Professional Documents
Culture Documents
Objectives. To estimate the health consequences of involuntary job loss among older workers in the United States.
Results. The effects of late-life involuntary job loss on both follow-up physical functioning and mental health were
negative and statistically significant (p ⬍ .05), even after baseline health status and sociodemographic factors were con-
trolled for. Among displaced workers, reemployment was positively associated with both follow-up physical function-
ing and mental health, whereas the duration of joblessness was not significantly associated with either outcome.
Discussion. The findings provide evidence of a causal relationship between job loss and morbidity among older
workers. This relationship is reflected in both poorer physical functioning and mental health for workers who experience
involuntary job loss. In addition to the economic consequences of worker displacement, there may be important health
consequences of job loss, especially among older workers.
S131
S132 GALLO ET AL.
for assessing the direction of causality, have become rea- The authors suggest that the observed age-related variations
sonably common during the past decade. These studies have in responses to unemployment may be due to differing role
generally shown that job loss has some negative impact on responsibilities throughout the life cycle. For instance, older
physical health, although the nature of this impact is diffi- workers approaching retirement may have fewer financial
cult to assess (Kasl & Jones, in press), given the diversity of commitments and less family responsibility than do middle-
physical health measures considered and the sensitivity of aged workers. Older workers may therefore face less pres-
results to the measures used. With reasonable agreement, sure to become reemployed, and may experience less stress
however, the longitudinal studies have demonstrated a neg- by remaining unemployed than younger workers (Warr et
ative effect of job loss on a range of measures likely related al., 1988). Nonetheless, this investigation did not use a com-
to physical health, including the number of reported medical parison group to contrast the unemployed men, which is
conditions, rates of medical services utilization, and pension necessary to describe the effects of job loss itself.
disability use (Ferrie, Shipley, Marmot, Stansfeld, & Smith, Given the limited number of studies on older workers, the
1998; Hamilton, Broman, Hoffman, & Renner, 1990; Hamil- health consequences of late-life job loss in the United States
ton, Hoffman, Broman, & Rauma, 1993; Kasl & Jones, in remain largely unknown. The objective of this research is
press; Morris & Cook, 1991; Westin, 1990). In terms of therefore to estimate the health effects of involuntary job
mental health, longitudinal data analysis has strongly sup- loss among older workers in the United States, using longi-
ences could bias estimates of the effect of job loss on post- a room; climbing several flights of stairs; climbing one
separation health. flight of stairs; getting up from a chair; getting in and out of
To construct the analysis sample, we created a continuous bed; bathing; dressing; eating; lifting or carrying over 10
monthly record of employment between interview dates for pounds; stooping, kneeling or crouching; pulling or pushing
the 4,792 workers eligible for the analysis, using retrospec- large objects; and extending arms above shoulders. Because
tive data collected at the 1994 survey. We next investigated response options for these items varied between the baseline
employment information for workers with intervals of un- and follow-up HRS interviews, comparable measurement of
employment. Respondents reporting job exit between sur- the level of difficulty in performing each task was not possi-
vey dates due to plant closure or layoff were characterized ble unless items were dichotomized to identify respondents
as experiencing involuntary job loss. Our sample was then with difficulty performing a particular activity (1 ⫽ respon-
limited to workers who were either continuously employed dent reported having some difficulty; 0 ⫽ otherwise). Di-
or had experienced involuntary job loss (n ⫽ 3939). Respon- chotomized values were summed to calculate a single
dents who reported job departures for other reasons (n ⫽ functioning score, ranging from 0 to 15, with higher values
853) at the follow-up interview were excluded from the reflecting poorer physical functioning.
analysis sample. Other reasons for job exit and their fre- Mental health was measured using a depression score
quencies were disability (n ⫽ 130), family reasons (n ⫽ based on 8 items from the 20-item Center for Epidemiologi-
reestimated the loadings. Factor loadings supported the In addition to baseline physical functioning, a number of
premise that the 15 items comprising the physical functioning other baseline physical health measures known from previ-
score are appropriately aggregated into a single dimension. ous literature to be predictive of various health outcomes,
A similar procedure was followed to investigate the valid- including physical functioning, were included in the models
ity of the 8-item depression score. Once more we observed no of post-loss physical functioning. This set of covariates in-
apparent factor structure from fitting the two-factor solution. cluded (a) obesity, which was determined as a body mass
Such findings indicate one underlying concept present in the index of at least 30, the most recent criterion for obesity in
depression score. This is not surprising when one considers this age group (Flegal, Carroll, Kuczmarski, & Johnson,
that the elements comprising our mental health measure were 1998), (b) whether a physician had ever told the respondent
drawn from a well-established empirical tool designed specif- that he or she had high blood pressure or hypertension, heart
ically to measure the phenomenon of depression. disease, or cancer or a malignant tumor, (c) heavy smoking
The use of a self-reported measure of physical functioning (at least 20 cigarettes per day), and (d) heavy drinking (at
and a depression battery, rather than a single, global measure least 3 drinks per day). The effect of these covariates on
of self-assessed physical or mental health, as the basis for es- mental health is less clear from previous literature. In addi-
timating health changes associated with job loss has empiri- tion, these covariates were not associated with follow-up
cal support. Numerous studies provide evidence that jobless mental health in this sample; their removal from the mental
ers in the sample who experience involuntary job loss alternatives was that the dependent variables were some-
relative to those who remain continuously employed. what nonnormally distributed (asymptotically declining).
We estimated Equation (1) using ordinary least squares Two alternative methods to estimate Equation (1) were
(OLS). As our data are longitudinal, the possibility of serial therefore investigated, each of which addressed the shape of
correlation exists. We therefore tested for serial correlation, the outcome variable’s distributions.
calculating Durbin’s h statistic (Durbin, 1970), a modified First, we estimated a log-linear variant of the model,
version of the Durbin–Watson statistic, that accounts for the transforming the dependent variables to their natural loga-
lagged dependent variables in our models. For the physical rithms (adding 1 to each value of the dependent variable to
functioning regression we find no evidence of serial correla- avoid losing observations with 0 health scores) and estimat-
tion. For the mental health regression, however, we may not ing the model with OLS. Second, we estimated the model
rule out serial correlation, as we were unable to calculate the h with Poisson regression, a maximum likelihood technique
statistic. (One element of the statistic involves a square root; for estimating the likelihood of rare events (in this case,
for mental health the number under the radical was negative.) physical disability or poorer mental health status). In both
An alternative test is proposed by Durbin (1970), but requires a cases, the results were not qualitatively different from the
second lag in the variables, which is not possible with two reported estimation results using least squares estimation.
waves of data. It should be noted that, even if the error co- Another important issue considered in this analysis was
Table 1. Means of Selected Regression Variables (N ⫽ 3116) among workers suffering involuntary job loss, longer unem-
Unweighted Weighted
ployment spells are suggestive of poorer physical function-
ing and mental health at follow-up. However, the observed
Variable M SD M SD effect was not statistically significant ( p ⬎ .05). Reemploy-
Involuntary job loss 0.067 0.250 0.067 0.209 ment was positively associated with physical functioning
Duration of joblessness (months)a 8.591 6.850 8.590 5.599 and mental health at the follow-up interview ( p ⬍ .05). This
Age 55.365 2.986 55.437 2.514 result suggests that reemployment may be protective against
Education (number of years) 12.583 3.046 12.896 3.387 the declines in both physical functioning and mental health
Male 0.516 0.499 0.526 0.417
associated with remaining unemployed. However, when
White 0.738 0.439 0.836 0.309
Married 0.749 0.433 0.738 0.367
considering the reemployment result, the possibility of re-
Blue-collar 0.292 0.455 0.267 0.370 verse causality may not be eliminated, because physical
Labor income 30,832 26,125 32,789 23,151 functioning and mental health were not assessed after job
Nonhousing net worth 117,992 336,570 128,868 291,623 loss and before reemployment for workers who had lost
aAmong workers experiencing involuntary job loss.
their jobs between the baseline and follow-up interviews.
DISCUSSION
Table 2. Unadjusted Associations Between Independent Variables and Health Measures ( N ⫽ 3116)
Physical function
score Mental health score
even after we controlled for a comprehensive set of health, untary job loss on mental health remains substantial and sig-
sociodemographic, and economic factors that might account nificant after we controlled for baseline measures of mental
for selection effects, suggesting that the observed associa- health, and sociodemographic and economic factors, sug-
tion is not due only to selection effects. gesting again that the observed negative effects of involun-
The evidence of the effect of involuntary job loss on tary job loss are not due only to selection effects.
mental health is even more heavily weighted toward the Few previous studies have had rich enough data to assess
causality interpretation. The unadjusted and adjusted associ- how the health effects of involuntary job loss may vary by
ations between involuntary job loss and post-loss mental subgroups of respondents, an important factor in the plan-
health are nearly identical, suggesting that the effect of job ning of targeted interventions to buffer negative health ef-
loss on post-loss mental health is not due to poorer baseline fects. In analyses not shown, interaction models were
mental health among workers experiencing job loss. In ad- estimated to test subgroup differences in the health conse-
dition, the association between baseline mental health and quences of involuntary job loss. Our results indicated that
involuntary job loss was small. Further, the effect of invol- both older and unmarried individuals may be especially vul-
S138 GALLO ET AL.
Table 3. Least Squares Regressions of Wave 2 Health: The Effect be significant for post-loss physical functioning, although
of Involuntary Job Loss (N ⫽ 3116) the interaction effect of marriage was in the expected (buff-
Physical functioning Mental health
ering) direction.
This research has also demonstrated that for displaced
Variable b (SE) b (SE) workers, reemployment is positively associated with post-
Intercept ⫺0.894 (0.558) 2.006*** (0.512) separation health. This was the case for both physical func-
Involuntary job loss 0.271** (0.111) 0.423*** (0.102) tioning and mental health. Our finding indicates that
Age 0.035*** (0.009) 0.001 (0.008) securing a new job may diminish the health impacts of job
Education ⫺0.032** (0.011) ⫺0.079*** (0.010) loss. In addition, our results are suggestive of an accumula-
Male ⫺0.182** (0.069) ⫺0.055 (0.062)
White 0.104 (0.078) ⫺0.212** (0.072)
tion effect of protracted unemployment. This was again in-
Married ⫺0.160** (0.066) ⫺0.161** (0.061) dicated for both health dimensions.
Blue-collar 0.083 (0.075) 0.001 (0.069) The study has limitations that merit recognition. Al-
Labor income ⫺0.022 (0.011) ⫺0.010 (0.001) though health status measures were ascertained prior to job
Nonhousing net worth ⫺0.007 (0.008) ⫺0.010 (0.008) loss, it is possible that respondents’ health status may have
Hypertension 0.111 (0.059) — — changed after the 1992 interview and prior to job loss. How-
Cancer 0.054 (0.129) — —
ever, it is unlikely that such changes in health would be sys-
Table 4. Least Squares Regressions of Wave 2 Health: The Effects of Unemployment Duration and Reemployment ( N ⫽ 209)a
Effect of Unemployment Duration Effect of Reemployment
Notes: b ⫽ unstandardized regression coefficient. The estimated coefficients on labor income and nonhousing net worth were rescaled for consistency
with dependent variable scale.
aN ⫽ 207 in the unemployment duration model.
this way due to the social stigma that may accompany layoff. D’Arcy, C. (1986). Unemployment and health: Data and implications. Ca-
In addition, there remains the potential for response rate nadian Journal of Public Health, 77, 124–131.
D’Arcy, C., & Siddique, C. M. (1985). Unemployment and health: An
bias, although HRS response rates are quite high for a na- analysis of Canada Health Survey data. International Journal of Health
tional survey. It has been found that initial HRS nonrespon- Services, 15, 609–635.
dents (who were later induced to respond) were more likely Diggle, P. J., Liang, K., & Zeger, S. L. (1994). Analysis of longitudinal
to be non-White and less likely to report poor physical data. New York: Oxford University Press.
Durbin, J. (1970). Testing for serial correlation in least-squares regression
health than were traditional respondents (Juster & Suzman, when some of the regressors are lagged dependent variables. Econo-
1995). In this study, we have controlled for both race and metrica, 38, 410–421.
health status. However, the potential response bias implied Farber, H. S. (1993). The incidence and costs of job loss, 1982–91. Brook-
by Juster and Suzman (1995) remains a concern for the gen- ings Papers: Microeconomics, 73–132.
eralizability of the reported results. Ferrie, J. E., Shipley, M. J., Marmot, M. G., Stansfield, S. A., & Smith, G. D.
(1998). An uncertain future: The health effects of threats in employ-
Finally, data were available for only the 2-year follow-up ment security in white-collar men and women. American Journal of
date; we could therefore assess only the short-term conse- Public Health, 88, 1030–1036.
quences of involuntary job loss. Future studies will be able Flegal, K. M., Carroll, M. D., Kuczmarski, R. J., & Johnson, C. L. (1998).
to estimate the longer run effects of involuntary job loss on Overweight and obesity in the United States: Prevalence and trends,
1960–1994. International Journal of Obesity, 22, 39–47.
health, as additional waves of the HRS become available.
ology, 132, 983–992. Warr, P. B., & Jackson, P. R. (1987). Adapting to the unemployment role: A
Swaim, P., & Podgursky, M. (1991). Displacement and unemployment. In longitudinal investigation. Social Science and Medicine, 25, 1219–1224.
J.T. Addison, (Ed.), Job displacement: Consequences and implications Warr, P. B., Jackson, P. R., & Banks, M. (1988). Unemployment and men-
for policy. Detroit, MI: Wayne State University Press. tal health: Some British studies. Journal of Social Issues, 44, 47–68.
Uchitelle, L., & Kleinfield, N. R. (1996, March 6). The downsizing of Westin, S. (1990). The structure of a factory closure: Individual responses
America: Analysis of Displaced Workers Survey, Bureau of Labor Sta- to job-loss and unemployment in a 10-year controlled follow-up study.
tistics, The New York Times, Section 1, p.1. Social Science and Medicine, 30, 1301–1311.
U.S. Department of Labor (1993). Taxable and reimbursible claims data.
Washington, D.C.: U.S. Government Printing Office.
Warr, P. B., & Jackson, P. R. (1984). Men without jobs: Some correlates of
age and length of unemployment. Journal of Occupational Psychology, Received March 31, 1999
57, 77–85. Accepted November 3, 1999