| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Department of Epidemiology and Public Health, Yale University School of Medicine, New Haven, Connecticut.
2 Hebrew Rehabilitation Center for the Aged, Boston, Massachusetts.
Address correspondence to William T. Gallo, PhD, Department of Epidemiology and Public Health, Yale University School of Medicine, One Church Street, 7th Floor, New Haven, CT 06510. E-Mail: william.gallo{at}yale.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Methods. Analyzing data from the first four waves (1992–1998) of the Health and Retirement Survey, we used longitudinal multiple regression in order to assess whether involuntary job loss between Wave 1 and Wave 2 was associated with depressive symptoms at Wave 3 and Wave 4. The study sample included 231 workers who had experienced job loss in the Wave 1–Wave 2 interval and a comparison group of 3,324 nondisplaced individuals. We analyzed the effect of job loss on depressive symptoms both in the full study sample and in subsamples determined by wealth.
Results. Among individuals with below median net worth, Wave 1–Wave 2 involuntary job loss was associated with increased depressive symptoms at Wave 3 and Wave 4. We found no effect of involuntary job loss for high net worth individuals at the later survey waves.
Discussion. Our findings identify older workers with limited wealth as an important group for which the potential effect of involuntary job separation in the years preceding retirement is ongoing (enduring) adverse mental health.
ACONSIDERABLE body of research has concluded that involuntary job loss is a salient life event capable of producing an acute adverse psychological response, primarily in the form of subclinical symptomatology (Brenner & Starrin, 1988
; Frese & Mohr, 1987; Gallo, Bradley, Siegel, & Kasl, 2000; Kaplan, Roberts, Camacho, & Coyne, 1987; Warr & Jackson, 1985). Two previous studies (Gallo et al., 2000; Siegel, Bradley, Gallo, & Kasl, 2003) used data from the Health and Retirement Survey (HRS) to examine mental health impacts of job loss among workers nearing retirement, demonstrating increases in depressive symptoms in the initial follow-up after separation (Gallo et al., 2000). Little is known, however, about the longer term psychological consequences of job loss among older workers. Understanding the longer term impact of job loss on psychological health among older workers is important, as researchers have linked persistent depression and its associated chronic stress to a multitude of harmful outcomes in this age cohort, including impaired immune function (Maes, Bosmans, & Meltzer, 1995; Reiche, Nunes, & Morimoto, 2004), coronary heart disease (Rozanski, Blumenthal, & Kaplan, 1999; Sesso, Kawachi, Vokonas, & Sparrow, 1998), and increased risk of mortality (Bruce, Leaf, Rozal, Florio, & Hoff, 1994).
There are numerous reasons why older workers may be at risk for long-term depression following involuntary job loss. As displaced older workers are often not yet age-eligible for private pension payouts and Social Security benefits, such individuals may face considerable financial distress in the period of unemployment, during which they may lose both income and noncash benefits, such as health insurance. Perhaps more importantly, older unemployed individuals may encounter limited reemployment prospects (Hipple, 1999); when they do secure new positions, they may suffer severe wage losses (Couch, 1998; Hipple), diminished occupational status, loss of seniority, and reduced health and pension benefits (Beckett, 1988). What is more, job loss severs workplace-identity factors (Joelson & Wahlquist, 1987), including social status and interaction (Iversen & Klausen, 1986; Jahoda, 1981), and disrupts the balance of time devoted to labor and leisure, which can contribute to family distress (Siegel et al., 2003). Finally, late-career unemployment interrupts the accumulation of wealth that will support consumption during retirement, which is especially important, given evidence that a considerable share of wealth accrual occurs in the decade prior to retirement (Bernheim, 1997; Mitchell & Moore, 1998).
Investigators in a variety of disciplines—most notably economics—have studied the enduring effects of a significant life experience, sometimes referred to as "scarring" (Oldehinkel, Van Den Berg, Bouhuys, & Ormel, 2003; Wilhelm, Parker, Dewhurst-Savellis, & Asghari, 1999). The majority of the economic research in this area has documented the long-term effect of unemployment on wages, lifetime earnings, and likelihood of future employment (Arulampalam, 2001; Chan & Huff Stevens, 2001; Huff Stevens, 1997; Ruhm, 1991). In addition, some recent research has assessed nonpecuniary (life satisfaction and self-esteem) scarring (Clark, Georgellis, & Sanfey, 2001; Goldsmith, Veum, & William Darity, 1996; Winkelmann & Winkelmann, 1998).
In the present study, we used data from the HRS to determine the association between involuntary job loss between Wave 1 (1992) and Wave 2 (1994) and depressive symptoms at Wave 3 (1996) and Wave 4 (1998). We assumed it unlikely that all individuals who experienced involuntary job loss would, years later, exhibit elevated depressive symptoms attributable to the job loss, but we did posit that the job loss would have the greatest long-term effect on the most economically vulnerable. We hypothesized that displaced workers of lower socioeconomic standing would show evidence of long-term mental health scarring. We anticipated that the putative long-term increase in depressive symptoms would be detectable in these strata, even after adjusting for a wide range of potentially confounding and moderating influences.
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
HRS data are collected at 2-year intervals and collection of 12 waves of data is planned. At the 1992 baseline, the HRS included a sample of 12,521 participants (response rate = 81.7%). Interviewers conducted in-home, face-to-face interviews with individuals born between 1931 and 1941 and their spouses. Certain subgroups (Blacks, Hispanics, and Florida residents) were oversampled. At all subsequent survey dates, respondents were interviewed by mail or telephone. The number of participants who were reinterviewed and the corresponding response rates are: Wave 2 (11,596 interviews, response rate = 89.1%); Wave 3 (11,200 interviews, response rate = 86.5%); Wave 4 (10,856 interviews, response rate = 84.4%). The Institute for Social Research at the University of Michigan collects HRS data, and the HRS is principally funded by the National Institute on Aging. The HRS is described in greater detail elsewhere (Juster & Suzman, 1995).
Analysis Sample
For this study, we analyzed a sample consisting of individuals who had experienced involuntary job loss between Waves 1 and 2 of the HRS and a comparison group of workers who were not displaced in this same Wave 1–Wave 2 interval. In order to be eligible for our analysis, individuals had to have been at risk for job loss at the 1992 baseline interview. At-risk individuals comprised those in the HRS birth cohort (51–61 years at baseline) who reported working for pay and who were not self-employed (n = 4,730). From this at-risk group, we first eliminated 732 individuals (15.5%) who had not responded to one or more of the follow-up surveys, leaving a potential sample of 3,998 persons. We then ascertained the extent of missing data due to nonresponse. We found that 378 persons (9.5%) lacked the follow-up data necessary to construct the depressive symptoms outcome variables, and 271 (6.7%) were missing one or more components of the physical function control. The remaining missing data (64 records, 1.6%) were distributed across the other explanatory variables. Because of its importance in predicting depressive symptoms, we imputed physical function, when missing, with a multiple imputation procedure based on established risk factors for physical functioning (Stuck et al., 1998) included in the HRS data. The final sample numbered 3,555 individuals, of whom 231 experienced involuntary job loss between Wave 1 and Wave 2 of the survey.
We compared respondents whom we had excluded because of missing data to members of the analytic sample by means of either a t test (for continuous variables) or a chi-square test (for categorical variables). Although the two groups were similar in the majority of attributes, survey participants excluded because of missing data were, on average, less educated (p <.01) and more likely to be male (p <.01), non-White (p <.05), married (p <.01), and to work in a blue-collar occupation (p <.01) than members of the analysis sample.
Outcome Measures
The outcome variable in this study was a summary measure of depressive symptoms, assessed at Waves 3 and 4 of the HRS. The measure of depressive symptoms was based on the shortened form (8 items) of the 20-item Center for Epidemiologic Studies–Depression scale (CES-D; Radloff, 1977). Of the eight CES-D-8 items, six negatively phrased statements reflect the presence of depressive symptoms (respondent felt depressed; felt everything s/he did was an effort; experienced restless sleep; "could not get going"; felt lonely; felt sad), and two positively phrased statements suggest the absence of depressive symptoms (respondent enjoyed life; was happy).
The response metric for the items comprising the depressive symptoms measure was modified between the baseline (1992) survey and later administrations. This change was instituted both to ease participant burden and to facilitate symptom response via telephone interview. At the baseline, HRS interviewers offered respondents a range of frequency in the occurrence of each item, represented by the 4-point response, whereas in later waves, they limited the range of frequency by modifying the stem of the question and presenting respondents with a 2-point response.
As an illustration, consider the symptom of feeling depressed. At the Wave 1 interview, interviewers asked participants, "Tell me how often you have experienced the following feelings during the past week: all or almost all of the time, most of the time, some of the time, or none or almost none of the time. During the past week, I felt depressed." At the Wave 2 interview and all subsequent interviews to date, interviewers asked participants, "Now think about the past week and the feelings you have experienced. Please tell me if each of the following was true for you much of the time during the past week. Much of the time during the past week, you felt depressed. Would you say yes or no?"
Although HRS investigators were assessing identical symptoms across waves, elicited frequencies varied. Research conducted by HRS investigators, summarized in a project technical report (Steffick, 2000), concluded that there is no simple conversion between the baseline response scale and the one used in later waves. In order to overcome this lack of comparability, in recent work we applied a linking algorithm based in item-response theory in order to generate equated depressive symptom scores for each respondent at each wave (Jones, 2001; Jones & Fonda, 2004). We used the resulting equated depressive symptoms scores in the current study. A full discussion of the equating algorithm and the associated technique appears in Jones and Fonda.
The equated measure had a baseline mean of 0.075 (SD = 0.57; range –0.43 to 2.33). It included a substantially broader range of values than the unequated (0–8) measure, with asymmetric noninteger intervals between values. Consistent with the underlying unequated (0–8) depression score, the distribution of equated scores was non-normal, with 48% of the values at baseline clustered at the minimum. Efforts to normalize the equated variable (e.g., log transformations, change scores, t transformations) proved ineffective. We deemed methods designed to accommodate non-Gaussian count data to be incompatible with the equated depression measure because of its noninteger intervals.
The reliability of the 8-item depressive symptoms measure underlying the equated measure was investigated both in our earlier, two-wave research (Gallo et al., 2000) and in the HRS report (Steffick, 2000). Cronbach's alpha coefficients computed by Steffick for the first three waves of data were.84,.83, and.81, suggesting adequate internal consistency. We also evaluated the factor structure of the 8-item measure in our earlier study (Gallo et al., 2000). Our findings, consistent with those of the majority of studies, indicated unidimensionality, or one underlying concept present in the depressive symptoms score, which supported our decision to use an aggregate score.
Exposure Variable: Involuntary Job Loss
We defined involuntary job loss as the loss of a job due to business/plant closing or layoff. We represented this by a binary variable, where 1 indicated involuntary job loss between Waves 1 and 2 of the HRS, and 0 indicated the absence of involuntary job loss in the same interval.
Other Explanatory Variables
We included covariates from a number of domains in our models. We selected factors based on existing evidence of their association with depressive symptoms (Dooley & Prause, 2004; Gallo, Royall, & Anthony, 1993; Gallo et al., 2000; Link & Dohrenwend, 1989; Penninx et al., 1998). We fixed time-invariant factors at the 1992 study baseline, and, at each survey wave, we updated variables in which changes could alter the job loss–depressive symptoms relationship. We also measured employment status design variables, which we used to account for labor force transitions, concurrently with the outcomes.
Time invariant factors
Variables fixed at the study baseline included: age (in years), gender (1 = female), race (1 = White), education (in years), and insurance status (1 = public or private health insurance); labor income, nonhousing net worth, and occupation class (white collar vs blue collar); and Wave 1 depressive symptoms. Blue-collar occupations included farming, forestry, and fishing; production and operations; and military. White-collar occupations included managerial and professional; sales, clerical, and administrative; and service occupations. Income included pretax individual labor earnings for 1991, which we dichotomized at its median value of $25,000 (1 = income > median). Net worth consisted of the sum of household nonhousing asset amounts as reported by respondents at the 1992 survey: 1992 values of checking and savings accounts; certificates of deposit, bonds, and Treasury bills; individual retirement accounts; stocks and mutual funds; vehicles; business equity; equity in real estate other than respondents' primary assets; and other reported nonhousing assets. We dichotomized net worth at its median value of $38,000 (1 = net worth > median). We dichotomized both income and net worth at the median in order to maximize statistical power for analysis of effect modification.
Time-varying factors
Variables updated at each survey wave included: marital status (1 = married or partnered); self-reported alcohol use (1 = respondent drinks), tobacco use (1 = respondent smokes), and physical activity (1 = respondent reports engaging in vigorous activity three or more times per week); self-reported heart attack myocardial infarction, nonskin cancer or malignant tumor, and stroke (1 = physician told respondent s/he had condition); obesity (1 = obese), determined as a calculated body mass index of at least 30 (Flegal, Carroll, Kuczmarski, & Johnson, 1998); and physical function (1 = function score > median), derived from self-reports of difficulty with 14 tasks (mobility, strength, and basic activities of daily living). We adapted the function measure from previous work on this cohort (Gallo et al., 2000). Higher values reflected more impaired function.
We used employment status variables in order to capture job status changes between survey waves, the importance of which has been highlighted in recent research (Dooley & Prause, 2004; Dooley, Prause, & Ham-Rowbottom, 2000; Grzywacz & Dooley, 2003). Measured concurrently with the outcomes, these variables accounted both for returns to work among the unemployed (which have been linked to changes in psychological health; Claussen, 1999; Dooley & Catalano, 1988; Gallo et al., 2000; Kessler, Turner, & House, 1988, 1989; Warr & Jackson, 1985) and for other labor force transitions. Represented by 4 dummy variables, which applied to all sample members regardless of Wave 1–Wave 2 job loss category, employment status controls indicated full-time employment (referent category), part-time employment, nonemployment, and retirement.
Relationships Among Explanatory Variables
We ran standard diagnostic analyses in order to investigate collinearity of explanatory variables in the final models. We first examined pairwise variable correlations. Next we assessed variance inflation factors, which reveal higher order collinear relationships not detectable in simple correlation analysis. The diagnostic results indicated a collinear relationship between pain and physical function. We thus omitted pain from the final set of explanatory variables.
We also tested for endogeneity of our exposure and found no evidence of this potential source of bias. In particular, we were concerned that job loss was determined by Wave 1 depressive symptoms, a lagged measure of the outcome variables. What motivated this concern was our finding that, in the stratum of low net worth individuals, people who had lost their job between Wave 1–Wave 2 had significantly (p <.05) higher baseline depressive symptoms than did members of the comparison group. In order to exclude the possibility that this baseline difference translated to biased estimates of the effect of job loss on later depressive symptoms, we used a Hausman specification test (Hausman, 1978) for endogeneity, a two-stage instrumental variable procedure. In the first stage, we estimated involuntary job loss as a function of baseline depressive symptoms, all exogenous explanatory variables, and an instrumental variable. The instrumental variable was a dummy variable specifying the geographic region (South or Midwest vs other regions) in which the participant lived. In the second step, we regressed our outcomes on job loss, baseline depressive symptoms, all exogenous variables, and a supplemental regressor, created from the residuals of the potentially endogenous variable (job loss) in the first step. The test results did not indicate endogeneity of involuntary job loss.
Analyses
We used longitudinal multivariable regression in order to evaluate whether the effect of job loss between Waves 1 and 2 of the HRS on depressive symptoms was measurable at 4-year and 6-year follow-up. Wave-specific models estimated follow-up depressive symptoms (Wave 3 or Wave 4) as a function of Wave 1–Wave 2 involuntary job loss, Wave 1 depressive symptoms, and adjustment variables. By covarying the outcomes with their lagged (Wave 1) values, we created a measure of residualized change (Cronbach & Furby, 1970).
We initially fit wave-specific, adjusted models for the full sample of participants. Among the full sample, we then tested for effect modification, in order to explore subgroup differences in the response to involuntary job loss. As we hypothesized that the most economically vulnerable sample members would be particularly susceptible to mental health scarring, we tested for variation in the response to job loss across socioeconomic strata by interacting a number of indicators of socioeconomic standing—including health insurance status, labor income, marital status, and household nonhousing net worth—with the job loss variable. Several studies (Broman, Hamilton, Hoffman, & Mavaddat, 1995; Dooley & Prause, 2004; Hamilton, Broman, Hoffman, & Renner, 1990; Turner, 1995) have also described differential responses to the experience of losing a job, either across demographic or socioeconomic groups. Because our findings revealed a differential response to job loss between individuals below and above the median of nonhousing net worth, we stratified the sample and refit the wave-specific models of job loss separately in the subsamples. We found no evidence of group-level variation in the effect of job loss for the remaining proxies.
We weighted all analyses in order to correct for the HRS oversampling of relevant subgroups. We weighted data by the person-level analysis weight, provided by the HRS, adjusted for the study's sample size.
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
|
|
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
) that identified increases in depressive symptoms associated with Wave 1–Wave 2 involuntary job loss at the initial (Wave 2) follow-up. Our present findings revealed that, among sample members with household wealth below the full-sample median, the impact of the Wave 1–Wave 2 job loss on depressive symptoms was discernible at both Wave 3 and Wave 4 after adjusting for a range of factors related to depression. The results of this study are largely supportive of our hypothesis. That is, in finding that involuntary unemployment was associated with long-term depressive symptoms in a group of individuals with limited wealth, we confirmed that the experience of late-career job displacement most predominantly affects workers of lower socioeconomic standing. However, of the several measures of lower socioeconomic standing assessed for interaction with job loss, net worth was the only proxy that indicated group-level variation in the impact of job loss. This implies that wealth, a cumulative measure of lifetime earnings, is more important in moderating the unemployment–depression relationship than are more current measures of financial position. The importance of wealth may be a function of its liquid components, such as savings, which are necessary in order to avoid the consequences of economic deprivation during the jobless period.
Our findings are also fundamentally consistent with those of a similar study of younger workers (Goldsmith et al., 1996) that identified long-term depression associated with joblessness as the mechanism responsible for scarring of self-esteem. This study, which used three waves of data from the National Longitudinal Survey of Youth, found elevated depression levels among workers who reported both unemployment and time out of the labor force two waves after employment separation.
Economic studies have proposed that longer term effects of job loss (or scarring) are primarily the result of reemployment in inappropriate or unstable positions, which carries higher risk of reexposure to displacement (Arulampalam, Gregg, & Gregory, 2001). Related epidemiologic research has linked reemployment in inadequate jobs to elevated depression (Grzywacz & Dooley, 2003), suggesting that postdisplacement employment transitions may mediate the effect of unemployment on longer term psychological health. Such a relationship may apply to our findings among lower net worth participants. These individuals, who typically possess modest savings and pension entitlements and are thus less capable than higher net worth individuals of financially sustaining their households during an extended period of unemployment, may be not be fully discriminating when considering new jobs during unemployment; they consequently accept unsuitable jobs, which ultimately influences their psychological well-being.
Descriptive analysis based on the follow-up surveys revealed that, in the low net worth stratum, individuals who reported part-time employment and who were displaced in the Wave 1–Wave 2 interval had significantly higher depressive symptoms than did part-time workers who were not previously displaced, possibly suggesting a potential mediating effect of transition to part-time employment. These descriptive findings could suggest that lower wealth individuals who were displaced in the Wave 1–Wave 2 interval were later involuntarily employed in part-time jobs, whereas the nondisplaced more willingly accepted part-time positions as bridge employment to retirement. Formal testing did not, however, indicate a strong mediating influence. In supplementary analyses (results not shown), we used two-stage instrumental variable estimation in order to explore the role of part-time employment as a mediator between unemployment and longer term depression in the lower net worth group. With the economic model of scarring as our basis, we first estimated follow-up part-time employment as a function of Wave 1–Wave 2 job loss, an instrument (age), and all exogenous variables, generating residuals. We then used the residuals from the first-stage model in order to instrument the potential mediator (part-time employment) in a second-stage model of follow-up (Wave 3 or Wave 4) depressive symptoms, otherwise specified in the same way as the models presented here in Tables 2 and 3. The estimated coefficient of part-time employment (instrumented) was strongly nonsignificant in the second-stage model, providing no evidence of mediation.
The inability to definitively attribute the depression scarring in the lower net worth group to postdisplacement employment transitions may mean that the relationship between unemployment and depression is more multifaceted than that between unemployment and long-term economic outcomes. It may also be indicative of shortcomings in the data. No observational data set can capture the complete range of life events occurring after job loss that can influence mental health. Our methodology may also be somewhat responsible. By selecting a comparison group whose members were eligible to make all of the labor force transitions available to the experimental group (with the exception of Wave 1–Wave 2 involuntary job loss), we added substantial complexity to our model. Thus, although our sample selection minimized the likelihood that our main findings were biased by a healthy worker effect—as would have been the case if we had used a comparison group of stably employed workers, which is common in shorter term studies of job loss—it made difficult the task of ascribing the long-term depression to any transition that was common to both exposed and nonexposed sample members. Future investigations should be directed toward more comprehensive modeling of the complex pathways to retirement in the setting of late-career job loss, building on earlier research efforts (Flippen & Tienda, 2000; Moen, 1996; Mutchler, Burr, Pienta, & Massagli, 1997; Szinovacz & Davey, 2004).
There are three potential limitations of note in this study. First, although we observed higher depressive symptoms at each follow-up among workers displaced within the first two HRS waves, we cannot know for certain whether the increase in depression that we attributed to job loss was actually sustained for the entire observation period. The term "scarring" must be interpreted in light of this shortcoming, which is common to all observational studies. Second, although the equated depressive symptoms data made possible the comparison of baseline depression to later measurements, they did not allow clinically meaningful implications from the results; rather, they limited the investigator to the inference of statistically significant differences. Even so, the 8-item unequated scale is, itself, an abbreviated version of the full 20-item CES-D battery, and, having no conventional cut-point, is equally restricted in its clinical application. Finally, it was impossible to fully capture the complex set of pathways in and out of the labor force taken by the heterogeneous sample analyzed in this study. Our employment status controls were therefore an aggregate approximation of actual labor force transition behavior. Nonetheless, the HRS provided the best source of data for a study with objectives such as ours.
Notwithstanding its limitations, this study makes an important contribution to researchers' understanding of the long-term effects of involuntary job loss in older workers. Our findings identify lower net worth individuals as a subgroup of older displaced workers at elevated risk for long-term depression following job loss, suggesting that the persistent impact of a stressful life event such as unemployment is most pronounced among the economically vulnerable. Researchers have not previously reported this in the aging literature. These individuals should be the target of occupational and mental health therapeutic interventions.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received for publication February 14, 2005. Accepted for publication July 13, 2005.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
hrswww/docs/userg/index.html.This article has been cited by other articles:
 |
|
 |
|
  W. T. Gallo, J. E. Brand, H.-M. Teng, L. Leo-Summers, and A. L. Byers Differential Impact of Involuntary Job Loss on Physical Disability Among Older Workers: Does Predisposition Matter? Research on Aging, May 1, 2009; 31(3): 345 - 360. [Abstract] [PDF]
|
|
|
|
 |
|
 S. L. Christ, D. J. Lee, L. E. Fleming, W. G. LeBlanc, K. L. Arheart, K. Chung-Bridges, A. J. Caban, and K. E. McCollister Employment and Occupation Effects on Depressive Symptoms in Older Americans: Does Working Past Age 65 Protect Against Depression? J. Gerontol. B. Psychol. Sci. Soc. Sci., November 1, 2007; 62(6): S399 - S403. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
