Unemployment Insurance: Factors Associated with Benefit Receipt
(07-MAR-06, GAO-06-341).
Unemployment Insurance (UI), established in 1935, is a complex
system of 53 state programs that in fiscal year 2004 provided
$41.3 billion in temporary cash benefits to 8.8 million eligible
workers who had become unemployed through no fault of their own.
Given the size of the UI program, its importance in helping
workers meet their needs when they are unemployed, and the little
information available on what factors lead eligible workers to
receive benefits over time, GAO was asked to determine (1) the
extent to which an individual worker's characteristics, including
past UI benefit receipt, are associated with the likelihood of UI
benefit receipt or unemployment duration, and (2) whether an
unemployed worker's industry is associated with the likelihood of
UI benefit receipt and unemployment duration. Using data from a
nationally representative sample of workers born between 1957 and
1964 and spanning the years 1979 through 2002, and information on
state UI eligibility rules, GAO used multivariate statistical
techniques to identify the key factors associated with UI benefit
receipt and unemployment duration. In its comments, the
Department of Labor stated that while there are certain
qualifications of our findings, the agency applauds our efforts
and said that this report adds to our current knowledge of the UI
program.
-------------------------Indexing Terms-------------------------
REPORTNUM: GAO-06-341
ACCNO: A48477
TITLE: Unemployment Insurance: Factors Associated with Benefit
Receipt
DATE: 03/07/2006
SUBJECT: Labor statistics
State programs
Statistical data
Unemployment insurance
Unemployment insurance benefits
DOL Unemployment Insurance Program
******************************************************************
** This file contains an ASCII representation of the text of a **
** GAO Product. **
** **
** No attempt has been made to display graphic images, although **
** figure captions are reproduced. Tables are included, but **
** may not resemble those in the printed version. **
** **
** Please see the PDF (Portable Document Format) file, when **
** available, for a complete electronic file of the printed **
** document's contents. **
** **
******************************************************************
GAO-06-341
* Results in Brief
* Background
* Certain Characteristics Are Associated with UI Benefit Recei
* Unemployed Workers with Higher Earnings, Younger Workers, Wo
* Unemployed Workers Who Have Higher Earnings or Are Younger o
* Unemployed Workers Who Received UI in the Past Are More Like
* Receiving UI Benefits, along with Other Factors, Is Associat
* Receiving UI Benefits Is Associated with Longer Unemployment
* Unemployed Workers with Lower Earnings and Less Education Te
* Unemployment Duration Is Not Associated with Past UI Receipt
* Certain Industries Are Associated with UI Benefit Receipt an
* Unemployed Workers from Mining and Manufacturing Are More Li
* The Relationship between Past and Current UI Receipt Is Stro
* Unemployed Workers from Construction and Manufacturing Have
* Certain Occupations Are Associated with UI Benefit Receipt a
* Concluding Observations
* Agency Comments
* Overview
* Data Used
* Econometric Model
* Results
* Industry-Interaction Specification
* UI Receipt Equation
* Unemployment Duration Equation
* Occupation-Interaction Specification
* Limitations of the Analysis
* Order by Mail or Phone
Report to the Chairman, Subcommittee on Human Resources, Committee on Ways
and Means, House of Representatives
United States Government Accountability Office
GAO
March 2006
UNEMPLOYMENT INSURANCE
Factors Associated with Benefit Receipt
GAO-06-341
Contents
Letter 1
Results in Brief 4
Background 5
Certain Characteristics Are Associated with UI Benefit Receipt and
Unemployment Duration 8
Certain Industries Are Associated with UI Benefit Receipt and Unemployment
Duration 22
Concluding Observations 31
Agency Comments 32
Appendix I Analysis of UI Benefit Receipt and Unemployment Duration 34
Overview 34
Data Used 34
Econometric Model 39
Results 43
Limitations of the Analysis 80
Appendix II Comments from the Department of Labor 82
Appendix III GAO Contact and Staff Acknowledgment 83
Bibliography 84
Related GAO Products 86
Tables
Table 1: Simulated Unemployment Duration for UI-Eligible Workers by
Current UI Receipt Status and Other Characteristics 21
Table 2: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers from Different Industries 23
Table 3: Simulated Likelihood of Receiving UI Benefits during Different
Periods of UI-Eligible Unemployment for Workers with Past UI Receipt, by
Industry 25
Table 4: Simulated Unemployment Duration for UI-Eligible Workers, by
Industry and UI Receipt Status 28
Table 5: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers from Different Occupations 29
Table 6: Simulated Likelihood of Receiving UI Benefits During Different
Periods of UI-Eligible Unemployment for Workers with Past UI Receipt, by
Occupation 30
Table 7: Simulated Unemployment Duration for UI-Eligible Workers, by
Occupation and UI Receipt Status 31
Table 8: Parameter Estimates for UI Receipt Equation from
Industry-Interaction Specification 45
Table 9: Parameter Estimates for Duration Equation from
Industry-Interaction Specification 48
Table 10: Parameter Estimates for UI Receipt Equation from
Occupation-Interaction Specification 52
Table 11: Parameter Estimates for Duration Equation from
Occupation-Interaction Specification 55
Table 12: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers during Successive Periods of Unemployment, by Past UI Receipt
Status 60
Table 13: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers from Different Industries 61
Table 14: Simulated Likelihood of Receiving UI Benefits during Different
Periods of UI-Eligible Unemployment for Workers with Past UI Receipt, by
Industry 63
Table15: Simulated Unemployment Duration for UI-Eligible Workers, by
Industry and UI Receipt Status 70
Table 16: Simulated Unemployment Duration for UI-Eligible Workers, by
Education Level and UI Receipt Status 72
Table 17: Simulated Unemployment Duration for UI-Eligible Workers, by
Race/Ethnicity and UI Receipt Status 73
Table 18: Simulated Unemployment Duration for UI-Eligible Workers, by
Union Status and UI Receipt Status 73
Table 19: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers from Different Occupations 76
Table 20: Simulated Likelihood of Receiving UI Benefits during Different
Periods of UI-Eligible Unemployment for Workers with Past UI Receipt, by
Occupation 78
Table 21: Simulated Unemployment Duration for UI-Eligible Workers, by
Occupation and UI Receipt Status 79
Figures
Figure 1: Incidence of UI Benefit Receipt from 1979 through 2002, for
Workers Born between 1957 and 1964 7
Figure 2: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Prior-Year Earnings 9
Figure 3: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Age 10
Figure 4: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Education Level 12
Figure 5: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers during Successive Periods of Unemployment, by Past UI Receipt
Status 15
Figure 6: Simulated Unemployment Duration for UI-Eligible Workers, by
Prior-Year Earnings and UI Receipt Status 18
Figure 7: Simulated Unemployment Duration for UI-Eligible Workers, by
Education Level and UI Receipt Status 20
Figure 8: Distribution of All Periods of UI Benefit Receipt across
Industries 24
Figure 9: Simulated Effect of Past UI Benefit Receipt on the Likelihood of
Receiving UI in Subsequent Periods of Unemployment, for Selected
Industries 26
Figure 10: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Age 65
Figure 11: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Prior-Year Earnings 67
Figure 12: Simulated Unemployment Duration for UI-Eligible Workers, by
Prior-Year Earnings and UI Receipt Status 75
Abbreviations
BLS Bureau of Labor Statistics
BPE base period earnings
CPI-U Consumer Price Index for All Urban Consumers
CPS Current Population Survey
HQE high quarter earnings
NLSY79 National Longitudinal Survey of Youth 1979
OLF out of the labor force
SIC Standard Industrial Classification
SMSA Standard Metropolitan Statistical Area
SOC Standard Occupational Classification
UI Unemployment Insurance
This is a work of the U.S. government and is not subject to copyright
protection in the United States. It may be reproduced and distributed in
its entirety without further permission from GAO. However, because this
work may contain copyrighted images or other material, permission from the
copyright holder may be necessary if you wish to reproduce this material
separately.
WBA weekly benefit amount
United States Government Accountability Office
Washington, DC 20548
March 7, 2006
The Honorable Wally Herger Chairman, Subcommittee on Human Resources
Committee on Ways and Means
House of Representatives
Dear Mr. Chairman:
Unemployment Insurance (UI), established in 1935, is a complex system of
53 programs that provide temporary cash benefits to eligible workers who
become unemployed through no fault of their own.1 Eligibility for UI
benefits, benefit amounts, and the length of time benefits are available
are determined by state law, within broad federal guidelines. Benefits are
financed through federal and state employer payroll taxes. In fiscal year
2004, employers paid about $39.3 billion in UI taxes, and 8.8 million
workers received UI benefits totaling $41.3 billion.
Decades of program experience and administrative data have resulted in a
firm understanding of the composition of UI caseloads and the overall cost
of the program. However, this understanding of the UI program has been
based on snapshots of the UI beneficiary population at any given time.
Additional research has provided limited information on the types of
workers who are likely to receive UI benefits and on how UI requirements
and benefits affect individuals' movement into and out of the workforce,
including how UI receipt affects the duration of unemployment. However,
because of the difficulty of tracking the same workers over time, the
circumstances that give rise to individual workers' use or nonuse of the
UI program and how this may, in turn, affect individuals' patterns of
unemployment over the course of their entire working careers are still not
well understood.
In 2005, we reported on the results of our analysis of a unique database
that tracked a single group of individuals over time.2 Examining this
database, we found that 85 percent of a nationally representative sample
of late baby boom workers (workers born between 1957 and 1964) had
experienced unemployment at least once between 1979 and 2002. Workers who
experienced unemployment were unemployed an average of five times over
this 23-year period. Moreover, we found that of those who were eligible
for UI benefits at least once, only 38 percent at some point received UI
benefits. About half of the workers receiving UI benefits received them
more than once. Finally, we reported that the rate at which unemployed
workers received UI benefits varied across industries.
1UI programs are administered by the 50 states, the District of Columbia,
Puerto Rico, and the Virgin Islands.
2GAO, Unemployment Insurance: Information on Benefit Receipt, GAO-05-291
(Washington, D.C.: Mar. 17, 2005).
As Congress reviews the ability of labor programs to meet the needs of the
workforce in the new century, it will be important to understand why fewer
than half of workers eligible for UI benefits receive them and the other
half do not, as well as what factors cause workers in some industries to
seek benefits multiple times over the course of their careers. In this
context, you asked us to determine (1) the extent to which characteristics
of individual workers, including a history of past UI benefit receipt, are
associated with the likelihood of UI benefit receipt and unemployment
duration, and (2) whether an unemployed worker's industry is associated
with UI benefit receipt or unemployment duration.
To answer these questions, we analyzed data from the National Longitudinal
Survey of Youth 1979 (NLSY79). This survey provides information that is
not typically available from other data sources. The dataset contains
information from ongoing periodic interviews with a nationally
representative sample of individuals who were born between 1957 and 1964.
At the time of our analysis, the database contained information from
interviews conducted between 1979 and 2002. There were 12,686 individuals
in the sample in 1979. The survey provides a wide range of detailed
information about these individuals, including their work histories,
income, family composition, and education. Using the dataset, we analyzed
a single birth cohort over time; therefore, our findings do not represent
the experience of workers of all ages during this time period.
Using this survey information and information on states' UI program
eligibility rules for each year from 1978 through 2002,3 we estimated
whether individuals from the sample were eligible for UI benefits
following a job separation. We identified 5,631 workers who met the
conditions for UI eligibility-a group that we refer to as "UI-eligible
workers"-who collectively experienced 15,506 separate periods of
unemployment during the study period (1979-2002).
3We considered an individual to be UI-eligible if that individual
experienced an involuntary job loss, reported receiving a minimum amount
of wages over a minimum period of time as defined by the state where the
individual lived, and was actively looking for new employment. Our method
of estimating eligibility tends to overestimate the number of UI-eligible
individuals. For a more complete discussion of our methodology, see
appendix I.
We used a multivariate statistical model to identify the key factors
associated with UI benefit receipt and unemployment duration for our
subsample of UI-eligible workers. The model allowed us to isolate the
effect of a particular characteristic by statistically controlling for a
number of other characteristics. In this report, we refer to the results
for individual characteristics in comparison with "otherwise similar
workers." By this phrasing, we intend to show that we have controlled for
all other characteristics that may be related to the characteristic being
studied. For example, the test of the effect of age on benefit receipt was
conducted while controlling, for example, for earnings and education-two
characteristics that are correlated with age. In addition, we modeled UI
benefit receipt and unemployment duration together to control for the
likely correlation that exists between these two outcomes.
To illustrate how changes in different characteristics affect the
likelihood of UI receipt and unemployment durations, we used the results
of our multivariate statistical model to simulate how changes in
observable characteristics affect the likelihood of UI receipt and
unemployment duration. The simulated results are calculated from our
statistical model estimates, holding selected characteristics constant, as
noted throughout the report. For example, to understand how changes in
workers' education affect their likelihood of receiving UI benefits, we
set the number of years of education at the same value for all workers in
our sample and then used the model estimates to simulate the likelihood of
UI receipt for each worker. We then calculated the average likelihood of
receiving UI benefits. We repeated this process for different years of
education. Unless otherwise noted, simulated likelihoods of UI receipt and
simulated unemployment duration are for workers experiencing unemployment
for the first time. See appendix I for a more complete discussion of our
methodology, including limitations of our analysis.
We assessed the reliability of the NLSY79 dataset and found it to be
sufficient for our analysis. Our work was conducted from May 2005 through
February 2006 in accordance with generally accepted government auditing
standards.
Results in Brief
Certain characteristics are associated with the likelihood of receiving UI
benefits and unemployment duration. Based on our analysis of workers
during the first half of their working lives, UI-eligible workers are more
likely than other workers to receive UI benefits if they have higher
earnings, are younger or have more years of education, or, most notably,
if they received UI benefits in the past. In particular, UI-eligible
workers who received UI benefits before are more likely than other workers
to receive UI benefits again and this likelihood increases each time they
are unemployed and receive UI. Other factors, including a high local
unemployment rate, increase the likelihood of receiving UI. UI-eligible
workers who receive UI benefits have longer periods of unemployment than
workers who do not receive benefits. Similarly, workers who have fewer
years of education, lower earnings, or no union membership experience
longer unemployment than workers who do not have these characteristics.
Workers who received UI benefits in the past, however, were unemployed
about as long as similar workers who had not received UI in the past.
UI-eligible workers from certain industries are more likely than other
workers to receive UI benefits and experience shorter unemployment
duration, although no clear industry trend emerged. Specifically, our
simulations show that
o The likelihood of receiving UI benefits during a first period
of unemployment is highest among workers from mining and
manufacturing. Furthermore, the likelihood of receiving UI
benefits when unemployed increases with each previous period of UI
receipt across all industries, and the most notable increase
occurs for workers from the public administration sector.
o The unemployment duration for first-time unemployed workers
from construction and manufacturing is significantly shorter than
the unemployment duration experienced by workers from other
industries. While unemployment duration varies across all
industries, this variation is not affected by whether workers were
unemployed in the past, or whether they received UI in the past.
In its comments, the Department of Labor stated that, while there are
certain qualifications of our findings, Labor applauds our efforts and
said that this report adds to our current knowledge of the UI program.
Labor also provided technical comments, which we incorporated where
appropriate.
Background
The UI program was established in 1935 and serves two primary objectives:
(1) to temporarily replace a portion of earnings for workers who become
unemployed through no fault of their own and (2) to help stabilize the
economy during recessions by providing an infusion of consumer dollars
into the economy. UI is made up of 53 state-administered programs that are
subject to broad federal guidelines and oversight. In fiscal year 2004,
these programs covered about 129 million wage and salary workers and paid
benefits totaling $41.3 billion to about 8.8 million workers.
Federal law provides minimum guidelines for state programs and authorizes
grants to states for program administration. States design their own
programs, within the guidelines of federal law, and determine key elements
of these programs, including who is eligible to receive state UI benefits,
how much they receive, and the amount of taxes that employers must pay to
help provide these benefits. State unemployment tax revenues are held in
trust by the Department of Labor (Labor) and are used by the states to pay
for regular weekly UI benefits, which typically can be received for up to
26 weeks. During periods of high unemployment, the Extended Benefits
program, funded jointly by states through their UI trust funds and by the
federal government through the Unemployment Trust Fund, provides up to 13
additional weeks of benefits for those who qualify under state program
rules. Additional benefits, funded by the federal government, may be
available to eligible workers affected by a declared major disaster or
during other times authorized by Congress.
To receive UI benefits, an unemployed worker must file a claim and satisfy
the eligibility requirements of the state in which the worker's wages were
paid. Although states' UI eligibility requirements vary, generally they
can be classified as monetary and nonmonetary. Monetary eligibility
requirements include having a minimum amount of wages and employment over
a defined base period, typically, about a year before becoming unemployed,
and not having already exhausted the maximum amount of benefits or benefit
weeks to which they would be entitled because of other recent
unemployment. In addition to meeting states' monetary eligibility
requirements, workers must satisfy their states' nonmonetary eligibility
requirements. Nonmonetary eligibility requirements include being able to
work, being available for work, and becoming unemployed for reasons other
than quitting a job or being fired for work-related misconduct. In all
states, claimants who are determined to be ineligible for benefits are
entitled to an explanation for the denial of benefits and an opportunity
to appeal the determination.
Since UI was introduced, researchers and those responsible for overseeing
the program have monitored the size, cost, and structure of the program
and its effects on individuals' movement into and out of the workforce,
including which types of workers receive UI benefits. Much of what is
known about the dynamics of the UI program has been based on snapshots of
the UI beneficiary population at any given time. Labor regularly gathers
UI program data from the states, including each state's eligibility
requirements, employers' UI tax rates, program revenues and costs, and
numbers of claims received and approved. An extensive amount of research
has been devoted to the effect of UI benefit receipt on unemployment
duration. Specifically, researchers have found that receiving UI benefits
increases unemployment duration. Much of this research is focused on
measuring how changes in the amount of UI benefits increase the amount of
time that an unemployed worker takes to find a new job.4 Although much is
known about UI caseloads and about the relationship between UI benefits
and unemployment duration, less is known about the patterns of UI receipt
among individual workers over their entire working careers.
What is known about the patterns of UI benefit receipt over an extended
period for individual workers comes from a few studies that are fairly
limited in scope. In one study, researchers analyzed 1980-1982 survey data
and found that among unemployed workers who were eligible for UI, younger
or female workers were less likely to receive UI benefits, while union
workers, workers from large families, or those with more hours of work
from their previous jobs were more likely to receive UI.5 In another
study, using UI administrative data from five states, researchers found
that between 36 and 44 percent of UI claims from 1979 to 1984 were from
workers who had received UI benefits more than once and that middle-aged
workers and workers with higher earnings were more likely to be repeat UI
recipients.6 Another study, based on survey data from the NLSY79, found
that 16 percent of young adults had received UI benefits more than once
between 1978 and 1991 and that as many as 46 percent of those who received
UI were repeat recipients.7 This study also found that workers who were
women or Hispanic or whose fathers had more years of education were less
likely to become repeat recipients than workers who were men or
non-Hispanic or whose fathers had fewer years of education.
4Alan B. Krueger and Bruce D. Meyer, "Labor Supply Effects of Social
Insurance," NBER Working Paper 9014 (Cambridge, Massachusetts: National
Bureau of Economic Research, 2002).
5Rebecca M. Blank and David E. Card, "Recent Trends in Insured and
Uninsured Unemployment: Is There an Explanation?" The Quarterly Journal of
Economics, vol. 106, no. 4 (1991).
6Bruce D. Meyer and Dan T. Rosenbaum, "Repeat Use of Unemployment
Insurance," NBER Working Paper 5423 (Cambridge, Massachusetts: National
Bureau of Economic Research, 1996), p. 20.
In 2005, we analyzed the NLSY79 to determine the extent to which
individual workers received UI benefits during their early working lives.8
We found that 38 percent of workers born between 1957 and 1964 received UI
at least once between 1979 and 2002, with almost half of these individuals
receiving UI benefits more than once. (See fig. 1.) We also found that the
rate at which unemployed workers received UI benefits varied across
industries, but we did not control for any of the other factors that may
have helped to explain this variation.
Figure 1: Incidence of UI Benefit Receipt from 1979 through 2002, for
Workers Born between 1957 and 1964
Note: Sampling errors were within plus or minus 5 percentage points at the
95 percent confidence level.
7See Brian P. McCall, "Repeat Use of Unemployment Insurance," in Laurie J.
Bassi and Stephen A. Woodbury, editors, Long-Term Unemployment and
Reemployment Policies (Stamford, Connecticut: JAI Press, Inc., 2000).
8 GAO-05-291 .
Certain Characteristics Are Associated with UI Benefit Receipt and Unemployment
Duration
Earnings, age, education, and most notably past UI benefit receipt are all
associated with the likelihood of receiving UI benefits for UI-eligible
workers. Education, earnings, and union membership, and current UI benefit
receipt, are associated with unemployment duration.
Unemployed Workers with Higher Earnings, Younger Workers, Workers with More
Education, or Those Who Received UI in the Past Are More Likely to Receive UI
Benefits
Unemployed workers are more likely to receive UI benefits if they have
higher earnings prior to becoming unemployed, are younger or have more
years of education, or have a history of past UI benefit receipt, when
compared to workers with similar characteristics.9 We found that past
experience with the UI program has a particularly strong effect on the
future likelihood of receiving UI benefits. In addition, UI-eligible
workers are more likely to receive UI when the local unemployment rate is
high. However, some characteristics, such as the weekly UI benefit amount
that a worker is eligible to receive, are not associated with a greater
likelihood of receiving UI benefits.
Unemployed Workers Who Have Higher Earnings or Are Younger or Have More Years
of Education Are More Likely to Receive UI
Unemployed workers who have higher earnings or are younger or who are more
educated are more likely to receive UI benefits than otherwise similar
workers. With respect to earnings,10 our simulations show that the
likelihood of receiving UI tends to increase as the amount earned in the
year prior to becoming unemployed increases (see fig. 2). For example, a
UI-eligible worker with earnings between $10,000 and $11,999 in the year
before becoming unemployed has a 36 percent likelihood of receiving UI,
whereas a worker who earned roughly twice as much (between $20,000 and
$24,999) has a 45 percent likelihood of receiving UI.11 The likelihood of
receiving UI is lowest among workers with the lowest earnings (i.e., less
than $10,000 in the year before becoming unemployed). There is generally
little difference in the likelihood of receiving UI among workers earning
$18,000 or more.
9The results described in this report are statistically significant at the
95 percent confidence level, unless otherwise noted. For a complete list
of findings from our multivariate statistical model of the key factors
associated with UI benefit receipt, see table 8 in appendix I.
10Earnings refers to base period earnings, which we define as the amount
of earnings received during the first four of the last five full calendar
quarters before a worker becomes unemployed. This definition is consistent
with the time frame states generally use to determine eligibility.
11The average and maximum earnings for the unemployed workers in our
sample are $15,524 and $597,950, respectively.
Figure 2: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Prior-Year Earnings
Note: Simulations are for the average likelihood of receiving UI during
first-time unemployment at different levels of earnings. The overall
average likelihood of receiving UI during first-time unemployment is 33
percent. See appendix I for methodology and estimation results.
This result confirms our 2000 finding that low-wage workers are less
likely to receive UI benefits than workers with higher earnings even when
they have worked for the same amount of time.12 Our current result also
controls for other worker differences, such as which industries the
workers were employed in or whether they were ineligible for benefits,
which we had not previously been able to rule out as explanations for the
variation in likelihood of receiving UI. The relationship between higher
earnings and a higher likelihood of receiving UI benefits is also
consistent with economic theory that predicts that workers with higher
earnings prior to becoming unemployed will be more reluctant to accept
lower reemployment wages and are therefore more likely to take advantage
of UI benefits as a way to subsidize their job search efforts.13
12GAO, Unemployment Insurance: Role as Safety Net for Low-Wage Workers Is
Limited, GAO-01-181 (Washington, D.C.: Dec. 29, 2000).
Concerning age, our simulations show that the likelihood of receiving UI
peaks at about age 25 and decreases thereafter (see fig. 3). More
specifically, a 25-year-old unemployed worker who is eligible for UI is
more than twice as likely to receive UI as an otherwise similar
40-year-old unemployed worker.
Figure 3: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Age
Note: Simulations are the average likelihood of receiving UI during
first-time unemployment at different ages. The overall average likelihood
of receiving UI during first-time unemployment is 33 percent. See appendix
I for methodology and estimation results.
13For economic theory concerning the relationship between job search and
unemployment insurance, see Dale T. Mortensen, "Unemployment Insurance and
Job Search Decisions," Industrial and Labor Relations Review, vol. 30, no.
4 (1977).
Previous studies have found that younger workers are less likely to
receive UI benefits than older workers.14 However, these previous studies
did not include as much information about workers' past unemployment and
UI benefit receipt histories as our current analysis. Therefore, because
older workers have more of this experience than younger workers, it is
possible that our analysis has controlled for the effect of this past
experience more completely than these previous studies, resulting in a
more precise estimate of the effect of age. We are unable to explain why
younger workers are more likely to receive UI benefits than otherwise
similar older workers. However, it is possible that older workers, who
have had more time to accumulate financial assets, may have more private
resources available to help them cope with unemployment than younger
workers.15 Alternatively, younger workers may be less optimistic about how
long it will take for them to become reemployed.
Unemployed workers with more years of education are more likely to receive
UI benefits than otherwise similar workers with fewer years of education.
Specifically, our simulations show that the likelihood of receiving UI
increases for each additional year of schooling that a UI-eligible worker
has completed before becoming unemployed (see fig. 4). For example, a
UI-eligible worker with a college education (one who has completed 16
years of schooling) when he or she becomes unemployed is almost one-fifth
more likely to receive UI than a UI-eligible worker with a high school
education (12 years of schooling).16
14See Blank and Card, and McCall.
15See Jonathan Gruber, "The Wealth of the Unemployed," October 2001,
Industrial and Labor Relations Review, vol. 55, no. 1.
16The average number of years of schooling completed by UI-eligible
workers, at the time when they became unemployed, is 12 years.
Figure 4: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Education Level
Note: Simulations are the average likelihood of receiving UI during
first-time unemployment at different education levels. The overall average
likelihood of receiving UI during first-time unemployment is 33 percent.
See appendix I for methodology and estimation results.
Although the impact of education on the likelihood of receiving UI
benefits has been analyzed in other research, this research found no
significant education effect.17 However, to the extent that workers with
more years of education are better able to access and understand UI
program rules, they may also be more likely to know when they are entitled
to benefits and to have the information that they need to file successful
benefit claims.
Other factors, including gender, marital status, job tenure, and the local
unemployment rate are also associated with UI benefit receipt. Controlling
for all other characteristics among this UI-eligible group,
o a woman is 29 percent more likely to receive UI benefits than a
man,
o a married worker is 13 percent more likely to receive UI than
an unmarried worker,
o a longer tenured worker is more likely to receive UI-for
example, a worker with 4 years of tenure at his or her most recent
job is 12 percent more likely to receive UI than a worker with 1
year of job tenure, and
o being in an area with high unemployment raises the likelihood
that an unemployed worker will receive UI-for example, a worker
living in an area with an unemployment rate of 9 percent is 10
percent more likely to receive UI than a worker living in an area
with an unemployment rate of 5 percent.
17See Blank and Card, p. 1185.
Our finding that women are more likely to receive UI benefits than
otherwise similar men differs from the results of previous research, which
generally found no statistically significant differences. Nevertheless,
our analysis controls for more worker characteristics than these previous
studies, and it is likely that we have more carefully isolated the effect
of gender from that of other characteristics related to gender, such as
workers' occupations or industries. It is not immediately clear why women
are more likely to receive UI benefits, however. We are likewise unable to
explain why married workers are more likely to receive UI benefits than
otherwise similar unmarried workers.18
Our findings on job tenure are consistent with previous research. However,
the higher likelihood of UI benefit receipt associated with more years of
job tenure is likewise difficult to explain. It might be that workers with
longer job tenures have acquired more skills that are not as easy to
transfer to another employer, relative to workers with less job tenure,
and anticipate longer job searches.
The higher likelihood of receiving UI benefits among workers living in
areas with higher unemployment is likely due to the higher number of
unemployed workers relative to available jobs, which may make workers more
willing to apply for UI benefits as they engage in what are likely to be
longer job searches.
In contrast to our findings above, a key UI program element, the weekly UI
benefit amount that UI-eligible workers are entitled to, is not associated
with a greater likelihood of receiving UI benefits. Using our model
estimates, we simulated increases in weekly UI benefit amounts of 10
percent and 25 percent and a decrease of 10 percent and found that these
changes had no effect on the likelihood of UI benefit receipt. This
finding is consistent with the work of others, who have found that
increases in the weekly benefit amount have mixed, but generally small
effects on UI benefit receipt.19 Collectively, these results suggest that
UI benefit levels have modest effects on individuals' decisions about
whether or not to receive UI benefits, after controlling for other
factors.
18We specifically tested for the effect of spousal income on the
likelihood of receiving UI to determine whether marital status was masking
some underlying effect of additional family income, and found this not to
be the case.
Unemployed Workers Who Received UI in the Past Are More Likely to Receive UI
during Subsequent Unemployment
Unemployed workers who have received UI benefits during a prior period of
unemployment are more likely to receive UI benefits during a current
period of unemployment than otherwise similar workers who never received
UI benefits (see fig. 5). For example, when workers experience their first
UI-eligible period of unemployment, their likelihood of receiving UI is 33
percent. During a second UI-eligible period of unemployment, the
likelihood of receiving UI is 48 percent for workers who received UI
during the first unemployment period but only 30 percent for workers who
did not receive UI. Furthermore, the likelihood that these UI-eligible
workers will receive UI benefits during successive periods of unemployment
increases each time that they receive UI benefits and decreases each time
that they do not.20
19See David E. Card and Phillip B. Levine, "Unemployment Insurance Taxes
and the Cyclical and Seasonal Properties of Unemployment," Journal of
Public Economics, vol. 53, no. 1 (1994); Patricia M. Anderson and Bruce D.
Meyer, "The Effect of Unemployment Insurance Taxes and Benefits on Layoffs
Using Firm and Individual Data," NBER Working Paper No. 4960, December
1994; and Robert H. Topel, "On Layoffs and Unemployment Insurance,"
American Economic Review, vol. 73, no. 4 (1983).
20As noted above, relatively few UI-eligible workers who receive UI
benefits receive them multiple times. See GAO-05-291 for a more complete
discussion of the incidence of repeat UI benefit receipt.
Figure 5: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers during Successive Periods of Unemployment, by Past UI Receipt
Status
Note: Simulations are the average likelihood of receiving UI during a
current unemployment period for two extreme cases: (1) workers who always
received UI benefits during previous unemployment and (2) workers who
never received UI during previous unemployment. The average likelihood of
receiving UI during first-time unemployment for all UI-eligible workers is
33 percent. See appendix I for methodology and estimation results.
This finding suggests that a worker's first unemployment experience has a
lasting and self-reinforcing effect. To the extent that workers know about
the UI program and whether or not they are eligible, receiving or not
receiving UI benefits may be a personal choice based on unobserved worker
characteristics or preferences. Alternatively, if workers do not have good
information about UI, those who receive UI benefits may know more about
the UI program than those who do not receive UI, and their knowledge about
the program could make it easier to apply for and receive benefits during
a subsequent period of unemployment.
Receiving UI Benefits, along with Other Factors, Is Associated with Unemployment
Duration
Overall, unemployed workers who receive UI benefits have longer
unemployment duration than otherwise similar workers who do not receive UI
benefits.21 Several other characteristics are also associated with
unemployment duration. Specifically, UI-eligible workers are more likely
to experience longer unemployment duration if they have lower earnings
before becoming unemployed or have fewer years of education. Other
characteristics associated with longer unemployment duration, after
controlling for other factors, include being African-American or female or
not belonging to a union. We found no relationship between past UI benefit
receipt and subsequent unemployment duration.
Receiving UI Benefits Is Associated with Longer Unemployment Duration
Whether or not an unemployed worker receives UI during a specific period
of unemployment has the strongest effect on how long that period of
unemployment is likely to last. Overall, UI-eligible workers who receive
UI benefits during a period of unemployment remain unemployed for about 21
weeks on average, whereas otherwise similar workers who do not receive UI
remain unemployed for about 8 weeks. This result is consistent with
economic theory that predicts that receiving UI benefits reduces the costs
associated with unemployment and allows workers to engage in longer job
searches.22 That is, an unemployed worker who receives UI benefits faces
less pressure to accept the first job offer they receive and can search
longer for a more desirable job than an unemployed worker who does not
receive UI. Another possible explanation for the strong association
between UI receipt and longer unemployment duration may be that workers
who expect to experience longer unemployment may be more likely to apply
for UI than those who expect to return to work quickly.
Unemployed Workers with Lower Earnings and Less Education Tend to Have Longer
Unemployment Duration
Unemployed workers with lower earnings tend to have longer unemployment
duration than otherwise similar workers with higher earnings. This finding
holds for workers who are receiving UI benefits, and for workers who are
not receiving UI benefits. Specifically, our simulations show that
UI-eligible workers who receive UI benefits and have relatively high
earnings ($30,000 and higher) in the year prior to becoming unemployed
have unemployment duration that is as much as 9 weeks shorter than workers
with earnings that are below $16,000.23 The results are similar for
UI-eligible workers who do not receive UI benefits (see fig. 6).
21For the parameter estimates of these and other variables included in our
multivariate statistical model of the key factors associated with
unemployment duration, see table 9 in appendix I. The variables reported
here are those that were statistically significant at the 95 percent
confidence level.
22See Mortensen.
23The average prior-year earnings amount for this sample is $15,524.
Figure 6: Simulated Unemployment Duration for UI-Eligible Workers, by
Prior-Year Earnings and UI Receipt Status
Note: Simulations are the median duration of unemployment during
first-time unemployment. Overall average duration is 21 weeks for
UI-eligible workers receiving UI benefits and 8 weeks for UI-eligible
workers not receiving UI benefits. See appendix I for methodology and
estimation results.
Our result is consistent with other research that has found that higher
previous earnings tend to reduce unemployment duration.24 Researchers have
suggested that the association between higher earnings and shorter
unemployment duration may be due, in part, to the higher cost of
unemployment for workers with higher earnings, relative to the cost for
workers with lower earnings.25 Specifically, the cost of unemployment in
terms of lost wages is greater for workers with higher earnings, because
they forego a higher amount of potential earnings in exchange for the time
they spend on unpaid activities, such as job search, home improvement, or
recreation.
24See Karen E. Needels and Walter Nicholson, An Analysis of Unemployment
Durations Since the 1990-1992 Recession, UI Occasional Paper 99-6,
prepared for the Department of Labor, 1999, p. 94.
25See Bruce D. Meyer, "Unemployment Insurance and Unemployment Spells,"
Econometrica, vol. 58, no. 4 (1990), p. 771.
Our model estimates also indicate that unemployed workers who have more
education tend to have shorter unemployment duration than otherwise
similar workers with less education. For example, simulations show that on
average, UI-eligible workers with a 4-year college education (16 years of
schooling) who receive UI benefits remain unemployed about 2 weeks less
than workers with a high school education (12 years of schooling).26 (See
fig. 7.) The results are similar for UI-eligible workers who do not
receive UI benefits. This finding is consistent with past research
indicating that less education is associated with longer unemployment
duration, because workers with less education have fewer work-related
skills.27
26The average number of years of schooling completed by UI-eligible
workers, at the time when they became unemployed, is 12 years.
27Needels and Nicholson, p. 6.
Figure 7: Simulated Unemployment Duration for UI-Eligible Workers, by
Education Level and UI Receipt Status
Note: Simulations are the median duration of unemployment during
first-time unemployment. Overall average duration is 21 weeks for
UI-eligible workers receiving UI benefits and 8 weeks for UI-eligible
workers not receiving UI benefits. See appendix I for methodology and
estimation results.
Unemployed workers' race or ethnicity, gender, union membership status,
and length of most recent job tenure are also associated with unemployment
duration. Specifically, simulations show that UI-eligible workers who are
African-American or women, who do not belong to labor unions, or who have
less years of job tenure before becoming unemployed tend to have longer
unemployment duration than otherwise similar workers. As seen in table 1,
these associations exist whether or not workers receive UI benefits.
Table 1: Simulated Unemployment Duration for UI-Eligible Workers by
Current UI Receipt Status and Other Characteristics
Unemployment duration (median weeks)
Worker characteristics Receiving UI benefits Not receivingUI benefits
Race or ethnicity
White 19 8
Hispanic 21 8
African-American 25 11
Gender
Male 20 8
Female 22 9
Union membership status
Union member 19 8
Not a union member 21 9
Tenure at most recent joba
10 years 20 8
1 year 21 8
Overall average duration 21 8
Source: Simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the median duration of unemployment during
first-time unemployment. See appendix I for methodology and estimation
results.
aSimulated decreases in median weeks of unemployment are less than 1 week
per additional year of tenure at most recent job, regardless of whether
workers received UI or not.
Our findings are generally consistent with prior research. In particular,
longer unemployment durations have been found to be associated with being
African-American, female, or not belonging to a union.28 Two possible
explanations for the differences in employment outcomes for
African-American workers include labor market discrimination, and limited
access to social networks that may enable these workers to find jobs more
quickly.29 Likewise, longer unemployment duration among female workers may
be due to labor market discrimination, or to differences in how they value
paid work versus nonemployment activities, relative to men.30 Likewise,
the associations between shorter unemployment duration and union
membership or longer job tenure may reflect the greater access of these
workers to reemployment opportunities than otherwise similar workers or
because of a greater likelihood of being recalled to their previous
jobs.31
28See Needels and Nicholson.
29See Antoni Calvo-Armengol, and Matthew O. Jackson, "The Effects of
Social Networks on Employment and Inequality," The American Economic
Review, Vol. 94, No. 3, (2004) for a discussion of the effects of
individuals' social networks on employment outcomes.
Unemployment Duration Is Not Associated with Past UI Receipt
Past UI receipt has no significant effect on subsequent unemployment
duration. Although receiving UI during a current period of unemployment is
associated with longer unemployment duration, past UI receipt does not
affect current unemployment duration. Specifically, simulations show that
unemployment duration tends to decrease by about the same amount
(typically, 1 week or less) from one unemployment period to the next,
regardless of whether a worker received UI benefits in the past or not,
and regardless of whether or not the worker receives UI benefits in the
current period.
Certain Industries Are Associated with UI Benefit Receipt and Unemployment
Duration
Unemployed workers in certain industries are more likely to receive UI
benefits and experience shorter unemployment duration than otherwise
similar workers from other industries. Simulations show that first-time
unemployed workers from mining and manufacturing are more likely to
receive UI than workers from other industries. Moreover, the strength of
the association between past and current UI benefit receipt varies across
industries. The increase in the likelihood of receiving UI from one
unemployment period to the next is highest for public administration and
is lowest for agriculture and construction. Furthermore, simulations
indicate that UI-eligible workers from industries with higher proportions
of unemployment periods that result in UI receipt are no more likely to
become repeat UI recipients than workers from other industries. With
respect to unemployment duration, UI-eligible workers from construction
and manufacturing have shorter unemployment duration than workers from
other industries.
30See Needels and Nicholson, and GAO, Women's Earnings: Work Patterns
Partially Explain Differences between Men's and Women's Earnings,
GAO-04-35 (Washington, D.C.: Oct. 31, 2003).
31See Needels and Nicholson. We did not control for the likely effect of
an expected job recall.
Unemployed Workers from Mining and Manufacturing Are More Likely to Receive UI
Benefits
Unemployed workers from mining and manufacturing are more likely to
receive UI than otherwise similar workers from other industries. For
example, first-time unemployed workers from the manufacturing industry are
about two-thirds more likely to receive UI benefits than workers from the
professional and related services industry (see table 2). Although
UI-eligible workers from mining are more likely to receive UI benefits
than workers from other industries, just 2 percent of the unemployment
periods that result in UI benefit receipt come from the mining industry.
(See fig. 8.)32
Table 2: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers from Different Industries
Simulated likelihood of receiving UI
Industry benefits (percent)
Mining 46
Manufacturing 40
Public administration 37
Wholesale and retail trade 35
Agriculture, forestry, and fishing 34
Business services 31
Construction 31
Finance, insurance, and real estate 31
Transportation and public utilities 29
Entertainment and recreation 26
services
Professional and related services 24
Personal services 23
All industries 33
Source: Simulations based upon GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI during
first-time unemployment for workers from different industries. The
parameter estimates for the mining, manufacturing, public administration,
wholesale and retail trade, agriculture, forestry, and fishing, business
services, and construction industries are statistically significant
relative to the professional and related services industry at the 95
percent confidence level. See appendix I for methodology and estimation
results.
32The percentages in table 2 and figure 8 are not comparable. The
percentages in table 2 represent an individual worker's likelihood of
receiving UI when UI-eligible unemployment occurs, whereas the percentages
in figure 8 compare the relative proportions of unemployment spells with
UI benefit receipt coming from different industries.
Figure 8: Distribution of All Periods of UI Benefit Receipt across
Industries
Note: Total does not equal 100 percent due to rounding.
The Relationship between Past and Current UI Receipt Is Strongest for Public
Administration
Unemployed workers who have received UI benefits in the past are more
likely to receive UI benefits during a current period of unemployment than
otherwise similar workers who never received UI benefits, across each
industry (see table 3). However, the increase in the likelihood of
receiving UI benefits associated with past UI benefit receipt is not the
same across all industries. Specifically, this effect is strongest for
workers from public administration and is weakest for workers from
agriculture and construction.33
33Although the association between past UI receipt and current UI receipt
is statistically significant for all industries combined, differences in
this association among industries were statistically significant only for
public administration, agriculture, and construction.
Table 3: Simulated Likelihood of Receiving UI Benefits during Different
Periods of UI-Eligible Unemployment for Workers with Past UI Receipt, by
Industry
Simulated likelihood of receiving UI benefits during
current UI-eligible unemployment period, given past UI
receipt (percent)
Second
Firstunemployment unemployment Thirdunemployment
Industry perioda period period
Mining 46 57 69
Manufacturing 40 52 65
Public 37 68 91
administration
Wholesale and 35 52 70
retail trade
Agriculture, 34 42 50
forestry, and
fishing
Business services 31 48 66
Construction 31 40 51
Finance, 31 64 91
insurance, real
estate
Transportation and 29 46 66
public utilities
Entertainment and 26 45 67
recreation
services
Professional and 24 39 58
related services
Personal services 23 38 56
All industries 33 48 64
Source: Simulations based upon GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI during a
first unemployment period, a second unemployment period with UI receipt
during the prior unemployment period, and a third unemployment period with
UI receipt during both prior unemployment periods. The positive effect
that each prior UI receipt period has on the likelihood of current UI
receipt is statistically significantly larger for the public
administration industry relative to the professional and related services
industry at the 95 percent confidence level, and smaller for the
agriculture and construction industries. The simulations also incorporate
the industry effects and the industry interactions with the number of
prior periods of unemployment. See appendix I for methodology and
estimation results.
aWorkers experiencing their first period of unemployment did not have past
UI receipt.
These results show that although UI-eligible workers in some industries
are more likely to receive UI benefits when they experience unemployment
for the first time, their likelihood of receiving UI benefits again when
they become unemployed a second or third time is not necessarily higher
than it is for workers from other industries. For example, the likelihood
of receiving UI benefits for workers from the manufacturing industry who
are unemployed for the first time is relatively high-about 40 percent.
This likelihood increases to 52 percent during a second period of
unemployment for workers who have already received UI benefits, and to 65
percent during a third period of unemployment for workers who received UI
each time they were unemployed. By comparison, the increase in the
likelihood of receiving UI between the first and third periods of
unemployment is higher for most other industries, especially public
administration. Specifically, the likelihood of receiving UI benefits for
public administration workers who are unemployed for the first time is 37
percent. This likelihood increases to 69 percent during a second period of
unemployment for workers who have already received UI, and to 92 percent
during a third period of unemployment for workers who received UI each
time they were unemployed. (See fig. 9.)
Figure 9: Simulated Effect of Past UI Benefit Receipt on the Likelihood of
Receiving UI in Subsequent Periods of Unemployment, for Selected
Industries
Note: Simulations are the average likelihood of receiving UI during a
first unemployment period, second unemployment period with UI receipt
during the prior unemployment period, and a third unemployment period with
UI receipt during both prior unemployment periods. The positive effect
that each prior UI receipt period has on the likelihood of current UI
receipt is statistically significantly larger for the public
administration industry relative to the professional and related services
industry at the 95 percent confidence level, and smaller for the
agriculture and construction industries. The simulations also incorporate
the industry effects and the industry interactions with the number of
prior periods of unemployment. See appendix I for methodology and
estimation results.
Administrative unemployment insurance data have shown that repeat UI
recipients tend to be from industries that are more seasonal, such as
manufacturing and construction. Our results, however, suggest that this is
not because workers with past UI receipt from these industries are more
likely to receive UI benefits when they become unemployed than otherwise
similar workers from other industries. Rather, it may be that workers from
such seasonal industries are unemployed more often on average than workers
from other industries, or that a larger proportion of unemployed workers
from such industries have collected UI previously.
Unemployed Workers from Construction and Manufacturing Have Fewer Weeks of
Unemployment
Unemployed workers from construction and manufacturing have shorter
unemployment duration than otherwise similar workers from other
industries. (See table 4.) Furthermore, simulations based on our model
estimates show that differences in unemployment duration across industries
exist whether or not UI benefits are received. Specifically, UI-eligible
workers from construction who receive UI benefits have the fewest weeks of
unemployment on average (17 weeks), when compared with workers from other
industries. Likewise, UI-eligible workers from construction who do not
receive UI benefits also have the fewest weeks of unemployment, on average
(6 weeks).
Table 4: Simulated Unemployment Duration for UI-Eligible Workers, by
Industry and UI Receipt Status
Simulated unemployment duration (median
weeks)
Not receiving UI
Industry Receiving UI benefits benefits
Construction 17 6
Mining 17 6
Business services 18 7
Manufacturing 19 7
Finance, insurance, and real
estate 21 8
Wholesale and retail trade 22 9
Public administration 23 9
Professional and related services 24 10
Entertainment and related services 24 10
Personal services 24 10
Agriculture, forestry, and fishing 26 11
Transportation and public
utilities 27 12
Overall average duration 21 8
Source: Simulations based upon GAO analysis of NLSY79 data.
Note: Simulations are the median duration of unemployment during
first-time unemployment. The parameter estimates for the construction and
manufacturing industries are statistically significant relative to the
professional and related services industry at the 95 percent confidence
level. See appendix I for methodology and estimation results.
Certain Occupations Are Associated with UI Benefit Receipt and Longer
Unemployment Duration
The likelihood of receiving UI benefits varies across occupations, but
generally not as much as it does across industries. Specifically,
UI-eligible managers are about one-fifth more likely to receive UI than
otherwise similar transportation equipment operators, and one-half more
likely to receive UI than professional and technical workers (see table
5).
Table 5: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers from Different Occupations
Simulated likelihood ofreceiving UI
Occupation benefits (percent)
Managers and administrators 39
Farmers, farm laborers, and foremen 38
Machine operators (nontransportation) 38
Craftsmen 35
Laborers (nonfarm) 34
Transportation equipment operators 33
Clerical and unskilled workers 33
Service workers (excluding private 28
household)
Sales workers 28
Professional and technical workers 25
Overall average 33
Source: Simulations based upon GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI during
first-time unemployment for workers from different occupations. The
parameter estimates for managers and administrators, farmers, farm
laborers, and foremen, machine operators, craftsmen, laborers,
transportation equipment operators, and clerical and unskilled workers are
statistically significant relative to professional and technical workers
at the 95 percent confidence level. See appendix I for methodology and
estimation results.
UI-eligible workers who have received UI benefits in the past are more
likely to receive UI benefits during a current period of unemployment than
UI-eligible workers who never received UI benefits, across each
occupation. Specifically, this effect is strongest for sales and service
workers and weakest for transportation equipment operators and craftsmen
(see table 6).34
34Although the association between past UI receipt and current UI receipt
is statistically significant for all occupations combined, differences in
this association among occupations were statistically significant only for
sales and service workers, and for transportation equipment operators and
craftsmen.
Table 6: Simulated Likelihood of Receiving UI Benefits during Different
Periods of UI-Eligible Unemployment for Workers with Past UI Receipt, by
Occupation
Simulated likelihood of receiving UI benefits
during current UI-eligible unemployment period,
given past UI receipt (percent)
First Second Third
unemployment unemployment unemployment
Occupation perioda period period
Managers and 39 52 65
administrators
Farmers, farm laborers, 38 54 70
and foremen
Machine operators 38 50 62
(nontransportation)
Craftsmen 35 46 56
Laborers (nonfarm) 34 45 58
Transportation equipment 33 42 51
operators
Clerical and unskilled 33 53 73
workers
Service workers 28 50 74
(excluding private
household)
Sales workers 28 66 94
Professional and 25 39 56
technical workers
Overall average 33 48 64
Source: Simulations based upon GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI during a
first unemployment period, a second unemployment period with UI receipt
during the prior unemployment period, and a third unemployment period with
UI receipt during both prior unemployment periods. The positive effect
that each prior UI receipt period has on the likelihood of current UI
receipt is statistically significantly larger for sales workers and
service workers relative to professional and technical workers at the 95
percent confidence level, and smaller for transportation equipment
operators and craftsmen. The simulations also incorporate the occupation
effects and the occupation interactions with the number of prior periods
of unemployment. See appendix I for methodology and estimation results.
aWorkers experiencing their first period of unemployment did not have past
UI receipt.
Unemployment duration also varies across occupations. UI-eligible
professional and technical workers have longer unemployment duration than
otherwise similar workers from other occupations. Specifically,
professional and technical workers have unemployment duration that is 5
weeks longer than average for workers receiving UI and 3 weeks longer than
average for workers not receiving UI (see table 7).35 Past experience with
UI benefit receipt has no significant effect on unemployment duration,
regardless of a worker's occupation.
35The largest differences between industries in median weeks of
unemployment are 10 weeks for workers receiving UI and 5 weeks for workers
not receiving UI.
Table 7: Simulated Unemployment Duration for UI-Eligible Workers, by
Occupation and UI Receipt Status
Simulated unemployment duration
(median weeks)
Occupation Receiving UI Not receiving UI
benefits benefits
Craftsmen 16 6
Sales workers 18 7
Machine operators (nontransportation) 19 7
Transportation equipment operators 20 8
Laborers (nonfarm) 20 8
Service workers (excluding private
household) 23 9
Managers and administrators 23 9
Clerical and unskilled workers 23 10
Farmers, farm laborers, and foremen 26 11
Professional and technical workers 26 11
Overall average duration 21 8
Source: Simulations based upon GAO analysis of NLSY79 data.
Note: Simulations are the median duration of unemployment during
first-time unemployment for workers from different occupations. The
parameter estimates for craftsmen and machine operators are statistically
significant relative to professional and technical workers at the 95
percent confidence level. See appendix I for methodology and estimation
results.
Concluding Observations
Although the UI program has existed for over 70 years and serves millions
of workers each year, little is known about workers who receive UI
benefits on a recurring basis or about workers who are eligible for UI
benefits but never receive them. We found that UI-eligible workers during
the first half of their working lives with certain demographic
characteristics and from certain industries have a greater likelihood of
receiving UI benefits multiple times and experiencing longer unemployment
durations than otherwise similar workers. Although our results are
generally consistent with past research, our analysis includes additional
information about workers' past experiences that provides new insight into
the factors that distinguish workers who receive UI benefits from those
who do not. In fact, the single most important factor associated with
eligible workers receiving benefits is whether or not they received
benefits during previous unemployment, suggesting that a worker's
perception of UI when they are faced with unemployment is key to whether
that worker will ever use the program. Moreover, it does not appear that
previous UI recipients from industries where UI benefit receipt is more
likely, such as construction and manufacturing, are any more likely to
receive benefits if unemployed again than similar workers from other
industries. Rather, it appears that workers from these industries are
simply more likely to face the choice of whether or not to file for UI
benefits more often than their counterparts in other industries. In
addition, while the patterns for UI receipt and unemployment duration we
identified for this group during the first half of their working lives may
not change significantly as they enter the second half of their working
lives, it remains to be seen whether the issues they face in the years
leading up to their retirement will reshape their use of the UI program.
Agency Comments
We provided a draft of this report to Labor officials for their review and
comment. Labor applauded GAO's efforts to determine the extent to which an
individual worker's characteristics are associated with the likelihood of
UI benefit receipt and with unemployment duration and noted that the study
adds to current knowledge of the UI program, particularly with regard to
the impact of past UI benefit receipt on current UI receipt. However,
Labor also noted that there are several issues related to our methodology
that may limit the utility of our findings for policymaking. While we
agree that there are limitations inherent in our methodology, we believe
that these limitations have been noted throughout the report, and that
they do not compromise the overall validity of our results. Nevertheless,
we have provided additional clarifications, as appropriate, to address
Labor's technical comments.
As agreed with your office, unless you publicly announce the contents of
this report earlier, we plan no further distribution of it until 30 days
from its date. At that time, we will send copies of this report to
relevant congressional committees, the Secretary of Labor, or other
interested parties. We will also make copies available to others upon
request. The report will be available at no charge on GAO's Web site at
http://www.gao.gov . If you or members of your staff have any questions
about this report, please contact me at (202) 512-7215. Other major
contributors are listed in appendix III.
Sincerely yours,
Sigurd R. Nilsen, Director Education, Workforce and Income Security Issues
Aand U Appendix I: Analysis of UI Benefit Receipt and Unemployment
Duration
Overview
We analyzed the factors affecting unemployment insurance (UI) benefit
receipt by statistically modeling the determinants of UI benefit receipt
and unemployment durations simultaneously. We model UI benefit receipt in
conjunction with unemployment durations to allow for correlations that may
exist between the two outcomes for a given individual. For example, an
unemployed person anticipating a lengthy unemployment period might be more
likely to receive UI benefits than a person expecting a short unemployment
period. Alternatively, the receipt of UI benefits may lengthen an
unemployment period by allowing an individual to spend more time looking
for new employment. In addition, our model controls for a number of
observable factors about each unemployed worker's situation, including
recent employment experience, prior unemployment and UI benefit receipt
experience, information about UI program factors, including benefit
levels, and demographic characteristics. The model was developed and
estimated by Dr. Brian McCall, Professor of Human Resources and Industrial
Relations, University of Minnesota, under contract to GAO.
This appendix describes (1) the data used in the analysis, including how
the data were prepared, (2) the econometric model that was estimated, (3)
the results from two specifications of the econometric model, and (4) the
limitations inherent in the analysis.
Data Used
We used the Bureau of Labor Statistics' (BLS) National Longitudinal Survey
of Youth 1979 (NLSY79) for our analysis. The NLSY79 is an ongoing
longitudinal survey of individuals who were between the ages of 14 and 22
in 1979, the first year of the survey.1 A primary focus of the NLSY79 is
on individuals' labor force patterns, and the data are collected at a very
detailed level. This detail allows us to track the weekly employment,
unemployment, and earnings histories of the individuals in the sample. The
NLSY79 also contains less detailed information about individuals' UI
receipt during unemployment.2 The NLSY79 does not contain direct
information about an individual's UI eligibility status, which is a
function of previous employment and earnings, among other things, and
varies by state of employment.3 We estimate an unemployed individual's UI
eligibility status using data that are available in the NLSY79.
1NLSY79 data begin in 1978. Interviews for the NLSY79 were conducted
annually until 1994, and biennially beginning in 1996. We used data
through 2002, which were the most recent NLSY79 data available.
2UI receipt information is provided on a monthly basis in the NLSY79.
Because this information is only given on a monthly basis, it cannot be
used to accurately measure the number of weeks of UI receipt during
unemployment.
There are three main reasons why the NLSY79 database provides the most
suitable data for our analysis. First, the longitudinal nature and level
of detail of the data allow us to control for an individual's history of
unemployment and UI receipt, which is a major contribution of this work.
Second, respondents were first surveyed at a young age, which reduces the
likelihood that we do not observe periods of unemployment and UI receipt
early in a person's working career. Third, the detailed data allow us to
estimate an individual's UI eligibility status, allowing us to focus our
analysis on unemployed individuals whom we estimated to be eligible for UI
benefits while also reasonably controlling for differences in UI program
rules across states. A few limitations to the NLSY79 database should be
mentioned. First, the sample began with 12,686 individuals in 1979, but
has decreased in size over time due to attrition.4 Second, the data are
self-reported and thus subject to recall error. We assessed the
reliability of the NLSY79 data by interviewing relevant BLS officials,
reviewing extensive NLSY79 documentation, and performing electronic tests
of the NLSY79 data for missing or corrupt information that might
negatively affect our analysis. On the basis of these reviews and tests,
we determined that the data were sufficiently reliable to be used in our
analysis.
We considered using administrative state UI data as an alternative to the
NLSY79. Although such administrative data could provide information about
all UI recipients in a state, these data could not provide information
about UI-eligible unemployed workers who did not receive benefits. Also,
because these data are not designed for research purposes, there is
limited information available about individuals that can be used to
control for differences, such as demographic characteristics. Finally,
there is also no nationally representative data source for administrative
UI data.
For each individual in the NLSY79 database, we created a detailed weekly
history of employment and unemployment, including whether UI benefits were
received during unemployment. Our definition of unemployment is not the
strict definition used in the BLS's Current Population Survey (CPS). We
define unemployment to include both the weeks in which an out-of-work
person is looking for work (the standard CPS unemployment definition) and
the weeks during which the individual reports being out of the labor force
(OLF). We did require that an individual spend at least 1 week actively
looking for work after a job loss to reduce the likelihood that the person
had permanently left the labor force. Other research has addressed the
effect that the UI program plays on the percentage of weeks of
nonemployment that a person reports that he or she was looking for work.5
3State UI programs determine eligibility using a number of criteria,
including the following conditions: (1) the unemployment must be the
result of a job loss that was not caused by the individual, (2) the
individual must have earned a specified amount of money during the time
preceding the unemployment, and (3) the individual must be actively
looking for new employment.
4See Center for Human Resource Research, Ohio State University, The
National Longitudinal Surveys NLSY79 User's Guide, prepared for the
Department of Labor, 2002.
For each unemployment period experienced by an individual, we estimate the
person's UI eligibility status. Although states determine UI eligibility
using a number of criteria, we focus on the following three: (1) the
unemployment must be the result of a job loss that was not caused by the
individual, (2) the individual must have earned a specified amount of
money during the time preceding the unemployment, and (3) the individual
must be actively looking for new employment. The NLSY79 provides the
information necessary to estimate whether these criteria are met by an
unemployed individual. For criterion 1, the NLSY79 provides information
about the reason that a job was lost. Only those unemployed individuals
who lost a job through no fault of their own were deemed to be
UI-eligible.6 For the monetary eligibility criterion 2, we compiled a
detailed set of UI eligibility and benefit criteria for each of the 50
states and the District of Columbia over the period 1978 to 2002.7 When
these criteria were combined with the NLSY79's detailed employment and
earnings histories, we were able to determine monetary eligibility for UI
with reasonable accuracy, as well as the weekly benefit amount and the
number of weeks of benefits a person was eligible to receive.8 For
criterion 3, we considered as UI-eligible only those unemployed
individuals who reported actively looking for work during at least 1 week
of their unemployment. We erred on the side of overestimating the
eligibility based on criterion 3, because individuals who self-report
information about nonemployment may not fully realize the impact that
"looking for work" versus "being out of the labor force" has on UI
eligibility, especially if they did not receive UI benefits. Although this
estimation method is not perfect, we believe that it captures some of the
most important features of UI eligibility. It is similar to the methods
used by other researchers.9
5R. Mark Gritz and Thomas MaCurdy, "Measuring the Influence of
Unemployment Insurance on Unemployment Experiences," Journal of Business
and Economic Statistics, vol. 15, no. 2, (1997), examined the role that UI
rules have on an individual's choice to report himself or herself as
unemployed (CPS definition) as opposed to out of the labor force. They
found that, in addition to having longer nonemployment periods, UI
recipients report being unemployed in the CPS sense for a greater
proportion of their nonemployment period.
6It appeared from the NLSY79 data that a number of respondents did not
differentiate between being laid off and being discharged or fired. As a
result, we include those who report being either laid off or discharged or
fired as satisfying the first UI eligibility rule. The NLSY79 reports a
number of other reasons for leaving a job, including having found better
work, low pay, pregnancy, illness, change of job by spouse or parents,
other family reasons, job's interference with school, the end of a
program, bad working conditions, and entrance into the armed forces.
7See U.S. Department of Labor. Employment and Training Administration,
Significant Provisions of State Unemployment Insurance Laws (Washington,
D.C., 1979-2002).
In addition to estimating the UI eligibility status of individuals at the
time of each of their unemployment periods, we also created the other
variables used in our analysis. The empirical model outlined in the
following subsection focuses on UI benefit receipt and unemployment
duration. UI benefit receipt during unemployment was determined using the
monthly measure provided in the NLSY79.10 The duration of unemployment, as
defined above, is measured in weeks from the week after a job was lost to
the week a new job was begun. We censor duration to be no longer than 100
weeks.
To isolate the impact that a variable has on the likelihood of UI benefit
receipt and unemployment duration, our model controls for a great number
of other factors that were observable at the start of, and throughout, the
person's unemployment. One set of variables relates to the employment
experience of the individual immediately preceding unemployment, including
industry and occupation of the lost job (measured at the one-digit
Standard Industrial Classification [SIC] and Standard Occupational
Classification [SOC] level), union status and tenure at the job lost,
earnings (base period earnings [BPE] and high quarter earnings [HQE]),
whether the job was lost because of a plant closing, and the calendar year
and month the unemployment began.11 We group both earnings measures into
brackets to allow for nonlinear effects. All dollar values are adjusted
for inflation to 2002 dollars using the BLS's Consumer Price Index for All
Urban Consumers (CPI-U). We also control for the state unemployment rate
during the month that unemployment began, and, in the duration equation,
for the time-varying state monthly unemployment rate over the period of
unemployment.
8Although UI eligibility is based upon the rules in the state where an
individual is employed, we used state of residence for our estimates
because state of employment was not available in the NLSY79. Thus, people
who work in one state but live in another may not be classified correctly.
However, we believe that only a small percentage of such data are
classified incorrectly and, thereby, our results should be only minimally
affected.
9See Gritz and MaCurdy, 1997, and McCall 2000 for examples.
10We consider only an individual's first period of unemployment with UI
receipt during a person's "benefit year." A benefit year is the 52-week
period during which UI benefits can be claimed.
A second set of variables summarizes UI program factors, such as the
weekly benefit amount (WBA) a person is eligible to receive, the number of
weeks of benefits a person is eligible to receive, whether the state has a
waiting period before benefits can be received, and whether permanent or
temporary extended benefits are in effect.12 We also control for the
percentage of new UI claims that are denied by a state (in the receipt
equation) and the percentage of continuing UI claims that are denied by a
state (in the duration equation). In the unemployment duration equation,
we also allow the parameter estimates for WBA, remaining weeks of
benefits, and extended benefits to vary over the period of unemployment.
This is done by interacting these variables with a cubic function of the
number of weeks unemployed. Again, all dollar values are adjusted for
inflation to 2002 dollars using the BLS's CPI-U.
A third group of variables relates to a person's history of unemployment
and UI benefit receipt as measured at the start of an unemployment period.
This group of variables includes the number of times the person had been
unemployed and the number of times a person had received UI benefits
previously (in the receipt equation) and whether or not the person had
been unemployed and whether or not the person had received UI benefits
previously (in the duration equation). We also interact these variables
with industry and occupation dummy variables to investigate whether
previous unemployment and UI receipt affect the likelihood of current UI
receipt and unemployment durations differently across industries. These
interactions with industry and occupation are done in separate
specifications of the model.
11The base period is the period of time during which earnings are counted
toward UI eligibility. It generally covers a year. We define the base
period as the first four of the last five completed calendar quarters.
High quarter earnings refers to the quarter of highest earnings during the
base period.
12Permanent extended benefits are triggered by high unemployment rates in
a state, and provide for up to 13 additional weeks of benefits to
UI-eligible individuals. Temporary extended benefits are available
periodically, as authorized by Congress.
A fourth group of variables relates to a person's demographic
characteristics at the time of unemployment. These include age, race,
gender, marital status, number of years of schooling, health limitations,
whether a spouse has used UI previously, family size, number of children,
number of children between the ages of 0 and 2, whether the person lives
with his or her parents, state of residence, and whether the person lives
in a Standard Metropolitan Statistical Area (SMSA) as opposed to a rural
area.
We limit our analysis to the nonmilitary sample of NLSY79 respondents.13
In addition, we drop individuals with insufficient information to estimate
UI eligibility with reasonable accuracy. Data for an individual were
included up to their first missed interview.14 Individuals without any
unemployment, and those without unemployment that was estimated to be
UI-eligible, were not used in the analysis. Also, individuals who were
missing data required by our econometric model were not used in the
analysis. This yielded a sample of 5,631 individuals who had been
unemployed and eligible for UI benefits at least once, resulting in a
total of 15,506 separate periods of UI-eligible unemployment.
Econometric Model
To investigate the key factors associated with UI benefit receipt,
including the role of prior UI benefit receipt (repeat UI recipiency), we
used a dynamic econometric model that jointly determines UI benefit
receipt and unemployment duration. As mentioned above, the reason for
modeling these outcomes jointly is to allow for the likely correlations
that exist between them.15 In addition to modeling UI receipt and
unemployment duration jointly, our model allows prior unemployment and
prior UI receipt to influence current UI receipt and unemployment duration
to allow for the correlations that possibly exist over time for an
individual.
13The NLSY79 began with 12,686 individuals in 1979, 1,280 of whom were
part of the military subsample. The majority of the military subsample of
the NLSY79 was eliminated in 1985.
14The NLSY79 attempts to reconnect with individuals that missed an
interview in the previous year.
15For one example of an economic model of how the receipt of UI benefits
can affect the expected length of unemployment by affecting a person's
reservation wage, see Mortensen.
We used a complementary log-log specification to model the probability of
UI receipt during an individual's kth unemployment period, k=1, ..., K,
as:
, where is a vector of exogenous variables measured at the start of the
kth unemployment period, all of which are assumed to be independent of the
unobserved random variable , which helps control for unobserved
heterogeneity. Variables in include demographic characteristics,
characteristics about the lost job, and UI program information. The vector
is a vector of endogenous variables pertaining to past unemployment and
past UI benefit receipt, which are measured at the start of an
individual's kth unemployment period and may be correlated to.
We modeled unemployment durations using a discrete-time hazard function,
which gives the probability of an event occurring during a discrete time
period, conditional upon not having experienced the event prior to that
time. This can be thought of as the escape rate from unemployment during a
specific time period. We assume that the conditional probability that an
individual's kth period of unemployment ends in the interval (m-1,m],
given that it exceeds m-1, where m indexes the number of weeks, follows a
complementary log-log specification:
,
for m=1, ..., M, and where is a vector of exogenous variables measured at
the start of the kth unemployment period, all of which are assumed to be
independent of the unobserved random variable , which helps control for
unobserved heterogeneity. Variables in include demographic
characteristics, characteristics about the lost job, and UI program
information. The vector consists of endogenous variables pertaining to
current UI benefit receipt, past unemployment, and past UI benefit
receipt, which are measured at the start of an individual's kth
unemployment period and may be related to. The parameter vector is the
baseline hazard function.
Letting i, i = 1, ..., I, index individuals and k index an individual's
unemployment periods, we define to be equal to 1 when individual i has a
kth unemployment period, and 0 otherwise. Also, we define to be equal to 1
when individual i's kth unemployment period is complete, and 0 otherwise.
Using this notation, individual i's contribution to the likelihood
function can be written:
where the vector of parameters is to be estimated and contains , , , and ,
where j = d,u. We assume that the distribution of the unobserved random
variables is such that there are 3 different types of individuals in the
population, with the fraction of each type equal to , where: . Combining
these possibilities, we write an individual's likelihood contribution as:
, where . Taking logarithms and summing over all individuals yields the
full log likelihood function for the sample: . This likelihood is computed
in FORTRAN and maximized using the BHHH algorithm.16
A number of features outlined above are simplifications of a more general
version of this model, and were introduced to help reduce the number of
parameters to be estimated by the model. First, the baseline hazard
function, , was assumed to be independent of the unemployment period
number, k. Second, the parameters associated with the exogenous (, j =
d,u) and endogenous (, j = d,u) variables were assumed to be independent
of the unemployment period number, k. Third, the unobserved random
variables (, j = d,u) were assumed to be independent of the unemployment
period number, k. Although this assumption is not as general as allowing
each individual to have different unobserved components over time, it does
help control for unobserved differences between individuals that may
influence UI receipt and unemployment durations.
Because of the complexity of the empirical model outlined above,
interpreting the parameter estimates is difficult.17 As a result, we use
the output from the model to simulate the effect that changes in certain
variables have on the likelihood of UI receipt and the duration of
unemployment for the average unemployed person in our sample. For example,
to understand differences in UI receipt and unemployment durations by
industry, we simulate the likelihood of UI benefit receipt and
unemployment duration for the average person in our sample for each of the
possible industries, and then compare the results. To do this, we use the
model's output to calculate every person's likelihood of UI receipt and
escape rate from unemployment-conditional upon receiving and not receiving
UI-assuming all were in the first industry grouping when they lost their
job. Averaging over all individuals yields the average probability of UI
receipt and the averaged (week by week) survivor function.18 The averaged
survivor function can be used to compute the expected median duration of
unemployment.19 We then repeat this process, successively, assuming that
all individuals were in another industry grouping when they lost their
jobs, until all industry groups have been simulated.
16See E. K. Berndt, B. H. Hall, R. E. Hall, and J. A. Hausman, "Estimation
and Inference in Nonlinear Structural Models," Annals of Economic and
Social Measurement, vol. 3, no. 4 (1974). The BHHH algorithm is a
quasi-Newton method for finding maximums.
17In addition to being a highly nonlinear model, the data were all
normalized to help the convergence of the parameter estimates.
The simulated average likelihood of UI benefit receipt and median
unemployment duration can then be compared across industries to estimate
the differences by industry. Using all individuals for each simulation,
and reporting results for the average unemployed person, helps insure that
differences in the simulation results (e.g., industry 1 versus industry 2)
reflect only the variables (industry 1, industry 2) being simulated.20 To
describe results that are not related to past experience with unemployment
and UI benefit receipt, we present simulations that are specific to
first-time unemployment-a simple and clearly defined scenario (the
observable trends also hold for unemployed individuals with prior
unemployment and UI receipt experience).
Results
We report parameter estimates from two specifications of our model. The
first specification includes interaction terms between industry and our
measures of past UI benefit receipt and past unemployment. These results
are presented in tables 8 and 9 for the UI benefit receipt equation and
the unemployment duration equation respectively. The second specification
includes interaction terms between occupation and our measures of past UI
benefit receipt and past unemployment. These results are presented in
tables 10 and 11 for the UI benefit receipt equation and the unemployment
duration equation respectively. We included the industry and occupation
interactions in separate specifications to avoid the issues brought about
by multicollinearity.21 Because the results for the noninteraction terms
are similar between the two specifications, we focus on those from the
industry-interaction specification (tables 8 and 9). After discussing
these results, we discuss the results for the occupation-interaction
specification (tables 10 and 11).
18The survivor function at time t for an event is the probability of not
having experienced that event prior to time t. The survivor function is
mathematically related to the escape rate (hazard rate).
19We chose median rather than mean because of the skewed nature of our
unemployment duration data.
20For our simulations, if we used only those individuals that reported
losing a job from a specific industry, as opposed to using all
individuals, it is likely that a portion of the differences we would
observe in the likelihood of UI benefit receipt and unemployment duration
would be due to differences in other observable factors between the
individuals from the different industry groups. For example, it may be
that professional services workers have higher average earnings than
agricultural workers, which would be earnings effect, not an
industry-specific effect.
Tables 8, 9, 10, and 11 are structured as follows. The first column in
each table lists the variable names; the second column, the parameter
estimates; the third column, the estimated standard errors; and the fourth
column, the t-statistics. The last column contains asterisks that signify
statistical significance. One asterisk (*) signifies statistical
significance at the 90 percent confidence level (t-statistics greater than
or equal to 1.65 in absolute value); two asterisks signify statistical
significance at the 95 percent confidence level (t-statistics greater than
or equal to 1.96 in absolute value) and three asterisks (***) signify
statistical significance at the 99 percent confidence level. Parameter
estimates discussed below are statistically significantly different from
zero at the 95 percent confidence level unless stated otherwise.22 To
conserve space, the tables do not present the parameter estimates for the
unobserved heterogeneity (and ), state, year, and month effects.
21We also tried running a specification of the model that included these
interactions for both industry and occupation. The parameter estimates and
simulations were generally similar to those for the two separate
specifications, but much of the statistical significance for individual
parameters was lost due to the correlation between industry and
occupation. However, a likelihood ratio test of the joint hypothesis that
the interaction terms for both industry and occupation are all equal to
zero is rejected at the 95 percent confidence level, suggesting that there
are both industry-specific and occupation-specific differences in the
effects of past unemployment and past UI receipt on the likelihood of
current UI receipt and current unemployment duration.
22A statistically insignificant result indicates that the effect of a
characteristic could not be precisely estimated using the sample data, and
does not necessarily prove that the characteristic is unimportant.
Table 8: Parameter Estimates for UI Receipt Equation from
Industry-Interaction Specification
Parameter Standard
estimate error t-statistic
Past unemployment and UI receipt
Number of previous UI receipt spells 0.714 0.086 8.26 ***
Number of previous unemployment spells -0.072 0.017 -4.27 ***
Industry
Agriculture, forestry, and fishing 0.438 0.211 2.07 **
Mining 0.868 0.242 3.59 ***
Construction 0.294 0.135 2.17 **
Manufacturing 0.672 0.108 6.20 ***
Transportation and public utilities 0.221 0.162 1.36
Wholesale and retail trade 0.475 0.109 4.36 ***
Finance, insurance, and real estate 0.292 0.174 1.68 *
Business services 0.310 0.142 2.19 **
Personal services -0.077 0.198 -0.39
Entertainment and recreation services 0.104 0.226 0.46
Public administration 0.560 0.188 2.98 ***
Valid missing -0.030 0.095 -0.31
Occupation
Managers and administrators 0.614 0.100 6.15 ***
Sales workers 0.122 0.138 0.88
Clerical and unskilled workers 0.296 0.088 3.39 ***
Craftsmen 0.261 0.092 2.85 ***
Machine operators (nontransportation) 0.187 0.090 2.07 **
Transportation equipment operators 0.244 0.115 2.12 **
Laborers (nonfarm) 0.075 0.101 0.75
Farmers, farm laborers, and foremen 0.134 0.191 0.70
Service workers (excluding private
household) 0.081 0.092 0.88
Industry * number previous UI receipt
spells
Agriculture, forestry, and fishing -0.301 0.101 -2.97 ***
Mining -0.115 0.201 -0.57
Construction -0.229 0.091 -2.53 **
Manufacturing -0.160 0.093 -1.72 *
Transportation and public utilities 0.033 0.120 0.27
Wholesale and retail trade 0.010 0.109 0.10
Finance, insurance, and real estate 0.565 0.344 1.64
Business services -0.005 0.122 -0.04
Personal services -0.011 0.161 -0.07
Entertainment and recreation services 0.160 0.227 0.70
Public administration 0.487 0.239 2.04 **
Valid missing 0.158 0.098 1.61
Industry * number of previous
unemployment spells
Agriculture, forestry, and fishing -0.043 0.035 -1.23
Mining -0.129 0.056 -2.29 **
Construction -0.040 0.022 -1.83 *
Manufacturing -0.057 0.019 -2.97 ***
Transportation and public utilities -0.001 0.028 -0.02
Wholesale and retail trade -0.047 0.020 -2.30 **
Finance, insurance, and real estate -0.013 0.032 -0.41
Business services -0.012 0.025 -0.48
Personal services 0.023 0.035 0.65
Entertainment and recreation services -0.052 0.045 -1.15
Public administration -0.061 0.040 -1.54
Valid missing -0.061 0.021 -2.90 ***
UI program variables
Weekly benefit amount (WBA) 0.064 0.058 1.09
Potential UI benefit duration -0.623 0.592 -1.05
Waiting week for UI benefits 0.053 0.124 0.43
Denial rate for new UI claims -1.448 0.836 -1.73 *
Extended UI benefits in effect 0.133 0.097 1.36
Personal characteristics
Years of education 0.569 0.128 4.44 ***
Armed Forces Qualifying Test score -0.295 0.097 -3.06 ***
African-American 0.005 0.059 0.08
Hispanic -0.084 0.086 -0.98
Hispanic * male 0.225 0.098 2.30 **
Married 0.167 0.050 3.35 ***
Age 20.541 4.510 4.55 ***
Age-squared -41.787 8.035 -5.20 ***
Male -0.357 0.051 -7.01 ***
Lives in SMSA (urban) -0.111 0.050 -2.24 **
Health limitations 0.041 0.110 0.38
Spouse used UI in past 0.270 0.100 2.70 ***
Spouse used UI in past * male -0.688 0.173 -3.97 ***
Live with parents -0.160 0.097 -1.65 *
Family size -0.478 0.177 -2.71 ***
Live with parents * family size 0.572 0.231 2.47 **
Children under age 2 0.091 0.060 1.51
Number of children -0.017 0.025 -0.68
Recent employment experience
Union member -0.003 0.048 -0.05
Tenure 0.140 0.029 4.85 ***
Tenure-squared -0.015 0.004 -4.20 ***
Lost job due to plant closing -0.263 0.085 -3.08 ***
State unemployment rate 0.314 0.150 2.09 **
Base period earnings brackets
Under $2,000 -1.450 0.278 -5.21 ***
$2,000-$3,999 -1.383 0.196 -7.05 ***
$4,000-$5,999 -1.177 0.168 -7.02 ***
$6,000-$7,999 -0.799 0.152 -5.25 ***
$8,000-$9,999 -0.780 0.140 -5.56 ***
$10,000-$11,999 -0.528 0.127 -4.16 ***
$12,000-$13,999 -0.565 0.126 -4.50 ***
$14,000-$15,999 -0.381 0.116 -3.28 ***
$16,000-$17,999 -0.279 0.109 -2.56 **
$18,000-$19,999 -0.282 0.108 -2.62 ***
$20,000-$24,999 -0.143 0.089 -1.61
$25,000-$29,999 0.015 0.086 0.17
High quarter earnings
$0-$999 -0.180 0.260 -0.69
$1,000-$1,999 0.076 0.196 0.39
$2,000-$2,999 0.333 0.153 2.18 **
$3,000-$3,999 0.443 0.129 3.44 ***
$4,000-$4,999 0.410 0.113 3.65 ***
$5,000-$5,999 0.270 0.103 2.63 ***
$6,000-$6,999 0.096 0.092 1.05
$7,000-$7,999 0.170 0.090 1.89 *
$8,000-$8,999 0.030 0.092 0.33
Year effects Included
Month effects Included
State effects Included
Unobserved heterogeneity effects Included
Source: GAO analysis of NLSY79 data.
Note: In the final column an asterisk signifies statistical significance
at the 90 percent confidence level, two asterisks signify statistical
significance at the 95 percent confidence level, and three asterisks
signify statistical significance at the 99 percent confidence level. The
notation X * Y signifies an interaction between the variables X and Y. The
omitted category for industry is professional and related services and for
occupation is professional and technical workers. The omitted category for
BPE is $30,000 and above and for HQE it is $9,000 and above. Sample
includes 5,631 individuals with a total of 15,506 unemployment spells. The
maximized log likelihood value is -63,438.514.
Table 9: Parameter Estimates for Duration Equation from
Industry-Interaction Specification
Parameter Standard
estimate error t-statistic
Past unemployment and UI receipt
Previous UI receipt 0.155 0.090 1.73 *
Previous unemployment 0.101 0.093 1.09
Industry
Agriculture, forestry, and fishing -0.088 0.198 -0.45
Mining 0.301 0.273 1.10
Construction 0.314 0.156 2.01 **
Manufacturing 0.213 0.107 1.99 **
Transportation and public utilities -0.135 0.233 -0.58
Wholesale and retail trade 0.069 0.104 0.67
Finance, insurance, and real estate 0.121 0.202 0.60
Business services 0.268 0.153 1.76 *
Personal services -0.031 0.191 -0.16
Entertainment and recreation services -0.024 0.299 -0.08
Public administration 0.029 0.236 0.12
Valid missing 0.005 0.084 0.06
Occupation
Managers and administrators -0.046 0.062 -0.74
Sales workers -0.024 0.070 -0.33
Clerical and unskilled workers -0.106 0.044 -2.43 **
Craftsmen 0.030 0.050 0.61
Machine operators (nontransportation) -0.025 0.048 -0.51
Transportation equipment operators 0.005 0.061 0.08
Laborers (nonfarm) -0.030 0.052 -0.59
Farmers, farm laborers, and foremen -0.055 0.105 -0.52
Service workers (excluding private
household) -0.112 0.044 -2.57 **
Industry * previous UI receipt
Agriculture, forestry, and fishing -0.004 0.162 -0.02
Mining 0.006 0.287 0.02
Construction -0.123 0.111 -1.10
Manufacturing -0.136 0.099 -1.37
Transportation and public utilities -0.007 0.144 -0.05
Wholesale and retail trade -0.128 0.110 -1.16
Finance, insurance, and real estate -0.165 0.213 -0.78
Business services -0.104 0.138 -0.75
Personal services 0.101 0.188 0.54
Entertainment and recreation services 0.088 0.221 0.40
Public administration 0.303 0.221 1.37
Valid missing -0.080 0.098 -0.82
Industry * previous unemployment
Agriculture, forestry, and fishing 0.053 0.196 0.27
Mining -0.455 0.295 -1.54
Construction -0.253 0.159 -1.60
Manufacturing -0.187 0.111 -1.69 *
Transportation and public utilities 0.139 0.239 0.58
Wholesale and retail trade -0.126 0.109 -1.16
Finance, insurance, and real estate -0.175 0.214 -0.82
Business services -0.276 0.163 -1.69 *
Personal services -0.199 0.199 -1.00
Entertainment and recreation services 0.012 0.305 0.04
Public administration -0.239 0.247 -0.97
Valid missing 0.224 0.090 2.48 **
UI program variables
Receiving UI -1.256 0.195 -6.45 ***
Weekly benefit amount (WBA) 0.031 0.035 0.90
WBA * receiving UI 0.067 0.059 1.14
Remaining UI benefit duration -0.014 0.009 -1.64
Waiting week for UI benefits 0.030 0.064 0.47
Denial rate for continuing UI claims 0.488 0.215 2.27 **
Extended UI benefits in effect 0.042 0.054 0.78
Personal characteristics
Years of education 0.235 0.069 3.40 ***
Armed Forces Qualifying Test score 0.251 0.056 4.49 ***
African-American -0.254 0.033 -7.74 ***
Hispanic -0.078 0.037 -2.13 **
Male -1.022 0.397 -2.57 **
Married -0.137 0.037 -3.73 ***
Married * male 0.294 0.049 6.05 ***
Age -5.371 2.796 -1.92 *
Age-squared 9.472 5.010 1.89 *
Age * male 7.759 2.905 2.67 ***
Age-squared * male -13.614 5.150 -2.64 ***
Lives in SMSA (urban) -0.040 0.027 -1.48
Health limitations -0.095 0.055 -1.73 *
Spouse used UI in past 0.136 0.051 2.68 ***
Live with parents -0.045 0.051 -0.87
Family size -0.118 0.089 -1.33
Live with parents * family size 0.114 0.136 0.84
Live with parents * family size * male 0.024 0.094 0.26
Children under age 2 -0.098 0.033 -2.93 ***
Number of children -0.004 0.014 -0.30
Recent employment experience
Union member 0.084 0.029 2.86 ***
Tenure 0.050 0.018 2.75 ***
Tenure-squared -0.008 0.002 -3.07 ***
Lost job due to plant closing -0.179 0.046 -3.87 ***
State unemployment rate (time varying) -0.030 0.007 -4.04 ***
Base period earnings brackets
Under $2,000 -0.389 0.103 -3.77 ***
$2,000-$3,999 -0.367 0.088 -4.18 ***
$4,000-$5,999 -0.360 0.080 -4.49 ***
$6,000-$7,999 -0.286 0.079 -3.64 ***
$8,000-$9,999 -0.239 0.074 -3.22 ***
$10,000-$11,999 -0.165 0.072 -2.29 **
$12,000-$13,999 -0.186 0.069 -2.71 ***
$14,000-$15,999 -0.137 0.066 -2.07 **
$16,000-$17,999 -0.125 0.064 -1.95 *
$18,000-$19,999 -0.101 0.066 -1.53
$20,000-$24,999 -0.039 0.056 -0.71
$25,000-$29,999 -0.135 0.053 -2.54 **
High quarter earnings
Under $1,000 -0.051 0.115 -0.44
$1,000-$1,999 0.103 0.096 1.07
$2,000-$2,999 0.060 0.083 0.72
$3,000-$3,999 0.088 0.074 1.20
$4,000-$4,999 0.031 0.066 0.46
$5,000-$5,999 0.021 0.062 0.33
$6,000-$6,999 0.009 0.057 0.17
$7,000-$7,999 -0.034 0.057 -0.60
$8,000-$8,999 -0.029 0.056 -0.51
Time interactions (t = number of weeks
unemployed)
UI receipt * Extended Benefits * (t-1) -1.536 0.740 -2.08 **
UI receipt * Extended Benefits *
(t-1)-squared 3.351 2.822 1.19
UI receipt * Extended Benefits *
(t-1)-cubed -0.179 0.246 -0.73
Remaining UI benefit duration * (t-1) 0.569 0.201 2.83 ***
Remaining UI benefit duration *
(t-1)-squared -3.172 2.624 -1.21
Remaining UI benefit duration *
(t-1)-cubed -0.256 0.956 -0.27
UI receipt * (t-1) 10.713 1.987 5.39 ***
UI receipt * (t-1)-squared -24.169 5.707 -4.24 ***
UI receipt * (t-1)-cubed 1.564 0.441 3.55 ***
UI receipt * WBA * (t-1) -1.052 0.683 -1.54
UI receipt * WBA * (t-1)-squared 2.425 2.065 1.17
UI receipt * WBA * (t-1)-cubed -0.141 0.163 -0.87
Year effects Included
Month effects Included
State effects Included
Unobserved heterogeneity effects Included
Source: GAO analysis of NLSY79 data.
Note: In the final column an asterisk signifies statistical significance
at the 90 percent confidence level, two asterisks signify statistical
significance at the 95 percent confidence level, and three asterisks
signify statistical significance at the 99 percent confidence level. The
notation X * Y signifies an interaction between the variables X and Y. The
omitted category for industry is professional and related services and for
occupation is professional and technical workers. The omitted category for
BPE is $30,000 and above and for HQE it is $9,000 and above. Sample
includes 5,631 individuals with a total of 15,506 unemployment spells. The
maximized log likelihood value is -63,438.514.
Table 10: Parameter Estimates for UI Receipt Equation from
Occupation-Interaction Specification
Parameter Standard
estimate error t-statistic
Past unemployment and UI receipt
Number of previous UI receipt spells 0.678 0.059 11.58 ***
Number of previous unemployment spells -0.094 0.016 -5.84 ***
Industry
Agriculture, forestry, and fishing -0.044 0.143 -0.30
Mining 0.302 0.171 1.77 *
Construction 0.004 0.095 0.04
Manufacturing 0.346 0.079 4.38 ***
Transportation and public utilities 0.256 0.106 2.41 **
Wholesale and retail trade 0.285 0.077 3.73 ***
Finance, insurance, and real estate 0.307 0.121 2.54 **
Business services 0.253 0.092 2.75 ***
Personal services 0.045 0.127 0.36
Entertainment and recreation services -0.090 0.157 -0.57
Public administration 0.426 0.122 3.48 ***
Valid missing -0.174 0.070 -2.48 **
Occupation
Managers and administrators 0.573 0.154 3.72 ***
Sales workers 0.101 0.202 0.50
Clerical and unskilled workers 0.351 0.122 2.88 ***
Craftsmen 0.437 0.126 3.46 ***
Machine operators (nontransportation) 0.534 0.121 4.42 ***
Transportation equipment operators 0.353 0.167 2.12 **
Laborers (nonfarm) 0.364 0.135 2.70 ***
Farmers, farm laborers, and foremen 0.549 0.239 2.29 **
Service workers (excluding private
household) 0.108 0.128 0.85
Occupation * number previous UI receipt
spells
Managers and administrators -0.155 0.082 -1.89 *
Sales workers 0.830 0.292 2.84 ***
Clerical and unskilled workers 0.141 0.097 1.45
Craftsmen -0.204 0.071 -2.89 ***
Machine operators (nontransportation) -0.100 0.069 -1.45
Transportation equipment operators -0.276 0.073 -3.79 ***
Laborers (nonfarm) -0.108 0.082 -1.31
Farmers, farm laborers, and foremen 0.035 0.182 0.19
Service workers (excluding private
household) 0.283 0.091 3.13 ***
Occupation * number of previous
unemployment spells
Managers and administrators 0.025 0.025 0.98
Sales workers -0.013 0.036 -0.35
Clerical and unskilled workers -0.009 0.021 -0.44
Craftsmen -0.007 0.020 -0.36
Machine operators (nontransportation) -0.060 0.021 -2.91 ***
Transportation equipment operators 0.020 0.027 0.73
Laborers (nonfarm) -0.049 0.024 -2.03 **
Farmers, farm laborers, and foremen -0.073 0.052 -1.39
Service workers (excluding private
household) -0.023 0.022 -1.02
UI Program variables
Weekly benefit amount (WBA) 0.068 0.058 1.16
Potential UI benefit duration -0.529 0.589 -0.90
Waiting week for UI benefits 0.050 0.121 0.41
Denial rate for new UI claims -1.482 0.828 -1.79 *
Extended UI benefits in effect 0.126 0.095 1.33
Personal characteristics
Years of education 0.537 0.128 4.19 ***
Armed Forces Qualifying Test score -0.265 0.096 -2.75 ***
African-American 0.008 0.058 0.14
Hispanic -0.084 0.086 -0.98
Hispanic * male 0.239 0.097 2.46 **
Married 0.164 0.050 3.30 ***
Age 18.880 4.432 4.26 ***
Age-squared -38.467 7.863 -4.89 ***
Male -0.362 0.051 -7.13 ***
Lives in SMSA (urban) -0.131 0.050 -2.64 ***
Health limitations 0.071 0.110 0.64
Spouse used UI in past 0.271 0.100 2.71 ***
Spouse used UI in past * male -0.696 0.170 -4.10 ***
Live with parents -0.170 0.097 -1.76 *
Family size -0.485 0.177 -2.74 ***
Live with parents * family size 0.577 0.232 2.49 **
Children under age 2 0.109 0.060 1.84 *
Number of children -0.028 0.024 -1.17
Recent employment experience
Union member 0.000 0.048 0.01
Tenure 0.130 0.029 4.50 ***
Tenure-squared -0.014 0.004 -3.71 ***
Lost job due to plant closing -0.250 0.085 -2.94 ***
State unemployment rate 0.354 0.148 2.40 **
Base period earnings brackets
Under $2,000 -1.423 0.278 -5.11 ***
$2,000-$3,999 -1.344 0.196 -6.85 ***
$4,000-$5,999 -1.135 0.168 -6.77 ***
$6,000-$7,999 -0.808 0.152 -5.32 ***
$8,000-$9,999 -0.778 0.140 -5.55 ***
$10,000-$11,999 -0.492 0.127 -3.87 ***
$12,000-$13,999 -0.564 0.126 -4.48 ***
$14,000-$15,999 -0.365 0.117 -3.12 ***
$16,000-$17,999 -0.292 0.109 -2.68 ***
$18,000-$19,999 -0.272 0.108 -2.51 **
$20,000-$24,999 -0.136 0.090 -1.50
$25,000-$29,999 0.019 0.086 0.22
High quarter earnings
$0-$999 -0.209 0.257 -0.81
$1,000-$1,999 0.080 0.195 0.41
$2,000-$2,999 0.339 0.152 2.23 **
$3,000-$3,999 0.441 0.129 3.43 ***
$4,000-$4,999 0.416 0.113 3.69 ***
$5,000-$5,999 0.275 0.104 2.65 ***
$6,000-$6,999 0.115 0.093 1.24
$7,000-$7,999 0.169 0.090 1.88 *
$8,000-$8,999 0.054 0.091 0.59
Year effects Included
Month effects Included
State effects Included
Unobserved heterogeneity effects Included
Source: GAO analysis of NLSY79 data.
Note: In the final column an asterisk signifies statistical significance
at the 90 percent confidence level, two asterisks signify statistical
significance at the 95 percent confidence level, and three asterisks
signify statistical significance at the 99 percent confidence level. The
notation X * Y signifies an interaction between the variables X and Y. The
omitted category for industry is professional and related services and for
occupation is professional and technical workers. The omitted category for
BPE is $30,000 and above and for HQE it is $9,000 and above. Sample
includes 5,631 individuals with a total of 15,506 unemployment spells. The
maximized log likelihood value is -63,453.973.
Table 11: Parameter Estimates for Duration Equation from
Occupation-Interaction Specification
Parameter Standard
estimate error t-statistic
Past unemployment and UI receipt
Previous UI receipt 0.111 0.076 1.47
Previous unemployment 0.270 0.095 2.83 ***
Industry
Agriculture, forestry, and fishing -0.031 0.081 -0.38
Mining -0.105 0.111 -0.95
Construction 0.052 0.051 1.03
Manufacturing 0.013 0.042 0.32
Transportation and public utilities -0.003 0.062 -0.05
Wholesale and retail trade -0.069 0.039 -1.75 *
Finance, insurance, and real estate -0.062 0.068 -0.92
Business services -0.007 0.050 -0.14
Personal services -0.205 0.058 -3.51 ***
Entertainment and recreation services -0.005 0.076 -0.07
Public administration -0.153 0.072 -2.14 **
Valid missing 0.190 0.033 5.71 ***
Occupation
Managers and administrators 0.117 0.251 0.47
Sales workers 0.340 0.188 1.81 *
Clerical and unskilled workers 0.087 0.123 0.71
Craftsmen 0.444 0.146 3.03 ***
Machine operators (nontransportation) 0.287 0.118 2.42 **
Transportation equipment operators 0.240 0.206 1.16
Laborers (nonfarm) 0.231 0.136 1.70 *
Farmers, farm laborers, and foremen -0.002 0.234 -0.01
Service workers (excluding private 0.123 0.122 1.01
household)
Occupation * previous UI receipt
Managers and administrators 0.056 0.147 0.38
Sales workers -0.074 0.241 -0.31
Clerical and unskilled workers -0.063 0.101 -0.63
Craftsmen -0.117 0.099 -1.18
Machine operators (nontransportation) -0.053 0.092 -0.58
Transportation equipment operators -0.141 0.118 -1.20
Laborers (nonfarm) 0.079 0.108 0.73
Farmers, farm laborers, and foremen 0.017 0.202 0.09
Service workers (excluding private -0.065 0.110 -0.59
household)
Occupation * previous unemployment
Managers and administrators -0.199 0.258 -0.77
Sales workers -0.387 0.197 -1.96 **
Clerical and unskilled workers -0.202 0.125 -1.62
Craftsmen -0.422 0.149 -2.84 ***
Machine operators (nontransportation) -0.338 0.120 -2.80 ***
Transportation equipment operators -0.222 0.213 -1.04
Laborers (nonfarm) -0.308 0.138 -2.24 **
Farmers, farm laborers, and foremen -0.085 0.238 -0.36
Service workers (excluding private -0.249 0.125 -2.00 **
household)
UI program variables
Receiving UI -1.247 0.195 -6.41 ***
Weekly benefit amount (WBA) 0.028 0.035 0.79
WBA * receiving UI 0.066 0.059 1.12
Remaining UI benefit duration -0.014 0.008 -1.68 *
Waiting week for UI benefits 0.030 0.064 0.47
Denial rate for continuing UI claims 0.474 0.214 2.21 **
Extended UI benefits in effect 0.042 0.054 0.77
Personal characteristics
Years of education 0.240 0.070 3.43 ***
Armed Forces Qualifying Test score 0.242 0.056 4.31 ***
African-American -0.255 0.033 -7.74 ***
Hispanic -0.080 0.037 -2.16 **
Male -0.994 0.399 -2.49 **
Married -0.133 0.037 -3.59 ***
Married * male 0.289 0.049 5.91 ***
Age -5.011 2.825 -1.77 *
Age-squared 8.897 5.061 1.76 *
Age * male 7.571 2.917 2.60 *
Age-squared * male -13.275 5.164 -2.57 **
Lives in SMSA (urban) -0.041 0.027 -1.51
Health limitations -0.096 0.055 -1.75 *
Spouse used UI in past 0.133 0.051 2.62 ***
Live with parents -0.042 0.051 -0.83
Family size -0.120 0.089 -1.35
Live with parents * family size 0.112 0.135 0.83
Live with parents * family size * Male 0.023 0.094 0.24
Children under age 2 -0.095 0.033 -2.83 ***
Number of children -0.004 0.014 -0.30
Recent employment experience
Union member 0.087 0.029 2.94 ***
Tenure 0.050 0.018 2.75 ***
Tenure-squared -0.007 0.002 -3.03 ***
Lost job due to plant closing -0.175 0.046 -3.80 ***
State unemployment rate (time varying) -0.029 0.007 -4.02 ***
Base period earnings brackets
Under $2,000 -0.387 0.103 -3.76 ***
$2,000-$3,999 -0.365 0.088 -4.16 ***
$4,000-$5,999 -0.359 0.080 -4.49 ***
$6,000-$7,999 -0.285 0.078 -3.65 ***
$8,000-$9,999 -0.236 0.074 -3.18 ***
$10,000-$11,999 -0.161 0.072 -2.24 **
$12,000-$13,999 -0.187 0.068 -2.73 ***
$14,000-$15,999 -0.133 0.066 -2.02 **
$16,000-$17,999 -0.134 0.064 -2.10 **
$18,000-$19,999 -0.097 0.066 -1.48
$20,000-$24,999 -0.048 0.055 -0.86
$25,000-$29,999 -0.134 0.053 -2.52 **
High quarter earnings brackets
$0-$999 -0.063 0.115 -0.55
$1,000-$1,999 0.096 0.096 0.99
$2,000-$2,999 0.053 0.083 0.63
$3,000-$3,999 0.084 0.074 1.14
$4,000-$4,999 0.027 0.066 0.41
$5,000-$5,999 0.022 0.062 0.36
$6,000-$6,999 0.011 0.057 0.19
$7,000-$7,999 -0.033 0.057 -0.58
$8,000-$8,999 -0.032 0.056 -0.57
Time interactions (t = number of weeks
unemployed)
UI Receipt * Extended Benefits * (t-1) -1.582 0.734 -2.16 **
UI Receipt * Extended Benefits * 3.496 2.805 1.25
(t-1)-squared
UI Receipt * Extended Benefits * -0.191 0.244 -0.78
(t-1)-cubed
Remaining UI benefit duration * (t-1) 0.572 0.201 2.85 ***
Remaining UI benefit duration * -3.185 2.626 -1.21
(t-1)-squared
Remaining UI benefit duration * -0.257 0.955 -0.27
(t-1)-cubed
UI receipt * (t-1) 10.722 1.976 5.43 ***
UI receipt * (t-1)-squared -24.286 5.667 -4.29 ***
UI receipt * (t-1)-cubed 1.577 0.438 3.60 ***
UI receipt * WBA * (t-1) -1.049 0.680 -1.54
UI receipt * WBA * (t-1)-squared 2.455 2.051 1.20
UI receipt * WBA * (t-1)-cubed -0.145 0.161 -0.90
Year effects Included
Month effects Included
State effects Included
Unobserved heterogeneity effects Included
Source: GAO analysis of NLSY79 data.
Note: In the final column an asterisk signifies statistical significance
at the 90 percent confidence level, two asterisks signify statistical
significance at the 95 percent confidence level, and three asterisks
signify statistical significance at the 99 percent confidence level. The
notation X * Y signifies an interaction between the variables X and Y. The
omitted category for industry is professional and related services and for
occupation is professional and technical workers. The omitted category for
BPE is $30,000 and above and for HQE it is $9,000 and above. Sample
includes 5,631 individuals with a total of 15,506 unemployment spells. The
maximized log likelihood value is -63,453.973.
Industry-Interaction Specification
UI Receipt Equation
Table 8 summarizes the parameter estimates for the UI receipt equation of
the industry-interaction specification. A positive parameter estimate for
a variable implies that an increase in the variable increases the
likelihood of UI benefit receipt. A negative parameter estimate implies
that an increase in the variable decreases the likelihood of UI benefit
receipt. For example, the parameter estimate for years of education is
0.569, meaning that unemployed individuals with more years of education
have a higher likelihood of receiving UI benefits than otherwise similar
individuals with fewer years of education. The single asterisk signifies
that the parameter estimate for years of education is statistically
significant at the 95 percent confidence level.
The results in table 8 show that the number of prior unemployment periods
and the number of prior UI benefit receipt periods are strong predictors
of an unemployed individual's likelihood of receiving UI benefits. The
parameter estimate for the number of prior unemployment periods is -0.072,
which indicates that each additional prior unemployment period experienced
by an individual reduces the likelihood of UI benefit receipt during
current unemployment. Alternatively, the parameter estimate for the number
of prior UI receipt periods is 0.714, which indicates that each additional
prior UI receipt period experienced by an individual increases the
likelihood of UI benefit receipt during current unemployment.
The fact that the parameter estimate for the number of previous UI receipt
periods is larger in absolute value suggests that this is the stronger of
the two effects. To illustrate the magnitude of the effects, table 12
presents simulations of the likelihood of UI receipt by past unemployment
and past UI receipt experience. According to the table, the average
simulated likelihood of UI receipt for unemployed individuals with one
previous unemployment period is 48 percent if UI was received in the
previous unemployment period, but only 30 percent if UI was not received
in the previous unemployment period.23 Thus, for individuals with one
previous unemployment period, the average likelihood of UI receipt is 60
percent higher (18 percentage points) for those who received UI benefits
in their previous unemployment period. The remainder of the table shows
that UI receipt exhibits significant occurrence dependence. Specifically,
an individual who does not receive UI benefits during unemployment becomes
less likely to receive them during future unemployment, while an
individual who does receive UI benefits during unemployment becomes more
likely to receive them during future unemployment.24 Our model and data do
not allow us to determine the underlying reasons for these associations.
There are several possible reasons for the strong relationship between
past UI receipt and current UI receipt, however. If unemployed people do
not know they are eligible for benefits, or think that UI benefits are not
worth the effort to apply, or are overoptimistic about finding employment,
then there may be a "learning effect" that results from having received UI
benefits which increases the likelihood of future use. Alternatively, if
people do not apply for benefits because of a misperception of UI as a
welfare program, then having received benefits once may soften such an
outlook and increase the likelihood of future use.
23Note that the average simulated likelihood of UI receipt for first-time
unemployed workers is 33 percent.
24See Brian P. McCall, "Repeat Use of Unemployment Insurance," in Laurie
J. Bassi and Stephen A. Woodbury, editors, Long-Term Unemployment and
Reemployment Policies (Stamford, Connecticut: JAI Press, Inc., 2000).
Table 12: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers during Successive Periods of Unemployment, by Past UI Receipt
Status
Simulated likelihood of receiving UI benefits
(percent)
Always received UI benefits Never received UI
Unemployment period previously benefits previously
First - 33
Second 48 30
Third 64 28
Fourth 78 25
Fifth 88 23
Sixth 94 21
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI by
unemployment period for two extreme cases: (1) individuals always received
UI benefits during previous unemployment, and (2) individuals never
received UI during previous unemployment. N/A denotes that there is no
applicable value. See accompanying text for details.
The results in table 8 also show that the likelihood of UI benefit receipt
varies by the industry of the job lost by unemployed individuals. The
industry variable is categorical in nature, so the parameter estimate for
a particular category is an estimate of the effect of being in that
category relative to an omitted category. The omitted category for
industry is professional and related services. Table 8 shows that
unemployed individuals from the mining, manufacturing, public
administration, wholesale and retail trade, agriculture, forestry and
fishing, business services, and construction industries are more likely to
receive UI benefits than similar individuals from the professional
services industry, because their parameter estimates are positive and
statistically significant. To illustrate the magnitudes of these
differences, table 13 presents the average simulated likelihood of UI
receipt by industry under the specific assumption of first-time
unemployment. The average simulated likelihood of UI receipt during
first-time unemployment is 45.6 percent for unemployed miners, but only
24.3 percent for unemployed professional service workers. Table 13 clearly
demonstrates that there are significant differences across industries in
unemployed individuals' likelihoods of UI benefit receipt during
first-time unemployment.
Table 13: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers from Different Industries
Simulated likelihood of receiving UI
Industry benefits (percent)
Mining 46
Manufacturing 40
Public administration 37
Wholesale and retail trade 35
Agriculture, forestry, and fishing 34
Business services 31
Construction 31
Finance, insurance, and real estate 31
Transportation and public utilities 29
Entertainment and recreation 26
services
Professional and related services 24
Personal services 23
All industries 33
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI during
first-time unemployment. The parameter estimates for the mining,
manufacturing, public administration, wholesale and retail trade,
agriculture, forestry, and fishing, business services, and construction
industries are statistically significant relative to the professional and
related services industry at the 95 percent confidence level. See
accompanying text for details.
To test whether or not the effects of previous experience with
unemployment and UI receipt differ by industry, we also included the
industry categories interacted with both the number of previous
unemployment periods and the number of previous UI receipt periods. As was
the case above, the parameter estimates are calculated relative to the
omitted professional and related services industry. The parameter
estimates for the industry interactions with the number of prior
unemployment periods indicate that unemployed individuals from the mining,
manufacturing, and wholesale and retail trade industries exhibit stronger
occurrence dependence than unemployed individuals from the professional
services industry.25 That is, each additional previous unemployment period
has a stronger negative effect on the likelihood of receiving UI benefits
for unemployed individuals from these three industries relative to similar
individuals from the professional services industry.26
The parameter estimates for the industry interactions with the number of
previous UI receipt periods show that unemployed individuals from the
agriculture and construction industries exhibit weaker occurrence
dependence than individuals from the professional and related services
industry.27 That is, each additional previous UI receipt period has a
weaker positive effect on the likelihood of receiving UI benefits for
unemployed individuals from these three industries relative to similar
individuals from the professional services industry. Unemployed
individuals from the manufacturing industry also have weaker occurrence
dependence, but the result is only statistically significant at the 90
percent confidence level. Unemployed individuals from the public
administration industry exhibit stronger occurrence dependence than
individuals from the professional services industry. A similar result
occurs for unemployed workers from the finance, insurance, and real estate
industry, but the result is only statistically significant at the 90
percent confidence level. The other industries showed no statistically
significant effects compared to those from the professional services
industry.28
To illustrate the magnitudes of these differences, table 14 presents the
average simulated likelihood of UI receipt by industry and by the number
of previous unemployment and UI receipt periods. Column 1 presents the
simulations for first-time unemployment (see table 13). Column 2 presents
the simulations assuming one prior unemployment period with UI receipt.
Column 3 presents the simulations assuming two prior unemployment periods,
both with UI receipt. Table 14 shows that, although unemployed individuals
from the mining and manufacturing industries have the highest average
simulated likelihoods of UI receipt for first-time unemployment, this is
not the case if individuals have received UI benefits previously. For
unemployed individuals with two prior UI receipt periods, those from the
public administration, wholesale and retail trade, entertainment services,
transportation, and business services industries are about as likely or
are more likely to receive UI benefits again than similar individuals from
the mining and manufacturing industries. Administrative unemployment
insurance data have shown that repeat UI recipients tend to be from
industries that are more seasonal, such as manufacturing and construction.
Our results, however, suggest that this is not because workers from these
industries who have received UI before are more likely to receive UI
benefits when they become unemployed than similar workers from other
industries. Rather, it may be that workers from such seasonal industries
are unemployed more often on average than workers from other industries,
or that a larger fraction of unemployed workers from such industries have
collected UI previously.
25As stated above, the occurrence dependence in this case relates to the
fact that an individual who does not receive UI benefits during
unemployment becomes less likely to receive them during future
unemployment.
26Although the results for some industries were not individually
statistically significant, a likelihood ratio test of the joint hypothesis
that all of the interaction terms between industry and past unemployment
experience are equal to zero is rejected at the 95 percent confidence
level.
27As stated earlier, occurrence dependence relating to previous UI receipt
means that an individual who receives UI benefits during unemployment
becomes more likely to receive them during future unemployment.
28However, a likelihood ratio test of the joint hypothesis that all of the
interaction terms between industry and past UI receipt experience are
equal to zero is rejected at the 95 percent confidence level.
Table 14: Simulated Likelihood of Receiving UI Benefits during Different
Periods of UI-Eligible Unemployment for Workers with Past UI Receipt, by
Industry
Simulated likelihood of receiving UI benefits during a
current UI-eligible unemployment period, given past UI
receipt (percent)
Second Third
Firstunemployment unemployment unemployment
Industry perioda period period
Mining 46 57 69
Manufacturing 40 52 65
Public 37 68 91
administration
Wholesale and 35 52 70
retail trade
Agriculture, 34 42 50
forestry, and
fishing
Business services 31 48 66
Construction 31 40 51
Finance, 31 64 91
insurance, and
real estate
Transportation and 29 46 66
public utilities
Entertainment and 26 45 67
recreation
services
Professional and 24 39 58
related services
Personal services 23 38 56
All industries 33 48 64
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI during a
first unemployment period, a second unemployment period with UI receipt
during the prior unemployment period, and a third unemployment period with
UI receipt during both prior unemployment periods. The positive effect
that each prior UI receipt period has on the likelihood of current UI
receipt is statistically significantly larger for the public
administration industry relative to the professional and related services
industry at the 95 percent confidence level, and smaller for the
agriculture and construction industries. The simulations also incorporate
the industry effects and the industry interactions with the number of
prior periods of unemployment. See accompanying text for details.
aWorkers experiencing their first period of unemployment did not have past
UI receipt.
Our model also controls for UI program factors, but the results in table 8
show that after controlling for other observable characteristics, these
factors had no statistically significant impact on an unemployed
individual's likelihood of UI receipt. These program factors include the
estimated amount of weekly benefits an unemployed individual was eligible
to receive, the estimated duration of those benefits, and the
state-specific denial rate for new UI claims.29 Weekly benefits and the
potential duration of benefits are functions of earnings, which we
controlled for (and are discussed below).
The parameter estimates in table 8 also show that a number of personal
characteristics are associated with an unemployed individual's likelihood
of UI benefit receipt, including education, age, and gender. For instance,
the parameter estimate on years of education is 0.569, which indicates
that each year of education increases an unemployed individual's
likelihood of receiving UI benefits. The direction of the age effect on
the likelihood of UI benefit receipt is difficult to interpret from the
parameter estimates in table 8, because it is included as a polynomial to
allow for nonlinear effects. Figure 10 presents a graph of the average
simulated likelihood of UI receipt by age for the specific case of
first-time unemployment. The graph shows that the likelihood of UI receipt
increases until about the age of 25 and then decreases thereafter. For
example, the average simulated likelihood of UI receipt during first-time
unemployment for 25-year-olds is 10 percentage points (39 percent) higher
than for 35-year-olds. While other research has found that older
individuals are more likely to receive UI benefits, other researchers
generally do not control for individuals' past unemployment and UI receipt
experience as completely as we did.30 Because age and experience with both
unemployment and UI receipt are correlated, age may act as a proxy for
these experience measures when they are not controlled for.
29Other researchers have found that the weekly benefit amount does not
affect UI receipt. See Gritz and MaCurdy.
30See McCall.
Figure 10: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Age
Note: Simulations are the average likelihood of receiving UI during
first-time unemployment at different ages. The overall average likelihood
of receiving UI during first-time unemployment is 33 percent. See
accompanying text for details.
Table 8 also shows that several measures relating to the recent employment
experience of unemployed individuals (excluding industry and occupation,
which are discussed elsewhere) affect an unemployed individual's
likelihood of UI benefit receipt. For instance, table 8 shows that an
unemployed individual's likelihood of receiving UI benefits increases with
earnings. We include two earnings measures: base period earnings and high
quarter earnings. Each measure is grouped in earnings brackets and entered
into the equation as a categorical variable to reflect nonlinear effects.
As was the case with industry, each estimated effect is relative to an
omitted category. For BPE the omitted earnings bracket is $30,000 and
above and for HQE the omitted bracket is $9,000 and above. The pattern of
parameter estimates for BPE shows that an unemployed individual is more
likely to receive UI benefits, the higher his BPE (at least up to
$20,000). The pattern of parameter estimates for HQE shows that an
unemployed individual is more likely to receive UI benefits if his HQE are
between $2,000 and $6,000. Figure 11 presents a graph of the average
simulated likelihood of receiving UI benefits by base period earnings for
the specific case of first-time unemployment. The level of HQE is varied
to maintain a ratio of HQE to BPE of about 25 percent to approximate
steady employment during the base period. The figure shows that unemployed
individuals who earned more than $14,000 in their base period had a
likelihood of UI receipt of over 40 percent, while individuals who earned
less than $6,000 had a likelihood of UI receipt of less than 20 percent.
Figure 11: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers, by Prior-Year Earnings
Note: Simulations are for the average likelihood of receiving UI during
first-time unemployment at different levels of earnings. The overall
average likelihood of receiving UI during first-time unemployment is 33
percent. See accompanying text for details.
Table 8 shows that employment experience measures other than earnings also
affect the likelihood of UI receipt. For instance, an individual's
likelihood of UI receipt increases with tenure up to 9 years, after which
it decreases. Also, an individual's likelihood of UI receipt increases as
the state unemployment rate increases. Interestingly, the parameter
estimate on the plant closing variable is a statistically significant
-0.263, indicating that unemployed individuals are less likely to receive
UI benefits if they lost their jobs because of a plant closing. Union
status does not have a statistically significant effect on an unemployed
individual's likelihood of UI receipt.
Unemployment Duration Equation
Table 9 summarizes the parameter estimates for the unemployment duration
equation of the industry interaction specification. A positive parameter
estimate implies that an increase in a variable increases the escape rate
from unemployment, thereby decreasing the duration of unemployment. A
negative parameter estimate implies that an increase in a variable
decreases the escape rate from unemployment, thereby increasing the
duration of unemployment. For example, the parameter estimate for years of
education is a statistically significant 0.235, which implies that
unemployed individuals with more years of education have higher escape
rates from unemployment than otherwise similar individuals with fewer
years of education. As a result, unemployed individuals with more years of
education will tend to have shorter unemployment durations than those with
fewer years of education.
We found that after controlling for other observable characteristics, the
single most important predictor of unemployment duration is whether or not
an individual receives UI benefits during the current unemployment period.
The parameter estimate on the dummy variable for UI receipt status is
-1.256, which implies that receiving UI benefits while unemployed reduces
an individual's escape rate from unemployment, thereby increasing
unemployment duration. Simulations show that the median duration of
unemployment is 8 weeks for individuals who do not receive UI benefits,
but 21 weeks when they do receive UI benefits. We also allowed the effect
of UI receipt to vary with the number of weeks of unemployment. These
results indicate that a UI recipient's escape rate from unemployment
increases until about the 33rd week of unemployment. After 33 weeks, the
escape rate decreases again until about the 72nd week, and then increases
until 100 weeks.31
The parameter estimates in table 9 show that having experienced prior
unemployment or prior UI receipt has no statistically significant effect
on unemployment duration. This result, however, is conditional upon
whether or not an individual currently receives UI benefits. The
unconditional effect of having previously received UI benefits is to
increase unemployment duration. As stated earlier, we found that
unemployed individuals who have previously received UI benefits are
significantly more likely to receive UI benefits during current
unemployment. Because those individuals who receive UI benefits during
unemployment have longer unemployment duration, the unconditional effect
of having previously received UI benefits is to increase unemployment
duration.
31Changes in the escape rate over an unemployment period are also affected
by the other time-interaction effects included in the specification.
However, these other effects do not affect the general shape of this
overall trend.
Table 9 also shows that there is an association between the industry from
which an individual lost a job and the duration of unemployment. As in the
UI receipt equation, the omitted category for industry is professional and
related services. Table 9 shows that unemployed individuals from the
construction and manufacturing industries have higher escape rates from
unemployment than otherwise similar individuals from the professional
services industry, because their parameter estimates are positive and
statistically significant. The parameter estimate for business services is
also positive, but is only statistically significant at the 90 percent
confidence level. The effects for the other industries are not
statistically significant relative to the professional services industry.
To illustrate the magnitudes of these differences, table 15 presents the
median simulated duration of unemployment by industry for the specific
case of first-time unemployment. The median duration is about 17 and 19
weeks, respectively, for unemployed individuals from the construction and
manufacturing industries who receive UI benefits, but is about 24 weeks
for those from the professional services industry.
Table 15: Simulated Unemployment Duration for UI-Eligible Workers, by
Industry and UI Receipt Status
Simulated unemployment duration (median
weeks)
Industry Not receiving UI
Receiving UI benefits benefits
Construction 17 6
Mining 17 6
Business services 18 7
Manufacturing 19 7
Finance, insurance, and real
estate 21 8
Wholesale and retail trade 22 9
Public administration 23 9
Professional and related services 24 10
Entertainment and related services 24 10
Personal services 24 10
Agriculture, forestry, and fishing 26 11
Transportation and public
utilities 27 12
Overall average duration 21 8
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the median duration of unemployment during
first-time unemployment. The parameter estimates for the construction and
manufacturing industries are statistically significant relative to the
professional and related services industry at the 95 percent confidence
level. See accompanying text for details.
To test whether or not the effects of previous experience with
unemployment and UI receipt on the duration of unemployment differ by
industry, we also included the industry categories interacted with the
indicators for both previous unemployment and previous UI receipt. As
stated above, the effects are relative to the omitted category of
professional and related services. The parameter estimates in table 9
indicate that there are no statistically significant differences across
industry types by previous experience with unemployment or previous UI
receipt, conditional upon current UI receipt status.32
Table 9 also shows that only one UI program factor (other than current UI
receipt) has a statistically significant impact on an individual's
unemployment duration. Specifically, individuals who are unemployed in
states with higher denial rates for continuing UI claims have higher
escape rates from unemployment. That is, these individuals tend to become
reemployed more quickly than those in states with lower denial rates.
The parameter estimates in table 9 show that a number of personal
characteristics affect an individual's unemployment duration, including
education, race, gender, and marital status. For example, the parameter
estimate on years of education is 0.235, which indicates that each year of
education increases an individual's escape rate from unemployment. The
simulations reported in table 16 show that unemployed individuals with 16
years of education (roughly a college education) have median unemployment
duration that is about 1.9 weeks shorter than unemployed individuals with
12 years of education when UI benefits are received, and 1.1 weeks when UI
benefits are not received. The parameter estimates for race show that
African-Americans have significantly lower escape rate from unemployment
than Hispanics, who in turn have slightly lower escape rates than whites.
Table 17 displays simulations of median unemployment duration by race for
the specific case of first-time unemployment. Simulations showed that the
age effect, although statistically significant, did not have much of an
impact on the median duration of unemployment. In table 9, the parameter
estimates for gender are difficult to interpret because gender is
interacted with other variables in our specification, including age.
Simulations show that unemployed men have median unemployment durations
that are about 2 weeks shorter than for unemployed women when UI benefits
are received; and about 1 week shorter when UI benefits are not received.
The parameter estimates for marital status show that married women tend to
have longer unemployment durations than do unmarried women and married men
tend to have shorter unemployment durations than do unmarried men.33
Although the age effects in table 9 are statistically significant,
simulations showed that age had minimal effect on the median duration of
unemployment.
32However, a likelihood ratio test of the joint hypotheses that all of the
interaction terms between industry and past unemployment experience are
equal to zero is rejected at the 95 percent confidence level. A similar
test of the joint hypothesis that all of the interaction terms between
industry and past UI receipt experience are equal to zero could not be
rejected at the 95 percent confidence level.
Table 16: Simulated Unemployment Duration for UI-Eligible Workers, by
Education Level and UI Receipt Status
Simulated unemployment duration (median
weeks)
Years of education when Not receiving UI
unemployment began Receiving UI benefits benefits
9 22 9
10 22 9
11 21 9
12 21 8
13 20 8
14 20 8
15 19 8
16 19 7
17 18 7
18 18 7
19 18 6
20 and higher 17 6
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the median duration of unemployment during
first-time unemployment. Overall average duration is 21 weeks for
UI-eligible workers receiving UI benefits and 8 weeks for UI-eligible
workers not receiving UI benefits. See accompanying text for details.
33In alternative specifications we explored whether an individual's
likelihood of UI benefit receipt and unemployment duration were affected
by spousal income in the previous year. We found that spousal income had
no statistically significant effect on an individual's likelihood of UI
benefit receipt, but did slightly increased the duration of unemployment.
Table 17: Simulated Unemployment Duration for UI-Eligible Workers, by
Race/Ethnicity and UI Receipt Status
Simulated unemployment duration (median weeks)
Race or ethnicity Receiving UI benefits Not receiving UI benefits
White 19 8
Hispanic 21 8
African-American 25 11
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the median duration of unemployment during
first-time unemployment. Overall average duration is 21 weeks for
UI-eligible workers receiving UI benefits and 8 weeks for UI-eligible
workers not receiving UI benefits. See accompanying text for details.
The last set of parameter estimates in table 9 relates to the recent
employment experience of unemployed individuals (excluding industry and
occupation, which are discussed elsewhere). Most of the parameter
estimates in this grouping are statistically significant at the 95 percent
level. Specifically, unemployed individuals who belonged to a union at the
job that was lost had a higher escape rate from unemployment than
otherwise similar individuals who were not union members. The simulations
in table 18 show that union members had median unemployment durations that
were 2 weeks shorter than nonunion members when UI benefits were received
and 1 week shorter when UI benefits were not received. Simulations also
show that an individual's unemployment duration decreases modestly with
job tenure until 7 years, after which it increases slightly.
Table 18: Simulated Unemployment Duration for UI-Eligible Workers, by
Union Status and UI Receipt Status
Simulated unemployment duration (median
weeks)
Union memberships status when Receiving UI Not receiving UI
unemployment began benefits benefits
Union member 19 8
Not a union member 21 9
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the median duration of unemployment during
first-time unemployment. Overall average duration is 21 weeks for
UI-eligible workers receiving UI benefits and 8 weeks for UI-eligible
workers not receiving UI benefits. See accompanying text for details.
Of our two measures of an individual's earnings, only the base period
earnings proved to have a statistically significant effect on the duration
of unemployment.34 The pattern of parameter estimates for BPE shows that
unemployed individuals with low BPE have lower escape rates from
unemployment than otherwise similar individuals with higher BPE. That is,
lower-earning individuals tend to have longer unemployment periods. Figure
12 graphs simulations of median unemployment duration by BPE for the
specific case of first-time unemployment.35 Individuals with BPE below
$6,000 tend to have longer unemployment duration than unemployed
individuals with higher BPE.
34Recall that each measure was broken into earnings brackets and entered
into the equation as a categorical variable. See tables 8,9,10, or 11 for
the brackets used. The omitted category for BPE is $30,000 and above and
the omitted category for high quarter earnings is $9,000 and above.
35For comparability, the simulations in figure 12 hold the ratio of HQE to
BPE as closely as possible to 25 percent.
Figure 12: Simulated Unemployment Duration for UI-Eligible Workers, by
Prior-Year Earnings and UI Receipt Status
Note: Simulations are the median duration of unemployment during
first-time unemployment. Overall average duration is 21 weeks for
UI-eligible workers receiving UI benefits and 8 weeks for UI-eligible
workers not receiving UI benefits. See accompanying text for details.
Occupation-Interaction Specification
We also estimated a specification of our model with interaction effects
between the occupation categories (as opposed to industry) and our
measures of past unemployment and past UI receipt experience. These
results are presented in tables 10 and 11. A comparison of these results
with those from tables 8 and 9 shows that the overall results of the two
specifications are very similar. Therefore, only the occupation estimates
will be discussed here.
Because occupation is included as a categorical variable, the parameter
estimates are relative to an omitted group, which is professional and
technical workers. The estimates in table 10 show that unemployed
managers, machine operators, craftsmen, laborers, transportation workers,
and clerical workers are more likely to receive UI benefits than similar
professional and technical workers. Table 19 presents the average
simulated likelihood of receiving UI benefits by occupation for the
specific case of first-time unemployment. Although the range is not as
wide as for industry (see table 13), the table shows that there are
differences in the likelihood of UI receipt by occupation.
Table 19: Simulated Likelihood of Receiving UI Benefits for UI-Eligible
Workers from Different Occupations
Simulated likelihood of receiving UI
Occupation benefits (percent)
Managers and administrators 39
Farmers, farm laborers and foremen 38
Machine operators (nontransportation) 38
Craftsmen 35
Laborers (nonfarm) 34
Transportation equipment operators 33
Clerical and unskilled workers 33
Service workers (excluding private
household) 28
Sales workers 28
Professional and technical workers 25
Overall average 33
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI during
first-time unemployment for workers from different occupations. The
parameter estimates for managers and administrators, farmers, farm
laborers, and foremen, machine operators, craftsmen, laborers,
transportation equipment operators, and clerical and unskilled workers are
statistically significant relative to professional and technical workers
at the 95 percent confidence level. See accompanying text for details.
The interactions between occupation and the number of previous
unemployment periods in table 10 indicate that unemployed machine
operators and laborers exhibit stronger occurrence dependence than
otherwise similar professional and technical workers.36 That is, each
additional previous unemployment period has a stronger negative effect on
the likelihood of receiving UI benefits for unemployed individuals from
these two occupations relative to similar professional and technical
workers.37
36As stated above, the occurrence dependence in this case relates to the
fact that an individual who does not receive UI benefits during
unemployment becomes less likely to receive them during future
unemployment.
The parameter estimates for occupation interacted with the number of
previous UI receipt periods show that unemployed transportation operators
and craftsmen exhibit weaker occurrence dependence than otherwise similar
professional and technical workers.38 That is, each additional previous UI
receipt period has a weaker positive effect on the likelihood of receiving
UI benefits for unemployed individuals from these two occupations relative
to otherwise similar individuals from professional and technical
occupations. Managers also showed weaker occurrence dependence, but this
estimate is only statistically significant at the 90 percent confidence
level. Unemployed sales workers and service workers exhibit stronger
occurrence dependence than otherwise similar professional and technical
workers. The other occupations showed no statistically significant effects
compared with professional and technical workers.39
To illustrate the magnitudes of these differences, table 20 presents the
average simulated likelihood of UI receipt by occupation and by the number
of previous UI receipt periods. Column 1 presents the simulations for
first-time unemployment (as in table 19). Column 2 presents the
simulations assuming one prior unemployment period with UI receipt. Column
3 presents the simulations assuming two prior unemployment periods, both
with UI receipt. Table 20 shows that although unemployed managers and
machine operators have among the highest average simulated likelihoods of
UI receipt for first-time unemployment, this is not the case if
individuals have received UI benefits previously. In the case of
unemployed individuals with two prior UI receipt periods, sales workers,
service workers, clerical workers, and farmers are about as likely, or are
more likely, to receive UI benefits than otherwise similar managers and
machine operators.
37Although the results for some occupations were not individually
statistically significant, a likelihood ratio test of the joint hypothesis
that all of the interaction terms between occupation and past unemployment
experience are equal to zero is rejected at the 95 percent confidence
level.
38As stated earlier, occurrence dependence relating to previous UI receipt
means that an individual who receives UI benefits during unemployment
becomes more likely to receive them during future unemployment.
39However, a likelihood ratio test of the joint hypothesis that all of the
interaction terms between occupation and past UI receipt experience are
equal to zero is rejected at the 95 percent confidence level.
Table 20: Simulated Likelihood of Receiving UI Benefits during Different
Periods of UI-Eligible Unemployment for Workers with Past UI Receipt, by
Occupation
Simulated likelihood of receiving UI benefits
during a current UI-eligible unemployment period,
given past UI receipt (percent)
First Second Third
unemployment unemployment unemployment
Occupation perioda period period
Managers and
administrators 39 52 65
Farmers, farm laborers,
and foremen 38 54 70
Machine operators
(nontransportation) 38 50 62
Craftsmen 35 46 56
Laborers (nonfarm) 34 45 58
Transportation equipment
operators 33 42 51
Clerical and unskilled
workers 33 53 73
Service workers
(excluding private
household) 28 50 74
Sales workers 28 66 94
Professional and
technical workers 25 39 56
Overall average 33 48 64
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the average likelihood of receiving UI during a
first unemployment period, a second unemployment period with UI receipt
during the prior unemployment period, and a third unemployment period with
UI receipt during both prior unemployment periods. The positive effect
that each prior UI receipt period has on the likelihood of current UI
receipt is statistically significantly larger for sales workers and
service workers relative to professional and technical workers at the 95
percent confidence level, and smaller for transportation equipment
operators and craftsmen. The simulations also incorporate the occupation
effects and the occupation interactions with the number of prior periods
of unemployment. See accompanying text for details.
aWorkers experiencing their first period of unemployment did not have past
UI receipt.
Table 11 shows that there is also an association between the occupation
from which an individual lost a job and the duration of unemployment.
Specifically, unemployed craftsmen and machine operators have higher
escape rates from unemployment than similar professional and technical
workers, because the estimates are positive and statistically significant.
The effects for the other occupations were not statistically significant
relative to professional and technical workers. To illustrate the
magnitudes of these differences, table 21 presents the median simulated
duration of unemployment by occupation for the specific case of first-time
unemployment. The median duration is under 20 weeks for unemployed
craftsmen and machine operators who receive UI, but is almost 26 weeks for
professional and technical workers.
Table 21: Simulated Unemployment Duration for UI-Eligible Workers, by
Occupation and UI Receipt Status
Simulated unemployment duration
(median weeks)
Receiving UI Not receiving UI
Occupation benefits benefits
Craftsmen 16 6
Sales workers 18 7
Machine operators (nontransportation) 19 7
Transportation equipment operators 20 8
Laborers (nonfarm) 20 8
Service workers (excluding private
household) 23 9
Managers and administrators 23 9
Clerical and unskilled workers 23 10
Farmers, farm laborers, and foremen 26 11
Professional and technical workers 26 11
Overall average duration 21 8
Source: GAO simulations based on GAO analysis of NLSY79 data.
Note: Simulations are the median duration of unemployment during
first-time unemployment for workers from different occupations. The
parameter estimates for craftsmen and machine operators are statistically
significant relative to professional and technical workers at the 95
percent confidence level. See accompanying text for details.
To test whether or not the effects of previous experience with
unemployment and UI receipt on the duration of unemployment differ by
occupation, we also included the occupation categories interacted with the
indicators for both previous unemployment and previous UI receipt. As
stated earlier, the effects are relative to the omitted category of
professional and technical workers. The parameter estimates in table 11
indicate that the interactions for prior unemployment are negative and
statistically significant for craftsmen, sales workers, machine operators,
laborers, and service workers. This suggests that unemployed workers from
these occupations have lower escape rates from unemployment relative to
professional and technical workers as the number of past unemployment
periods increases.40 The parameter estimates for the interactions between
occupation and past UI receipt showed no individual statistical
significance.41
Limitations of the Analysis
Although our analysis was performed using the most appropriate dataset and
methodology available, there are a number of limitations to the analysis
that could not be avoided and should be highlighted. Although the NLSY79
is the best available dataset for our purposes, it lacks some information
that could have improved our analysis. It does not provide information
about whether an unemployed individual attempted to collect UI benefits or
not, only whether the individual did collect benefits. It also does not
provide information about whether an individual was aware of his or her
eligibility for benefits. As a result, we had to estimate each unemployed
individual's UI-eligibility status. An unemployed worker's awareness of
the UI program and knowledge of its basic rules could have a large impact
on his or her decision to apply for benefits. This awareness may also be
correlated with other observable characteristics (education and earnings,
for example). Not controlling for awareness may affect the estimates of
such variables.
The NLSY79 also lacks information about an unemployed worker's former
employer that could help estimate UI receipt and unemployment duration.
Although our results control for industry, firms within an industry have
different labor turnover patterns that result in different UI tax rates
through experience rating.42 The lack of perfect experience rating may
even encourage firms to use temporary layoffs and recalls as a way of
managing its labor force during demand fluctuations.43 An individual who
works for a firm with high labor turnover or with a high UI tax rate may
be more aware of the UI program and, thus, more likely to receive
benefits.
40However, a likelihood ratio test of the joint hypothesis that all of the
interaction terms between occupation and past unemployment experience are
equal to zero could not be rejected at the 95 percent confidence level.
41In addition, a likelihood ratio test of the joint hypotheses that all of
the interaction terms between occupation and past UI receipt experience
are equal to zero could not be rejected at the 95 percent confidence
level.
42Experience rating describes the practice of making a firm's UI tax rate
a function of the amount of UI benefits paid to its former employees.
Another limitation of the NLSY79 is that it includes only information
about the specific group of individuals who were between the ages of 14
and 22 in 1979. Thus, any findings based on the NLSY79 are specific to
this group and do not represent the experiences of workers of all ages
during the 1979-2002 period.
A methodological limitation is that we assume that the time between
unemployment spells is fixed. One might expect individuals who have been
unemployed and received UI benefits to change their subsequent work
behavior, either to increase or decrease their chances of using the
program in the future. For example, a person who received UI benefits
while unemployed may search for more stable employment in order to reduce
the likelihood of experiencing a layoff in the future. We do not
incorporate such possibilities into our model because this would require a
third equation to model employment duration, which would be a more complex
and time-consuming analysis.
43See Martin Feldstein, "Temporary Layoffs in the Theory of Unemployment,"
The Journal of Political Economy, vol. 84, no. 5 (1976).
Appendix II: Comments from the Department of Labor Appendix II: Comments
from the Department of Labor
Appendix III: GAOA Appendix III: GAO Contact and Staff Acknowledgment
GAO Contact
Sigurd R. Nilsen, Director, (202) 512-7215
Staff Acknowledgments
In addition to the contact named above, Brett Fallavollita, Assistant
Director, Regina Santucci, James Pearce, Bill Bates, Gale Harris, Gene
Kuehneman, Jonathan McMurray, Edward Nannenhorn, Dan Schwimer, Shana
Wallace, and Daniel G. Williams made major contributions to this report.
Other Acknowledgments
We contracted with Dr. Brian McCall from the University of Minnesota for
analysis of the NLSY79 and other technical assistance.
Bibliography
Anderson, Patricia M., and Bruce D. Meyer. "The Effect of Unemployment
Insurance Taxes and Benefits on Layoffs Using Firm and Individual Data,"
NBER Working Paper No. 4960. Cambridge, Massachusetts: National Bureau of
Economic Research, 1994.
Berndt, E. K., B. H. Hall, R. E. Hall, and J. A. Hausman. "Estimation and
Inference in Nonlinear Structural Models," Annals of Economic and Social
Measurement, Vol. 3, No. 4 (1974): 653-665.
Blank, Rebecca M., and David E. Card. "Recent Trends in Insured and
Uninsured Unemployment: Is There an Explanation?" The Quarterly Journal of
Economics, Vol. 106, No. 4 (1991): 1157-1189.
Calvo-Armengol, Antoni and Matthew O. Jackson. "The Effects of Social
Networks on Employment and Inequality." The American Economic Review, Vol.
94, No. 3 (2004): pp. 426-454.
Card, David E., and Phillip B. Levine. "Unemployment Insurance Taxes and
the Cyclical and Seasonal Properties of Unemployment," Journal of Public
Economics, Vol. 53, No. 1 (1994): 1-29.
Feldstein,Martin. "Temporary Layoffs in the Theory of Unemployment," The
Journal of Political Economy, Vol. 84, No. 5 (1976): 937-958.
Gritz, R. Mark, and Thomas MaCurdy. "Measuring the Influence of
Unemployment Insurance on Unemployment Experiences," Journal of Business
and Economic Statistics, Vol. 15, No. 2 (1997): 130-152.
Gruber, Jonathan. "The Wealth of the Unemployed," Industrial and Labor
Relations Review, Vol. 55, No. 1 (2001): 79-94.
Krueger, Alan B., and Bruce D. Meyer. "Labor Supply Effects of Social
Insurance," NBER Working Paper 9014. Cambridge, Massachusetts: National
Bureau of Economic Research, 2002.
McCall, Brian P. "Repeat Use of Unemployment Insurance," in Laurie J.
Bassi and Stephen A. Woodbury, editors, Long-Term Unemployment and
Reemployment Policies. Stamford, Connecticut: JAI Press, Inc., 2000.
Meyer, Bruce D. "Unemployment Insurance and Unemployment Spells,"
Econometrica, Vol. 58, No. 4 (1990): 757-782.
Meyer, Bruce D., and Dan T. Rosenbaum. "Repeat Use of Unemployment
Insurance," NBER Working Paper 5423. Cambridge, Massachusetts: National
Bureau of Economic Research, 1996.
Mortensen, Dale T. "Unemployment Insurance and Job Search Decisions,"
Industrial and Labor Relations Review, Vol. 30, No. 4 (1977): 505-517.
Needels, Karen E., and Walter Nicholson. An Analysis of Unemployment
Insurance Durations since the 1990-1992 Recession. Prepared for the
Department of Labor. 1999.
O'Leary, Christopher J., and Stephen A. Wandner, editors. Unemployment
Insurance in the United States: Analysis of Policy Issues. Kalamazoo,
Michigan: W. E. Upjohn Institute for Employment Research, 1997.
Topel, Robert H. "On Layoffs and Unemployment Insurance," The American
Economic Review, Vol. 73, No. 4 (1983): 541-559.
Related GAO Products Related GAO Products
Unemployment Insurance: Better Data Needed to Assess Reemployment Services
to Claimants. GAO-05-413 . Washington, D.C.: June 24, 2005.
Unemployment Insurance: Information on Benefit Receipt. GAO-05-291 .
Washington, D.C.: March 17, 2005.
Women's Earnings: Work Patterns Partially Explain Difference between Men's
and Women's Earnings. GAO-04-35 . Washington, D.C.: October 31, 2003.
Unemployment Insurance: States' Use of the 2002 Reed Act Distribution.
GAO-03-496 . Washington, D.C.: March 6, 2003.
Unemployment Insurance: Enhanced Focus on Program Integrity Could Reduce
Overpayments. GAO-02-820T . Washington, D.C.: June 11, 2002.
Unemployment Insurance: Increased Focus on Program Integrity Could Reduce
Billions in Overpayments. GAO-02-697 . Washington, D.C.: July 12, 2002.
Unemployment Insurance: Role as Safety Net for Low-Wage Workers Is
Limited. GAO-01-181 . Washington, D.C.: December 29, 2000.
(130465)
GAO's Mission
The Government Accountability Office, the audit, evaluation and
investigative arm of Congress, exists to support Congress in meeting its
constitutional responsibilities and to help improve the performance and
accountability of the federal government for the American people. GAO
examines the use of public funds; evaluates federal programs and policies;
and provides analyses, recommendations, and other assistance to help
Congress make informed oversight, policy, and funding decisions. GAO's
commitment to good government is reflected in its core values of
accountability, integrity, and reliability.
Obtaining Copies of GAO Reports and Testimony
The fastest and easiest way to obtain copies of GAO documents at no cost
is through GAO's Web site ( www.gao.gov ). Each weekday, GAO posts newly
released reports, testimony, and correspondence on its Web site. To have
GAO e-mail you a list of newly posted products every afternoon, go to
www.gao.gov and select "Subscribe to Updates."
Order by Mail or Phone
The first copy of each printed report is free. Additional copies are $2
each. A check or money order should be made out to the Superintendent of
Documents. GAO also accepts VISA and Mastercard. Orders for 100 or more
copies mailed to a single address are discounted 25 percent. Orders should
be sent to:
U.S. Government Accountability Office 441 G Street NW, Room LM Washington,
D.C. 20548
To order by Phone: Voice: (202) 512-6000 TDD: (202) 512-2537 Fax: (202)
512-6061
To Report Fraud, Waste, and Abuse in Federal Programs
Contact:
Web site: www.gao.gov/fraudnet/fraudnet.htm E-mail: fraudnet@gao.gov
Automated answering system: (800) 424-5454 or (202) 512-7470
Congressional Relations
Gloria Jarmon, Managing Director, JarmonG@gao.gov (202) 512-4400 U.S.
Government Accountability Office, 441 G Street NW, Room 7125 Washington,
D.C. 20548
Public Affairs
Paul Anderson, Managing Director, AndersonP1@gao.gov (202) 512-4800 U.S.
Government Accountability Office, 441 G Street NW, Room 7149 Washington,
D.C. 20548
www.gao.gov/cgi-bin/getrpt? GAO-06-341 .
To view the full product, including the scope
and methodology, click on the link above.
For more information, contact Sigurd Nilsen at (202) 512-7215 or
nilsens@gao.gov.
Highlights of GAO-06-341 , a report to the Chairman, Subcommittee on Human
Resources, Committee on Ways and Means, House of Representatives
March 2006
UNEMPLOYMENT INSURANCE
Factors Associated With Benefit Receipt
Certain characteristics are associated with the likelihood of receiving UI
benefits and unemployment duration. UI-eligible workers that GAO studied
are more likely to receive UI benefits if they have higher earnings prior
to becoming unemployed, are younger, have more years of education, or if
they have a history of past UI benefit receipt when compared with
otherwise similar workers. GAO found that past experience with the UI
program has a particularly strong effect on the future likelihood of
receiving UI benefits. However, some characteristics, such as receiving a
higher maximum weekly UI benefit amount, are not associated with a greater
likelihood of receiving UI benefits. UI-eligible workers who receive UI
benefits have longer unemployment duration than workers with similar
characteristics. Also, UI-eligible workers are more likely to experience
longer unemployment duration if they have lower earnings before becoming
unemployed or have fewer years of education. Other characteristics
associated with longer unemployment duration include being
African-American, female, or not belonging to a union. GAO found no
relationship between past UI benefit receipt and subsequent unemployment
duration.
UI-eligible workers from certain industries are more likely than similar
workers in other industries to receive UI benefits and experience shorter
unemployment duration. Specifically, GAO's simulations show that the
likelihood of receiving UI benefits during a first period of unemployment
is highest among workers from the mining and manufacturing industries.
Furthermore, the likelihood of receiving UI benefits when unemployed
increases with each previous period of UI receipt across all industries,
and the most notable increase occurs in public administration. First-time
unemployed workers from construction and manufacturing experience
significantly shorter unemployment duration than workers from other
industries.
Simulated UI Benefit Receipt Rates for UI-Eligible Workers during
Successive Periods of Unemployment, by Past UI Receipt Status
Unemployment Insurance (UI), established in 1935, is a complex system of
53 state programs that in fiscal year 2004 provided $41.3 billion in
temporary cash benefits to 8.8 million eligible workers who had become
unemployed through no fault of their own. Given the size of the UI
program, its importance in helping workers meet their needs when they are
unemployed, and the little information available on what factors lead
eligible workers to receive benefits over time, GAO was asked to determine
(1) the extent to which an individual worker's characteristics, including
past UI benefit receipt, are associated with the likelihood of UI benefit
receipt or unemployment duration, and (2) whether an unemployed worker's
industry is associated with the likelihood of UI benefit receipt and
unemployment duration. Using data from a nationally representative sample
of workers born between 1957 and 1964 and spanning the years 1979 through
2002, and information on state UI eligibility rules, GAO used multivariate
statistical techniques to identify the key factors associated with UI
benefit receipt and unemployment duration.
In its comments, the Department of Labor stated that while there are
certain qualifications of our findings, the agency applauds our efforts
and said that this report adds to our current knowledge of the UI program.
*** End of document. ***