| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
ORIGINAL ARTICLES |
From the Center for Society and Population Health (B.C.A.), University of Texas School of Public Health, Houston, Texas; Department of Public Health Sciences (P.M.), University of Toronto, Toronto, Canada; The Health Institute (H.C., W.H.R.), New England Medical Center, Boston, Massachusetts; Center for Aging and Human Development (C.F.P.), Duke University, Durham, North Carolina; and Center for Policy Studies (G.D.), Northwestern University, Chicago, Illinois.
Address reprint requests to: Benjamin C. Amick III, PhD, Associate Professor, University of Texas Health Science Center, School of Public Health, PO Box 20186, 1200 Herman Pressler, Houston, TX 77225. Email: bamick{at}sph.uth.tmc.edu
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
METHODS: A working cohort was created using the U.S. Panel Study of Income Dynamics. Information on psychosocial and physical work conditions were imputed using the Job Characteristics Scoring System exposure matrix for the period 1968 through 1991 to construct working life courses. Deaths were ascertained from 1970 through 1992.
RESULTS: Working in low-control jobs for a working life was associated with a 43% increase in the chance of death (OR, 1.43, 1.13–1.81) assuming a 10-year time lag. No significant effect was found for high-strain work (ie, high psychosocial job demands and low job control), but a relationship was found between passive work (ie, low psychosocial job demands and low job control) and mortality (OR, 1.35, 1.06–1.72). No significant risk of death was found for psychosocial or physical job demands, job security, or work-related social support. Retirement (OR, 2.85, 1.59–5.11) and unemployment (OR, 2.26, 1.65–3.10) transitions and baseline disability (OR, 1.38, 1.06–1.79) predicted mortality.
CONCLUSIONS: The results support the importance of job control to health. The passive work effect suggests that job content may be important in shaping a worker’s health over the life course. Future research should focus on modeling stressors over the life course to capture the dynamic interplay of life transitions, stressor intensity and duration and the role of health in the interplay.
Key Words: life course work exposures, • job strain, • mortality.
Abbreviations: CI = confidence interval;; JCSS = Job Characteristics Scoring System;; JDL = job-decision latitude;; OR = odds ratio;; PJD = psychological job demands;; PSID = Panel Study of Income Dynamics.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Three psychosocial work conditions are considered to increase mortality risk: 1) low levels of job control, 2) high levels of psychological job demands, and 3) few or no workplace social supports (6–8). These three conditions are often combined to define hazardous psychosocial work environments. Karasek and Theorell (9–11) hypothesize that workers who do not have enough job control to meet job demand requirements experience job strain. In the job-strain model, psychological job demands are cross-classified with job control to create four work states: active work (high demands and high control), passive work (low demands and low control), low job strain (low demands and high control), and high job strain (high demands and low control). High strain increases morbidity and mortality risk (7, 12). Johnson and Hall (13) and Johnson et al. (14) extended this model by combining low levels of social support with low job control and high job demands (high strain) and coined the term "iso-strain" to describe this work state. Iso-strain increases morbidity and mortality risk (13, 14). Despite the growing acceptance of these combined effect models, enough prospective studies have found no effect to raise concern (14–21). New approaches, such as the effort-reward imbalance model, offer alternative conceptualizations to the Karasek job-strain framework but remain too new to evaluate (22). Comparisons of the Karasek job-strain model with the effort-reward model have been made with mixed results (20). Still others have examined the independent contributions to mortality of job demands, job control, and support, finding strong evidence for the importance of only job control (17, 19, 20, 23). Broadening the job-strain perspective to consider a person’s working life course may enrich psychosocial work stress research and at the same time bring the working life years more centrally into aging and life course research.
Inconsistent job-strain effects could be the result of psychosocial work conditions measured with no attempt to estimate cumulative or lifetime exposures (6, 17, 24). Using only a baseline measure assumes that all the pertinent information about a person’s psychosocial work experiences has been captured at this one time point. A baseline measure may be capturing something totally different for younger, healthier, or less experienced workers compared with older workers with accumulated illness, thus creating inconsistent findings. In the only systematic study of cumulative job-strain exposure and mortality, Johnson et al. (17) found that whether a worker worked 5, 10, 15, or 20 years, the risk of cardiovascular mortality was significantly increased only for low-control jobs (no effect was found for job strain). Johnson’s innovative study included only Swedish men and relied on retrospective recall of occupational history over a 25-year period. Despite these limitations, the study supports a cumulative exposure approach that considers a person’s working life course.
A second advantage to a life course perspective is that it suggests possible alternative psychosocial explanations for increased mortality risk to be incorporated into work-stress research. A person participates in many roles and experiences multiple role transitions that can contribute to mortality (3). In particular, the retirement transition is one important predictor of mortality that has not been included in any psychosocial work research despite its demonstrated relationship to mortality (25). Another set of significant life transitions is marriage, divorce, and widowhood—important contributors to mortality (26). Although adjustment for marital status is often done, no attention has been paid to divorce or widowhood in work and health research. In most research on psychosocial work conditions and mortality, even the transition from employment to unemployment and the state of unemployment have not been assessed.
In this article, a person’s lifetime exposure to psychosocial work conditions was modeled over the working life course, and its relationship to mortality was assessed in a representative sample of U.S. workers. We hypothesized that a career characterized by an extensive percentage of the working life in high job strain (high psychosocial job demands and low job control) increases the risk of death. In addition, we examined the independent effects of one physical and four psychosocial job conditions. We hypothesized that a career characterized by a working life with low job control, low social support, high psychosocial or high physical demands, and high job insecurity increases the risk of death. The contribution of these various psychosocial work conditions to mortality was examined in the context of multiple life course psychosocial stressors.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
The Analysis Sample
The sample excluded any person who never worked or only worked 1 or 2 years between 1968 and 1991. Whether a person was working was defined using multiple criteria. If a person said they were working in the past year (full time, part time, or temporary), then he or she was classified as working. If a person was missing information on work status in any year, then he or she was classified as working if annual working hours were greater than 500 hours or individual labor income was greater than $1000 (adjusted to 1992 dollars). Only household heads and spouses were included in this analysis because much of the information in the household interview was not collected for other working household members. Heads and spouses were excluded if they entered the study before age 18 or after age 62. This age exclusion was intended to avoid ages where working may not be the primary life course activity. The upper age restriction allowed us to follow a person a minimum of 3 years until age 65, the standard retirement age.
Death was ascertained through yearly reports of the reasons for nonresponse. In the majority of instances, a surviving household member reported deaths at the next annual interview. Death ascertainment for those living alone came from a variety of sources including a surviving contact person, administrators of the individual’s estate, and other family members. Death could not be ascertained in all instances. Unfortunately, the lack of unique identifiers (eg, Social Security number) and a National Death Index only extending back to 1979 precludes further death ascertainment. Comparisons of the PSID sample death rates with national data show that the PSID tends to reasonably approximate the National Center for Health Statistics death rates except for younger ages (33). A total of 726 deaths were observed in the analytic sample.
The Jobs Characteristics Scoring System Job Exposure Matrix
Psychosocial and physical work conditions (hereafter referred to as exposures) were estimated using the Job Characteristics Scoring System (JCSS), a job exposure matrix developed from worker self-report exposure information contained in three Quality of Employment Surveys (1969, 1972, and 1977) (25). The JCSS assigns to individuals psychosocial and physical work condition information in data sets where no self-report psychosocial work data exists but occupation (ie, three-digit census occupation code) information exists (10, 24, 34). This allows testing the Karasek job-strain model. One physical exposure and four psychosocial measures were imputed using a standard multivariate imputation algorithm (34). The five scales are defined below along with the number of self-report items used to create the scale (in parenthesis).
The JCSS validity has been discussed elsewhere (10, 24, 34, 35). The job exposure information was linked to the PSID occupational data using the three-digit 1970 census occupation code. Because the Quality of Employment Survey participants did not represent all 424 occupations listed in the 1970 census, some occupations (N = 20) were not assigned exposure values in the PSID. This resulted in 3154 person-years of information being excluded. No additional occupational information was ascertained between interviews, thus any event that occurred after the interview but did not endure until the next interview (ie, a temporary job change) could not be considered.
Constructing Cumulative Lifetime Exposure Measures
Cumulative lifetime exposure was estimated as a time-varying measure. To estimate a cumulative lifetime exposure, the percentage of time a person spent in each quartile of the exposure distribution (high, medium high, medium low, and low) was calculated. The quartiles were defined based on all person-years of information. The percentage of time spent in each quartile of the exposure distribution was recalculated year-by-year taking into account all information up to and including that year. This created up to 24 person-years of lifetime exposure information. If a person was not working in any year, the information from the prior year was carried forward until death, new exposure information was encountered, or 5 years past the last job when the individual’s information became censored.
To illustrate this method, consider the information displayed in Table 1. Column 1 displays the study year. In column 3, the imputed job-decision latitude scale values are listed. The occupation shown in column 2 provides the linkage to the JCSS job exposure matrix that allowed imputation. The last four columns represent year-by-year cumulative calculations. For example, in 1972, 2 of the 5 years the person had been working were spent in the medium-low quartile for a job-decision latitude scale value of 0.4. In 1975 and 1976, the information from 1974 was carried forward and used to predict mortality. Using the statistical procedures described below, this method allowed estimation of the hazard of death for a person who spent 100% of his or her working life in an exposure quartile compared with the reference category (always the "high" quartile). We hypothesized that the lowest quartile was most hazardous for job-decision latitude, job security, and work support, whereas the highest quartile was the most hazardous for both job demand measures.
|
The median split approach may lead to Type I errors (36), so the recommended quotient method was also used (11). In this method, psychosocial job demands was divided by job-decision latitude to create a continuous measure where a high value indicates high strain and a low value indicates low strain. Because PJD had a lower range of values (4.6–49.2) compared with JDL (15.8–97.93), the job-strain quotient measure can hypothetically vary between 0.05 and 3.11. Higher values indicate more demands and less control (high strain), and lower values indicate less demands and more control (low strain). The disadvantage to this method was that it did not allow effect estimation of the four theoretically meaningful work states. Because of our interest in estimating cumulative lifetime exposure, we estimated the percentage of time in each of four quartiles of the job-strain quotient distribution following the procedures described above. We hypothesized that the highest quartile is the most hazardous (high job strain) compared with the lowest quartile (low job strain).
An advantage of the cumulative lifetime exposure method was that it maximized the information in the PSID. It allowed an individual to contribute a minimum of 3 and up to 24 years of exposure information to the estimation. The average number of years a person contributed was 9.7. It addressed the missing exposure information problem when a person was not working (midcensoring). When someone stopped working, he or she continued to contribute information to the model on both work status and exposure.
If you follow a person for a specified period of time, cumulating exposure as described above, and then the person stops working, there is a question about how long to follow that person to see whether death occurs. This demanded specification of lag times. If someone is working and then he or she stops working because of job loss, health, childbirth, or family responsibilities, this person may stay out of work for extended periods of time before returning, if ever, to work. This required stipulating an exposure lag and a point where exposure information for an earlier period in the working life becomes censored. Similarly, if a person retires and will no longer work, how long do you follow to observe a death transition? To specify a lag, we assumed that the psychosocial work exposure effects would persist for a period of up to 5 years after a person reported not working. At some point, the psychosocial and material effects of not working would begin to be more significant to poor health. To capture these effects, unemployment status and family income were measured yearly. This applies to situations where a person stops working for extended periods during his or her working life course as well as at the end of the working life course (ie, retirement). The 5-year lag allows comparison with prior mortality research on the PSID (33). To test this lag assumption, we also estimate a 10-year lag period and hypothesize that the family income and not working effects will be more important contributors to the mortality experience compared with psychosocial work exposures.
Covariates
Five sociodemographic variables were assessed. Two were fixed (gender and race), and three were time varying. Gender (male = 1) and race (black = 1) were coded as dummy variables. Age, family income, and family size were assessed for the year preceding the current interview year. Yearly family income was standardized to 1992 dollars. Family size was entered yearly as an adjustment for family income. Current interview year was also introduced to enable estimation of a temporal effect and allow approximation of instantaneous hazard rates in discrete event history models (37). Because current year and years since first job were highly correlated (r = .99), we could not include in multivariate models as a left censoring adjustment years since first job. Age and year were marginally related (r = .24).
Unemployment status was ascertained yearly from a work-status question (unemployed = 1). Unemployment status was used to adjust for the "healthy worker effect" (38). Retirement status was ascertained yearly from a work-status question and coded as an indicator variable (1 = retired). Retirement and unemployment were assessed for the year preceding the current interview year. Finally, we assessed baseline health defined as self-reported work disability at first year of study entry (disabled = 1). Disability status was ascertained from a question asking whether a person had a physical or nervous condition that limits the type or amount of work. A person’s baseline health was carried forward. Because health data were not collected for PSID household heads and spouses for all study years, multivariate models with baseline health were reduced by 5911 person years with a 5-year lag and 6029 with a 10-year lag.
Education, an important sociodemographic variable, was excluded because it did not significantly predict mortality in this sample of working adults after adjustment for age, race, gender, and year in preliminary analyses. Furthermore, as expected, education was highly correlated with job exposure variables (its correlation with job-decision latitude is 0.43), so including it in models might be an over-adjustment reflecting a major pathway through which education affects mortality in the working age population: how education sorts people into jobs.
Statistical Analyses
Discrete time-event history analysis was used to examine whether and when a death transition occurs across the interview years. The hazard of death, h(t), is the probability that death will occur in year t1, given exposure information cumulated to the preceding interview year (t0). The conditional probability is a function of time-varying and fixed covariates and can be represented as a logit function: Logit h(t) = XA + Z(t)B, where X is a vector of fixed covariates, Z(t) is a vector of time-varying covariates, and A and B represent vectors of parameters to be estimated.
Estimation of hazard of death was done using a file that pooled together person-years of observation. Although using person-years rather than individuals as the unit of analysis inflates the number of observations, standard error estimates and statistical significance tests were calculated in an appropriate manner (37). To adjust for the interdependence of observations, we clustered on person and estimated standard errors using robust variance estimation (39). Each person year has the following information: whether or not the person died in the current interview year, cumulated lifetime psychosocial and physical exposure information, and information on time-varying and time-independent covariates.
Sampling weights were used in all estimations (except when reporting actual N values) to adjust for differential initial selection probabilities and attrition (28). Because weights differ across time periods, time-varying weights for each interview year were used in the analysis.
In preliminary analyses with all information in the model, interactions between covariates were estimated. Only the age by retirement and age by race interactions were significant at the 0.05 level and were included in the final model.
In multivariate analysis, three models were estimated: 1) exposure and age, race, gender, and year, 2) exposure, age, race, gender, year, family income, family size, retirement, unemployment, retirement by age interaction, and race by age interaction, and 3) exposure, age, race, gender, year, family income, family size, retirement, unemployment, retirement by age interaction, and race by age interaction and baseline disability. All models were estimated with 5- and 10-year lags. Analyses were done with STATA (40).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
|
Multivariate Risk Models
The effects for the Karasek job-strain model do not vary in any meaningful way across all adjustments (Table 3). The hazard of death for high strain compared with active work was 1.36 (CI, 0.92–2.00, p = .122) for the full model with baseline disability (model 3 in 5-year lag) and 1.23 (CI, 0.85–1.79, p = .282) for the comparable 10-year lag. The hazard for low strain was nonsignificant across all adjustments. The hazard for passive work was 1.35 (CI, 1.06–1.7, p = .016) for the 10-year lag (model 3) and consistently significant across all adjustments. The high-strain risk remained nonsignificant (OR, 1.32, CI, 0.84–2.08, p = .235) with low strain the reference group in a 5-year lag model.
|
|
In all models, two interactions were significant: 1) age and race and 2) age and retirement. A black person was more likely to die at a younger age compared with a white person, yet there seemed to be a survivor effect for blacks over the course of this panel study. For example, in Table 3 model 3 with a 5-year lag, the hazard was 0.97 (CI, 0.95–0.99, p = .007). In Table 3 model 3 with a 5-year lag, a younger worker was at a greater risk of death if he or she transitioned to retirement (OR, 0.92, CI, 0.90–0.94, p < .0001). In general, being black or retired placed a person at a greater risk of dying than being nonblack or nonretired, however the risk varied depending on the age of the individual.
Table 5 shows the independent hazard rates for each psychosocial and physical job condition. Three series of adjusted risks are shown for each exposure. The first series adjusts for age, gender, race, and year; the second adds household income, family size, unemployment and retirement status, and age/retirement and age/race interactions; the third adds baseline disability. These adjustments were estimated twice using 5- and 10-year lags. These data indicate that spending a working life in a job with low-control resources increases a person’s hazard of death 43% to 50% after adjustment for sociodemographics, life stressors, and baseline health (5-year lag: OR, 1.5, CI, 1.17–1.91, p = .001; 10-year lag: OR, 1.43, CI, 1.13–1.81, p = .003). The hazard dropped for workers who spent their working life in jobs with medium-low control (5-year lag: OR, 1.35, CI, 1.01–1.83, p = .042; 10-year lag: OR, 1.33, CI, 1.02–1.73, p = .035). The increased hazards of death for the lowest two job-decision latitude quartiles were consistent for all adjustments. Psychological and physical job demands, work-related social support, and job security had few significant effects that became nonsignificant after adjustment for other life stressors and income or baseline health. (The full tables with all covariate effects are available from the first author.)
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Although unable to confirm the primary Karasek job-strain model hypothesis, the analysis produced new results about passive work. High-strain work was not a significant predictor of mortality using the standard estimation method (ie, using median splits of job-decision latitude and psychological job demands to create four work states). Other studies using a job exposure matrix have reported similar nonsignificant results (15–18). The use of the job-strain quotient, which is intended to reduce Type I error associated with the median split method, supported the general conclusion of no job-strain effect. Surprisingly, the job-strain quotient yielded a consistent effect for a person in the medium-high quartile, but only using the 10-year lag assumption. This may partially reflect both the low job control effect as well as the low psychosocial job demands effect represented by the passive work state; as both decrease, the job-strain quotient gets smaller.
This research found, for the first time, that a working life in passive jobs increased mortality risk compared with active jobs. Although this relationship has been specified in Karasek’s two-dimensional model as the negative element in the active learning axis (with active work the good work state promoting active learning), often it has not been explicitly tested. O’Hanlon (41) suggested that passive work may require a type of cognitive vigilance that elicits the same neurohormonal responses as high-strain work situations. Passive work could represent work depleted of meaningful content. This alienating work could result in social disengagement and/or adoption of high-risk behaviors that lead to mortality (42–44). Passive work could induce more sedentary behavior (10) that increases mortality risk, especially cardiovascular mortality. Unfortunately, the PSID does not contain information on these lifestyles, so the pathways cannot be explored.
Passive work might be a surrogate for work that is not physically active. The increased mortality risk might be due to the absence of a protective physical activity effect. Physical job demands are inversely correlated with job-decision latitude (r = -0.31). In examining physical job demand mean scores for the four work states, no differences were observed. It could be that the limited statistical adjustment for health does not capture important health effects early in a person’s working life course that may cause transitions out of high-strain work into passive work. Despite these limitations and those noted above (absence of lifestyle data to explore pathways), these data suggest that in addition to the amount of job control a person has during a working life, the meaningfulness of work may be an important contributor to the mortality experience.
Although the job exposure matrix does address one of the major problems in psychosocial work-exposure research (self-report biases), it has limitations that could explain the lack of some significant findings. The matrix may lead to misclassification because important intra-occupation variability in psychological job demands is unmeasured (7). This could result in high-demand jobs being classified as low-demand jobs, placing a person in the passive rather than the high-strain work state (24). Furthermore, the JCSS was developed using self-report job information from the 1960s and 1970s. The changes in the meaning of work and significant period effects, especially for variables like job security, have not been measured and may account for the lack of observed significance.
One consequence of approaching psychosocial work exposure through the lens of aging over the life course has been an emphasis on estimating a cumulative exposure. Our exposure measure was grounded in life course conceptualizations of role duration (3). The emphasis of our work was on the duration in a hypothetical hazardous work state. This works adds to the small number of studies that have attempted to capture a cumulative burden using income (45) or the allostatic load concept (46). An implication of this approach could be that a single point in time measure of exposure may be inadequate to capture exposure-duration effects. It remains a fundamental question as how to estimate psychosocial work exposure over the life course. An average exposure may be an adequate measure because it utilizes all the available information. It may be that the peak or highest exposure, the lowest, the most recent, or the first exposure may be substituted for a more complex cumulative measure. Future research should explore this fundamental question: whether cumulative estimation of exposure offers an advantage over a single point in time measure.
The absence of lifestyle (ie, smoking, substance use and abuse, exercise, and diet) risk factors known to predict mortality limits the study’s conclusions. Evidence on the causal relationship of job strain to these lifestyles has been mixed (9). Only a single prospective study has examined the relationship between job strain and smoking, finding that it was changes in job control, not job strain, that predicted changes in smoking (47). The evidence for drinking is mixed (48), whereas the evidence for substance abuse does highlight the importance of an interaction between job demands (either physical or psychosocial) and job control (49). Studies of obesity, body mass index, and diet are far too limited to draw any conclusion (9). Yet, the importance of smoking, substance use and abuse, diet, and sedentary behavior to mortality limits the generalizability of the findings. However, other prospective research such as the Whitehall study has shown significant relationships between job control and mortality with robust adjustment for these lifestyle factors (19, 20, 23).
Acute or chronic health states can also affect and be affected by work. Below, we address these problems in modeling. Certainly, health contributes to the nature of the human experience through the working life course (21), but more important, morbidity contributes to an individual’s frailty and makes the individual more vulnerable to risk of death. Without inclusion of these determinants of the body’s frailties, the conclusions remain limited. Similarly, early childhood experiences affect intellectual functioning, which can affect adult health (50). Early childhood and adolescent experiences need to be considered when estimating the unique contribution of psychosocial work conditions to mortality (51). The lack of any data on early childhood experience is a further limitation.
There are several important findings that hold relevance for aging and life course research. First, the observed interaction between age and race substantiates earlier research showing that aging was protective for blacks and supports the utility of the PSID for aging and life course research (52, 53). Young blacks were more likely to die compared with young whites. For researchers studying aging phenomenon or processes that focus on later adult life, these finding suggest that the adult work role contributes to an individual’s health as he or she ages, even beyond retirement. This was true when other significant life stressors (becoming unemployed or retiring) were considered. Future work should continue to explore the long-term contributions of work-role stressors to health and aging and, in particular, at what points during the life course these factors are most significant.
The differential work status and income effects using the two lag assumptions point to the importance of expanding research beyond role transitions to examine role intensity and duration (3). The longer a person occupied the nonwork role the more significant it became for his or her health. This could be the direct result of: the psychological impact of not working, the time necessary for the adoption of adverse health-risk behaviors to cope with not working, or the biological consequences of aging for individuals who retire. A comparison of the hazard rates for retirement between the 5- and 10-year lag models may illustrate the importance of role duration and intensity. The high hazard rates for retirement using the 5-year lag assumption indicate that this is an intense experience that affects mortality. The hazard rate drops substantially for the 10-year lag, suggesting the retirement effect becomes attenuated. As a person ages, the retirement role became less significant (albeit still highly predictive of mortality).
The interaction of age and retirement suggests that younger workers entering retirement may do so because of changes in health that increases long-term mortality risk. Moen (54) has shown that health determines retirement. However, it is equally plausible that having to exit a significant adult social role early in adult life may cause health problems. As a person ages, retirement becomes a more accepted role. This would account for the diminishing retirement effect with age. Although Kasl and Jones (25) recently suggested that there is little evidence to support an adverse health effect, on average, for retirement, our findings suggest the importance of viewing retirement as a life stressor that can occur at various life course stages.
Although there was no significant interaction between age and income, the income effect on mortality depended on the time lag. In the short term (5 years), while a person was working or just retired, psychosocial job conditions, health, and the unemployment and retirement transitions were more important to negative health. But over time, income exerted a protective effect. In other research using the PSID, McDonough (33) showed that income drops contributed to the middle-class American mortality experience. The ability to avoid these income drops could be the reason for the protective income effect. Why it requires durations >5 years for the protective effects of income to be observed in the context of the continued negative effects of the other life stressors remains an important question to answer.
A longstanding problem in the work and health research area has been the inattention to temporality. Temporality relates to the timing of events during a person’s life but also the social period during which the events are experienced. When a person experiences a stressor in his early 30s, it may have a different meaning than in the 50s. A person may have different expectations about work in the early part of the career compared with later in the working life. Similarly, the timing of a promotion during a person’s career can be more stressful if it does not comply with age-based norms. Our analysis of the age-retirement interaction may be another example of this normative expectation affect. Finally, the meaning of work experiences may differ for a generation that experienced the recessions of the late 1970s and early 1980s compared with the boom of the middle 1990s. An aging and life course analysis directs this more-focused analysis of age, period, and cohort effects.
These findings support several important directions for future research. First, the research demonstrates the value of large longitudinal studies. Future efforts need to consider more complete life course models. For example, the widowhood transition is an important life course transition with known mortality effects (26). However, it is not only important to consider role transitions, but also role intensity and duration. Although the JCSS is dated, it provided the best available information to model the intensity of the person’s work experience and relate it to mortality. Similar efforts need to be made with modeling the unpaid work role and unemployment duration effects. Certainly, a new job exposure matrix would enhance efforts to accurately estimate life course effects for work stressors in the 21st century.
Second, by only considering total mortality, we do not know whether the job-control effects were a result of cardiovascular disease-related deaths or whether there was a range of causes of death psychosocial work exposures affect. Clearly, to test specific etiological hypotheses requires cause of death information. The PSID is currently coding cause-of-death information.
Third, perhaps the most critical need is to address the fundamental problem of health as an intervening variable throughout a person’s life course. Health, both positive and negative, determines the intensity in which we participate in roles and the duration we stay in the role. In other research using the PSID, we have shown that health affects transitions to unemployment and exits from the labor force (55). Yet, estimating lifetime exposure effects precludes examination of health as an intervening variable. We were restricted in how yearly health information could be used because introducing health as a time-varying covariate would be an over control (56). Future work needs to estimate health as both confounder and intermediate variable and model how the intensity and duration of life stressors predict health and how health predicts the intensity and duration of life stressors (57, 58).
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Received for publication July 25, 2000.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  S Laine, D Gimeno, M Virtanen, T Oksanen, J Vahtera, M Elovainio, A Koskinen, J Pentti, and M Kivimaki Job strain as a predictor of disability pension: the Finnish Public Sector Study J Epidemiol Community Health, January 1, 2009; 63(1): 24 - 30. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. P. Daaleman and G. H. Elder Jr Family Medicine and the Life Course Paradigm J Am Board Fam Med, January 1, 2007; 20(1): 85 - 92. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A Kouvonen, M Kivimaki, M Elovainio, J Pentti, A Linna, M Virtanen, and J Vahtera Effort/reward imbalance and sedentary lifestyle: an observational study in a large occupational cohort Occup. Environ. Med., June 1, 2006; 63(6): 422 - 427. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. A. S. Wickrama, F. O. Lorenz, S.-A. Fang, W. T. Abraham, and G. H. Elder Jr. Gendered Trajectories of Work Control and Health Outcomes in the Middle Years: A Perspective from the Rural Midwest J Aging Health, December 1, 2005; 17(6): 779 - 806. [Abstract] [PDF]
|
|
|
|
|
|
D Gimeno, B C Amick III, R V Habeck, J Ossmann, and J N Katz The role of job strain on return to work after carpal tunnel surgery Occup. Environ. Med., November 1, 2005; 62(11): 778 - 785. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Ala-Mursula, J. Vahtera, A. Linna, J. Pentti, and M. Kivimaki Employee worktime control moderates the effects of job strain and effort-reward imbalance on sickness absence: the 10-town study J Epidemiol Community Health, October 1, 2005; 59(10): 851 - 857. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. A. Sloan and J. Wang Disparities Among Older Adults in Measures of Cognitive Function by Race or Ethnicity J. Gerontol. B. Psychol. Sci. Soc. Sci., September 1, 2005; 60(5): P242 - P250. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Kouvonen, M. Kivimaki, S. J. Cox, T. Cox, and J. Vahtera Relationship Between Work Stress and Body Mass Index Among 45,810 Female and Male Employees Psychosom Med, July 1, 2005; 67(4): 577 - 583. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T Chandola, J Siegrist, and M Marmot Do changes in effort-reward imbalance at work contribute to an explanation of the social gradient in angina? Occup. Environ. Med., April 1, 2005; 62(4): 223 - 230. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Kouvonen, M. Kivimaki, M. Virtanen, J. Pentti, and J. Vahtera Work stress, smoking status, and smoking intensity: an observational study of 46 190 employees J Epidemiol Community Health, January 1, 2005; 59(1): 63 - 69. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. S. Schernhammer, S. E. Hankinson, B. Rosner, C. H. Kroenke, W. C. Willett, G. A. Colditz, and I. Kawachi Job Stress and Breast Cancer Risk: The Nurses' Health Study Am. J. Epidemiol., December 1, 2004; 160(11): 1079 - 1086. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. Gimeno, F. G Benavides, B. C Amick III, J. Benach, and J. M. Martinez Psychosocial factors and work related sickness absence among permanent and non-permanent employees J Epidemiol Community Health, October 1, 2004; 58(10): 870 - 876. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. S. Williams, S. S. Ghose, and R. W. Swindle Depression and Other Mental Health Diagnoses Increase Mortality Risk After Ischemic Stroke Am J Psychiatry, June 1, 2004; 161(6): 1090 - 1095. [Abstract] [Full Text] [PDF]
|
<
