The Life Course of Unemployment and Midlife Health

Abstract

Objectives: We estimate associations between unemployment trajectories from ages 27-49 and physical and mental health at age 50. Methods: Data are from the U.S. National Longitudinal Survey of Youth, 1979 (N=6434). Group-based trajectory models are used to identify unemployment trajectories. Generalized linear models with a modified Bolck, Croon, and Hagenaars (BCH) correction are used to regress health on unemployment trajectory groups. Results: We identified “Consistently Low (70%),” “Decreasing Mid-Career (18%),” and “Persistently High (12%)” unemployment trajectories. Experiencing Decreasing Mid-Career or Persistently High trajectories was associated with worse physical and mental health at age 50 than Consistently Low trajectories. Experiencing a Persistently High trajectory was associated with worse physical and mental health than a Decreasing Mid-Career trajectory. Discussion: Timing and likelihood of unemployment are associated with midlife health. Mid-Career unemployment is associated with worse physical and mental health at age 50, but not to the same degree as Persistently High unemployment.

Keywords

employment lifecourse social factors

Introduction

Across the life course, unemployment is associated with poorer health, including a higher risk of mortality and lower levels of psychological well-being (Amick et al., 2002; Brand, 2015). Moreover, unemployment risk is an increasingly pervasive part of American employment (Hollister, 2011; Kalleberg, 2009). A single unemployment experience is associated with events that have consequences for later-life health, including the loss of employee benefits, lower social status, lost current and lower future income, and increased risky health behaviors (Burgard et al., 2012; Burgard & Lin, 2013; Krug & Eberl, 2018; Young, 2012). In fact, unemployment does not only undermine health when it is experienced: an unemployment experience “scars” future health, including lower levels of life satisfaction, poorer mental health, and worse self-rated health at time spans ranging from two to nearly 30 years after the unemployment event (Clark et al., 2001; Krug & Eberl, 2018; Strandh et al., 2014; Tøge & Blekesaune, 2015). Additionally, prior research has found a relationship between cumulative unemployment experiences and heart attacks and well-being (Booker & Sacker, 2012; Dupre et al., 2012). Yet, to our knowledge, research has not simultaneously examined people’s differential experiences of unemployment over time—both whether and when they experience unemployment risk as they age or what we refer to as unemployment trajectories—and midlife health.

There is ample reason to expect that risk of unemployment varies across the life course and across the population, and that differences in the likelihood of experiencing unemployment will be associated with health inequality at midlife. Blacks and Latinos, those with a high school degree or less, and men have a greater likelihood of both single and repeated bouts of unemployment, suggesting that while unemployment is increasingly common, it is also unequally distributed in the population (Brand, 2015). Multiple job loss events have been associated with higher risk of acute cardiovascular events and poorer health at age 40 (Dupre et al., 2012; Frech & Damaske, 2012), and there is recent evidence that persistent unemployment risk may be associated with a greater risk to health (Herber et al., 2019; Janlert et al., 2015).

Furthermore, when unemployment occurs likely also matters for health, but existing research provides limited evidence as to whether and how the timing of unemployment is associated with enduring consequences for health. Unemployment risks are often high in one’s late youth and early adulthood and unemployment risks rise again in later life (Bell & Blanchflower, 2011; Wanberg et al., 2016). The health impacts of unemployment experiences may vary across the life-span, although some evidence suggests they do not (Rowley & Feather, 1987). On the one hand, compared with later-life unemployment, early-life unemployment has been associated with greater mortality risk and depression (Roelfs et al., 2011; Strandh et al., 2014), suggesting earlier unemployment may exert a long-term impact on health. On the other hand, a German panel study found that the negative physical health effects of unemployment duration were much larger in midlife (around age 55) than young adulthood (around age 25) (Stauder, 2019), suggesting that unemployment in middle age may be most consequential.

Most prior studies examine either (1) aggregate-level unemployment statistics and health (Norström et al., 2014; Peng et al., 2021), (2) individual-level cohorts comparing unemployed to employed persons (Silver et al., 2021), (3) panel studies that examine the relationship between multiple unemployment bouts and later health outcomes (Dupre et al., 2012; Frech & Damaske, 2012) or (4) panel studies that examine the relationship between either early or later-life unemployment and later-life physical and mental health outcomes (Böckerman & Ilmakunnas, 2009; Roelfs et al., 2011; Stauder, 2019; Strandh et al., 2014). No study, to our knowledge, has conducted a prospective, longitudinal analysis over the individuals’ career trajectories to tease out differences in the mental and physical health consequences of both the timing and likelihood of unemployment. The current study addresses this gap in research by implementing a latent trajectory approach to classifying unemployment experiences over the working years and analyzing health outcomes in midlife based on these latent classifications.

Unemployment may worsen health in part because it is associated with a loss of the health benefits of stable employment, a perceived loss in social status, as well as increases in risky health behaviors and financial strain (Burgard & Lin, 2013; Krug & Eberl, 2018). After an unemployment period, there is reason to expect lasting changes in wages, family formation, occupational status, employment status, healthy behaviors, and health status (Brand, 2006; Gangl, 2006; Kalousova & Burgard, 2014). Thus, we would expect to see unemployment may undermine one’s access to resources that have been associated with health, including those related to family and financial status, access to stable work, and health behaviors and health status. If this is the case, identifying these resources could be important for interventions.

Finally, research demonstrates that health selection affects any investigation of how unemployment undermines health (Burgard et al., 2007; Stauder, 2019; Strully, 2009). Unemployment and health share a bidirectional relationship, where poor health can lead to unemployment, even as unemployment undermines health (Haas, 2006; Rueda et al., 2012). Poor mental and physical health in childhood or young adulthood is associated with poorer educational, labor market, and health outcomes later in life (Haas, 2006; Haas et al., 2011). Both alcoholism and suicide attempts are associated with unemployment, even though prior hospitalizations, cardiovascular disease, and cancer do not appear to increase risk of unemployment (Vågerö & Garcy, 2016). And there is some evidence from a panel study in Finland that suggests that the selection effects from poor health may, in fact, explain the relationship between unemployment and later-life health, although the evidence for this remains limited (Böckerman & Ilmakunnas, 2009). At the same time, the association between unemployment and poorer health is robust (Rueda et al., 2012), with research taking selection effects into consideration finding unemployment associated with poorer physical and mental health (Frech & Damaske, 2012; Stauder, 2019).

The purpose of this study is to identify common patterns, or group-based trajectories, of unemployment risk over time, and to estimate associations between unemployment trajectories and health at midlife after accounting for demographic and early-life variables. We do this using a cohort of late “Baby Boomers” from the U.S. National Longitudinal Survey of Youth 1979 (NLSY79). Building on previous research finding that more time spent unemployed is associated with greater risk of mortality and poorer health (Garcy & Vågerö, 2012; Silver et al., 2021), and that unemployment exerts a scarring effect on mental health over time (Strandh et al., 2014), we expect that groups with a higher likelihood of unemployment early in their careers or throughout their careers will experience worse mental and physical health at midlife than those with lower unemployment risk across their working years. We further expect that many of these associations will be attenuated after adjusting for household resources and health behaviors at midlife.

Data and Methods

Data and Sample

The National Longitudinal Survey of Youth—1979 cohort (NLSY79)—is a nationally representative cohort of 12,686 youth ages 14–21 in the USA in 1979 who were interviewed annually through 1994 and biennially beginning in 1996 (US Bureau of Labor Statistics, 2020). The NLSY data are well suited for this study, as respondents provide detailed employment histories at each interview and participate in health modules administered at ages 40 and 50. The NLSY retention rate is also quite high, with 69% of the original sample remaining in the most recently released wave of data in 2018 (National Longitudinal Surveys, 2019b).

To create our sample, we first excluded 2700 respondents from oversampled groups that were dropped from the survey in 1984 (the military oversample) or 1990 (the economically disadvantaged oversample) (US Bureau of Labor Statistics, 2020). Because we were interested in examining the effects of unemployment on health among those who remained vulnerable to job loss across the study period, we further limited our study to those who were ever employed between ages 27–50 (182 excluded), who were not in active-duty military (as they are classified as “out of work” rather than employed or unemployed in the NLSY, 301 excluded), and remained in the labor force (either employed or unemployed) at the time the age 50 health modules were assessed (1115 excluded). Of the 8388 respondents remaining, 6381 reported their health at age 50, 53 were present for the health module but had missing values, 1346 were lost to attrition prior to age 50, and 605 had died by age 50. Because respondents lost to attrition could not report whether they remained in the workforce at age 50, it is not known whether these respondents would have been included or excluded from the analyses. As shown in Appendix A, a multinomial logistic regression predicting mortality and non-mortality attrition (relative to retention to age 50) indicated that mortality was significantly (p<.05) more likely among men, Black respondents, those with health conditions that limited work or a job search at the baseline 1979 interview, and those whose mothers did not graduate high school (as a proxy for family-of-origin socioeconomic status). Non-mortality attrition was more likely among non-US natives, and less likely among Black respondents, those with work-limiting health conditions, and those whose mothers did not graduate high school. We control for these variables in multivariate analyses, but results are likely to be less generalizable among those more likely to attrit. For all other missing data, we impute values using the mi impute suite in Stata 16 and compare results to those found using listwise deletion (results are similar and available upon request).

Measures

Unemployment

At each interview at or near ages 27, 29, 31, 33, 35, 37, 39, 41, 43, 45, 47, and 49, respondents reported the number of weeks in the last calendar year they spent employed, unemployed (not working and seeking work), and out of work (not working and not seeking work). We focus on this age range in order to control for completed educational attainment at age 25, and continue through age 49 in order to measure health during the age 50 health module, the most recently collected health module in the NLSY79 data. Respondents born in 1957 reported their age 27 unemployment experiences in 1984, those born in 1958 report their age 27 unemployment in 1985, and so on, until 2012, when those born in 1964 reported their unemployment experiences at about age 49. We considered respondents to be unemployed at age X if they reported at least 1 week of unemployment in the last year. Short-term unemployment (less than 15 weeks) was the most common form of unemployment during nearly every year of this period (Kroft et al., 2016). Results were consistent when defining unemployment as at least 2 weeks (“one paycheck”) or at least 4 weeks (“one month”) in the last calendar year (see Supplementary analyses for details).

Physical and Mental Health

The Short-Form Survey, or SF-12 (Brazier et al., 1992; Gandek et al., 1998) is a well-validated instrument used to assess overall mental and physical health in adult populations, and is administered to all NLSY79 respondents at ages 40 and 50. We use the age 50 SF-12 physical and mental health scores, which are measured between 2008 and 2014, at the wave closest to the respondent’s 50^th birthday. The SF-12 includes 12 questions assessing respondents’ energy, pain, mobility, self-rated health, and emotional wellness. A proprietary method is used to assign separate scores for mental health and physical health; scores are scaled so that an average score is 50 and the standard deviation is 10, with higher scores indicating better physical or mental health (National Longitudinal Surveys, 2019a, p. 19).

Demographic and Early-Life Control Variables

All control variables were measured prior to age 27, the first year we began measuring unemployment trajectories (Nagin, 2005), and are included because previous research has established their relationships with health at midlife. These included gender, race-ethnicity, US nativity, whether the respondent’s mother completed high school, whether the respondent had a health condition that limited work at age 25, and educational attainment at age 25 (Crimmins & Saito, 2001; Frech & Damaske, 2019; Haas, 2006; Melvin et al., 2014).

Age 50 Confounding Variables

At age 50, we controlled for post-trajectory variables likely to be associated with midlife health outcomes. Household and financial resources included marital status, average hours worked between ages 27 and 49, and household wealth at age fifty, measured using net worth (Boen et al., 2020; Frech et al., 2021; Zhang & Hayward, 2006). Because net worth is highly skewed, we add a constant to make all values positive, and then log this value. Health behaviors and health status included frequency of binge drinking, whether the respondent smokes or is physically inactive, BMI (calculated from self-reported height and weight and centered at 25), hours of sleep on a typical weeknight, whether the respondent lacked health insurance, and mental health in models predicting physical health, as well as physical health in models predicting mental health (Prince et al., 2007; Rogers et al., 2010).

Data Analysis

We use group-based developmental trajectories (see Nagin, 2005) to identify common patterns or trends in the timing and likelihood of experiencing unemployment across ages 27–49. Group-based trajectories identify groups of individuals experiencing similar patterns of change or stability over time for an outcome of interest. Quantitative measures of model fit and prior research and theory are used to compare models and select a single model of best fit, comparing fit statistics across the number of possible groups and the shape of each trajectory (linear, quadratic, cubic, or quartic). The goal of GBTM is to identify variability in trajectories across a population, and to make meaningful comparisons across groups. After estimating a series of unconditional models using only age and unemployment variables to determine the best-fitting number of groups and polynomial order of each group-based trajectory, the second stage of the analysis uses group membership as an explanatory variable for SF-12 physical and mental health scores, adjusting first for pre-trajectory control variables, and then for confounding variables measured at age 50. As is well-known in latent classification with distal outcomes, the naïve use of predicted latent groupings as manifest variables leads to a systematic underestimation of the strength of association with the distal outcome (Bolck et al., 2004). Thus, we implement the Bolck, Croon, and Hagenaars (BCH) approach to address the latent nature of unemployment groups by performing a pseudo-ML estimation of the generalized linear model adjusting the log-likelihood function by classification weights, estimating variances using a clustered sandwich estimator (Vermunt, 2010). All analyses are conducted using Stata 16.

Results

Descriptive Statistics

Descriptive statistics for all variables other than unemployment are provided in Table 1, and the proportion of respondents experiencing any employment by age is shown in Figure 1. The proportion of respondents reporting unemployment in the last year varied by age, as shown in Figure 1. At age 27, 22% reported at least 1 week of unemployment in the last 12 months, and this number declined gradually with time to 9% by age 45, before increasing slightly to 12% by age 49. Group-based trajectory models are used in the next stage of analyses to identify heterogeneity within these overall trends in unemployment over time.

Table 1.

Descriptive statistics for all model variables, N=6434.

	Mean (sd) or percent
Dependent variables, age 50
SF-12, physical health	50.99 (8.32)
SF-12, mental health	53.78 (7.89)
Pre-trajectory control variables
Male	51%
Non-black, non-Latino (reference)	49%
Black, non-Latino	30%
Latino, non-Black	19%
Non–US native	7%
Mother did not complete high school	43%
Health limits work, age 25	3%
Educational attainment in years, age 25	12.80 (2.2)
Post-trajectory confounding variables
Average hours worked while employed, ages 27–49	39.08 (11.24)
Net worth, logged	14.85 (.26)
Married	57%
Never married	15%
Divorced	23%
Widowed	5%
Did not binge drink in last 30 days	86%
Binge drinks <1 time a week	7%
Binge drinks 1+ times a week	7%
Daily or occasional smoker	23%
Sleep hours on weekdays	6.66 (1.30)
BMI	29.08 (5.96)
Physically inactive	29%
No health insurance	17%

Note. Descriptive statistics are averaged across 20 imputed datasets using the “misum” command in Stata 16. Post-trajectory confounding variables are measured at age 50 unless otherwise noted.

Figure 1.

Proportion unemployed from ages 27-49, N=6434.

Group-Based Trajectories of Unemployment

After creating group-based trajectories for 3, 4, and 5 groups and comparing linear, quadratic, and cubic forms, we selected the three-group model with a cubic polynomial order for two groups (Consistently Low and Decreasing Mid-Career), and a quadratic order for the remaining group (Persistently High), as the model of best fit. Nagin (2005) recommends using BIC statistics (negative values closer to zero indicate better fit) and the average probability of correct placement (APP, values of >0.70 for each group indicate better fit) to aid in model selection. The polynomial orders are selected by increasing the order until no longer statistically significant and retaining the highest significant polynomial. As shown in Table 2, BIC statistics suggested the five-group model was a better fit than three or four groups; however, the APPs for three of the groups in the five-group model fell below the recommended 0.70 threshold, indicating individuals were not consistently a good fit for their assigned groups. In the four-group model, one group fell below the recommended threshold of 0.70, making the three-group model the model of best fit. As shown in Table 3, the three-group model averages an APP of .873, with all three-group APPs above 0.70. The final three-group model is shown in Figure 2.

Table 2.

Measures of model fit comparing 3, 4, and 5 group-based trajectories of unemployment.

Number of groups	BIC, person-years (N=77,208)	BIC, persons (N=6434)	Lowest APP	Across group APP
3	−27774.56	−27758.41	0.789	0.873
4	−28534.98	−28511.37	0.675	0.824
5	−27702.87	−27673.05	0.555	0.706

Note. BIC = Bayesian Information Criterion; APP = Average probability of correct placement to assigned group-based trajectory. Results are averaged across 20 imputed datasets.

Table 3.

Characteristics of group-based unemployment trajectories, N=6434.

	Percent, %	APP	Mean bouts unemployment, ages 27–49 (sd)	Mean work hours, ages 27–49 (sd)
Consistently Low	70	.869	.67 (.77)	40.54 (10.83)
Decreasing Mid-Career	18	.895	3.41 (1.51)	37.09 (10.71)
Persistently High	12	.789	5.37 (1.83)	32.36 (11.77)
All	100	.873	1.63 (1.93)	39.08 (11.24)

Note. APP = Average probability of correct placement to assigned group-based trajectory. Results are averaged across 20 imputed datasets.

Figure 2.

Group-based trajectories of unemployment from ages 27-49, N=6434.

Unemployment trajectories varied from ages 27 to 49 across Consistently Low (70%), Decreasing Mid-Career (18%), and Persistently High (12%) trajectories. We named trajectories for when in their careers they were most likely to be unemployed and for their likelihood of unemployment over time. The Consistently Low trajectory was most common and had the lowest risk of unemployment at every age between 27 and 49. At age 27, nineteen percent of Consistently Low respondents experienced at least a week of unemployment in the last twelve months, and this number declined with time, to six percent by age 49. In contrast, Decreasing Mid-Career respondents most often experienced unemployment prior to age 40, with the highest levels—35% unemployed—at age 29, before declining to ten percent at age 49. At greatest risk of unemployment at every age was the Persistently High group, with nearly half experiencing unemployment at age 27, before declining to only 38% at ages 37 and 39, then increasing again to 50% by age 49. Overall, unemployment experiences varied significantly from what is suggested by the overall trends in Figure 1. The next series of analyses estimates associations between these three-group–based trajectories and mental and physical health at age 50.

Group-Based Trajectories of Unemployment and Health

Table 4 presents results from generalized linear models predicting physical and mental health at age 50 using group membership as the main explanatory variable. Results were averaged across 20 imputed datasets. Following Vermunt (2010) and Bolck et al. (2004), respondents were assigned to a group based on modal assignment, replicated three times for each possible assignment, and weighted using the inverse of the misclassification matrix to partially adjust for the probabilistic nature of group membership. Both mental and physical health are treated as continuous, and results can be interpreted as linear regression coefficients. We also rotate the reference category from Consistently Low (as tabled) to Decreasing Mid-Career to compare the health of the Decreasing Mid-Career versus Persistently High trajectories.

Table 4.

Estimated coefficients from generalized linear model with BCH weighting to predict physical and mental health at age 50 across group-based trajectories of unemployment, N=6434.

	Model 1				Model 2
	Physical health		Mental health		Physical health		Mental health
Unemployment trajectory (ref = Consistently Low)
Decreasing Mid-Career	−1.40	***	−1.41	***	−.78	***	−.80	*
Persistently High	−2.04	***^	−3.38	***^	−.95	***	−2.13	***^
Pre-trajectory control variables
Male	1.52	***	2.01	***	.42	**	1.36	***
Black (ref= non-Latino, non-black)	.11		1.12	***	1.14	***	1.59	***
Latino	.36		1.13	***	.62		1.19	***
Non–US native	1.30	***	−.25		1.01	***	−.43
Mother did not graduate HS	−.61	***	.05		−.42	**	.17
Health limits work, age 25	−4.23	***	−.34		−3.71	***	.24
Educational attainment in years (centered at 12), age 25	.44	***	.06		.19	***	−.15	*
Age 50 confounding variables
Average weekly hours while employed, ages 27–49 (centered at 40)					.06	***	.07	**
Net worth, logged					.71		.27
Never married (ref = Married)					−.68	**	−.40	*
Divorced					−.64	***	−.82	***
Widowed					−.74		−1.55
Binge drank less than 1x/week in last month (ref = did not binge drink)					1.46	**	−1.26	***
Binge drank 1+ times per week in last month					.46		−1.05	**
Smoker					−1.58	***	−.84	***
Hours of sleep, centered at 7					.79	***	.97	***
BMI, centered at 25					−.27	***	−.04	**
Physically inactive					−2.54	***	−.83	**
Respondent lacks health insurance					−.27	***	−1.05	***
Mental health SF-12 score					.05	***	---	---
Physical health SF-12 score					---	---	.05	***
Constant	50.55	***	52.81	***	40.04	***	48.03	***

Note. *p<.05, **p<.01, ***p<.001, two-tailed tests. Reference category is Lower unemployment.

^=differs from Decreasing Mid-Career unemployment trajectory at p<.05. Results are averaged across 20 imputed datasets using mi estimate in Stata 16. Observations are replicated for each possible group assignment and weighted according to the misclassification matrix using the BCH method. Variances are estimated using a clustered sandwich estimator.

Model 1 of Table 4 estimates physical health SF-12 scores at age 50 across the Consistently Low, Decreasing Mid-Career, and Persistently High trajectories, with Consistently Low treated as reference. On average, a Decreasing Mid-Career unemployment trajectory was associated with SF-12 physical health scores 1.40 points lower than those on a Consistently Low unemployment trajectory, and a Persistently High unemployment trajectory was associated with a score 2.04 points lower than Consistently Low. The difference in health between Decreasing Mid-Career versus Persistently High was also significantly different, suggesting that early experiences of unemployment were less harmful to midlife health than a Persistently High risk. On average, men, non-US natives, and those with higher educational attainment reported higher physical health scores, and those with work-limiting health conditions at age 25 and those whose mothers did not graduate high school reported lower scores. Race-ethnicity was not associated with health at age 50 controlling for other variables.

The second column of Table 4 estimates mental health differences across Consistently Low, Decreasing Mid-Career, and Persistently High unemployment trajectories. As with physical health, the Consistently Low and Persistently High trajectories were associated with significantly lower mental health SF-12 scores, and the Decreasing Mid-Career and Persistently High trajectories also differed significantly, with Decreasing Mid-Career trajectories reporting better midlife mental health. Men and Black and Latino respondents also reported better mental health, controlling for all other variables.

The first column in Model 2 of Table 4 adds post-trajectory confounding variables that may contribute to midlife health to the model predicting physical health. Net of pre-trajectory and post-trajectory variables, Decreasing Mid-Career and Persistently High unemployment continued to be associated with worse physical health at midlife. However, the difference between Decreasing Mid-Career and Persistently High unemployment was reduced to non-significance. Of the variables added in Model 2, being never married, divorced, smoking, having higher BMI, physical inactivity, and lacking health insurance were associated with worse physical health, while averaging longer work hours while employed, greater sleep hours, occasional binge drinking (less than once a week in the last month), and higher mental health SF-12 scores were associated with better physical health at midlife.

The final column of Table 4 adds post-trajectory confounding variables to the model predicting mental health SF-12 scores at age 50. Relative to those on a Consistently Low unemployment trajectory, those with a Decreasing Mid-Career or Persistently High trajectory averaged significantly worse mental health after adjusting for these variables. Notably, unlike physical health, the difference between Decreasing Mid-Career and Persistently High remained statistically significant. In addition, the never married, the divorced, those binge drinking in the last month, smokers, those with higher BMI, and the physically inactive reported worse mental health, while averaging higher work hours, longer sleep on weekdays, and higher SF-12 physical health scores were associated with better mental health at age 50.

Supplemental Analyses

In Appendix B, we conducted a series of robustness checks for the findings presented in Table 4. We first considered our threshold of unemployment, to test whether our findings were consistent when considering at least 4 weeks of unemployment in the last year, rather than 1 week of unemployment. Our findings were similar in magnitude and significance when using a 4-week threshold. Second, because those with work-limiting health conditions are likely to experience both more unemployment and poorer mental and physical health, we replicated our analyses excluding those with work-limiting health conditions at age 49 (n=1062 excluded). Our results remained similar when excluding these respondents. Finally, because respondents in our sample may have experienced different economic circumstances based on their age, we stratified our analyses by birth year, comparing those born between 1957 and 1960 to those born between 1961 and 1964. Those born after 1960 reported slightly better physical and mental health in fully adjusted models, but experienced similar unemployment trajectories (see Appendix C) and similar health associations with unemployment trajectories (see Appendix B).

Discussion

Our study extends prior research on the lasting relationship between unemployment and poor health (Dupre et al., 2012; Garcy & Vågerö, 2013; Stauder, 2019; Young, 2012) by simultaneously considering the timing and likelihood of unemployment to identify group-based trajectories of unemployment from ages 27 to 49 and their relationship with physical and mental health at midlife. Our findings are consistent with prior research finding that those with less unemployment experience better mental and physical health. We add to these studies by demonstrating differences between Decreasing Mid-Career and Persistently High unemployment trajectories at age 50 for physical and mental health, with those experiencing a Persistently High trajectory reporting significantly worse mental health than other groups in fully adjusted models. We also find evidence of enduring unemployment tolls: both Decreasing Mid-Career unemployment and Persistently High unemployment risk are associated with poorer physical health in fully adjusted models.

Our modeling strategy allows us to account for the early-life course variables associated with poorer midlife mental and physical health when examining the relationships between health and the unemployment trajectories. By adjusting for these early-life variables, we can better isolate the relationships between the unemployment trajectories and health. In doing so, we find that the relationship between both Decreasing Mid-Career and Persistently High trajectories and poorer physical and mental health is significant at age 50, net of early-life factors. This advances prior cross-sectional research and points to the lasting importance of unemployment risk, while also showing who is most at risk of the cumulative disadvantages associated with poorer midlife health. Health limitations at age 25 and a mother who did not graduate high school were associated with worse physical health at age 50. In contrast, we found men had better physical and mental health than women at age 50; Black and Latino respondents had better mental health at age 50 than non-Black and non-Latino respondents; and Non-US (compared to US) natives and those with higher educational attainment (compared to those with fewer years of education) at age 25 had better physical health at age 50.

Our findings that the physical and mental health differences across the group-based trajectories of unemployment were reduced by the age 50 confounding variables suggests some of the scarring effects of unemployment may operate through employment-based resources and health behavior characteristics. These findings are in line with previous research demonstrating that unemployment is associated with a loss of health insurance, uptake of unhealthy behaviors, and declines in job quality upon reemployment (Damaske, 2021; Kalousova & Burgard, 2014), which in turn appear to be associated with worsened health at midlife. We see similar patterns for mental health where the relationship between both the Decreasing Mid-Career and the Persistently High unemployment trajectories is reduced, but remains significant, with the addition of the age 50 confounding variables. This suggests that the long-term toll over job insecurity may remain present for these groups and may not diminish with age. More research is needed to investigate this.

The significance of the age 50 confounding variables may be good news for policy experts, as it may point to areas of potential intervention for the physical and mental health costs of unemployment. Lacking health insurance, smoking, and a lack of physical activity were associated with poorer physical and mental health at age 50. Some variables were protective of both physical and mental health, including averaging higher work hours, not binge drinking, and sleeping more. Consistent with prior research, mental and physical health were mutually reinforcing (Frech & Damaske, 2019; Prince et al., 2007).

Our research includes several limitations. We control for, but do not exclude those with work-limiting health conditions or consider the significance of differences in the severity or timing of onset of work-limiting health conditions. While it stands to reason that these respondents will report more frequent unemployment and labor force exits than those who do not report these limitations, our findings are consistent when excluding those who report a work-limiting health condition at age 49 (see Supplemental analyses and Appendix B). Work-limiting health conditions are also an incomplete marker of the potential reverse causality between health and unemployment—poor health and health behaviors like heavy alcohol use, smoking, or other conditions may be associated with unemployment in ways that are not reflected in our models. Injury and illness cause people to leave the workforce, often in ways we are unable to address using the limited health measures available in the NLSY data before the respondents turn 40. As a consequence, our results may underestimate the differences in health at age 50. Controlling for early-life work-limiting health conditions partially adjusts for, but does not eliminate, the concern for reverse causality. Last, our control variables are measured at two points in time—pre-trajectory and post-trajectory, preventing us from assessing dynamic changes in marital status, work hours, occupation, or health behaviors on midlife health.

Using a life course approach allows us to see the reach of unemployment risk into middle age and midlife health, as we compare those whose risk of unemployment sharply declines after the age of 30 with those whose risk remains higher throughout their main working years. Our findings about mental health risk may be most troubling, as they suggest that unemployment risk leaves a lasting mark on workers. As job insecurity rises for most workers (Kalleberg, 2011), our findings suggest it will be increasingly important to identify the paths workers take to middle age and whether those paths were riddled with unemployment risk. Looking ahead, the age 50 measures point to important areas for potential policy intervention suggesting that policies aimed at improving access to full-time work and to its attendant benefits (e.g., health insurance), as well as those promoting healthy behaviors may be able to counter the negative effects of unemployment. Since the Great Recession, long-term unemployment has increased; future research should consider the duration of unemployment in addition to whether and when an unemployment bout occurred, as long-term unemployment may prove to be more damaging to one’s health (Kroft et al., 2016; Strandh et al., 2014). Moreover, in the United States, nearly 75% of workers receive health insurance through their employers (Damaske, 2021), which suggests that the longitudinal strains of unemployment may be greater here than in other countries; assessing whether and how these risks may vary in countries where healthcare is not tied to employment would be an important next step.

Footnotes

Acknowledgments

Many thanks to Jason Houle, Jamie Lynch, Steve McClaskie, and Léa Pessin.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We gratefully acknowledge support from The National Institutes of Health (NIH-NICHD) and the Population Research Institute at Pennsylvania State University (RO3HD088806, P2CHD041025).

ORCID iDs

Adrianne Frech

Adrienne Ohler

Multinomial Logistic Regression Predicting Mortality and Attrition by Age 50 Relative to Retention,Relative Risk Ratios,N = 9986.

	Mortality versus Retention		Attrition versus Retention
Female	.57	***	.92
Black (ref = non-Black, non-Latino)	1.34	**	.56	***
Latino	.87		.92
Non–US native	1.12		1.65	***
Health limits work, 1979	1.77	***	.67	*
Mother did not graduate high school	1.44	***	.74	***
Constant	.09	***	.27	***

^*p<.05, **p<.01, ***p<.001, two-tailed tests.

Supplemental Models Predicting Physical and Mental Health at Age 50 Across Group-Based Trajectories of Unemployment

	Model 2 of Table 4, N=6434				At least four weeks unemployed at age X, N=6434				Excluding those with work-limiting health conditions at age 49, N=5372				By age group, N=6434
	Physical health		Mental health		Physical health		Mental health		Physical health		Mental health		Physical health		Mental health
Unemployment trajectory (ref = Consistently Low)
Decreasing Mid-Career	−.78	***	−.80	*	−.80	***	−.53	***	−.58	*	−.76	*	−.86	***	−.78	***
Persistently High	−.95	***	−2.13	***^	−1.30	***^	−1.79	***^	−.89	**	−1.80	***^	−.98	**	−2.41	***^
Born 1961–1964 (ref = born 1957–1960)	---	---	---	---	---	---	---	---	---	---	---	---	.18	***	.52	***
Decreasing Mid-Career X Born 1961–1964	---	---	---	---	---	---	---	---	---	---	---	---	.17		.01
Persistently High X Born 1961–1964	---	---	---	---	---	---	---	---	---	---	---	---	.05		.46

Group-Based Trajectories of Unemployment by Age Group

Figure C1.

Group-based trajectories of unemployment among respondents born from 1957 to 1960, N= 3058.

Figure C2.

Group-based trajectories of unemployment among respondents born from 1961 to 1964, N=3376.

References

Amick

B. C.

McDonough

Chang

Rogers

W. H.

Pieper

C. F.

Duncan

(2002). Relationship between all-cause mortality and cumulative working life course psychosocial and physical exposures in the United States labor market from 1968 to 1992. Psychosomatic Medicine, 64(3), 370–381. https://doi.org/10.1097/00006842-200205000-00002

Bell

D. N. F.

Blanchflower

D. G.

(2011). Young people and the great recession. Oxford Review of Economic Policy, 27(2), 241–267. https://doi.org/10.1093/oxrep/grr011

Böckerman

Ilmakunnas

(2009). Unemployment and self-assessed health: Evidence from panel data. Health Economics, 18(2), 161–179. https://doi.org/10.1002/hec.1361

Boen

Keister

Aronson

(2020). Beyond net worth: Racial differences in wealth portfolios and black–White health inequality across the life course. Journal of Health and Social Behavior, 61(2), 153–169. https://doi.org/10.1177/0022146520924811

Bolck

Croon

Hagenaars

(2004). Estimating latent structure models with categorical variables: One-step versus three-step estimators. Political Analysis, 12(1), 3–27. https://www-jstor-org.web.bisu.edu.cn/stable/25791751

Booker

C. L.

Sacker

(2012). Psychological well-being and reactions to multiple unemployment events: Adaptation or sensitisation? Journal of Epidemiology & Community Health, 66(9), 832–838. https://doi.org/10.1136/jech.2010.126755

Brand

J. E.

(2006). The effects of job displacement on job quality: Findings from the Wisconsin longitudinal study. Research in Social Stratification and Mobility, 24(3), 275–298. https://doi.org/10.1016/j.rssm.2006.03.001

Brand

J. E.

(2015). The far-reaching impact of job loss and unemployment. Annual Review of Sociology, 41(1), 359–375. https://doi.org/10.1146/annurev-soc-071913-043237

Brazier

J. E.

Harper

Jones

N. M.

O’Cathain

Thomas

K. J.

Usherwood

Westlake

(1992). Validating the SF-36 health survey questionnaire: New outcome measure for primary care. Bmj: British Medical Journal, 305(6846), 160–164. https://doi.org/10.1136/bmj.305.6846.160

10.

Burgard

S. A.

Brand

J. E.

House

J. S.

(2007). Toward a better estimation of the effect of job loss on health. Journal of Health and Social Behavior, 48(4), 369–384. https://doi.org/10.1177/002214650704800403

11.

Burgard

S. A.

Kalousova

Seefeldt

K. S.

(2012). Perceived job insecurity and health: The Michigan recession and recovery study. Journal of Occupational and Environmental Medicine, 54(9), 1101–1106. https://doi.org/10.1097/JOM.0b013e3182677dad

12.

Burgard

S. A.

Lin

K. Y.

(2013). Bad jobs, bad health? How work and working conditions contribute to health disparities. American Behavioral Scientist, 57(8), 1105–1127. https://doi.org/10.1177/0002764213487347

13.

Clark

A. E.

Clark

A. E.

Georgellis

Sanfey

(2001). Scarring: The psychological impact of past unemployment. Economica, 68(270), 221–241. http://ezproxy.uakron.edu:2048/login?url=http://search.ebscohost.com/login.aspx?direct=true&db=bth&AN=4549960&site=ehost-live

14.

Crimmins

E. M.

Saito

(2001). Trends in healthy life expectancy in the United States, 1970-1990: Gender, racial, and educational differences. Social Science & Medicine, 52(11), 1629–1641. https://doi.org/10.1016/s0277-9536(00)00273-2

15.

Damaske

(2021). The tolls of uncertainty: How privilege and the guilt gap shape unemployment in America. Princeton University Press. https://press.princeton.edu/books/hardcover/9780691200149/the-tolls-of-uncertainty

16.

Dupre

George

Liu

Peterson

(2012). THe cumulative effect of unemployment on risks for acute myocardial infarction. Archives of Internal Medicine, 172(22), 1731–1737. https://doi.org/10.1001/2013.jamainternmed.447

17.

Frech

Damaske

(2012). The relationships between mothers’ work pathways and physical and mental health. Journal of Health and Social Behavior, 53(4), 396–412. https://doi.org/10.1177/0022146512453929

18.

Frech

Damaske

(2019). Men’s income trajectories and physical and mental health at midlife. American Journal of Sociology, 124(5), 1372–1412. https://doi.org/10.1086/702775

19.

Frech

Houle

Tumin

(2021). Trajectories of unsecured debt and health at midlife. SSM - Population Health, 15 (September), 100846. https://doi.org/10.1016/j.ssmph.2021.100846

20.

Gandek

Ware

J. E.

Aaronson

N. K.

Apolone

Bjorner

J. B.

Brazier

J. E.

Bullinger

Kaasa

Leplege

Prieto

Sullivan

(1998). Cross-validation of item selection and scoring for the SF-12 health survey in nine countries: Results from the IQOLA project. Journal of Clinical Epidemiology, 51(11), 1171–1178. https://doi.org/10.1016/S0895-4356(98)00109-7

21.

Gangl

(2006). Scar effects of unemployment: An assessment of institutional complementarities. American Sociological Review, 71(6), 986–1013. https://doi.org/10.1177/000312240607100606

22.

Garcy

A. M.

Vågerö

(2012). The length of unemployment predicts mortality, differently in men and women, and by cause of death: A six year mortality follow-up of the Swedish 1992–1996 recession. Social Science & Medicine, 74(12), 1911–1920. https://doi.org/10.1016/j.socscimed.2012.01.034

23.

Garcy

A. M.

Vågerö

(2013). Unemployment and suicide during and after a deep recession: A longitudinal study of 3.4 million Swedish men and women. American Journal of Public Health, 103(6), 1031–1038. https://doi.org/10.2105/AJPH.2013.301210

24.

Haas

S. A.

(2006). Health selection and the process of social stratification: The effect of childhood health on socioeconomic attainment. Journal of Health and Social Behavior, 47(4), 339–354. https://doi.org/10.1177/002214650604700403

25.

Haas

S. A.

Glymour

M. M.

Berkman

L. F.

(2011). Childhood health and labor market inequality over the life course. Journal of Health and Social Behavior, 52(3), 298–313. https://doi.org/10.1177/0022146511410431

26.

Herber

G.-C.

Ruijsbroek

Koopmanschap

Proper

van der Lucht

Boshuizen

Polder

Uiters

(2019). Single transitions and persistence of unemployment are associated with poor health outcomes. BMC Public Health, 19(1), 740. https://doi.org/10.1186/s12889-019-7059-8

27.

Hollister

(2011). Employment stability in the U.S. labor market: Rhetoric versus reality. Annual Review of Sociology, 37(1), 305–324. https://doi.org/10.1146/annurev-soc-081309-150042

28.

Janlert

Winefield

A. H.

Hammarström

(2015). Length of unemployment and health-related outcomes: A life-course analysis. European Journal of Public Health, 25(4), 662–667. https://doi.org/10.1093/eurpub/cku186

29.

Kalleberg

A. L.

(2009). Precarious work, insecure workers: Employment relations in transition. American Sociological Review, 74(1), 1–22. https://doi.org/10.1177/000312240907400101

30.

Kalleberg

A. L.

(2011). Good jobs, bad jobs: The rise of polarized and precarious employment systems in the United States, 1970s-2000s. Russell Sage Foundation.

31.

Kalousova

Burgard

S. A.

(2014). Unemployment, measured and perceived decline of economic resources: Contrasting three measures of recessionary hardships and their implications for adopting negative health behaviors. Social Science & Medicine, 106 (April), 28–34. https://doi.org/10.1016/j.socscimed.2014.01.007

32.

Kroft

Lange

Notowidigdo

M. J.

Katz

L. F.

(2016). Long-term unemployment and the great recession: The role of composition, duration dependence, and nonparticipation. Journal of Labor Economics, 34(S1), S7–S54. https://doi.org/10.1086/682390

33.

Krug

Eberl

(2018). What explains the negative effect of unemployment on health? An analysis accounting for reverse causality. Research in Social Stratification and Mobility, 55 (June), 25–39. https://doi.org/10.1016/j.rssm.2018.03.001

34.

Melvin

Hummer

R. A.

Elo

I. T.

Mehta

(2014). Age patterns of racial/ethnic/nativity differences in disability and physical functioning in the United States. Demographic Research, 31(17), 497–510. https://doi.org/10.4054/DemRes.2014.31.17

35.

Nagin

(2005). Group-based modeling of development. Harvard University Press.

36.

National Longitudinal Surveys . (2019a). Appendix 19: SF-12 health scale scoring. https://www.nlsinfo.org/content/cohorts/nlsy79/other-documentation/codebook-supplement/nlsy79-appendix-19-sf-12-health-scale

37.

National Longitudinal Surveys . (2019b). Retention & reasons for noninterview. https://www.nlsinfo.org/content/cohorts/nlsy79/intro-to-the-sample/retention-reasons-noninterview

38.

Norström

Virtanen

Hammarström

Gustafsson

P. E.

Janlert

(2014). How does unemployment affect self-assessed health? A systematic review focusing on subgroup effects. BMC Public Health, 14(1), 1310. https://doi.org/10.1186/1471-2458-14-1310

39.

Peng

Chen

Guo

(2021). Macroeconomic conditions and health-related outcomes in the United States: A metropolitan and micropolitan statistical area-level analysis between 2004 and 2017. Health Economics. Advance online publication. https://doi.org/10.1002/hec.4420

40.

Prince

Patel

Saxena

Maj

Maselko

Phillips

M. R.

Rahman

(2007). No health without mental health. Lancet, 370(9590), 859–877. https://doi.org/10.1016/S0140-6736(07)61238-0

41.

Roelfs

D. J.

Shor

Davidson

K. W.

Schwartz

J. E.

(2011). Losing life and livelihood: A systematic review and meta-analysis of unemployment and all-cause mortality. Social Science & Medicine, 72(6), 840–854. https://doi.org/10.1016/j.socscimed.2011.01.005

42.

Rogers

Everett

Saint Onge

J. M.

Krueger

P. M.

(2010). Social, behavioral, and biological factors, and sex differences in mortality. Demography, 47(3), 555–578. https://doi.org/10.1353/dem.0.0119

43.

Rowley

K. M.

Feather

N. T.

(1987). The impact of unemployment in relation to age and length of unemployment. Journal of Occupational Psychology, 60(4), 323–332. https://doi.org/10.1111/j.2044-8325.1987.tb00264.x

44.

Rueda

Chambers

Wilson

Mustard

Rourke

S. B.

Bayoumi

Raboud

Lavis

(2012). Association of returning to work with better health in working-aged adults: A systematic review. American Journal of Public Health, 102(3), 541–556. https://doi.org/10.2105/AJPH.2011.300401

45.

Silver

S. R.

Quay

(2021). Employment status, unemployment duration, and health-related metrics among US adults of prime working age: Behavioral risk factor surveillance system, 2018–2019. American Journal of Industrial Medicine. Advance online publication. https://doi.org/10.1002/ajim.23308

46.

Stauder

(2019). Unemployment, unemployment duration, and health: Selection or causation? The European Journal of Health Economics: HEPAC: Health Economics in Prevention and Care, 20(1), 59–73. https://doi.org/10.1007/s10198-018-0982-2

47.

Strandh

Winefield

Nilsson

Hammarström

(2014). Unemployment and mental health scarring during the life course. The European Journal of Public Health, 24(3), 440–445. https://doi.org/10.1093/eurpub/cku005

48.

Strully

K. W.

(2009). Job loss and health in the U.S. labor market. Demography, 46(2), 221–246. https://doi.org/10.1353/dem.0.0050

49.

Tøge

A. G.

Blekesaune

(2015). Unemployment transitions and self-rated health in Europe: A longitudinal analysis of EU-SILC from 2008 to 2011. Social Science & Medicine, 143, 171–178. https://doi.org/10.1016/j.socscimed.2015.08.040

50.

US Bureau of Labor Statistics . (2020). NLSY79 data overview. https://www.bls.gov/nls/nlsy79.htm

51.

Vågerö

Garcy

A. M.

(2016). Does unemployment cause long-term mortality? Selection and causation after the 1992-96 deep Swedish recession. European Journal of Public Health, 26(5), 778–783. https://doi.org/10.1093/eurpub/ckw053

52.

Vermunt

J. K.

(2010). Latent class modeling with covariates: Two improved three-step approaches. Political Analysis, 18(4), 450–469. https://doi.org/10.1093/pan/mpq025

53.

Wanberg

C. R.

Kanfer

Hamann

D. J.

Zhang

(2016). Age and reemployment success after job loss: An integrative model and meta-analysis. Psychological Bulletin, 142(4), 400–426. https://doi.org/10.1037/bul0000019

54.

Young

(2012). Losing a job: The nonpecuniary cost of unemployment in the United States. Social Forces, 91(2), 609–634. https://doi.org/10.1093/sf/sos071

55.

Zhang

Hayward

M. D.

(2006). Gender, the marital life course, and cardiovascular disease in late midlife. Journal of Marriage and Family, 68(3), 639–657. https://doi.org/10.1111/j.1741-3737.2006.00280.x