| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
a Department of Epidemiology and Public Health, Yale University School of Medicine
William T. Gallo, Department of Epidemiology and Public Health, Yale University School of Medicine, 60 College Street, New Haven, CT 06520-8034 E-mail: william.gallo{at}yale.edu.
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Methods. Using longitudinal data from the 1992 and 1994 waves of the Health and Retirement Survey, the authors estimated multivariate models to assess the impact of involuntary job loss on subsequent alcohol use. Two outcomes were investigated: reported daily alcohol consumption and onset of drinking. The analysis sample included 207 workers who experienced involuntary job loss between survey dates and a comparison group of 2,866 continuously employed workers.
Results. After baseline alcohol consumption and a variety of socioeconomic and illness-related covariates were controlled, involuntary job loss was not associated (p > .05) with number of daily drinks consumed at follow-up. However, among those who did not consume alcohol at baseline, individuals who suffered involuntary job loss were twice as likely as continuously employed individuals to start drinking by follow-up (OR = 2.01; CI = 1.06–3.80). The majority of those who began drinking at follow-up reported drinking less than 1 drink per day.
Discussion. The findings provide evidence of a significant relationship between job loss and subsequent alcohol use among baseline nondrinkers. However, the magnitude of the changes in drinking was quite modest.
THE impact of job loss among older workers on health behavior generally, and on alcohol use specifically, is of special concern for those who care for and conduct research on older populations. In the early 1990s, researchers began to document a rise in the proportion of displaced workers older than 55 years of age (Gardner 1993
, Gardner 1995). Other studies have described protracted jobless spells (Swaim and Podgursky 1991) and substantial earnings losses (Couch 1998) among this population, outcomes that may be especially damaging given that they occur in the critical period in which such workers are preparing financially for retirement.
Significant changes in employment status can produce behavioral changes, including increased alcohol use, as older workers cope with the stress of such a change in employment. Increases from light or moderate to heavy drinking can have detrimental long-term health effects for this population; evidence shows that heavy drinking is associated with increased mortality for both older (Graham and Schmidt 1998; Thun et al. 1997) and younger (Poikolainen 1995) people.
The literature examining the impact of unemployment on alcohol consumption has produced inconsistent results in terms of the magnitude and direction of postseparation changes in alcohol use. Several studies have documented increases in alcohol consumption associated with unemployment (Cobb and Kasl 1977; Crawford, Plant, Kreitman, and Latchman 1987; Ettner 1997; Janlert and Hammarstrom 1992; Leventman 1981; Plant 1979). Some have argued that this association exists because drinking behavior is related to subsequent job loss (Forcier 1988; Kasl and Jones 2000; Mullahy and Sindelar 1995). In support of this argument, several longitudinal analyses have found heavy drinking to be predictive of future job loss (Catalano, Dooley, Wilson, and Hough 1993; Morris, Cook, and Shaper 1992; Mullahy and Sindelar 1995). However, other studies have concluded that job loss predicts subsequent increased alcohol consumption, even after preseparation drinking behavior (Catalano et al. 1993; Janlert and Hammarstrom 1992; Plant 1979) and additional factors that might predict both drinking behavior and job loss (Ettner 1997) are controlled.
In contrast to studies concluding that job loss is associated with increased alcohol consumption, several longitudinal analyses have found that job loss is associated with reduced alcohol consumption (Giesbrecht, Markele, and MacDonald 1982; Iversen and Klausen 1986). Further, a number of studies have found no association between job loss and alcohol use (Broman, Hamilton, Hoffman, and Mavaddat 1995; Cook, Cummins, Bartley, and Shaper 1982; D'Arcy 1986; Morris et al. 1992).
In general, reviews of the literature (Forcier 1988; Hammarstrom 1994; Kasl, Rodriguez, and Lasch 1998) have concluded that the relationship between alcohol use and unemployment is unclear. In addition to the lack of consensus in this literature, many of the previous studies have been conducted on relatively small samples outside the United States, where norms regarding both employment and alcohol use may differ considerably from those in the United States. Moreover, existing research has studied younger workers. We are aware of no study that examines the effects of job loss on alcohol use among older individuals.
Therefore, our objective in this study was to assess the impact of involuntary job loss on alcohol consumption among older workers in the United States. Using longitudinal data from the Health and Retirement Survey (HRS), we estimated the impact of job loss on changes in subsequent drinking behavior. In view of the complexity of the association between job loss and subsequent alcohol use, we estimated two outcomes at follow-up: reported daily alcohol use and the onset of drinking behavior among individuals who did not drink at baseline. As part of the analysis, we addressed the question of whether baseline alcohol consumption is predictive of job loss. This is a companion study to our earlier study (Gallo, Bradley, Siegel, and Kasl 2000), which assessed the effect of involuntary job loss on changes in physical functioning and mental health.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
).
The analysis sample for this study (N = 3,073) consisted of 207 workers who experienced involuntary job loss between survey waves and a comparison group of 2,866 individuals who remained continuously employed over the 1992–1994 study period. The sample was selected from a broader group of 4,792 respondents who were employed at the 1992 baseline and had been working for 3 years without interruption for the same employer. The 3-year tenure screen, which is consistent with prior studies of involuntary job loss (Couch 1998; Gallo et al. 2000), reduces sample heterogeneity by eliminating seasonal workers and the marginally employed. The sample in this study differed from that of our earlier study by just 43 observations. This difference was the result of missing values in a new variable that measured declines in income between survey waves. (Wave 2 income data were missing for 43 eligible sample members and thus precluded our creating this new variable.) Further detail concerning sample selection has been previously reported (Gallo et al. 2000).
Dependent Variables
Drinking behavior at Wave 2 was measured in two ways: as a dichotomous measure of whether the respondent reported drinking any alcohol or not at Wave 2, and as a four-level categorical variable measuring the number of drinks consumed per day at Wave 2. The dichotomous measure of drinking at Wave 2 was ascertained by a single question asked of each respondent: "Do you ever drink any alcoholic beverages such as beer, wine, or liquor?" Responses were coded 1 if respondents stated that they drank alcohol at follow-up and 0 if they did not. The ordered categorical measure of Wave 2 drinking was ascertained with an additional question asked only of persons who reported that they drank alcoholic beverages: "In general, do you have less than one drink a day, one to two drinks a day, three or four drinks a day, or five or more drinks a day?" Responses were coded as 1, 2, 3, or 4, depending on which category was chosen by respondents to describe their drinking behavior.
Independent Variables
The principal independent variable in the analyses was a dichotomous dummy variable for involuntary job loss, measured as job exit due to either plant closing or layoff between survey dates. Given the 3-year sample selection requirement, our measure of involuntary job loss is equivalent to the U.S. Government's definition of job displacement.
Sociodemographic independent variables consisted of age, gender (male = 1), years of education, race (White = 1), marital status (married = 1), and occupation class (blue collar = 1). Economic variables included baseline labor income, nonhousing household net worth, and two additional variables measuring negative changes in income and assets between survey waves. Measures for baseline labor income and net worth have been described elsewhere (Gallo et al. 2000).
Inspection of plots of the dependent variable against each of the continuous covariates (education, income, net worth) revealed a linear association. The continuous covariates were therefore not categorized. However, we rescaled income (divided by 10,000) and net worth values (divided by 100,000) for consistency with the dependent variable scales.
Baseline alcohol consumption was coded identically to the dependent variable, number of daily drinks at Wave 2, measured categorically. Each model also contained a control for having a history of problem drinking, ascertained from baseline responses to four questions that comprise the CAGE questionnaire, a validated instrument (Ewing 1984; Mayfield, McLeod, and Hall 1974) designed to detect the existence of alcoholism. Consistent with most published studies using the CAGE questionnaire, we classified individuals who provided positive responses to at least two of the four CAGE questions as problem drinkers.
Following the approach used by Catalano and colleagues 1993, we used an interaction term measuring the joint occurrence of job loss and having a history of problem drinking to investigate whether the models of follow-up drinking should be fit separately for problem and nonproblem drinkers. The estimated coefficient on this interaction term was not statistically significant in any of our models (p > .05), suggesting that pooling problem drinkers and nonproblem drinkers was appropriate. A control variable was retained identifying individuals with a history of problem drinking.
To account for the possibility that the relationship between involuntary job loss and follow-up drinking might be confounded by comorbidity factors, we tested baseline mental health, measured by an eight-item scale adapted from the CES-D and described elsewhere (Gallo et al. 2000), and a set of controls for the onset of major diseases for inclusion in our models. Disease onset variables included hypertension, cancer, heart disease, and diabetes.
Data Analyses
The unadjusted associations between daily drinking at follow-up and independent variables (not shown), including involuntary job loss, were measured by differences of means. Statistical significance was determined with t tests.
We tested two multivariate models to assess the effect of involuntary job loss on the number of drinks consumed per day at follow-up among participants who reported that they consumed some alcohol at baseline. Consistent with the traditional epidemiological approach in prospective cohort studies, the first model included all respondents who reported that they drank some alcohol at Wave 1, regardless of their drinking status at Wave 2 (n = 2,015). Individuals who were not asked how many drinks they consumed at Wave 2 because they reported no alcohol consumption at Wave 2 (n = 260) were assigned a zero for the dependent variable. The result was an additional category in the outcome variable (measured 0 to 4), describing Wave 2 number of daily drinks consumed. We fit this model using ordinary least squares (OLS) regression.
We estimated the second model using selection correction methodology (Heckman 1979) to adjust for censoring of the dependent variable. Censoring occurred in our design because the 260 baseline drinkers who reported not drinking at follow-up had invalid values for the four-category dependent variable. Because nondrinkers at follow-up may differ from drinkers in unobserved characteristics, inclusion of Wave 2 nondrinkers with a zero assigned for number of daily drinks in the estimation of follow-up drinking could result in biased and inefficient estimates of the effect of involuntary job loss. On the other hand, excluding nondrinkers without accounting for the possibility of selection effects may also influence the empirical estimates. Specifically, our concern was that unobserved factors affecting the probability of drinking at Wave 2 were associated with unobserved factors related to the daily use of alcohol. The Heckman 1979 procedure accounts for this possibility, adjusting standard errors and producing consistent estimated regression parameters.
The Heckman technique involves a two-step estimation process: a first-step binary selection estimation produces a supplemental variable that accounts for sample selection bias in a second-step regression of follow-up drinking. In the first-step probit (results not shown), the response variable was a dichotomous dummy variable measuring whether the respondent reported drinking at follow-up (1 = respondent drinks at follow-up). Right-hand-side variables included the dummy for involuntary job loss, all control variables, and an additional variable to help with identification. The over-identifying variable used was a dummy variable that measured whether respondents reported beginning to take medication to treat congestive heart failure between surveys (1 = takes medication by Wave 2). This variable was strongly (negatively) associated with the sample selection criterion, respondent drinks at follow-up, but was not associated with reported daily drinking at follow-up.
In the second-step estimation, we restricted the sample to contain only those who reported drinking at Wave 2 (n = 1,755), and the response variable was reported daily number of drinks at follow-up. Independent variables included involuntary job loss, number of daily drinks at baseline (measured 1–4), and the set of controls from the first-step probit. The second-step regression also contained a supplemental regressor, the Inverse Mills Ratio, which was calculated from the estimated probit coefficients. The Inverse Mills Ratio was included to detect the presence of selection bias. The results of the first-step selection estimation are available from the authors upon request.
Several other empirical techniques for estimating the effect of involuntary job loss on reported number of drinks per day at follow-up were explored. Because the number of drinks consumed was ascertained by HRS interviewers using an ordered, categorical response, we first considered ordered logistic regression. Results derived from this analysis were qualitatively similar to those we derived using OLS; however, our data did not meet the proportional odds assumption (Agresti 1990, Agresti 1999) required for accurate interpretation of ordered logistic models. Multinomial logistic regression, a technique used for nonordered categorical outcomes, and not requiring proportional odds, was also investigated. Nonetheless, interpretation of the results from these models is difficult, and this technique did not ultimately explain the estimated relationships better than the approaches chosen. Finally, we estimated another censored regression model using the Tobit technique. In the Tobit estimation we corrected for both the lower censoring and the upper truncation of the dependent variable. The estimates from the Tobit model were very similar to those produced by the OLS technique.
Because several longitudinal studies have indicated that alcohol consumption is predictive of future job loss (Catalano et al. 1993; Morris et al. 1992; Mullahy and Sindelar 1995), we investigated whether reported baseline daily drinking was associated with involuntary job loss in our sample. We used several approaches to test for simultaneity of involuntary job loss and baseline drinking. Bivariate and multivariate logistic regressions of job loss were both estimated; in neither case did baseline daily drinking predict involuntary job loss (p > .05). We also used a more sophisticated test, the Hausman Specification Test (Hausman 1978), to examine this relationship. The test results did not indicate that involuntary job loss was endogenously determined. The results of the Hausman Test, including the first-stage logistic predictors of job loss, are available from the authors upon request.
In addition to these models of the number of drinks consumed per day at Wave 2, we also modeled the onset of drinking among baseline nondrinkers (n = 1,058). In the onset model, the Wave 2 drinking outcome was estimated as a function of involuntary job loss, sociodemographic, and illness-related covariates. Because the outcome was relatively rare, odds ratios were used as reasonable estimates of relative risks. We tested the statistical significance of the odds ratios with Wald statistics.
We ran standard collinearity diagnostics to investigate correlations among independent variables. Variable correlations were examined pairwise and did not indicate collinearity. Variance inflation factors (VIF) and tolerance statistics were also examined. Once again, we found no evidence of collinearity. Most variance inflation factors (VIF) were close to 1. The largest VIF was 1.45, on the education variable, a factor that is within the acceptable range. However, in an effort to produce more parsimonious models of follow-up drinking and drinking onset, we eliminated several variables that were not significant in bivariate analyses, and were further judged not to be confounders in multivariate analyses, from the final multivariate models.
We conducted both the selection-corrected estimation and the logistic analyses using SAS 6.12 and LIMDEP 7.0 from Econometric Software, Inc. We ran all analyses using weighted data to correct for the HRS oversampling of relevant subgroups. Data were weighted by the person-level analysis weight provided by the HRS.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
Alcohol Consumption at Baseline and Follow-Up
At baseline, 65.6% of respondents reported drinking some quantity of alcohol (Table 2 ). Among these baseline drinkers (n = 2,015), the majority (76.4%) reported drinking less than one drink per day; only 7.4% reported drinking three or more drinks per day. The overall profile of drinking behavior was similar at follow-up. About 61% reported some drinking at follow-up. Among these respondents (n = 1,881), the majority (75.4%) reported drinking less than one drink per day, and 7.1% reported drinking at least three drinks per day. Onset of drinking was experienced by 11.9% (n = 126) of the 1,058 individuals who reported not drinking at baseline. About 92% (n = 116) of those reporting onset reported very light drinking (i.e., less than one drink per day at follow-up).
|
|
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
) suggests that job loss might lead to reduced alcohol consumption. Consistent with several earlier studies investigating the relationship between unemployment and alcohol use, we found no evidence that involuntary job loss influenced the number of drinks consumed per day subsequent to the job loss among baseline drinkers. Our results did, however, indicate that involuntary job loss increased the likelihood of drinking onset among respondents who reported that they did not currently drink alcohol at baseline. Specifically, we found that, among baseline nondrinkers, individuals who suffered involuntary job loss were twice as likely to report some alcohol use at follow-up as those who remained continuously employed.
In interpreting the drinking onset finding, it is important to recognize that the measure of drinking we used to describe onset is a static measure of drinking behavior. In other words, a negative response to the question of whether one currently consumes alcohol implies little about prior alcohol use. Although a proportion of the cohort of baseline nondrinkers was undoubtedly composed of lifetime nondrinkers, this should not be assumed of all members of this cohort. Thus, in some cases drinking onset was actually resumption of alcohol use.
A second caveat related to the onset finding is also warranted. For most of the baseline nondrinkers who began drinking by follow-up, consumption was light. As we reported previously, 92% of incident drinkers reported drinking less than one drink per day, a result that suggests that, even among job losers, the magnitude of the change in drinking behavior was quite modest. On the basis of evidence that light to moderate drinking reduces mortality risk relative to no drinking (Liao, McGee, Cao, and Cooper 2000), especially in elderly persons (Scherr et al. 1992), our finding may indicate that the changes observed subsequent to job loss may have positive, rather than negative, long-term effects on older workers' health.
We cannot know the long-term effect of changes in alcohol use on health in this population. Nevertheless, the onset finding is particularly important for social gerontologists interested in the influence of life changes on the health behavior of older individuals. Our results suggest that substantial employment transitions, such as job displacement, can result in altered health behaviors, demonstrated in this research by changes in alcohol use. Although observed changes were generally modest, this remains a notable finding, especially among older individuals, whose health habits are commonly assumed to be well established.
Our earlier study (Gallo et al. 2000) reported an association between involuntary job loss and adverse changes in both physical functioning and mental health. To examine whether such changes were the result of alcohol use following job loss, we investigated the simultaneity of alcohol use and the previous health outcomes as an auxiliary component of this research. To test for simultaneity, we used a two-stage instrumental variable procedure, in which follow-up alcohol use was first estimated as a function of job loss, and follow-up health (physical or mental) was then estimated as a function of follow-up alcohol use. This model did not reveal statistical associations or indicate simultaneity, suggesting that the health effects observed in our earlier study were not the result of changes in alcohol use. Such a result seems reasonable, given both the modest level of alcohol consumption observed and the fact that health changes due to alcohol consumption, especially physical changes, would not likely be detectable at such an early stage following job loss. The simultaneity hypothesis may be better tested when a full release of the next wave of HRS data is made available.
This study's findings should be considered in light of its limitations. First, the data on alcohol use are self-reported, and may be biased by the perceived social desirability of drinking behavior. This concern has been well documented in previous studies of self-reported alcohol use (Babor, Stephens, and Marlatt 1987; Midanik 1988). The potential bias of self-reported drinking behavior would be particularly problematic for interpreting our results if participants who experienced involuntary job loss were more apt to overstate their drinking at Wave 2 than were continuously employed respondents. However, a more plausible line of reasoning is that workers who have lost their jobs might understate their alcohol use. Hence, the results regarding drinking onset are likely to be biased toward the null hypothesis. A related concern is the inherent measurement error of self-reported data. Respondents might naturally vacillate from reports of not drinking to consuming less than one drink per day and thus be mistakenly classified as experiencing drinking onset. Nonetheless, it is unlikely that such misclassification would be associated with job loss.
Second, the data do not thoroughly describe respondents' drinking behavior. For instance, HRS data do not distinguish between beer, wine, and hard liquor intake. Further, respondents reporting less than one drink per day likely represent a variety of drinking behavior, from a drink every other day to one drink per year. Although these variations in the dependent variable are not fully captured in the analysis, substantial debate remains on whether the type of alcohol consumed or differing degrees of light drinking (Poikolainen 1995) have important health consequences.
Finally, the data offer limited information on respondents' drinking histories. For this reason we were unable to distinguish individuals currently not drinking from those who had never consumed alcohol (i.e., lifetime nondrinkers). Although one might expect the behavioral response of job loss to vary between these groups, we were unable to explore this hypothesis.
Despite its limitations, this study is important for a number of reasons. It is the first study to examine this issue on a national sample of older workers in the United States. Further, unlike many previous efforts to assess the effect of unemployment on drinking behavior, this study benefited from longitudinal data with which the potentially causal link between job loss and subsequent alcohol use could be assessed. Lastly, the results provide evidence of an association between job loss and onset of drinking among those who reported they did not currently drink alcohol, a finding that has not previously been reported among this population. Assessing the health effects of such onset of drinking in this cohort would require substantially longer-term follow-up, a feasible topic for future research using the HRS.
|
Acknowledgments |
|---|
This research was supported in part by NIMH Grant MH14235.
Received for publication January 31, 2000. Accepted for publication August 2, 2000.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  G. C. Gee, J. Liang, J. Bennett, N. Krause, E. Kobayashi, T. Fukaya, and Y. Sugihara Trajectories of Alcohol Consumption among Older Japanese Followed from 1987 1999 Research on Aging, July 1, 2007; 29(4): 323 - 347. [Abstract] [PDF]
|
|
|
|
 |
|
 S. E. Locke Presidential address: psychosomatic medicine and biodefense preparedness--a new role for the american psychosomatic society. Psychosom Med, September 1, 2006; 68(5): 698 - 705. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
