policy makers to focus retraining programs on workers
belonging to particular demographic groups.

It is reasonable, however, to utilize information local labor market conditions in
order to predict industry switching. Obvious possible predictors include the local
unemployment rate, the growth or decline in the employment levels over time, and the
growth or decline in employment in different industries over time. In addition, it may be
possible that workers who live in highly populated areas have more re-employment
opportunities in their former industries since they have access to larger labor markets in
which to search for re-employment.

Another factor which may inﬂuence whether workers switch industries as a result of
displacement is the relative importance of their former industry in their local labor market.
If their former industry is “concentrated” in their area (i.e. their former industry employs a
large fraction of the local labor force), they may be less likely to switch industries upon re-
employment, particularly if layoffs in that industry tend to be of a small scale (since a
relatively large number of jobs in their industry will still exist in their area). Alternatively,
if workers are displaced as part of a mass layoff from a relatively non-concentrated industry,
they may be more likely to switch industries upon re-employment since relatively few jobs
will remain in their industry in their area. In the rest of this section, I analyze these and

other possible inﬂuences upon industry switching.

74

4.1 Data. During every other wave of the January Current Population Survey
(CPS), workers who have identiﬁed themselves as having lost a job in the last ﬁve years
due to their plant closing, their employer going out of business, or a layoff from which they
were not recalled, are asked a series of supplementary questions about their unemployment
spell and the characteristics of their former and (if applicable) current jobs. This
supplement, known as the Displaced Worker Survey (DWS), has been conducted biennially
since 1984 and provides detailed demographic and earnings information on a large,
nationwide sample of displaced individuals.

I use the 1992 DWS for the analysis in this paper, but place three restrictions upon
the DWS data. First, following the federal deﬁnition of a displaced worker, I include in the
sample only workers in the DWS who had at least three years tenure prior to their
displacement. Second, in order to control for the effects of local labor market conditions, I
limit the sample to workers who live in metropolitan statistical areas (MSAS) and who did
not move after displacement.8 Third, in order to ensure that workers had enough time after
displacement to conduct a job search, the sample is further restricted to include only

workers who were displaced between January 1987 and December 1990.9

 

8 Unfortunately, the DWS does not ask workers about the geographic area in which they were employed prior
to displacement. However, the survey does ask "since displacement, has [the reference person] moved to a
different city or county to look for work or to take a different job?" As this question does not reﬂect whether
workers changed MSAs as a result of displacement, it would be erroneous to include workers who moved in
the analysis, since the labor market characteristics of their former MSA are unobserved. I ran several tests to
analyze the effect of excluding movers from the sample. I found no signiﬁcant effect of moving upon whether
a worker industry switching upon re-employment. In addition, results from estimates of industry switching
which included the movers did not differ substantially from the results of a similar regression which excluded
the movers.

9 The DWS asks workers about the year but not the month of their displacement. It is unlikely that workers
displaced in late 1991 or January 1992 had enough time to conduct a job search before the January 1992
survey. Together, the three exclusion restrictions drop 2127 of the original 2880 observations from the sample.
The exclusion of workers with fewer than three years of tenure drops 1527 of the original 2880, the exclusion

75

In addition to the DWS, I include data from two other sources: County Business
Patterns, an annual data set containing employment and payroll information, disaggregated
by county and industry, for establishments covered by the Federal Insurance Contributions
Act (FICA);lo and Employment and Earnings, which is published by the Bureau of Labor
Statistics (BLS) and contains unemployment rate statistics for most MSAs in the CPS. I use
Employment and Earnings because although unemployment ﬁgures obtained from the CPS
are reliable at the state level, the CPS is insufﬁciently large to accurately estimate MSA-
level unemployment rates.ll

In order to measure the relative importance of workers’ industries in their local labor
markets, I use the County Business Patterns to calculate the employment shares (ES) for
each of the 50 detailed industries and 203 MSAs deﬁned in the CPS/DWS. The formula I
use to calculate the employment shares is given by:

ES = (number of workers in industry j in MSA k) / (number of workers in MSA k)12

 

of those displaced in 1991 or 1992 drops 442 of the remaining 1353 observations, and the exclusion of movers
drops 158 of the remaining 911 observations from the sample.

'0 The basic data items in the CBP are extracted from the Standard Statistical Establishment List, a ﬁle of all
known single and multi-establishment companies maintained and updated by the Bureau of the Census. The
Annual Company Organization Survey is the source of data for multi-location fums, while data on single-
location ﬁrms are obtained from various programs conducted by the Census Bureau (County Business
Patterns, 1993).

” State-level unemployment rates were assigned to the 25 MSAs not covered in Employment and Earnings.
Although primarily from the same sources, the BLS data differ from CPS data at the sub-state level because
the BLS adjusts statewide CPS unemployment rates based on administrative statistics (compiled from reports
ﬁled by establishments covered under state unemployment insurance laws) in order to obtain sub-state
unemployment estimates. Comparisons of MSA-level unemployment rates calculated using the CPS outgoing
rotation groups and the BLS Employment and Earnings data reveal that the CPS unemployment ﬁgures were
consistently much smaller (by an average of about two percentage points) than the BLS ﬁgures.

'2 Alternative measures of industry geographic concentration are available. See Diamond and Simon (1990),
and Simon (1988) for analyses which use alternative measures in another context.

76

The ES ﬁgures are use to capture the inﬂuence of the industrial structure of

displaced workers’ local labor markets.

4.2 Sample Characteristics. Table 2 presents the mean values of the variables
used in this study and thus describes many of the characteristics of the displaced workers
under consideration. As shown, about 92 percent of the workers in this sample were re-
employed at the time of the survey, and about 63 percent had switched industries upon re-
employment. Almost all of the workers had been displaced from full-time jobs where they
had, on average, worked almost nine years. The average displaced worker spent about 17
weeks unemployed, was male, white, married, and about 41 years old. The highest
percentage of workers, 42 percent, held high school diplomas.

The average displaced worker in this sample came from an MSA which experienced
a slight decrease in the unemployment rate during the year before their displacement (about
.10). In addition, the average displaced worker came from an industry which employed
about seven percent of the local labor force at the time of displacement, although this
employment share was declining slightly during the year before displacement occurred.

The highest percentage of workers, 36 percent, were displaced from the
manufacturing industry. This represents a slight departure from earlier years of the DWS,
when declines in manufacturing accounted for roughly one-half of all displacements (Flaim

and Seghal, 1985 and Horvath, 1987).

77

4.3 Estimations and Predictions. The DWS data provide information on many
characteristics which can help to predict the industry switching outcome and equation (1)
provides the model I use to assess the possible impacts of these characteristics on industry

switching.

(1) switch=b0+bl X+ b2Z+ b3 ES+u

The variable switch is equal to one if a worker switched detailed industry upon re-
employment. Whether a worker switches industries upon re-employment is dependent upon
employment share (ES), local economic conditions at the time of displacement (Z),
including the unemployment rate in the workers’ local areas, the proximity of the workers’
labor market to other metropolitan areas, and the change in employment in the workers’
industries over time. Whether a worker switches industry is also likely to depend upon both
demographic and former industry characteristics (X).

Results from estimating equation (1) are presented in Table 3. Although I include
demographic characteristics as controls in all speciﬁcations except column (5), I do not
report their coefﬁcients in Table 3 because the demographic characteristics do not explain
much of the variation in switch, particularly once occupational controls are included in the
regressions.

Apart from employment share, labor market characteristics have generally
insigniﬁcant impacts upon switch. The employment share variable has a negative and

signiﬁcant sign, implying that a one percentage point increase in employment share is

78

associated with approximately a one percentage point decrease (or, alternatively, 1.7
percent decrease) in the probability that the average worker switches industries. A one
standard deviation increase in employment share is associated with approximately a 12
percent decrease in the probability of switching industries for the average worker. Thus, the
larger the fraction of the local labor force employed in the worker’s former industry, the less
likely it is that the worker will switch industries upon re-employment. This negative result
holds throughout all speciﬁcations in Table 3. '3

It may be possible that the employment share variable is actually capturing industry
effects, since each industry will vary in its relative concentration in different labor markets.
It may also be the case that the impact of employment share differs across different
industries, or alternatively, that the impact of former industry varies with levels of
employment share. In order to test these possibilities, I add dummy variables for workers’
former industries to the regression speciﬁcation in column (5) of Table 3. I also include
interactions of the industry and employment share variables. Results from estimating these
speciﬁcations are reported in Table 4.

Column (1) of Table 4 reports the coefﬁcients from a specification which includes
six industry dummies (services is the excluded category). Relative to workers displaced
from the service industries, workers displaced from the agriculture and mining;
manufacturing; trade; and ﬁnance, insurance and real estate (FIRE) industries are less likely

to switch industries upon re-employment, although only the impact of the FIRE industries

 

'3 1 also ran speciﬁcations which included a variable for the number of weeks the individual spent unemployed
after displacement. The estimates suggest that an additional week of unemployment is associated with a .32
percent increase in the probability that the average worker switches industries upon re-employment
(signiﬁcant at the ﬁve percent level). I do not report the results of these speciﬁcations because weeks
unemployed would not be of much use for ex ante targeting of displaced worker assistance programs.

79

was statistically signiﬁcant. Workers displaced from FIRE industries were about 42 percent
more likely to switch industries than those displaced from service industries. Workers
displaced from the construction and transportation, communications, and public utilities
industries were also less likely to switch industries than service workers. However, the
difference in industry switching for the construction industry was insigniﬁcant. Note that
employment share remains a strong predictor of industry switching even when industrial
controls are included in the speciﬁcations.

In Column (2) of Table 4, 1 report estimated partial effects of employment share on
industry switching for all seven industries. The partial effects result from a speciﬁcation
which drops the industry dummies and instead includes interactions between the industry
dummies and the employment share variables, thereby allowing the impact of employment
share to vary across industries. The partial impact of employment share is signiﬁcant for
several industries. For example, for the average worker in the construction industry, a one
percentage point increase in employment share is associated with a 4.70 percentage point
decrease in the probability of switching industries. In column (3), I report the percent
change in the average probability of switching industries. Thus, a one percentage point
increase in employment share is associated with a 8.55 percent decrease in the probability
of switching industries for the average worker in construction. For the average worker in
the transportation, communications, and public utilities industry, a once percentage point
change in employment share is associated with a 22.34 percent change in the probability of

switching industry. The results presented in Table 4 illustrate that not only does

80

employment share affect industry switching, but the effect of employment share differs
across industry.

Figures 1 and 2 further examine the impact of employment share by providing both
actual and predicted probabilities of switching industries, disaggregated by levels of
employment share. The top panel of Figure 1 shows the probability of industry switching
for workers with different employment share values. Thus, for the workers displaced from
industries which employed less than three percent of the local labor force at the time of their
displacements, the probability of switching industries was a little higher than 70 percent.
For workers whose former industry employed a larger ﬁaction of the local labor force at the
time of their displacements, the probability of switching was lower. The bottom panel of
Figure 1 shows a regression of the mean probability of industry switching (disaggregated by
levels of employment share) on employment share. The negative slope and large t-statistic
imply that employment share has a negative and signiﬁcant impact on worker industry
switching.

In order to test the accuracy of industry switch predictions, I report predicted and
actual probabilities of switching in Figure 2. The predictions were made by estimating
industry switch regressions using data for half of the sample (randomly chosen). These
estimated equations were then used to predict the probability of industry switching for
workers in the other half of the sample. Although the predictions match the actual values
fairly well, they are generally higher than the actual probabilities of switching. Exceptions

include the particularly large underestimation of the probability of switching industries for

81

workers displaced from industries which employed less than .03 of the labor force at the

time of displacement.

5. Conclusion

The limited research that has evaluated the effectiveness of displaced worker
training programs has not found consistent evidence to support the idea that formal
training signiﬁcantly improves either the earnings or employment of displaced workers.
This result, combined with both the high cost of providing formal training to displaced
workers, and policy makers’ and researchers’ interest in improving the effectiveness and
efﬁciency of programs for displaced workers, has led some researchers to suggest that
formal training only be offered to select groups of workers (Leigh, 1990; NCEP, 1995).
One recent study suggests that training may have a stronger impact on earnings and
employment for workers who switch industries upon re-employment (Corson, et al.,
1993). This paper has been an attempt to discern which, if any, factors can be used to
identify displaced workers who are likely to switch industries upon re-employment.

I ﬁnd that workers displaced from industries which are concentrated in their local
area are less likely than other workers to switch industries upon re-employment and that
this impact differs across industries. For example, a one percentage point increase in the
employment share of the average worker’s former industry is associated with
approximately a 1.7 percent decrease in the probability that the worker will switch
industries upon re-employment. In addition, the effect of employment share differs across

industries. After controlling for industry concentration, I ﬁnd that, relative to workers

82

displaced from service industries, workers displaced from the transportation,
communications, and public utilities industries are more likely to switch industries.

Thus, my estimates show that it is possible to use industry and local labor market
information to partially predict which workers are more likely to switch industries upon
re-employment. In combination with information available from local industry councils,
from local employment offices, and from interviews with individual workers, the results
presented here can be used to by policy makers and assistance providers to assess the
appropriate policy response to displacements. Since workers displaced from concentrated
industries are less likely to switch industries upon re-employment, resources designated for
these workers may be better spent on relatively inexpensive job search assistance programs
than on training programs which focus upon skills required for jobs in different industries.
On the other hand, workers displaced from non-concentrated industries are more likely to
switch industries upon re-employment, and assistance which focuses upon re-training

workers to learn skills useful in different industries may be more appropriate for them.

TABLE 1

Federal Pr_ograrns for Displaced Workers

 

 

 

 

 

 

Program Year Services Number Annual
Enacted Provided Served Funding
Annually Allocation1
Trade Adjustment Assistance 1962 Cash 27,309 $75,000,000
payments, (FY 1993)
Training,
JSA2
Job Training Partnership Act: 1988 Cash 312,252 $517,000,000
Title III (EDWAA) payments, (FY 1992)3 (FY 1993)
Training, JSA
Employment Service 1933 J SA 21,346,336 $811,000,000
Unemployment Insurance 1935 Cash 7,800,000 $21,900,000,000
Program payments (FY 1992)

 

 

 

 

 

 

 

1 Source: National Commission for Employment Policy (1995). Allocated funds may exceed actual expenditures
because most programs serve only a small fraction of eligible individuals. In addition, the funds allocated to the
Employment Service and Unemployment Insurance programs are available for both displaced and non-displaced
workers.

2 JSA = Job Search Assistance.

3 PY==Program year. PY 1992 covers the period July 1, 1992 until June 30, 1993.

83

TABLE 2

Variable Means

 

Employed at time of the 1992 DWS

Switched industry upon re-employment *2

Weeks unemployed

Employment share of former industry in year of displacement

Change in unemployment rate in MSA
(year before displacement to year of displacement)

Another MSA within 25 miles *

Years of tenure on former job

Full time (> 35 hours/week) on former job *

Less Than a High School Diploma

High School Graduate

Some College

Bachelors Degree or more

Age

Male *

White ‘

Married *

Former Industry - Agriculture and Mining *
Former Industry - Construction “

Former Industry - Manufacturing *

Former Industry - Transportation, Communications, Public Utilities *
Former Industry - Trade *

Former Industry - Finance. Insurance, Real Estate *
Former Industry - Services *

Number of observations

 

1 Standard deviations are reported in parentheses.

.92 (.27) 1
.63 (.48)
17 (21)
.07 (.08)

-.10 (.87)

.37 (.48)
8.72 (6.55)
.96 (.20)
.13 (.33)
.42 (.49)
.25 (.44)
.19 (.40)
41 (10)
.60 (.49)
.88 (.32)
.69 (.46)
.02(.14)
.10(30)
.36 (.48)
.06(.24)
.22(.42)
.08 (.28)
.15 (.39)

753

2 Variables denoted with an asterisk are dummy variables equal to one if the condition holds and zero

otherwise.

84

TABLE 3

Switch Effects

 

Dependent Variable: Switched Industries upon Re-Employment1

 

 

 

(l) (2) (3) (4) (5)
Employment Share of Former -1.09"2 -1 .09" -l .08" -1.12"‘* -1.20"
Industry in year of displacement (.25) (.25) (.25) (.44) (.43)
Percent Change in Industry Employment - -.01 .09 -. 10 -.12
(year before displacement (.10) (.10) (.10) (.10)
to year of displacement)
Change in unemployment rate in MSA - - -.01 -.02 -.03
(year before displacement (.03) (.03) (.02)
to year of displacement)
Percent Change in MSA Employment - - .46“ .30 .29
(year before displacement (.26) (.22) (.22)
to year of displacement)
Another MSA within 25 miles - - .04 .05 .06

(.04) (.04) (.04)

Demographic Controls yes yes yes yes no
Schooling Controls yes yes yes yes no
Year of Displacement Controls yes yes yes yes yes
Former Industry and Occupation no no no no yes
Controls
Adjusted R-squared .03 .03 .03 .1 1 .11

 

1 n = 693. All coefﬁcients result from linear probability (OLS) regressions. White (heteroskedasticity robust) standard
errors are reported in parentheses. Demographic controls include age, age-squared, full-time on former job, tenure on
former job, marital status, male, white, and schooling. The coefficients from these variables are not reported in the table,
but are available from the author. Sample selection tests resulted in no signiﬁcant change in the reported coefﬁcients.
Column (6) includes a control for full-time on former job.

2 " Signiﬁcant at the 5 percent level. “ Signiﬁcant at the 10 percent level.

85

TABLE 4

Former Industry Effects on Switching

Dependent Variable: Switched Industries upon Re-Employment1

 

 

 

in 121 or
Partial Impact Partial Impact Percent Change in
of Industry of Employment Share, Probability of Switching
by industry for average worker
in industry, given
one percentage point
change in Employment
Share
Employment Share of Former -l.20"""2 - -
Industry in year of displacement (.43)
Services - - 17 -.24
(.68)
Agriculture and Mining .11 1.70 1.85
(. 10) (3.28)
Construction -. 19 -4.70** -8.55**
(.08) (1.21)
Manufacturing .01 -2.74** -3.56**
(.06) (.72)
Transportation, Communications, -.28** -ll.l7** -22.34"‘*
and Public Utilities (.09) (2.15)
Trade .01 -1.01" -1.91
(.09) (.27)
Finance, Insurance, and Real Estate -.42** 905" -27.42“‘*
(-08) (2.31)
Adjusted R-squared .ll .10 -

 

1 n = 693. All coefficients result from linear probability (OLS) regressions. White (heteroskedasticity robust) standard
errors are reported in parentheses. All regressions also include all variables in column 6 of Table 3. The coefﬁcients from
these variables are not reported in the table, but are available from the author.

2 ** Signiﬁcant at the 5 percent level. * Signiﬁcant at the 10 percent level.

86

Probability ol SIIIChIng Industry

Actual and Predicted Industry Switch Probability

disaggregated Dy employ-em snare

HGUREI

Industgy Switch Rates by Level of Emploment Share

 

1
3- 06 9-12 15-18 21- 24
Employment Snare in Year of Displacement

 

 

 

5 0‘3 0'6 02's 12 1'5 15 21 2'4 27
= -97 t- statistic = 3
Emp10yment Share in Year of Displacement

87

FIGURE 2

Actual and Predicted Industry Switch Probabilities

 

 

 

 

 

 

 

 

 

 

Z I Actual DCODaDlllty [j predrcte: Drcbamlxy
E .8 r
’U
S
9dr
neg -7
Eco
:m .6~ T7
3‘4—
m0
:3 5 _‘ [—— ._.
sf ‘
m A
28 4
r—< 7‘ __
(30.0 3 L
a—IU l
UGJ
<13 1
.... .2 ‘
‘38
(UL
UCL '1 _.
Q)
s . l
8 O ‘ 0-03 .06-09 1215 .18-.2 » 24
C": .03-06 09-12 .15-.18 .21-.24

Employment Share in Year of Displacement

88

BIBLIOGRAPHY

BIBLIOGRAPHY

Corson, Walter, Decker, Paul, Gleason, Phillip, and Walter Nicholson (1993)
International Trade and Worker Dislocation: Evaluation of the Trade Adjustment
Assistance Program, Princeton, NJ: Mathematica Policy Research, Inc.

County Business Patterns (1993) County Business Patterns, 1990-1991 on CD-ROM,
prepared by Data Access and Use Branch, Data User Services Division, Bureau of the
Census. Washington: The Bureau.

Diamond, Charles A. and Curtis J. Simon (1990) “Industrial Specialization and the
Returns to Labor,” Journal of Labor Economics, 8(2): 175-201.

F allick, Bruce Chelimsky (1993) “The Industrial Mobility of Displaced Workers,”
Journal of Labor Economics, 11(2):302-323.

Flaim, Paul and Ellen Seghal (1985) "Displaced Workers of 1979-83: How Well Have
They F ared?" Monthly Labor Review, 108(6):3-16.

Horvath, Francis W. (1987) "The Pulse of Economic Change: Displaced Workers of
1981-85," Monthly Labor Review, 110(6): 3-12.

Jacobson, Louis S., LaLonde, Robert, and Daniel G. Sullivan (1994) “The Returns from
Classroom Training for Displaced Workers,” Working Paper 94-27, Federal Reserve
Bank of Chicago.

---------- (1993) "Earnings Losses of Displaced Workers," American Economic Review,
83(4):685-707.

LaLonde, Robert J. (1995) “The Promise of Public Sector-Sponsored Training
Programs,” Journal of Economic Perspectives.

Leigh, Duane E. (1990) Does Training Work for Displaced Workers? A Survey of the
Existing Evidence. Kalamazoo, Michigan: W.E. Upjohn Institute.

National Commission for Employment Policy (1995) Understanding Federal Training
and Employment Programs. Washington DC: US GPO.

Podgursky, Michael and Paul Swaim (1987) "Job Displacement and Earnings Loss:

Evidence From the Displaced Worker Survey," Industrial and Labor Relations Review,
41(1):17-29.

89

90

Simon, Curtis (1988) “Frictional Unemployment and the Role of Industrial Diversity,”
Quarterly Journal of Economics, 103 :715-728.

Stock, Wendy A. (1996) “The Effect of Geographic Industry Concentration on the Re-
Employment Experiences of Displaced Workers,” Unpublished Manuscript, Michigan
State University.

Yu, Peter (1993) Brieﬁng: Northwest Economic Adjustment Initiative, Washington: The
National Economic Council, The White House.

United States Department of Agriculture and United States Department of Interior (1994)
Final Supplemental Environmental Impact Statement on Management of Habitat for
Late-Successional and Old-Growth Forest Related Species Within the Range of the
Northern Spotted Owl. Washington: GPO.

Chapter 3
COMPARISON GROUP USE AND THE EARNINGS LOSSES
OF DISPLACED WORKERS
1. Introduction

Interest in both the size of earnings losses for displaced workers and the factors
that contribute to those earnings losses has been the focus of much attention in the
displacement literature. Recent research has attempted to measure how demographic
characteristics, industries and occupations, and economic conditions in local labor
markets inﬂuence the size and distribution of displaced worker earnings losses.

The most commonly used data on displaced workers comes from the Displaced
Worker Surveys (DWS), which have supplemented the January Current Population
Surveys (CPS) biennially since 1984. Individuals in the CPS who are identiﬁed as having
lost their jobs in the ﬁve years prior to the survey because of plant shutdowns, mass
layoffs, or slack work are asked supplemental questions about their current and former
earnings, job characteristics, and experiences since displacement. Thus, the DWS
provides detailed demographic and earnings information on a large, nationwide sample of
individuals and has been used extensively to study the experiences of displaced workers.

One of the major disadvantages of the DWS is its lack of comparable information
on the earnings of non-displaced workers, which has made it difﬁcult to use the DWS to
measure displaced workers’ earnings losses as the difference between their actual

earnings growth (or decline) and what that earnings growth would have been had

91

92

displacement not occurred. Put more simply, the DWS does not readily include the
information necessary to obtain a comparison group of non-displaced workers. Various
techniques have been invoked to obtain information on the earnings of non-displaced
workers, most notably the use of non-DWS/CPS data, as well as the use of earnings data
taken from the outgoing rotation groups of the CPS.

In this paper, I ﬁrst review previous attempts to obtain comparison groups and to
estimate the earnings losses of displaced workers. Second, I evaluate the potential gains
and biases introduced by comparison groups and propose a method for obtaining a
comparison group by using the average earnings of workers in the CPS. Finally, using
data on workers in the 1992 DWS and workers in the 1986 through 1992 CPS, I compare
earnings loss estimates made while using the comparison group against loss estimates
made while simply comparing the pre- and post-displacement earnings of displaced
workers.

Although the ideas presented here are applicable in more general contexts, I focus
attention throughout the paper on the problem of estimating the impact of industry
switching on displaced worker earnings losses. Many researchers have found that
displaced workers who switch industries upon re-employment experience larger earnings
losses, but my particular interest lies in determining how estimates of the impact of
industry switching on earnings losses vary depending upon ﬁrst, whether a comparison
group is used to measure earnings losses and second, how the earnings of the comparison

group change over time and with industry switching.

93

The paper proceeds as follows: Section 2 provides motivation for using a
comparison group and reviews previous research on displaced worker earnings losses;
Section 3 presents a model of earnings losses in order to evaluate potential gains and
problems introduced by utilizing a comparison group; Section 4 suggests a method of
obtaining a comparison group and compares estimates of earnings losses made both with

and without the comparison group. The paper concludes in Section 5.

2. Previous Research and Use of Comparison Groups

A displaced worker who experiences lower earnings upon re-employment suffers an
absolute loss in earnings. One can measure the relative earnings loss (or gain) for the
displaced worker by comparing the change in earnings experienced by a similar, but non-
displaced, worker with the change in earnings experienced by the displaced worker. As an
example, if non—displaced workers saw an increase in their earnings over the period under
consideration, only the use of a comparison group will capture the fact that displaced
workers are not only absolutely worse off, but they are also worse off relative to non-
displaced workers. This distinction is particularly important if both non—displaced workers
and displaced workers suffer earnings losses over the same period. Although displaced
workers will have negative earnings changes, so will non-displaced workers, and in relative
terms, the displaced workers may be better off (or at least less worse off). A similar
argument can be made for displaced workers who experience higher earnings upon re-

employment.

94

Table 1 provides a simpliﬁed explanation of the importance of using a comparison
group by comparing the real earnings losses (or gains) for a hypothetical worker using two
earnings loss measures: losses measured with a comparison group and losses measured
without a comparison group. The rows represent three different groups of hypothetical
workers: displaced, non-displaced whose real earnings fall over time, and non-displaced
whose real earnings rise over time. Columns (1) and (2) show earnings for the three groups
at the time of displacement (t-s) and at the time of re-employment (t). Columns (3), (4), and
(5) show the earnings losses of displaced workers measured both with and without a
comparison group. The earnings loss ﬁgures differ substantially depending upon ﬁrst,
whether a comparison group is used, and second, how the real earnings of the comparison
group change over time. Losses measured without a comparison group are 200, while
losses measured with a comparison group are either 100 or 300, depending on whether the
earnings of the comparison group fell or rose over time, respectively.

Unfortunately, completely controlled experiments to measure the impact of
displacement are not possible. Because workers cannot be observed with two different
levels of earnings at the same time, researchers cannot estimate the impact of displacement
on earnings by simply comparing the workers’ net earnings with displacement and the
workers’ net earnings without displacement. Nor can we estimate the impact of
displacement via a controlled experiment which randomly assigns identical workers into

displacement.

95

Various methods have been used to measure earnings losses despite the non-
experirnental nature of the information available. Table 2 provides a summary of recent
ﬁndings on the distribution of earnings losses for displaced workers, as well as a summary
of the types of comparison groups, if any, utilized to measure earnings losses.l

The ﬁndings summarized in Table 2 are somewhat varied, primarily because each
piece of research differs in focus and scope. For example, many researchers conclude that
women tend to lose more from displacement (Ruhm, 1994; Madden, 1988; Podgursky and
Swaim, 1987; Maxwell and D’Amico, 1986), but some studies focus their analysis only on
males (Canington and Zaman, 1993; Shapiro and Sandell, 1985). Ruhm (1991) focuses on
the impact of advance notice of displacements on earnings losses while di la Rica (1992)
focuses upon pre—displacement earnings losses. There is general consensus among the
studies that high tenure workers experience relatively large eamings losses, which reﬂects
the loss of speciﬁc human capital associated with displacement. In addition, Carrington
(1993) ﬁnds that workers with high experience levels also experience large earnings losses,
which leads to concern about whether earnings losses are primarily ﬁrm- or industry-

speciﬁc.

 

' The literature on displaced workers is extensive and many studies do not focus primarily on the earnings
losses of displaced workers. In Table 2, I exclude case studies, research conducted prior to 1985, and
research which did not focus upon estimating the distribution of earnings losses among displaced workers.
For example, Kruse (1988) focuses on the duration of joblessness, Kletzer (1989) uses displaced workers to
estimate the returns to seniority, and Flairn and Seghal (1985), Horvath (1987), and Gardner (1993)
primarily discuss the incidence of displacement.

96

2.1 Use of Comparison Groups: Non-DWS Data. Jacobson, LaLonde, and
Sullivan (1993) use a unique longitudinal data set obtained from Social Security records in
Pennsylvania which contains information on both displaced and non-displaced workers.
Despite the advantage of measuring earnings over long periods of time both before and after
displacement, the data set has very little information on worker demographics (limited to
age and gender). The data set also lacks a variable which explicitly identiﬁes displaced
workers. Both unemployed workers and those who left the Pennsylvania wage and salary
work force are observed to have zero earnings, which allows the authors only an estimate of
which workers are truly displaced. Furthermore, in order to reduce sample attrition, the
authors only include workers in the sample if they received wage or salary earnings during
each year. This systematic exclusion of workers who remained unemployed for long
periods (and who may suffer different earnings losses ﬁ'om other workers) leads to
underestimation of earnings losses.

Ruhm (1991) uses the Panel Study of Income Dynamics (PSID) to measure long-
terrn effects of displacement. The PSID contains a time series of data on earnings and
demographic characteristics for heads of households over the period from 1969 to 1982. Its
longitudinal structure and observations on both displaced and non-displaced workers make
the PSID an adequate data set for addressing the comparison group problem. Despite this
data advantage, Ruhm’s study has several drawbacks. Like Jacobson, LaLonde, and
Sullivan (1993), Ruhm includes in his sample only those workers who were displaced in

period t and re-employed in all periods t+1 through t+5, which systematically omits

97

workers who remained unemployed for long periods. Ruhm also fails to control for whether
a worker switches industries upon re-employment; instead he includes only dummies for
whether a worker was employed in the manufacturing industry prior to displacement.
Maxwell and D’Amico (1986) use the NLSY to measure the differential impact of
displacement across gender. Using data on non-displaced workers in the NLSY, they
estimate earnings equations for the periods one year and three years after displacement.
These earnings equations are then used to predict “non-displacement” earnings for
displaced workers. Thus, Maxwell and D’Arrrico use a comparison group to obtain
predictions of what displaced workers’ earnings would have been had displacement not
occurred. They ﬁnd that although females had lower probabilities of being displaced than
males in all industries and occupations except agriculture and sales, females had higher

unemployment rates and larger earnings losses than males.

2.2 Use of Comparison Groups: DWS Data. Both of the studies discussed above
utilize longitudinal data which provides information on the earnings of both displaced
and non-displaced workers at the time of re-employment (t) and the time of displacement
(t-s). Although displaced workers in the DWS are asked about their current and their pre-
displacement earnings, individuals in the outgoing rotation groups of the corresponding

CPS are asked only about their current earnings.2 This leaves researchers using the DWS

 

2 Individuals in their fourth or eighth month of participation in the CPS (the outgoing rotation groups) are
asked about their current earnings. Thus, for any one month of the CPS, earnings information is available
for approximately one-fourth of the CPS sample.

98

with information only about the time t earnings of non-displaced workers. Researchers
who want to use the CPS to construct a comparison group of non-displaced workers must
somehow estimate the time t-s earnings of non-displaced workers.

Alternatively, researchers using the DWS who do not use a comparison group can
simply measure earnings losses as the difference between displaced workers’ pre- and
post-displacement earnings. Carrington and Zaman (1994), Ruhm (1994), Carrington
(1993), Addison and Portugal (1989), Howland and Peterson (1988), and Podgursky and
Swaim (1987) all use this methodology.

Using the 1986 DWS, Seitchik and Zornitsky (1989) obtain a comparison group
from the CPS by measuring the relative earnings position of displaced workers across
quartiles of the wage and salary distribution of individuals in the outgoing rotation groups
of the CPS. Although their method utilizes a comparison group, it only allows a
comparison of a worker’s change in standing in the earnings distribution, which is a very
imprecise measure of displaced worker earnings losses. The Seitchik and Zornitsky measure
also masks potentially large, intra-quartile movements in earnings and overstates losses for
those who moved from the upper end of one quartile to the lower end of the next (and vice
versa).

Madden (1988) uses CPS household identiﬁcation numbers to match the records of
individuals who were in both the January 1984 CPS and the January 1983 CPS. This
matched data includes two time periods of information on individuals, as well as an

identiﬁer and extra information for displaced workers in the 1984 DWS and thus generates

99

a comparison group of non-displaced workers in the CPS. However, in a replication of
Madden’s study, Swaim and Podgursky (1991) argue that Madden’s matching technique
produces a non-random sample of displaced workers because it includes only workers who
were displaced within one year of the 1984 survey date, effectively eliminating the longer-
terrn displaced. Furthermore, because the CPS repeatedly samples houses/addresses and not
speciﬁc individuals, Madden’s method also only allows the inclusion of workers who did
not move as a result of displacement since those individuals in the 1984 DWS who state
that they had moved as a result of displacement cannot be identiﬁed in the 1983 CPS.3
Although di la Rica (1992) focuses upon the size and distribution of the pre-
displacement earnings losses of displaced workers, her study also utilizes a comparison
group, which consists of workers in the 1984 and 1985 outgoing rotation groups of the CPS.
She uses the 1986 DWS, but includes only workers displaced as part of a plant closing in
1984 or 1985. In order to measure the impact of displacement on pre-displacement
earnings, she regresses earnings at the time of displacement for both groups of workers on
several demographic and industrial characteristics, including a dummy for displacement.
She ﬁnds that, on average, displaced workers’ earnings are about 11 percent below the
earnings of similar non-displaced workers, suggesting that the earnings losses of displaced

workers begin to mount even before displacement occurs.4 She also ﬁnds that white collar

 

3 The method of obtaining a comparison group that I propose in section (IV) suffers from a similar data
restriction. However, I ran several tests to assess the effect of excluding movers from my analysis and
found that it had little impact on the results.

4 Jacobson, LaLonde, and Sullivan (1993) reach a similar conclusion.

100

workers tended to suffer eight percent larger pre-displacement losses than blue collar
workers, but there were no differential effects of gender or industry. For females, belonging
to a union reduced pre-displacement earnings losses. Note that since the di la Rica study
focuses upon only pre-displacement earnings, only one time period of earnings data from
the CPS is necessary to obtain an adequate comparison group.

Farber (1993) uses the earnings of workers in the CPS outgoing rotation groups in
order to measure the size of the average earnings losses of displaced workers. He utilizes
the 1984 through 1992 DWS and the CPS outgoing rotation groups for 1982 to 1992.
However, he only includes in his study displaced workers who had lost their jobs within
two years of the DWS survey date (for example, he includes workers displaced in 1990 and
1991 from the 1992 DWS, although the 1992 DWS provides information on workers
displaced in 1987, 1988, and 1989 as well). Like the Madden (1988) study discussed above,
this exclusion of workers displaced in earlier years systematically eliminates the longer-
term displaced, thus making his loss estimates downward biased.5 He estimates earnings
regressions for both displaced and non-displaced workers while including a dummy for
displacement, a dummy for the survey year, and an interaction between the displacement
and survey year dummies. By summing the coefﬁcients on these dummies, F arber
computes four different intercept terms: average displaced earnings in the survey year,

average displaced earnings in the year of displacement, and average non-displaced earnings

 

5 In one sense, this exclusion may actually improve Farber’s estimates. The pre-displacement earnings
information in the DWS is retrospective in nature. Thus, his exclusion of workers displaced further from
the survey date eliminates some of the measurement error, due to recall problems, in the data. See Horvath
(1982) for a discussion of bias in retrospective surveys.

101

for the two periods. This method, which is similar to the one I propose below, allows
Farber to measure the impact of displacement as a difference-in-differences: the difference
between earnings growth for the average displaced worker and earnings growth for the
average non-displaced worker. He does not, however, measure how that impact varies
across different demographic groups or across different industries. In a different section of
the paper, he measures earnings losses for displaced workers as the difference between pre-
and post-displacement earnings and ﬁnds that older workers and workers with high tenure
tend to experience larger losses. As noted by Pencavel (1993), F arber’s comparison group
estimator provides estimates of earnings losses which are very close in size to his estimates
obtained without a comparison group.

To summarize, in most cases, the ﬁndings from studies which utilize comparison
groups do not differ substantially from those of studies that do not use comparison groups.
For example, the studies conducted by Jacobson, LaLonde and Sullivan (1993), Swaim and
Podgursky (1991), and Seitchik and Zornitsky (1989), utilize comparison groups and ﬁnd
higher earnings losses for workers who change industries or occupations upon re-
employment. Several studies which do not utilize comparison groups (Addison and
Portugal (1989), Podgursky and Swaim (1987), and Shapiro and Sandell (1985)) also ﬁnd
larger earnings losses for industry and/or occupation changers. Using a comparison group,
Maxwell and D’Amico (1986) ﬁnd larger losses for women. Podgursky and Swaim (1987)
reach the same conclusion without using a comparison group. These results suggest that the

use of comparison groups may be unnecessary for measuring the impact of many

102

characteristics upon the earnings losses of displaced workers. However, as I discuss below,
whether or not one should utilize a comparison group depends upon the nature of the data at
hand, the parameter(s) of interest, and the assumed differences between the comparison
group and the displaced workers. For example, when trying to estimate the impact of
industry switching upon earnings losses while using the DWS, researchers do not have easy
access to an adequate comparison group. Thus, they must weigh the potential costs and
beneﬁts of either comparing either (i) the earnings losses of displaced industry switchers
against those of other displaced workers who did not change industries, or (ii) the earnings
losses of displaced industry switchers against the earnings changes of an external group of

non-displaced workers.

3. A Model of Earnings Losses

In order to further examine the importance and limitations of constructing a
comparison group, I build a series of simple models of earnings losses and compare the
differences in outcome and interpretation that each model implies. The most basic model
of earnings losses for displaced workers is given by equation (1), which represents the
difference between a given displaced worker’s current earnings and his earnings on his

former job.

(1) Ma: xl3+ 37+“

103

The dependent variable, Ayd, represents the change in earnings from time t-s, the
time of displacement, until time t, the time of re-employment.6 The vector X contains
demographic information including age, gender, race, tenure on former job, and
educational attainment. Although it could represent any variable of interest, in this
particular context S is a dummy variable representing whether workers changed industries
from time t-s to time t.7

The coefﬁcient [3 captures the effect of demographic characteristics on the
earnings losses of displaced workers relative to other displaced workers (for example,
male displaced workers relative to female displaced workers or black displaced workers
relative to white displaced workers). Similarly, y captures the effect of S on earnings
losses measured relative to other displaced workers.

Suppose that the earnings changes for non-displaced workers are given by
equation (2). The dependent variable, Ayn, represents the change in earnings from time t-s
to time t for non-displaced workers. The X vector and the S variable are deﬁned as in
equation (1). Thus, equation (2) provides earnings information for both industry changers

and those who remain in the same industry from time t-s to time t.

(2) Ayn: XBW‘ SY'+U‘

 

6 For example, when using the 1992 DWS, time t represents the year 1992 and time t-s can range from
1987 to 1992.

7 For example, S could represent whether workers participated in a training program from time t-s to time t,
or whether workers belong to a certain industry or metropolitan statistical area (MSA) at time t or time t-s.

104

It follows that a measure of earnings losses deﬁned as the difference in earnings

growth for displaced workers relative to non-displaced workers is given by equation (3).

(3) <4y.-Ay.)= XB“ + St“ +u“

Estimates of B and y in equation (1) can be interpreted as estimates of ([3“ + B‘)
and (7“ + 7‘), respectively. If [i and y (or (13“ + 13‘) and ('y“ + y‘)) are the parameters of
primary interest, there is no need to use the comparison group of non-displaced workers.
For example, let S be equal to one if the displaced worker participated in a training
program available only to displaced workers. A positive estimate of y from equation (1)
would indicate that participants in the training program experienced smaller earnings
losses than other displaced workers. Since the earnings changes for displaced workers are
only compared to those of other displaced workers, no comparison group of non-
displaced workers is necessary; instead, one could view the displaced workers who did
not participate in training as a comparison group for the displaced workers who
participated in training.

Alternatively, the estimates of B“ and y“ (or, ([3 - 13‘) and (y - y‘)) in equation (3)
may be the parameters of interest. For example, let S be equal to one if the individual was
displaced from the services industry. A positive estimate of y from equation (3) would

indicate that workers who were displaced from the services industry experienced smaller

105

earnings losses than other displaced workers. Note that the term loss as applied to
equation (3) is deﬁned as the difference between a displaced worker’s earnings growth
and what that earnings growth would have been in the absence of displacement. If
researchers using the DWS are interested in the parameters [3“ and y“, they must ﬁnd a
proxy for Ayn, since it is unobserved in the CPS/DWS data.

It is important to note that whether or not a researcher actually needs to use a
comparison group depends not only upon the parameter she is interested in estimating,
but also upon how that parameter varies for displaced and non-displaced workers. In the
ﬁrst example above, the training program is assumed not to affect the earnings of non-
displaced workers, so that y‘ would be equal to zero in equation (2). Use of the earnings
of non-displaced workers would not add information to the model of training program
impact, since it would result in an estimate of y‘ ‘ = (y + O) = y .

Table 3 forrnalizes this notion. Like Table 1, Table 3 provides earnings change
comparisons for displaced workers measured against a comparison group of non-
displaced workers. In addition, Table 3 speciﬁes the impact of the outcome S.

In this case, assume that S represents whether a worker switched industries from
time t-s to time t. For displaced workers, previous research (summarized in Table 2) has
shown that switching industries is associated with speciﬁc human capital losses, long
unemployment spells, and large earnings losses. Alternatively, one would think that non-
displaced workers would only voluntarily switch industries if their earnings rose as a

result of switching (assuming away industry effects which both contribute to

106

displacements and encourage non-displaced workers to switch industries).8 In Table 3,
the impact of S is allowed to vary across displaced and non-displaced workers, which
allows for measurement of the differential effect of the parameter of interest, 7‘ ‘.

Column (1) shows the earnings for both displaced workers and the comparison
group of non-displaced workers, which are both equal to 1000 at time t-s, the time of
displacement. At time t, one can observe the earnings of both non-displaced and re-
employed displaced workers, as shown in column (2). The re-employment (time t)
earnings of displaced workers with S equal to zero are 900 and the time t earnings of non-
displaced workers with S equal to zero are 1100, which implies an earnings loss of 200
for those displaced workers with S equal to zero. This loss measure, which would be
captured by estimating equation (3), is shown in column (5). Similarly, earnings losses
for displaced workers with S equal to one are 350 (the difference between the earnings
change for displaced workers with S equal to one and the earnings change for non-
displaced workers with S equal to one). The differential impact of S, which would be
captured by y“ in equation (3), is -150 since the loss for those with S equal to zero is 150
smaller than for those with S equal to one.

Note that if one simply compares the earnings changes of both types of displaced

workers (those with S equal to zero and those with S equal to one) with the earnings

changes of only the non-displaced workers who have S equal to zero, the differential

 

8 The increase in earnings for non-displaced workers who switch industries is not as important as the idea
that earnings for non-displaced workers who switch industries are diﬂerent ﬁom other non-displaced
workers at time t.

107

effect of S can also be captured without using the comparison group (i.e. by estimating y
in equation (1)). This can be seen by comparing the earnings changes in column (3) with
those in column (6). In this instance, the differential impact of switching (-100) is the
same using either equation (1) or equation (3), since equation (3) would merely subtract
the same amount (7‘) from y‘ ‘ for all displaced workers.

The point of the example above is this: if earnings changes can be observed for
both non-displaced workers with S equal to one and non-displaced workers with S equal
to zero, as in equation (3), an estimate of the differential impact of S on earnings losses,
deﬁned as in equation (3), can be obtained by using a comparison group of non-displaced
workers; however, if variation, due to S, in the earnings changes of non—displaced
workers cannot be observed, the same estimate of the impact of S can be obtained without
using the comparison group, as in equation (1).9

Thus, when using data like the DWS, researchers must choose the loss measure
and estimation method based upon the parameter of interest and upon assumptions about
the differences in earnings changes between the comparison group and the displaced
workers. If the earnings of non-displaced workers do not vary (or, perhaps worse, cannot

be observed to vary) with the variable of interest, the differential impact of that variable

 

9 Note that the result in the example above makes the assumption that the probability that S is equal to one

is the same for displaced and for non—displaced workers, i.e. a worker who is displaced and has S equal to
one would have been just as likely to have S equal to one if he had not been displaced. This is unrealistic if
some industry switching occurs as a result of displacement and/or is more likely for displaced workers.
This suggests that one would need to control for the different probabilities of switching for displaced and
non-displaced workers when using a comparison group to estimate earnings losses. However, the
assumption is likely to be true for variables like industry, occupation, and MSA at the time of
displacement.

108

should be the same using either equation (1) or equation (3). For example, it may be
reasonable to assume that displaced workers only switch industries as a result of
displacement and in the absence of displacement, they would not have switched
industries. To estimate the differential impact of switching on earnings losses in this case,
one could either compare the earnings changes of all displaced workers against the
earnings changes of only the non-displaced workers who did not switch industries, or one
could simply compare the earnings changes of displaced industry switchers against the
earnings changes of displaced workers who did not switch industries. Either method
would give the same estimated differential impact of switching, but the estimate made
without the comparison group of non-displaced workers would be less costly to obtain for

those using the DWS.10

4. Comparison of Earnings Loss Measures Using A CPS Comparison Group

In this section, I compare earnings loss estimates made using the basic models in
both equation (1) and equation (3) using data on workers in the 1992 DWS. I use the
average earnings changes of non-displaced workers in the CPS outgoing rotation groups for
the years 1987 to 1992 as a benchmark against which to compare the earnings changes of

displaced workers. For each year of the CPS (1987 to 1992), I compute the average

 

‘0 The idea that one can observe variation in the earnings changes of non-displaced workers across the
variable of interest is important because, as I point out below, in theory my proposed method of obtaining a
comparison group from the CPS allows a measure of the variation in the earnings of non-displaced workers
across many factors, including age, gender, marital status, and switching industries from time t-s to time t.
In practice, however, the measure of such variation in the earnings of non-displaced workers is not attained
without a tradeoff of sample size.

109

earnings of workers, disaggregated by MSA, industry, occupation, full-time (greater than 35
hours per week), and demographic cells. Thus, I compute a different average earnings ﬁgure
for each MSA/industry/occupation/ﬁrll-time/demographic combination in the CPS. For
each year, these earnings ﬁgures are them matched to workers in the DWS who share the
same MSAlindustry/occupation/full-time/demographic characteristics. The change in
earnings (from the time of displacement until the survey date) for a displaced worker can
then be compared with the change in eamings (over the same time period) for an average

non-displaced worker sharing the same characteristics.

4.1 Potential Problems and Biases. The above method of obtaining a comparison
group has two primary weaknesses. First, merging the earnings ﬁgures from the CPS with
individuals in the DWS results in many cases where no MSA/industry/occupation/full-
time/demographic match occurs. Because of this, displaced workers who do not have
similar counterparts in the outgoing rotation groups of the CPS could not be used to
estimate earnings losses. Although it is possible to avoid this problem by computing
average earnings at more aggregated levels, this aggregation would result in less variation in
the earnings of non-displaced workers across important characteristics. As I discussed
above, if variation in the earnings changes of non-displaced workers across the outcome of
interest cannot be observed, the same estimates of the impact of that parameter can be
obtained without using the comparison group at all. Thus, it is necessary to balance the level

of disaggregation for computation of average earnings and the size of resulting sample of

110

matched displaced workers. In the analysis below, I utilize data on workers in the DWS
who had matching counterparts in the CPS only across MSA/industry/occupation/ﬁrll-time
characteristics.

Second, as is the case with any comparison group, if the expected earnings changes
of displaced workers differ systematically from those of non-displaced workers, estimates

of earnings losses made using this method are biased.'1

4.2 Sample Characteristics. Following the federal deﬁnition of a displaced
worker, I include in my sample workers in the DWS who had at least three years of job
tenure prior to their displacement]2 In order to both control for local labor market
conditions and to merge the average eamings ﬁgures obtained from the CPS to the workers
in the DWS sample, I limit the sample to workers who live in metropolitan statistical areas

(MSAS) and who did not move after displacement.l3 Finally, in order to ensure that workers

 

” Jacobson, LaLonde, and Sullivan (1993) and de la Rica (1992), ﬁnd that displaced workers start
experiencing relative earnings decreases two to three years prior to the time of displacement. This result might
lead one to conclude that displaced workers have characteristics which would lead them to have lower
earnings than other workers, regardless of displacement. However, Jacobson, LaLonde, and Sullivan conclude
that because both the earnings and earnings-related characteristics of displaced and non—displaced workers
were generally similar prior to time t-s-n (n periods prior to displacement), in the absence of displacement or
some other adverse event, the earnings of the two groups would have remained similar throughout the period
of their study (which ranges from ﬁve years prior to displacement until six years afterwards). Thus, the
difference between the earnings of displaced workers and similar non-displaced workers seen at time t-s can
be interpreted primarily as “losses due to events that led to workers’ displacements” and not primarily due to
differences between displaced workers and non-displaced workers (Jacobson, LaLonde, and Sullivan, 1993: p.
691)

12 The official federal deﬁnition of a displaced worker is one who lost a job after three or more years of
tenure, owing to slack work, abolition of the job, mass layoffs, or plant closure (F lairn and Seghal, 1985).

‘3 Workers who moved to ﬁnd re-employment are excluded from the estimates since their former MSA is not
identiﬁed in the DWS, making it impossible to match them with non-displaced workers in their former MSA.
Unfortunately, the DWS does not ask workers about the geographic area in which they were employed prior to

111

had enough time after displacement to conduct a job search, the sample is further restricted
to include only workers who were displaced between January 1987 and December 1990.14

Table 4 presents the mean values of the variables used in this study. The columns
represent three different groups: (1) all displaced workers in the sample; (2) the subsample
of displaced workers who had matching counterparts in the CPS outgoing rotation groups
and could therefore be included in the earnings loss regressions which utilize the
comparison group; and (3) workers in the 1992 CPS outgoing rotation groups. Among all
displaced workers in the sample, those who became re-employed experienced earnings
losses (measured as the difference between pre- and post-displacement earnings) of about
$44 per week. The sample is approximately 60 percent male, 88 pei‘cent white, and on
average 41 years old with 8.7 years of job tenure. Approximately 63 percent of the re-
employed workers switched industries upon re-employment.

For most variables, the means differ only slightly between the entire sample
(column (1)) and the “matched” subsample (column (2)). Notable differences between the

two groups include: the current and former weekly earnings variables, which are slightly

 

displacement. However, the survey does ask "since displacement, has [the reference person] moved to a
different city or county to look for work or to take a different job?" As this question does not reﬂect whether
workers changed MSAs as a result of displacement, it would be erroneous to include workers who moved in
the analysis. I ran several tests to assess the impact of excluding movers from the sample and found no
signiﬁcant effect of moving on earnings losses.

1’ The DWS asks workers about the year but not the month of their displacement. It is unlikely that workers
displaced in late 1991 or January 1992 had enough time to conduct a job search before the January 1992
survey. Together, the three exclusion restrictions drop 2127 of the original 2880 observations from the sample.
The exclusion of workers with fewer than three years of tenure drops 1527 of the original 2880, the exclusion
of those displaced in 1991 or 1992 drops 442 of the remaining 1353 observations, and the exclusion of movers
drops 158 of the remaining 911 observations ﬁ'om the sample.

112

higher for the subsample; the switched detailed industry upon re-employment variable,
which was about six percentage points lower for the subsample; the change in MSA
unemployment rate variable, which was slightly larger for the subsample; and the
proportion of the group living within 25 miles of another MSA, which was six percentage
points higher for the subsample.

A comparison of columns (2) and (3) reveals that the average weekly earnings of
individuals in the CPS who had matching counterparts in the DWS were slightly higher
than those of other workers in the CPS. This is unlikely to affect the earnings loss estimates
made using equation (3), however, unless the diﬂerence in average earnings for the workers
in the CPS who had matching counterparts in the DWS are systematically different those of
workers in the entire CPS.15 In addition, workers in the CPS were on average three years
younger than displaced workers in the DWS sample, and were less likely to be male, white,

or married.

4.3 Estimation of earnings losses. Detailed models of the earnings loss measures

suggested in equations ( l) and (3) are presented in equations (4) and (5), respectively.

(4) (yijkmft ‘ yijkmf. t-s) = 130 + 51 Xit + 132 SWitChit + 53 ijm,t-s + 134 Dt-s + “it

 

‘5 There is slight contamination in the non-displaced earnings values since displaced workers could be
included in the CPS. As noted by Farber (1993), since only a small fraction of workers are displaced in any
year (he estimates the fraction to be about 10 percent), the distortion is likely to be minimal. However, the
contamination will slightly bias the earnings loss measure downward.

113

(5) (Yijkmm ' Yijhmr. t-s) ' (ijmf.t ' ijmf.t-s) = 130” + [31” xii + 132” SWitChit + [33” ijm.t-s + 134” Dt-s + “it

The variables yiJ-kmf, . and Yijkmr, H represent the logs of the 1992 and pre-
displacement earnings, respectively for displaced workers. The variables Yin,“ and ijmf, ,_
3 represent the logs of average weekly earnings for non-displaced workers in industry j,
occupation k, MSA m, and full-time status f in 1992 and in the year of displacement,
respectively. The difference (yijkmf, , - yijkmﬂ t_,) measures the change in actual earnings for
displaced workers. The difference (Y hunt. , - jkmf. H) measures the change in weekly
earnings for the average worker who is in the same industry, occupation, and MSA, and
who works a similar amount of weekly hours as the displaced worker. If (Y “mt, , - ijmf, t,,)
is positive, it reﬂects that on average, non-displaced workers in the given MSA, industry,
occupation, and full-time status experienced earnings growth between the time of
displacement and January, 1992. If the difference (Y M“, t - Mr, H,) is negative, workers
who were not displaced also suffered earnings losses on average during the period.'6 The
X, and ijmgS vectors represent demographic and industry/area characteristics, respectively.
The D,_S vector represents dummies for the year of displacement.

Estimates of equations (4) and (5) are presented in Table 5. Each speciﬁcation in
columns (1) through (4) is estimated ﬁrst for the sample of displaced workers only

(columns denoted with “D”) and then for the comparison group subsample (columns

 

'6 For the sample used in this analysis, there are 157 (out of 483 total) cases where the average earnings of the
comparison group fell over time, representing 110 MSA/industry combinations.

114

denoted with “C”).'7

Estimates made without utilizing the comparison group (equation (4)) suggest that
displaced workers who switched industries upon re-employment experienced 11 to 15
percent smaller earnings changes (11 to 15 percent larger losses) than displaced workers
who remained in the same detailed industry upon re-employment. Estimates from the model
which utilizes the comparison group (equation (5)) suggest that workers who switched
industries upon re-employment experienced 14 to 21 percent larger losses than workers who
did not switch industries upon re-employment. Thus, as expected, the estimates of the
switch effect made while using the comparison group are biased if one assumes that
displaced workers would not have switched industries had they not been displaced. Relating
this back to the example in Table 3, since the average earnings ﬁgures are computed for the
workers’ displacement industries, they only allow comparisons relative to non-displaced

workers with S equal to zero. 18 However, the comparison group estimates in Table 5 reduce

 

l 7 In order to correct for possible re-employment selection in the earnings loss estimates, 1 utilized the standard
Heckman (1979) procedure. Two additional variables, a dummy for the presence of children in the household,
and a variable which counts the number of earners in the household, were used to identify the re-employment
regression. The resulting estimates were not signiﬁcantly different from those presented. The most likely
reason for this is that 92 percent of the sample was re-employed at the time of the 1992 DWS, primarily since
the sample includes workers who were displaced at least one year prior to the survey date and generally had
ample time to become re-employed.

'8 It would be possible to measure average earnings of non-displaced workers who “switch” industries from
time t-s to time t by computing the difference between the average earnings of non-displaced workers in
one industry at time t and the average earnings of non-displaced workers in a different industry at time t-s
(the industries would correspond to the displaced workers’ current and former industries, respectively).
However, although the DWS asks workers about their job tenure, the regular CPS does not. Thus, the non-
displaced “switchers” earnings measure would be likely to introduce bias into the estimates since the
relative time in the industry for the two groups of workers (displaced and non-displaced) would differ. For
the displaced workers who switched industries, the time in industry would be less than ﬁve years and for
the non-displaced workers, the time in industry could be much longer.

115

some of this possible bias since the change in non-displaced earnings values used in
estimating equation (5) include cross-sections of both non-displaced workers who switched
industry from time t-s to time t, and individuals who did not change industries over the
period. In addition, if nondisplaced workers who switch industries generally experience
wage gains, the effects of switched industry would be even larger than those reported in the
“C” columns in Table 5.

Alternatively, the impact of weeks unemployed was roughly the same using either
earnings loss measure, reﬂecting the idea that weeks unemployed is not an outcome which
varies for non-displaced workers. Thus, using the comparison group does not add
information when estimating the impact of weeks unemployed on earnings losses.

Other notable quantitative differences between the two loss measures occur for the
estimated impact of male, white, and the presence of another MSA within 25 miles. The
impact of the change in MSA unemployment rate is qualitatively different for the two

measures and is only marginally signiﬁcant when the comparison group is not used.

5. Conclusion

In this paper, I utilize simple hypothetical examples to show that the appropriateness
of using a comparison group depends upon the parameter of interest and upon the
assumptions about the differences between the comparison group and the displaced
workers. 1 then propose a method of obtaining a comparison group by computing the

average earnings (conditional upon several worker characteristics) of non-displaced workers

116

in the outgoing rotation groups of the CPS. Earnings loss estimates made both with and
without the comparison groups differ along several parameters. Thus, I conclude that
although the use of a comparison group is important for measuring the earnings losses of
displaced workers, it is neither a prerequisite for obtaining reliable qualitative estimates, nor
a guarantee that estimates will not be biased, particularly when longitudinal data on
displaced and non-displaced workers is not available, as is the case for researchers using the
DWS.

I also present a review of previous attempts to measure the earnings losses of
displaced workers, paying particular attention to differences in estimates which result when
comparison groups of non-displaced workers are utilized. I ﬁnd that in most cases, research
which does not use comparison groups has reached the same general conclusions as
research which utilizes comparison groups.

Future research on the use of comparison groups to measure displaced worker
earnings losses could utilize longitudinal data (from the PSID, perhaps) to measure the
relative earnings loss outcomes using three different earnings loss measures: (1) a
comparison of pre- and post-displacement earnings for displaced workers, (2) use of the
change in average earnings of non-displaced as a benchmark against which to compare the
earnings changes of displaced workers (the method proposed in this paper), and (3) the use
of actual earnings changes of non-displaced workers as a benchmark against which to
compare the earnings changes of displaced workers. One advantage of using true

longitudinal data like the PSID would be the earnings information for both displaced and

117

non-displaced workers. This is important since recent research has found that the earnings
of otherwise similar displaced and non-displaced workers differ substantially at the time of
displacement. One disadvantage of using the PSID, however, would be its small size and

scope.

Comparison of Earnings Loss Measurement Methodsl

TABLE I

 

 

 

 

 

 

(l) (2) (3) (4) (5)
Real Real Real Weekly Real Weekly Real Weekly
Weekly Weekly Earnings Earnings Earnings Loss
Earnings Earnings Loss without Loss with with
time t-s time t comparison comparison comparison
group group group
Type of Worker2
CASE A CASE B
Displaced 1000 800 200 100 300
CASE A 1000 900
Non-displaced
(earnings fall)
CASE B 1000 1100
Non-displaced
(earnings rise)

 

 

 

 

 

 

 

1 The table compares the real earnings losses (or gains) for a hypothetical worker under two earnings loss measures:
losses measured without a comparison group and losses measured with a comparison group. The table also compares
the earnings loss measures with a comparison group under two situations (cases): Case A, when the real earnings of
non-displaced workers in the industry fall over period t-s to period t, and Case B, when the real earnings of non-

displaced workers in the industry rise over the period.

2 Calculations assume that the three types of workers are identical apart from displacement (that is, the earnings of all
three groups would have moved together had displacement at time t not occurred).

118

 

TABLE 2

Comparison Group Use and Findings from Studies which measure the

 

Distribution of Earnings Losses for Displaced Workersl

 

 

 

 

 

 

 

 

 

Studies Which Use Comparison Group: Primary Data Groups of workers found to
Comparison Groups Earnings loss Measure Sources experience the largest losses
Jacobson, LaLonde, Sullivan non-displaced workers in Social Security high tenure; workers displaced in
(1993) Pennsylvania Social Records in depressed local communities;
Security administrative data Pennsylvania industry changers
set: difference between 1974-1986

Farber ( 1993)

di la Rica (1992)

Ruhm(l99l)

Swaim and Podgursky (I991)

Seitchik and Zornitsky (1989)

actual and expected
earnings

workers in the CPS:
compares average earnings
changes for displaced and
non-displaced

workers in the CPS:
compares pre-displacement
earnings of displaced with
earnings of workers in the
CPS

non-displaced workers in
the PSID: compares the
difference in estimated
earnings of displaced and
non-displaced

workers in the CPS:
compares the difference in
estimated earnings of
displaced and non-
displaced

workers in the CPS:
compare the change in
relative position in the
earnings distribution of
non-displaced workers

 

DWS 1984-1992

DWS 1986

PSID 1968-1979

DWS 1984

DWS 1986

does not present earnings loss
differentials for different
demographic groups, but instead
focuses upon the size of average
losses

pre-displacement losses higher for
white collar workers; pre-
displacement losses lower for
women in unions

long term losses exist; did not
measure demographic or industry
characteristics on losses

industry or occupation changers; less
educated; workers displaced during
depressed economic conditions

industry changers

1 Full citations for each of these studies is provided in the bibliography. The table excludes case studies, studies
conducted prior to 1985, and studies which did not focus upon earnings losses.

119

TABLE 2 (cont’d).

 

Studies Which use

 

 

Comparison Group:

 

 

Primag Data

Groups of workers found to

 

 

 

 

 

 

 

Comparison Groups Earnings loss Measure Sources experience the largest losses
Madden (1988) workers in the CPS: DWS 1984 blue-collar; industry changers; white
compares the difference in women; black men; less educated
estimated earnings of
displaced and non-
displaced
Maxwell and D'Amico (1986) non-displaced workers in NLSY low wage; women
the NLSY: difference
between predicted and
actual earnings
Studies Which do not use Earnings loss Measure Primagy Data Groups of workers found to
Comparison Groups Sources experience the largest losses

 

 

 

Carrington and Zaman (1994)

Ruhm (1994)

Carrington (1993)

Topel (1990)

none: compare pre- and
post-displacement earnings

none: compares post-
displacement earnings of
displaced

none: compares pre- and
post-displacement earnings

none: compares pre-and
. . 1
post-displacement earnings

 

DWS 1984-1988
(males only)

DWS 1988-1990

DWS 1984-1988

PSID 1968-1985
DWS 1984 1986
(males only)

workers from unionized, high wage
industries with large ﬁrms; high
tenure workers in manufacturing and
mining

workers who did not receive
advance notice of displacement;
workers with long unemployment
spells; minorities; women; workers
displaced during depressed
economic conditions; workers in
highly unionized industries

high tenure workers; highly
experienced workers displaced
during depressed economic
conditions; workers displaced from
declining industries; industry and
occupation changers

industry or occupation changers;
high experience; union workers;
older workers

1 Although Topel (1990) does not utilize a comparison group for the analysis in the body of his paper, he states in a
footnote that in order to adjust earnings losses to reﬂect wage growth that would have occurred in the absence of
displacement, he imputed wage growth for displaced workers in the PSID using the earnings of non-displaced workers
within in the same “job”. This resulted in earnings loss ﬁgures which were larger, but it did not substantially alter other
conclusions in his study (Topel, 1990; p. 194).

120

TABLE 2 (cont’d).

 

Studies Which do not use

 

Comparison Groups

 

 

Groups of workers found to
experience the largest losses

 

 

 

Addison and Portugal (1989)

Howland and Peterson (1988)

Podgursky and Swaim (1987)

Shapiro and Sandell (1985)

Earnings loss Measure Primary Data
Sources

none: compare pre- and DWS 1984

post-displacement earnings

none : compare pre- and DWS 1984

post-displacement earnings

none: compare pre- and DWS 1984

post-displacement earnings

none - predicted earnings of
displaced used as
benchmark for post-
displacement earnings

121

NLS-mature men

high tenure; those with long
unemployment spell; industry or
occupation changers

workers displaced in depressed local
communities; older; less educated;
high tenure blue collar

blue-collar; women; workers with
substantial speciﬁc human capital
investments; workers displaced in
depressed local communities; male
blue-collar industry or occupation
changers; male white-collar industry
changers.

high tenure; workers displaced
during depressed economic
conditions; older workers; blacks;
occupation changers

TABLE 3

Estimating the Differential Impact of Outcome S '

 

 

 

 

 

 

 

 

 

 

(l) (2) (3) (4) (5) (6)
Real Real Change in Change in Earnings Earnings
Weekly Weekly Earnings Earnings Loss Loss relative
Earnings Earnings Displaced Non- relative to to
time t-s time t Displaced comparison comparison
group2 group
with s = o3
y... Y: Ayu Ayn Ayn - Ayr. Ayr - AYuls-o
Displaced 1000 900 -100 - 200 200
S=0
Non-Displaced 1000 1100 - 100 - -
S=0
Displaced 1000 800 -200 - 350 300
S=1
Non-displaced 1000 l 150 - 150 - -
S=l
Differential - - = -100‘ y’ = 50 y“ = -150 y = -100
impact of S = ('y" + 7’)

 

 

 

 

 

 

 

 

1 The table compares the real earnings losses (or gains) for workers while allowing the impact of S, the variable of
interest, to vary for displaced and non-displaced workers.

2 Calculations assume that the displaced and non-displaced workers are identical apart from displacement (that is, the
earnings of the groups with the same values for S would have moved together had displacement at time t-s not

occurred).

3 Calculations compare the earnings changes of displaced workers only to non-displaced workers with S equal to zero.

4 See text for a discussion of the parameters 7, y‘, and y“.

122

 

TABLE 4

Mean Values of Variables 1

 

 

 

Re-employed

Current weekly earnings (1992)

Weekly earnings on former job

Change in weekly earnings

Weekly earnings of average non-displaced worker (1992)

Weekly earnings of average non-displaced worker at time
of displacement

Change in weekly earnings for average non-displaced
worker

Weeks unemployed
Switched detailed industry upon re-employment *2

Change in Unemployment rate in MSA
(year of displacement to year after displacement)

Another MSA within 25 miles *
Years of tenure on former job
Age

Male "‘

White *

Married *

Number of observations

 

 

(r)
.92 (.27)
459 (258)
505 (308)

-44 (219)

17 (21)
.63 (.48)

.42 (1.06)

.37 (.48)
8.72 (6.55)
41 (10)
.60 (.49)
.88 (32)
.69 (.46)

753

L2)
.92 (.27)
472 (267)
518 (317)
44 (220)
539 (228)

483 (215)

56 (167)

17 (22)
.57 (.50)

.56 (1.02)

.43 (.50)
8.70 (6.48)
41 (10)
.58 (.49)
.88 (.32)
.68 (.47)

525

525 (321)

38 (11)
.56 (.50)
.86 (.35)
.62 (.49)

132554

1 Standard deviations are reported in parentheses. Column (1) presents means for all displaced workers in the sample.
Column (2) presents means for re-employed displaced workers who had a matching counterpart in the CPS and were
therefore used in earnings loss regressions which utilized the comparison group. Column (3) presents means for full-

time workers in the 1992 CPS outgoing rotation groups.

2 Variables denoted with an asterisk are dummy variables equal to one if the condition holds and zero otherwise.

123

Displaced Only (D)
Comparison Group (C)

 

Switched Industries

Tenure on former job

33

Age

White

Male

Married

Weeks Unemployed

Change in MSA
Unemployment Rate (year of
displacement until year
afterwards)

Another MSA within 25
miles

Constant

Full-time/Part-time Controls
Schooling Controls

Year of Displacement
Controls

Former Industry and
Occupation Controls

Adjusted R-Squared

Comparison of Earnings Loss Estimates ‘

TABLE 5

 

 

(l) (2) (3) (4)

D C D c D c D c
..13*"*2 -.16*"‘ -.11** -.14" -.11** -.14** -.15** -21"
(.04) (.05) (.04) (.05) (.04) (.05) (.04) (.06)
-.01** -.01* -.01" -.01 -.01" -.01* -.o1" -.01*
(.003) (.004) (.003) (.004) (.003) (.004) (.003) (.005)
-.005 -.017 -.003 -.014 -.005 -.015 -.003 -.031
(.01) (.05) (.01) (.02) (.01) (.02) (.01) (.02)
.04 .13* 04 .14* 04 .15* 05 .13
(.06) (.08) (.05) (.08) (.05) (.08) (.06) (.09)

-.14** -.11" -.14" -.11** -.13" -.11** -.12" -.11
(.04) (.05) (.04) (.05) (.04) (.05) (.05) (.07)
.001 .01 .003 .01 -.002 .01 -.01 -.07
(.04) (.06) (.04) (.06) (.04) (.06) (.04) (.06)

- - -004" -.003" -.004" -.003" -.004" -.003**
(.001) (.001) (.001) (.001) (.001) (.001)
- - - - 05* -.02 04* -.03
(.03) (.04) (.03) (.04)
- - - - -.11** -.07 -.13** -.09
(.04) (.06) (.04) (.07)
.41 .57 .44 .54 .43 .60 .05 1.19

(.34) (.49) (.34) (.49) (.34) (.49) (.69) (1.35)
yes yes yes yes yes yes yes yes
yes yes yes yes yes yes yes yes
yes yes yes yes yes yes yes yes

no no no no no no yes yes
.18 .09 .20 .10 .21 .11 .25 .12

 

l The dependent variable for columns labeled with D is: log (current earnings) - log (former earnings). The dependent
variable for columns labeled with C is: change in log (displaced earnings) - change in log (average earnings). See text

for discussion.

2 " Statistically signiﬁcant at the 5 percent level. " Statistically signiﬁcant at the 10 percent level.

3 All speciﬁcations also include age-squared, which has a coefﬁcient of zero and is insigniﬁcant in all specifications.

124

BIBLIOGRAPHY

BIBLIOGRAPHY

Addison, John T. and Pedro Portugal. (1989) “Job Displacement, Relative Wage
Changes, and Duration of Unemployment,” Journal of Labor Economics, 7(3):281-302.

Carrington, William J. (1993) “Wage Losses for Displaced Workers: Is It Really the Firm
that Matters?” Journal of Human Resources, 28(3):435-462.

---------- and Zaman, Asad. (1994) “Interindustry Variation in the Costs of Job
Displacement,” Journal of Labor Economics, 12(2):243-275.

di la Rica, Sara. (1992) “Displaced Workers in Mass Layoffs: Pre-Displacement
Earnings and Unions Effect,” Industrial Relations Section Working Paper No. 303,
Princeton University.

Flaim, Paul and Ellen Seghal (1985) "Displaced Workers of 1979-83: How Well Have
They Fared?" Monthly Labor Review, lO8(6):3-16.

F arber, Henry S. (1993) “The Incidence and Costs of Job Loss: 1982-91,” Brookings
Papers: Microeconomics 1993 73-119.

Gardner, Jennifer M. (1993) Displaced Workers 1987-1991 . Bulletin 2427. Bureau of
Labor Statistics. July.

Heckman, James J. (1979) "Sample Selection Bias as Speciﬁcation Error."
Econometrica, 47(1): 1 53-1 61 .

Horvath, Francis W. (1987) "The Pulse of Economic Change: Displaced Workers of
1981-85," Monthly Labor Review, 1 10(6): 3-12.

---------- (1982) "Forgotten Unemployment: Recall Bias in Retrospective Data," Monthly
Labor Review, 105(3):40-44.

Howland, Marie and George E. Peterson (1988) "Labor Market Conditions and the Re-
employment of Displaced Workers." Industrial and Labor Relations Review, 42(1):109-
122.

Jacobson, Louis S., LaLonde, Robert, and Daniel G. Sullivan (1993) "Earnings Losses of
Displaced Workers," American Economic Review, 83(4):685-707.

125

126

Kletzer, Lori Gladstein (1989) “Returns to Seniority After Permanent Job Loss,”
American Economic Review, 79(3):536-543.

Kruse, Douglas L. (1988) “International Trade and the Labor Market Experience of
Displaced Workers,” Industrial and Labor Relations Review 41(3):402-417.

Madden, Janice F. (1988) “The Distribution of Earnings Losses Among Displaced
Workers,” Journal of Human Resources, 23(1):93-107.

Maxwell, Nan L. and Ronald J. D’Amico (1986) “Employment and Wage Effects of
Involuntary Job Separation: Male-Female Differences,” American Economic Review,
76(2):373-377.

Pencavel, John (1993) Comment on Henry S. Farber, “The Incidence and Costs of Job
Loss: 1982-91,” Brookings Papers: Microeconomics 1993 124-128.

Podgursky, Michael and Paul Swaim (1987) "Job Displacement and Earnings Loss:
Evidence From the Displaced Worker Survey," Industrial and Labor Relations Review,
41(1):17-29.

Ruhm, Christopher J. (1994) "Advance Notice, Job Search, and Post-displacement
Earnings," Journal of Labor Economics, 12(1):1-28.

---------- (1991) “Are Workers Permanently Scarred by Job Displacements?” American
Economic Review, 81(1):319-324.

Seitchik, Adam and Jeffery Zornitsky (1989) From One Job to the Next: Worker
Adjustment in a Changing Labor Market. Kalamazoo, Michigan: W.E. Upjohn Institute.

Shapiro, David and Steven H. Sandell (1985) “Age Discrimination in Wages and
Displaced Older Men,” Southern Economic Journal, 52(1):90-102.

Swaim, Paul and Michael Podgursky (1991) "The Distribution of Economic Losses

Among Displaced Workers: A Replication," The Journal of Human Resources,
26(4):742-755.

Topel, Robert (1990) “Speciﬁc Capital and Unemployment: Measuring the Costs and
Consequences of Job Loss,” Carnegie-Rochester Conference Series on Public Policy,
3 3: 1 81-214.

"lllllllllllllll“