Content uploaded by Elizabeth Savoca
Author content
All content in this area was uploaded by Elizabeth Savoca on Dec 13, 2017
Content may be subject to copyright.
The Journal of Mental Health Policy and Economics
J. Mental Health Policy Econ. 3, 199 –207 (2000)
DOI: 10.1002/mhp.102
The Civilian Labor Market Experiences of
Vietnam-Era Veterans: The Influence of
Psychiatric Disorders
Elizabeth Savoca1∗and Robert Rosenheck2
1Department of Economics, Smith College, Northampton, MA, USA
2VISN 1 Mental Illness Research, Education, and Clinical Center (MIRECC), and Department of Psychiatry, Yale University, New Haven, CT, USA
Abstract
Background: Most research on the civilian labor market experi-
ences of veterans has focused on the extent to which the skills and
experience acquired in the military are rewarded in the civilian
employment sector. While studies have been mindful of the need
to analyze this question in a multivariate framework, controlling
for other factors that might independently affect labor market
outcomes, they have met this goal with limited success. As a
result, an important element of the employment and wage deter-
mination process—psychiatric health—has been absent from this
literature.
Aims of the study: Using a nationally representative survey
of Vietnam-era veterans, this study analyzes the contribution
of psychiatric health toward explaining differences in the post-
service civilian wages, hours worked, and employment probabil-
ities among male veterans.
Methods: The analysis is based on data from the National
Survey of the Vietnam Generation, a survey, completed in the
late 1980s, of persons who were on active duty during the
years of the Vietnam War, 1964– 1975. Three outcome vari-
ables are studied—the hourly wage rate, usual hours worked
per week, and a 0–1 indicator for whether the respondent is
currently working. Lifetime diagnoses of four categories of men-
tal disorders—major depression, anxiety disorders, substance
abuse/dependence, and combat-related posttraumatic stress disor-
der (PTSD)—were constructed from the US NIMH Diagnostic
Interview Schedule, administered by the survey. The employ-
ment probability equation was estimated using probit; the hourly
earnings and hours worked equations via ordinary least squares
conditioned on being employed.
Results: The study finds that PTSD significantly lowered the
likelihood of working and, for those veterans who were working,
their hourly wages. On average, a veteran with a lifetime diagno-
sis of PTSD was 8.5 percentage points less likely to be currently
working than was a veteran who did not meet diagnostic criteria.
Among those who were employed, veterans with PTSD earned,
on average, $2.38 less per hour ($3.61 in 1999 U.S. dollars). Anx-
iety disorders and major depression had nearly as large an effect
on employment rates, as did PTSD. Major depression was also
found to have lowered hourly wages by an average of $6.77 per
hour ($10.17 in 1999 US dollars). However, psychiatric health
did not affect typical hours worked per week.
* Correspondence to: Elizabeth Savoca, Wright Hall, Smith College,
Northampton, MA 01063, USA.
Email address: esavoca@smith.edu
Discussion: This study contributes new information to several lit-
eratures. Previous research on the extent to which PTSD interferes
with readjustment to civilian life has focused on quality-of-life
outcomes such as overall well-being, physical health, and home-
lessness. Previous research on mental health and earnings has
focused on annual earnings. This study makes hourly wage com-
parisons, a closer measure of productivity differences since they
represent differences in pay for the same input of time. Finally,
this study demonstrates that the effects of psychiatric health are
as important as the influence of non-health characteristics that
are thought to signal earnings potential in the civilian labor mar-
ket (education and experience). These findings, however, may
not apply generally. The importance of PTSD may be specific
to veterans of the Vietnam War and may not pertain to persons
suffering non-combat-related PTSD.
Implications for Health Care Provision and Use and Health
Policy Formulation: The magnitude of our estimates implies
potentially large benefits from developing effective treatments for
PTSD and from insuring access to these treatments.
Implications for Future Research: Future research should
examine the relationship between work and PTSD in the gen-
eral population and should explore the indirect effects of mental
health, such as its effects on the post-service educational attain-
ment and occupational choices of veterans. Copyright 2000
John Wiley & Sons, Ltd.
Received 15 March 2000; accepted 26 March 2000
The civilian labor market experiences of veterans have
been of long-standing interest to labor market scholars.
Much research has focused on the extent to which the
skills and experience acquired in the military are rewarded
in the civilian employment sector. This interest is driven
by policy efforts to compensate veterans sufficiently for
their services. In their review of the literature and in their
own analysis, Mangum and Ball conclude that the rewards
vary by length of military service, era of service, education
levels, and military occupation.1A newer generation of
work has established that the observed differences between
veteran and non-veteran civilian wages reflect, in part, a
nonrandom selection process that leads to the enlistment
of persons whose average civilian earnings potential differ
from the average in the general population. When this
process is accounted for in wage comparisons, the results
suggest that male veterans from all eras incur a civilian
wage penalty.2,3,4
199
Copyright 2000 John Wiley & Sons, Ltd.
Most studies have been mindful of the need to analyze
these questions in a multivariate framework, controlling
for other factors that might independently affect earnings
and that might also be correlated with covariates related to
veteran status and military experience. Because of data lim-
itations, however, studies have met this goal with varying
levels of success. This paper examines an often-neglected
element of the employment and wage determination pro-
cess—psychiatric health. Research based on large-scale
general population surveys, known for the sophisticated
design of their psychiatric screening instruments, has
shown that certain psychiatric diseases significantly lower
income and employment opportunities.5,6,7,8,9,10 Only one
study has used these data to consider the consequences of
emotional distress for the civilian labor market experiences
of veterans, a particularly vulnerable population.11
Using a uniquely rich source of demographic and health
data on Vietnam-era veterans, this article analyzes the con-
tribution of psychiatric health toward explaining differences
in the post-service civilian wages and employment prob-
abilities among male veterans. We find that, relative to
non-health characteristics, many psychiatric diseases have a
surprising large influence on the earnings and employment
rates of veterans.
Methods
Sample
Our empirical analysis is based on data from the National
Survey of the Vietnam Generation (NSVG), a survey,
completed in the late 1980s, of persons who were on active
duty during the years of the Vietnam War, 1964–1975.12
The primary goal of the survey was to assess the prevalence
and effects of psychological problems among Vietnam-era
veterans. The survey’s method of psychiatric assessment
closely parallels two concurrent general population surveys,
the Epidemiological Catchment Area Survey (ECA) and
the National Comorbidity Survey (NCS). The NSVG is
better suited for our purposes, however, for it combines
clinically based assessments of psychiatric health with
detailed information on wartime experiences that might
exacerbate the consequences of mental illness.
The NSVG is also unique among the three in providing
data on hourly wages. Assuming that a person’s wage
rate is determined by his productive effort, in population-
based surveys hourly wage comparisons are regarded as
the closest measure of productivity differences, since they
represent differences in pay for the same input of time.
The annual earnings measures from the ECA and the NCS
include non-labor income (welfare, disability and social
security benefits, e.g.) and can vary across individuals
because of differences in hours worked. Strictly speaking,
however, hourly wage comparisons tell us whether the job
performance of persons suffering from mental disorders is
valued less by employers. Hence differences in pay rates
may not be due solely to differences in productive effort but
also may reflect pure labor market discrimination against
persons with psychiatric diseases. More direct measures
of employee productivity, however, are only available in
specialized data sets for industries and firms where output
is easily quantifiable.*
The NSVG administered the National Institute of Mental
Health’s Diagnostic Interview Schedule (DIS), a highly
structured interview designed for use by laypersons and
non-specialists to simulate clinical diagnoses of psychiatric
disorders.14 The survey derived two diagnostic variables for
each of several specific psychiatric diseases. One variable
indicates whether the individual met psychiatric criteria
for an illness at any time in his life (referred to as a
lifetime diagnosis); the other indicates whether a person
with a lifetime diagnosis presented symptoms within six
months of the interview date (referred to as a current
diagnosis).
Reliability studies of the DIS have shown it to be
far preferable to standard survey measures based on the
respondent’s self-assessment of overall emotional health.
Self-reported measures display systematic biases that vary
with gender, education and ethnicity. Such biases have been
shown to confound multivariate estimates of the indepen-
dent effects of socioeconomic background and the effects
of health status on wages and employment rates.15,16 How-
ever, in validity studies, the DIS diagnoses have been
shown to have high rates of disagreement with diagnoses
based on direct physician appraisals, particularly for those
diseases that are relatively rare in the male population.17
Savoca has shown that when psychiatric diagnoses are
entered into a regression at a highly disaggregated level,
these discrepancies lead to substantive differences in the
estimated inferences about the earnings effects of psy-
chiatric diseases, depending on which type of psychi-
atric assessment is the benchmark of the analysis, the
DIS or a direct clinical exam.18 To mitigate this prob-
lem we condense the diseases into three categories: major
depressive episode, anxiety disorders (obsessive compul-
sive disorder, panic disorder, generalized anxiety disor-
der), and substance abuse and dependence (alcohol and
drugs).
In our analysis, we also include survey diagnoses of
posttraumatic stress disorder (PTSD), based on the Missis-
sippi Scale for Combat-Related PTSD, an indicator that was
found to have the highest concordance with direct clinical
appraisals.12
The NSVG also differentiates veterans along two war-
related characteristics: whether or not a respondent served
in the war theater (Southeast Asia) and whether or not a
respondent was exposed to high levels of war-zone stress.
To construct the latter indicator, the survey developed
an index of war-zone stress based on the respondent’s
degree of exposure to combat, to abusive violence inflicted
on enemy soldiers or civilians, and to deprivation, and
* Berndt et al.13, for example, analyze the productivity effects of mental
disorders for employees at a large insurance claims processing company
using, as their measure of productivity, employer records on the average
number of claims processed per day for each employee in their sample.
200 E. SAVOCA AND R. ROSENHECK
Copyright 2000 John Wiley & Sons, Ltd. J. Mental Health Policy Econ. 3, 199 –207 (2000)
on whether the respondent was taken as a prisoner of
war. Correlations between these war-related characteristics
and the incidence of specific psychiatric diseases reveal
that, regardless of where they served, veterans exposed
to low levels of war-zone stress had prevalence rates
similar to the estimates derived from the ECA for the
general population. Veterans exposed to high levels of
stress, however, displayed significantly higher rates of
virtually all types of psychiatric disorder.19
Table 1 reports summary statistics on all of the variables
used in our analysis for both the full sample and the sample
of employed veterans. Indeed, the data from both samples
show that veterans exposed to high levels of war-zone
stress not only have higher rates of combat-related PTSD
but also substantially higher rates of most other types
of psychiatric disorders. With a few exceptions the other
variables need no additional explanation. Hourly wage rates
are directly available for workers who reported their pay
on an hourly basis. For workers who reported weekly,
monthly, or annual earnings, instead, the hourly wage was
constructed from information on usual hours worked per
week. The number of chronic medical conditions is a self-
reported measure of active physical health problems. Non-
labor income refers to income brought into the household
Table 1. Sample descriptive statistics. means (standard deviations)
Full sample Employed sample
Vete r an s Ve ter ans
exposed to exposed to
high levels of high levels of
All veterans war-zone stress All veterans war-zone stress
Dependent variables
Employed 0.872 (0.334) 0.844 (0.364) — —
Hourly wage — — 15.484 (64.432) 15.889 (76.665)
Hours worked per week — — 44.973 (10.300) 45.168 (10.836)
Independent variables
Health statusa
Anxiety disorder 0.166 (0.372) 0.301 (0.459) 0.139 (0.346) 0.244 (0.430)
Substance abuse/dependence 0.411 (0.492) 0.491 (0.501) 0.391 (0.488) 0.458 (0.499)
Major depression 0.056 (0.230) 0.145 (0.352) 0.040 (0.196) 0.100 (0.301)
Posttraumatic stress disorder 0.222 (0.416) 0.526 (0.500) 0.190 (0.393) 0.465 (0.500)
Number of chronic health conditions 2.205 (1.329) 2.630 (1.496) 2.083 (1.249) 2.468 (1.424)
Military experience
Theater veteran 0.718 (0.450) 1.000 (0.000) 0.738 (0.440) 1.000 (0.000)
Length of service
Less than 1 year 0.011 (0.105) 0.008 (0.092) 0.007 (0.085) 0.007 (0.082)
One to 3 years 0.746 (0.435) 0.770 (0.422) 0.763 (0.426) 0.786 (0.411)
Four to 20 years 0.144 (0.351) 0.162 (0.369) 0.139 (0.346) 0.157 (0.365)
Twenty years or more 0.099 (0.299) 0.060 (0.237) 0.091 (0.288) 0.050 (0.219)
Exposed to high levels of war-zone stress 0.248 (0.432) 1.000 (0.000) 0.240 (0.427) 1.000 (0.000)
Demographic variables
Minority 0.512 (0.500) 0.563 (0.497) 0.501 (0.500) 0.548 (0.498)
Age 41.879 (6.313) 40.540 (4.165) 41.349 (5.277) 40.358 (3.826)
Currently married 0.739 (0.439) 0.688 (0.464) 0.763 (0.425) 0.732 (0.443)
Years of schooling 13.390 (2.457) 13.182 (2.360) 13.527 (2.401) 13.234 (2.349)
Non-labor income 7666.769 8 469.193 7 971.055 9 280.936
(9850.598) (10 638.228) (9945.916) (10 997.450)
Region of residence
Northeast 0.156 (0.363) 0.173 (0.379) 0.160 (0.366) 0.177 (0.383)
Midwest 0.183 (0.387) 0.179 (0.384) 0.183 (0.387) 0.194 (0.396)
Southeast 0.373 (0.484) 0.341 (0.475) 0.362 (0.481) 0.311 (0.464)
Mountain and Pacific 0.265 (0.442) 0.290 (0.454) 0.273 (0.446) 0.304 (0.461)
Puerto Rico 0.023 (0.150) 0.017 (0.129) 0.022 (0.146) 0.013 (0.115)
Job characteristicsb
Data — — 2.855 (1.868) 3.177 (1.954)
People — — 5.593 (2.442) 5.786 (2.330)
Things — — 4.457 (2.694) 4.602 (2.615)
Sample size 1417 352 1247 299
aThe mental health variables indicate the presence (1) or absence (0) of a lifetime diagnosis.
bThe job characteristics are coded on a scale of 0 to 6, 7, or 8, with zero denoting the most complex tasks. The variable Data ranges from synthesizing
(0) to comparing data (6); the variable Things ranges from setting up (0) to handling (7) things (equipment, machines, tools, etc); the variable People
ranges from mentoring (0) to taking instructions (8).
MENTAL HEALTH AND VETERANS’ WORK AND PAY 201
Copyright 2000 John Wiley & Sons, Ltd. J. Mental Health Policy Econ. 3, 199 –207 (2000)
by other family members.* In addition to this and other
demographic variables which tend to differentiate high-
wage from low-wage workers (education, age, ethnicity,
and marital status), the wage and hours worked analyses
also include variables that reflect job characteristics. Fol-
lowing a coding scheme developed by the Dictionary of
Occupational Titles,20 we classified occupations according
to the complexity with which the worker must deal with
data, people, and things, with the lowest score denoting the
highest level of complexity.
Model Specification
The earnings equation is an augmented version of the
human capital earnings function.21 The dependent variable
is the natural log of the hourly wage. Potential lifetime
work experience is measured by age, education by years of
schooling. To the extent that workers are matched to jobs
according to both measured and unmeasured productivity,
the job trait variables are intended to control for differ-
ences in skills that remain after controlling for education
and potential experience. These variables are potentially
important control variables since the survey year (1986)
coincided with a period during which labor demand shifted
markedly toward workers with more technically oriented
skills.22 We also include regional dummies to control for
differences in labor market conditions across geographic
areas. Finally, in keeping with the literature, we allow for a
potential difference in the returns to military versus civilian
work experience by including a series of dummy variables
for length of military service.
The hours worked and employment probability equations
are modeled as labor supply decisions. They include vari-
ables that may influence a person’s potential market wage:
age, ethnicity, marital status, education, health, military
experience, and region of residence. Labor supply decisions
may also be influenced by non-labor sources of income.
The mental health variables included in all three equ-
ations indicate the presence or absence of lifetime diag-
noses. Preliminary estimation investigated the possibility
of interaction effects among the four disorders. Six of the
interaction terms included in our regressions involved two-
way interactions of each disorder with the other. We also
interacted substance abuse with education. This latter deci-
sion was guided by the findings of Mullahy and Sindelar,
who showed that alcoholism had its most depressing effects
on the earnings of relatively less-educated men.6In the
hours worked and work probability equations none of the
interaction effects was statistically significant, either jointly
or individually. In the log-wage equation all but one, sub-
stance abuse interacted with depression, was statistically
insignificant, both individually and jointly. Consequently,
the log-wage equation reported here includes only this
interaction effect.
* The ideal measure of non-labor income would also include the respon-
dent’s own non-labor sources of income (interest income, rental income,
transfer payments, etc). Unfortunately, we are unable to derive the amount
from these sources.
Since observations on wages and hours worked were
restricted to employed persons, we investigated the poten-
tial for sample selection bias using conventional meth-
ods: Heckman’s two-step procedure and full informa-
tion maximum likelihood. We found little practical differ-
ence between the ordinary regression coefficients and the
coefficient estimates corrected for sample selection bias.
Furthermore, in the sample selection models, the estimated
correlations between the error in the work probability
equation and the errors in the log-wage and hours worked
equations were statistically insignificant.
However, in Monte Carlo simulations, Leung and Yu
have shown that when there is high collinearity among
the regressors, the OLS procedure outperforms the sample
selection procedure in several important respects, even
when the sample selection model is the true model.23
Summary measures of multicollinearity indicate a high
degree in our data.* Hence, we have decided to take a
conservative approach in estimation and interpretation. We
focus our paper on the OLS results and apply our findings
for hours worked and the hourly wage only to veterans who
work, not necessarily to the general population of veterans.
The employment probability equation is estimated on the
full sample using probit.
We also examined the issue of the exogeneity of sev-
eral of the regressors in our earnings and hours worked
equations. Basic econometric intuition suggests that the
OLS estimates, however large or small, may reflect a bias
resulting from a possible simultaneous relationship. From a
human capital perspective, expectations of high wages may
induce greater personal efforts to maintain mental health.
From a psychological perspective adverse economic cir-
cumstances may lead to the development of psychiatric
disorders.25 Moreover, studies have shown a strong cor-
relation between mental health and, respectively, marital
stability and educational attainment5,7,26 —correlations so
strong that one might reasonably suspect that unobservable
personal characteristics may be driving the results rather
than the measured covariates themselves. If this were so,
then causal inferences drawn from OLS estimates would
be suspect.
To resolve this issue, we carried out a Hausman test
of the joint exogeneity of years of schooling, marital sta-
tus, and the four mental health indicators. The instruments
included variables for the educational attainment of each
of the respondent’s parents and variables which reflect the
respondent’s childhood environment and genetic predis-
position toward psychiatric illness—whether the respon-
dent was an only child, whether the respondent lived with
* We used two diagnostic procedures for detecting collinearity. One,
auxiliary regressions of the inverse Mills ratio on the regressors in the
hours worked and log-wage equations, yielded an R2of 0.88 in both
equations. These values are much higher than the R2s in the hours
worked and log-wage equations, 0.06 and 0.08, respectively. The other,
the condition number for the moment matrix of the regressors, was
approximately 1827 for both the hours worked and log-wage equations.
Both numbers are far in excess of 30, the value that is considered the
threshold for serious collinearity problems.24
202 E. SAVOCA AND R. ROSENHECK
Copyright 2000 John Wiley & Sons, Ltd. J. Mental Health Policy Econ. 3, 199 –207 (2000)
both parents until age 16, and indicators for the extent
of mental illness, substance abuse, and criminal behavior
in the respondent’s parents during his childhood. The chi-
square statistics were 3.868 (p-value =0.695)and 4.470
(p-value =0.613)for the hourly wage and hours worked
equations, respectively. Hence we are unable to reject the
null hypothesis of joint exogeneity.
Since the hourly wage equation is specified in a log-
linear form, the coefficient on a continuous regressor, such
as years of schooling, age, and non-labor income, can be
interpreted as the rate of return to an additional unit of that
regressor. To interpret the effect of a dummy variable, D,
on the hourly wage we compute eγ−1, the proportionate
difference in the hourly wage for a person for whom D=1
versus a person for whom D=0, where γis the coefficient
of Din the log-wage equation. To derive the impact on
the hourly wage of a dummy variable, D1, whose effect
interacts with another dummy variable, D2, we compute
eγ+βD2−1, the proportionate difference in the hourly wage
for a person for whom D1=1 versus a person for whom
D1=0 evaluated at the mean value of D2, weighted by
β, the coefficient on the interaction term in the log-wage
equation.
Results
The first column of Table 2 reports results from the probit
estimation of the employment probability equation. All four
types of psychiatric disorder have statistically significant
negative effects on the probability of employment. The
findings for substance abuse and dependence conform to
the results from numerous other studies of the employment
rates of adult men.5,6,7,9The finding that depression has an
even larger negative effect on the likelihood of working is
consistent with Ettner, Frank, and Kessler, whose analysis
of the NCS predicts that out of 13 specific psychiatric
Table 2. Estimates of hourly wages, hours worked and the probability of employment
Employment Hours worked Log of hourly
Variables probability per week wage
Health statusa
Anxiety disorder −0.415∗(0.129) 1.229 (0.918) 0.267∗(0.087)
Substance abuse/dependence −0.233∗∗ (0.106) 0.189 (0.614) −0.083 (0.059)
Major depression −0.390∗∗ (0.184) −2.158 (1.581) −0.970∗(0.265)
Substance abuse ×depression — — 0.943∗(0.308)
Posttraumatic stress disorder −0.498∗(0.132) −0.356 (0.871) −0.171∗∗ (0.082)
Number of chronic health conditions −0.165∗(0.037) −0.515∗∗ (0.247) −0.004 (−0.023)
Military experience
Theater veteran 0.600∗(0.125) 0.589 (0.714) 0.050 (0.067)
Length of service
One to 3 years 0.484 (0.446) 0.350 (3.416) 0.489 (0.323)
Four to 20 years 0.363 (0.458) 0.428 (3.501) 0.415 (0.331)
Twenty years or more 0.461 (0.476) −0.313 (3.673) 0.109 (0.347)
Exposed to high levels of war-zone stress −0.045 (0.139) 0.926 (0.775) 0.019 (0.073)
Demographic variables
Minority −0.316∗(0.106) 2.391 (5.918) −0.049 (−0.858)
Age (÷10) 2.574∗(0.749) −1.633 (0.603) 1.050∗∗∗ (0.559)
Age squared (÷100) −0.324∗(0.078) −1.312 (0.655) −0.105∗∗∗ (0.062)
Currently married 0.468∗(0.108) 1.516∗∗ (0.699) 0.098 (0.065)
Years of schooling 0.068∗(0.021) 0.081 (0.131) 0.048∗(0.012)
Non-labor income (÷1000) 0.000 (0.000) −0.000∗∗∗ (0.000) —
Region of residence
Northeast 0.226 (0.323) 2.404 (2.126) 0.588∗(0.201)
Midwest 0.054 (0.315) 3.356 (2.121) 0.612∗(0.200)
Southeast 0.080 (0.300) 3.684 (2.046) 0.496∗∗ (0.193)
Mountain and Pacific 0.304 (0.308) 2.765 (2.062) 0.614∗(0.195)
Job characteristicsb
Data — 0.086 (0.194) −0.026 (0.018)
People — −0.746∗(0.161) 0.002 (0.015)
Things — 0.024 (0.121) −0.045∗(0.011)
Intercept −5.034∗(1.851) 40.272∗(13.937) −1.627 (1.316)
Sample Size 1417 1247 1247
Log-likelihood −394.149
Rsquared 0.060 0.081
Note: The reference group is white veterans without a history of any psychiatric disorder, who served outside Southeast Asia, who
were exposed to low levels of war-related stress, and who served for less than one year. They lived in Puerto Rico and were not
married at the time of the survey.
a,bSee notes to Table 1. ∗p-value <0.01; ∗∗ 0.01 ≤p-value <0.05; ∗∗∗ 0.05 ≤p-value <0.10.
MENTAL HEALTH AND VETERANS’ WORK AND PAY 203
Copyright 2000 John Wiley & Sons, Ltd. J. Mental Health Policy Econ. 3, 199 –207 (2000)
diseases major depression has the largest negative effect on
male employment rates.9Anderson and Mitchell, however,
find that depression is the only disorder that has no effect
on the labor force participation rates of the working-age
males included in the ECA.11
It is difficult to place the result for anxiety disorders
within the literature since most studies of specific disease
effects have focused on depression and alcohol and drug
abuse.5,6,27–31 Ettner, Frank, and Kessler, who control for
five different types of anxiety disorder simultaneously, find
that none individually affects male employment rates.9In
bivariate comparisons, however, Eaton et al. and Blazer
et al. find much higher rates of panic and generalized
anxiety disorders among welfare recipients and others that
are financially dependent on the government— persons who
have the most difficulty holding a job.32,33
In column 2 we see that none of the psychiatric disorders
has a statistically significant effect on the usual number of
hours worked per week. With the exception of dysthymia,
Ettner et al. obtain the same result for the men surveyed
in the National Comorbidity Survey. Berndt et al.French
and Zarkin, and Greenberg et al. have shown, however,
that persons with anxiety disorders and other emotional
problems tend to have higher rates of absenteeism.34–36
Together these results suggest that survey questions about
usual hours worked fail to elicit enough information about
the respondent’s work habits to reveal the full effects of
psychiatric illnesses on labor supply.
Turning to the last column of Table 2 we find that
depression and posttraumatic stress disorder have statis-
tically significant negative effects on male hourly wages.
Veterans suffering from PTSD are predicted to have 16%
lower hourly wages than nonsufferers. Substance abuse
and depression have a strong positive interaction effect.
Evaluated at the means, the coefficients imply that a man
suffering from major depression has a 45% lower hourly
wage than that of a nonsufferer (t =−2.14). Substance
abusers are predicted to earn 4.4% lower pay rates than
those of other workers, although that effect is statisti-
cally insignificant (t =−0.78). The statistically insignif-
icant coefficient for substance abuse/dependence comes as
no surprise. Evidence on the wage effects of substance use
has been inconclusive, varying not only in practical sig-
nificance but also in sign. In studies based on the New
Haven site of the ECA, Mullahy and Sindelar find that
alcohol abuse/dependence significantly lowers the income
of full-time male workers.6,7French and Zarkin find a
nonlinear relationship between alcohol consumption and
wages where heavy drinkers and those who abstain earn
lower annual earnings than do moderate drinkers.37 Regis-
ter and Williams, Gill and Michaels, and Kaestner find that
drug use is associated with higher wages among young
adults.30,31,38
We did not expect to find a positive association between
anxiety disorders and hourly wages. Blazer et al.finda
strong negative association between occupational status and
a diagnosis of generalized anxiety disorders, the most com-
mon type of anxiety disorder in our sample. That is, highly
educated, high-wage workers have much lower prevalence
rates.33 In its Diagnostic and Statistical Manual of Men-
tal Disorders, the American Psychiatric Association writes
that impairment in occupational functioning is typically
mild for generalized anxiety disorders and for most panic
disorders but can be quite severe for persons with obses-
sive–compulsive disorder.39 No study has found that per-
sons with anxiety disorders are more productive on the job.
In both the employment and wage equations the coeffi-
cients on the standard demographic control variables have
the expected signs. Each additional year of schooling raises
hourly earnings by 4.8%. Hourly earnings initially rise with
age but eventually the payoff to experience drops and after
age 50 it becomes negative. This is a familiar pattern in
almost all cross-sectional age–earnings profiles. Minori-
ties earn 4.8% less than whites with identical backgrounds,
married men earn 10% more than do other men, although
these effects are not statistically significant. It is interest-
ing to note that veterans placed in jobs that require skill in
the use of machinery are amply rewarded. Those in jobs
requiring the highest skill level earn 37% more than do vet-
erans at the lowest skill level. Variables that are associated
with high market earnings potential also raise the probabil-
ity of working. There are large regional variations in wage
rates but not in employment probabilities. Length of mili-
tary service appears to have no effect on the likelihood of
working or on hourly wages. Only physical health, mari-
tal status, sources of other income, and degree of personal
interactions on the job have statistically significant effects
on usual hours worked, all in the expected direction.
Studies of the difficulty Vietnam-era veterans faced in
adjusting to civilian life emphasize the importance of
actual exposure to combat experiences instead of service
in Southeast Asia per se.12 Some of our findings confirm
this view. Veterans who served in Southeast Asia, ‘theater’
veterans, were actually more likely to be employed but
having served in the war theater did not give rise to either
a wage premium or penalty. A diagnosis of combat-related
PTSD, however, significantly reduced both the likelihood
of working and the hourly rate of pay.
Although exposure to war-zone stressors is a significant
risk factor for posttraumatic stress disorder, not all veter-
ans exposed to high levels of war-zone stress met the full
criteria for this disease. To assess the potentially indepen-
dent effects of extensive exposure on success in the civilian
labor market we added a binary variable to our specifica-
tion, which equals one if exposure to war-related stress was
high, zero if not. As shown in Table 2, this variable was
not statistically significant. Hence, although one-fourth of
the veterans in our sample were exposed to high levels of
war-zone stress, among them only those veterans who also
met diagnostic criteria for psychiatric disorders were at a
serious disadvantage in the civilian labor market.
Assessing the Practical Significance of the Results
To compare the effects of health to some important
non-health factors it is helpful to consider the effects of
204 E. SAVOCA AND R. ROSENHECK
Copyright 2000 John Wiley & Sons, Ltd. J. Mental Health Policy Econ. 3, 199 –207 (2000)
Table 3. The effects of two standard deviations and category changes in selected explanatory variables
on the probability of employment and the hourly wage rate
Difference in Percentage difference
employment in the hourly
Variable probabilities wage rate
Minority versus white −0.045∗4.76
Currently married versus not currently married 0.078∗10.27
Lifetime diagnosis of at least one substance abuse
disorder versus none
−0.034∗∗ −4.41
Lifetime diagnosis of at least one anxiety disorder versus
none
−0.071∗30.54∗
Lifetime diagnosis of major depression versus none −0.070∗∗∗ −45.23∗
Lifetime diagnosis of posttraumatic stress disorder
versus none
−0.086∗−15.75∗∗
Number of chronic medical conditions one s.d. above
the mean versus one s.d. below the mean
−0.063∗−0.97
Age one s.d. above the mean versus one s.d below the
mean
−0.025 21.32∗∗
Years of schooling one s.d above the mean versus one
s.d. below the mean
0.048∗25.88∗
Notes: The predicted probability of employment for an individual with characteristics Xis computed according to
(X ˆ
β) where denotes the cumulative distribution function of the standard normal density and ˆ
βthe probit estimates
reported in Table 2. The predicted proportionate difference in the hourly wage is computed as eβ−1. The effect of
a given change in a variable is computed setting the values of all others to their sample means. Standard errors are
computed according to the delta method.
∗p-value ≤.01. ∗∗ .01 ≤p-value ≤.05. ∗∗∗ .05 ≤p-value ≤.10.
comparable changes in the explanatory variables. Such
comparisons are reported in Table 3. To explain the entries
in this table consider the effect of a two standard devia-
tion change in years of schooling, reported in the last row.
Setting all other variables equal to their sample means,
the difference between the probability of employment for
a veteran with years of schooling one standard deviation
above the mean versus a veteran with years of schooling
one standard deviation below the mean is 4.8 percentage
points. The entry in the next column indicates that the vet-
eran with the higher years of schooling will earn 26 percent
more per hour. The entries for dummy variables report the
difference in employment probabilities and hourly wages
when the category switches from 1 to 0. For example,
evaluated at the sample means of all other variables, the
probability that a veteran, who is currently married, will
be employed is 7.8 percentage points higher than the like-
lihood of employment for other veterans. His hourly wage
is 10 percent higher.
By these measures the most important determinants of
the probability of employment are marital status and health,
particularly the presence or absence of PTSD, anxiety
disorder and major depression. The effect of PTSD is
almost twice the magnitude of the effect of a two-standard-
deviation change in years of schooling. The importance of
health, however, is not confined to mental health status.
The effect of physical health is nearly one-third larger than
the effect of schooling.
These measures also imply that psychiatric health is of
great importance in determining hourly wages. Working
veterans suffering from major depression, on average, earn
45 percent less per hour than do veterans who do not suffer
from this disease. This is nearly double the effect of a
two standard deviation change in education. Posttraumatic
stress disorder lowers hourly earnings, on average, by 16
percent, an amount that is 60 and 75 percent of the effects
of education and experience, respectively.
Summary and Conclusions
In the 1980s, the US Congress passed legislation mandating
the National Vietnam Veterans Readjustment Study. The
primary objective of the study was to assess the prevalence
of post-war psychological problems, with a particular focus
on posttraumatic stress disorder (PTSD), and to determine
the extent to which these problems interfered with read-
justment to civilian life. Analyses of the data collected in
this study have contributed much to our understanding of
the factors, both pre-war and war related, that lead to the
development of PTSD symptoms in adults.40–42 Other stud-
ies have highlighted the relationship between PTSD and
quality-of-life outcomes such as overall well-being, phys-
ical health, and homelessness.43–45 This paper contributes
to this literature with a systematic look at the long-term
effects of psychiatric distress on the civilian labor market
experiences of male Vietnam-era veterans.
We find that combat-related PTSD significantly lowers
the likelihood of working and, for those veterans who are
working, the hourly wage rate. By our measure, PTSD is,
in fact, the most important determinant of the probability
of employment. On average, a veteran with a lifetime
diagnosis of PTSD is 8.6 percentage points less likely to
be currently working than is a veteran who did not meet
MENTAL HEALTH AND VETERANS’ WORK AND PAY 205
Copyright 2000 John Wiley & Sons, Ltd. J. Mental Health Policy Econ. 3, 199 –207 (2000)
diagnostic criteria for this disease. This effect is nearly
double the difference between the likelihood of working
for a veteran with 16 years of schooling versus one with
11 years. Among those who are employed, veterans with
PTSD earn, on average, 16 percent less per hour.
We also find that other psychological diseases, in partic-
ular anxiety disorders and major depression, have nearly
as large an effect on employment rates as does PTSD.
Their effects are more important than the influence of non-
health characteristics that are thought to signal high earn-
ings potential in the civilian labor market (education and
age/experience). Psychiatric health also affects on-the-job
productivity as measured by the hourly wage rate. A work-
ing veteran who has suffered a major depressive episode
incurs a substantial earnings loss, a 45 percent reduction in
his hourly rate of pay.
Psychiatric health may assume added importance in the
labor supply decisions of veterans as opposed to the gen-
eral population because of the availability of VA disability
compensation. However, Rosenheck, Frisman, and Sindelar
have shown that the work disincentive effects of VA dis-
ability benefits are modest.46 Hence our estimated effects
of mental illness on labor supply and earnings most likely
reflect its debilitating effects on work effort. The size of our
estimates implies potentially large benefits from providing
veterans with effective treatments for these diseases.
References
1. Mangum SL, Ball DE. The transferability of military-provided occu-
pational training in the post-draft era. Industrial and Labor Relations
Review 1989; 42: 230– 245.
2. Angrist J. Lifetime earnings and the Vietnam-era draft lottery: evi-
dence from social security administrative records. American Eco-
nomic Review 1990; 80: 313 –336.
3. Angrist J. Estimating the labor market impact of voluntary military
service using social security data on military applicants. Economet-
rica 1998; 66: 249– 288.
4. Angrist J, Krueger A. Why do world war II veterans earn more than
nonveterans? Journal of Labor Economics 1994; 12: 74 –97.
5. Mullahy J, Sindelar J. Lifecycle effects of alcoholism on education,
earnings, and occupation. Inquiry 1989; 26: 272 –82.
6. Mullahy J, Sindelar J. Alcoholism, work, and income. Journal of
Labor Economics 1993; 11: 494 –520.
7. Mullahy J, Sindelar J. Alcoholism and income: the role of indirect
effects. The Millbank Quarterly 1994; 72: 359 –375.
8. Frank RG, Gertler P. An assessment of measurement error bias for
estimating the effect of mental distress on income. The Journal of
Human Resources 1991; 26: 154 –164.
9. Ettner SL, Frank RG, Kessler RC. The impact of psychiatric dis-
orders on labor market outcomes. Industrial and Labor Relations
Review 1997; 51: 64– 81.
10. Kessler RC, Frank RG. The impact of psychiatric disorders on work
loss days. Psychological Medicine 1997; 27: 861– 873.
11. Anderson KH, Mitchell J. Effects of military experience on men-
tal health problems and work behavior. Medical Care 1992; 30:
554– 563.
12. Kulka RA, Schlenger WE, Fairbank JA, Hough RL, Jordan BK,
Marmar CR, et al.Contractual Report of Findings from the National
Vietnam Veterans Readjustment Study. Volume 1: Executive Summary,
Description of Findings, and Technical Appendices. Research Trian-
gle Park, NC: Research Triangle Institute; 1988.
13. Berndt ER, Bailit HL, Keller MB, Verner JC, Finkelstein SN. Health
care use and at-work productivity among employees with mental
disorders. Health Affairs 2000; 244: 256.
14. Robins LN, Helzer JE, Croushan J, Ratcliff KS. National Institute
of Mental Health Diagnostic Interview Schedule: its history, char-
acteristics, and validity. Archives of General Psychiatry 1981; 38:
381.
15. Savoca E. Measurement error in self-evaluations of mental health:
implications for labor market analysis. In: Economics and Mental
Health. Frank RG, Manning WG (eds). Johns Hopkins University
Press: Baltimore, MD, 1992; 130–151.
16. Savoca E. Controlling for mental health in earnings equations: what
do we gain and what do we lose? Health Economics 1995; 4:
399– 410.
17. Helzer J, Robin LN, McEvoy LT, Spitznagel EL, Stoltzman RK,
Farmer A, et al. A comparison of clinical and diagnostic inter-
view schedule diagnoses. Archives of General Psychiatry 1985; 42:
657– 666.
18. Savoca E. Measurement errors in binary regressors: an application
to measuring the effects of specific psychiatric diseases on earn-
ings. Health Services and Outcomes Research Methodology 2000; 1:
149– 164.
19. Kulka RA, Schlenger WE, Fairbank JA, Hough RL, Jordan BK,
Marmar CR, et al.Trauma and the Vietnam War Generation .New
York: Brunner and Mazel, 1990.
20. United States Department of Labor, Employment and Training
Administration. Dictionary of Occupational Title. 4th edn, revised.
Bernan: Lanham, MD, 1991.
21. Polachek SW, Siebert WS. The Economics of Earnings . Cambridge
University Press: Cambridge, 1993.
22. Katz LF, Murphy KM. Changes in relative wages, 1963–1987.
Quarterly Journal of Economics 1992; 107: 35 –78.
23. Leung SF, Yu S. On the choice between sample selection and the
two-part model. Journal of Econometrics 1996; 72: 197 –229.
24. Belsley D, Kuh E, Welsch R. 1980. Regression Diagnostics: Identi-
fying Influential Data and Sources of Collinearity. Wiley: New York.
25. Ezzy D. Unemployment and mental health: a critical review. Social
Science and Medicine 1993; 37: 41– 52.
26. Berndt ER, Koran LM, Finkelstein SN, Gelenberg AJ, Kornstein SG,
Miller IM, Thase ME, Trapp GA, Keller MB. Lost human capital
from early-onset chronic depression. American Journal of Psychiatry
2000; 157: 940– 946.
27. Anderson G. A sober look at the cost of intoxication. Contemporary
Policy Issues 1992; 10: 111–115.
28. Greenberg PE, Stiglin LE, Finkestein SN, Berndt ER. The economic
burden of depression in 1990. Journal of Clinical Psychiatry 1993;
54: 405– 418.
29. Antonuccio DO, Thomas M, Danton WG. A cost-effectiveness anal-
ysis of cognitive behavior therapy and fluoxetine (prozac) in the
treatment of depression. Behavior Therapy 1997; 28: 187– 210.
30. Register CA, Williams DR. Labor market effects of marijuana and
cocaine use among young men. Industrial and Labor Relations
Review 1992; 45: 435– 448.
31. Gill A, Michaels R. Does drug use lower wages? Industrial and
Labor Relations Review 1992; 45: 419 –433.
32. Eaton W, Dryman A, Weissman MN. Panic and phobia. In: Psychi-
atric Disorders in America. Robins L, Regier D (eds). Free Press:
New York, 1991; 155–179.
33. Blazer DG, Hughes D, George LK, Swartz M, Boyer R. Generalized
anxiety disorders. In: Psychiatric Disorders in America. Robins L,
Regier D (eds). Free Press: New York, 1991; 180–203.
34. Berndt ER, Bailit HL, Keller MB, Verner JC, Finkelstein SN. Health
care use and at-work productivity among employees with mental
disorders. Health Affairs 2000; 19: 244–256.
35. French MT, Zarkin GA. Mental health, absenteeism, and earnings at
a large manufacturing worksite. The Journal of Mental Health Policy
and Economics 1998; 1: 161–172.
36. Greenberg PE, Sisitsky T, Kessler RC, Findelstein SN, Berndt ER,
Davidson JRT, Ballenger JC, Fyer AJ. The economic burden of anx-
iety disorders in the 1990s. Journal of Clinical Psychiatry 1999; 60:
427– 435.
37. French MT, Zarkin GA. Is moderate alcohol use related to wages?
evidence from four work sites. Journal of Health Economics 1995;
14: 319– 344.
38. Kaestner R. New estimates of the effect of marijuana and cocaine use
on wages. Industrial and Labor Relations Review 1994; 47: 454 –470.
39. American Psychiatric Association. Diagnostic and Statistical Manual
of Mental Disorders 4th edn, revised. American Psychiatric Associ-
ation: Washington, DC, 1987.
206 E. SAVOCA AND R. ROSENHECK
Copyright 2000 John Wiley & Sons, Ltd. J. Mental Health Policy Econ. 3, 199 –207 (2000)
40. Fontana A, Spoonster Schwartz L, Rosenheck RA. Posttraumatic
stress disorder among female Vietnam veterans: a causal model of
etiology. American Journal of Public Health 1997; 87: 169 –175.
41. King DW, King LS, Foy DW, Gudanowski DM. Prewar factors in
combat-related posttraumatic stress disorder: structural equation mod-
eling with a national sample of male and female Vietnam veterans.
Journal of Consulting and Clinical Psychology 1996; 64: 520 –531.
42. Fontana A, Rosenheck RA. Posttraumatic stress disorder among
Vietnam theater veterans: a causal model of etiology in a commu-
nity sample. Journal of Nervous and Mental Diseases 1994; 182:
677– 684.
43. Taft CT, Stern AS, King LA, King DW. Modeling physical health and
functional health status: the role of combat exposure, posttraumatic
stress disorder, and personal resource attributes. Journal of Traumatic
Stress 1999; 12: 3– 23.
44. Zatzick DF, Marmar CR, Weiss DS, Browner WS, Metzler TJ, Gold-
ing JM, et al. Posttraumatic stress disorder and functioning and
quality of life outcomes in a nationally representative sample of
male Vietnam veterans. American Journal of Psychiatry 1997; 154:
1690– 1695.
45. Rosenheck RA, Fontana A. A model of homelessness among male
veterans of the Vietnam War generation. American Journal of Psy-
chiatry 1994; 151: 421– 427.
46. Rosenheck RA, Frisman L, Sindelar J. Disability compensation and
work among veterans with psychiatric and nonpsychiatric impair-
ments. Psychiatric Services 1995; 46: 359– 366.
MENTAL HEALTH AND VETERANS’ WORK AND PAY 207
Copyright 2000 John Wiley & Sons, Ltd. J. Mental Health Policy Econ. 3, 199 –207 (2000)