Health Effects of Unemployment Benefit Program Generosity

Abstract

Objectives. We assessed the impact of unemployment benefit programs on the health of the unemployed.

Methods. We linked US state law data on maximum allowable unemployment benefit levels between 1985 and 2008 to individual self-rated health for heads of households in the Panel Study of Income Dynamics and implemented state and year fixed-effect models.

Results. Unemployment was associated with increased risk of reporting poor health among men in both linear probability (b = 0.0794; 95% confidence interval [CI] = 0.0623, 0.0965) and logistic models (odds ratio = 2.777; 95% CI = 2.294, 3.362), but this effect is lower when the generosity of state unemployment benefits is high (b for interaction between unemployment and benefits = −0.124; 95% CI = −0.197, −0.0523). A 63% increase in benefits completely offsets the impact of unemployment on self-reported health.

Conclusions. Results suggest that unemployment benefits may significantly alleviate the adverse health effects of unemployment among men.

An extensive body of research has linked job loss to poorer physical and mental health¹ and higher risk of premature death.² Recent literature has focused on establishing the causal nature of this association,^2–8 but few studies have explored whether specific social programs modify the health effects of job loss. Understanding the impact of policies is useful for identifying intervention approaches to reduce the harms associated with unemployment, but they may also reveal some of the mechanisms explaining the association between job loss and health. Job loss is associated with a substantial loss in earnings.⁹ If earnings losses are the primary mechanism linking job loss to health, we would expect generous unemployment benefit programs to mitigate some of the negative consequences of job loss on health. On the other hand, unemployment benefits may be less effective if job loss influences health primarily through nonfinancial mechanisms, such as the loss of a time structure for the day, decreased self-esteem, chronic stress,¹⁰ or changes in health-related behavior.

A few studies have investigated the association between unemployment benefit receipt and self-reported health measures.^11–13 For example, Rodriguez¹¹ analyzed self-reported health data from Britain, Germany, and the United States and found that unemployed workers in receipt of unemployment benefits do not have statistically higher likelihood of reporting poor health compared with the employed, while unemployed workers receiving no benefits are in worse health than these 2 groups. She concluded that benefit receipt moderates the association between unemployment and poor subjective health. Similarly, McLeod et al.¹⁴ found that unemployed US workers not receiving benefits are more likely to report poor health than employed workers, but the health of unemployed workers in receipt of benefits does not statistically differ from the health of employed workers. The association between receiving benefits and health was most pronounced among low-skilled unemployed workers, who appear to gain substantially from receipt of cash benefits.

A key caveat in these studies is that they do not account for selection into benefit receipt, a bias that could lead to either over- or underestimation of effects. For example, if those who lose their jobs are healthier and more likely to be eligible for and receive unemployment benefits, the health benefits of unemployment benefits will be overestimated. During the recent recession, for example, non-Hispanic White race, higher educational level, and being married, characteristics associated with better health, also predicted receipt of benefits among long-term unemployed workers.¹⁵ On the other hand, job losers in poor health may anticipate longer-term spells of unemployment and therefore may be more likely to claim unemployment benefits than healthier individuals who expect to quickly find new employment. While 61% of workers in manufacturing and 66% of workers in construction were receiving benefits in the period 2008 to 2011, only 52% of professional and management workers and 49% of workers in the retail trade industry were receiving benefits in the same period.¹⁵ These findings suggest that selection is a serious source of potential bias in the relationship between unemployment benefit receipt and health, though the direction of bias is unclear.

In the United States, the Federal–State Unemployment Insurance Program provides temporary wage replacement for eligible workers who become unemployed through no fault of their own. Although all states must follow general rules established at the federal level relating to coverage and eligibility, each state operates its own program. As a result, there is considerable variation in the generosity of unemployment benefit programs across states and over time. An approach to account for selection is to exploit these variations in the generosity of unemployment benefit programs to understand their effects on the health of workers. The assumption is that changes in unemployment benefit policy are uncorrelated with a worker’s health or other characteristics, as individuals have no control over the policy at the time they experience job loss. Variations in unemployment benefit generosity across states and over time, therefore, offer a unique natural experiment to estimate the impact of this policy on the health of unemployed workers.

In a recent study, Cylus et al. exploited these variations to assess whether unemployment benefits moderate the relationship between aggregate unemployment rates and suicide,¹⁶ which are known to increase during recessions.^17,18 Findings from this study suggest that more generous unemployment benefits are associated with a weaker effect of recessions on suicide. However, this study was based on aggregate data and did not estimate whether unemployment benefits reduced the negative impact of job loss among unemployed workers or whether benefits might in fact lead to improvements in mental health among both employed and unemployed workers, for example, by reducing the stress associated with the fear of job loss.¹⁹ Likewise, it is not clear whether results for suicide are applicable to self-rated health, a measure that combines elements of both physical and mental health, and a strong predictor of mortality.²⁰

In this study, we assessed the impact of unemployment benefit programs on the health of the unemployed. We hypothesized that income from unemployment benefits reduces psychological and physical morbidity among displaced workers such that individuals losing their job at a time of more generous unemployment benefit policies will suffer fewer health consequences than comparable individuals losing their jobs during years of lower benefit generosity. By focusing on unemployment benefit program generosity at the state level, we circumvent the bias generated by selection into benefits in the aforementioned studies.^21,22 To identify this effect, we exploited variation in state unemployment benefit program generosity across US states and linked these to longitudinal individual-level data.

METHODS

We used data from the 1984–2009 waves of the Panel Study of Income Dynamics (PSID).²³ The PSID is a longitudinal household study that collects data on individual characteristics, including employment status and self-reported health. The survey was conducted annually from 1968 through 1997, after which it shifted to biennial interviews.

Outcome Variable of Interest

The outcome variable of interest was self-reported health, first included in the PSID survey in the 1984 wave, which ranges from “excellent” (1) to “poor” (5). We collapsed this variable into a dichotomous indicator of poor health, in which “fair” (4) or “poor” (5) health are equal to 1. Other individual-level data included whether an individual reported joblessness in the year before the health assessment (t-1), age, gender, and the natural log of family income, lagged to avoid simultaneity with joblessness (i.e., assessed at t-2). Although the main results did not materially differ after doing so, we excluded 3673 observations (person-years) in which maximum available benefits were larger than household income in the previous year, as these individuals were unlikely to meet eligibility criteria. We also excluded 1803 observations with missing data. Our final sample consisted of 12 855 heads of household aged 18 to 65 years participating in PSID, amounting to 66 795 person-years.

We linked PSID data to state-level data on maximum state unemployment insurance benefits obtained from the US Department of Labor Employment and Training Administration. Maximum benefit generosity is reported as the maximum allowable amount per week (in US dollars) and the maximum number of weeks a worker is entitled to receive benefits. We multiplied these 2 values to obtain the maximum total allowable benefit level a worker is entitled to receive in a given year and state, adjusted to constant 1999 US dollars using the Bureau of Labor Statistics Consumer Price Index.²⁴ Finally, we included data on state unemployment rates for the working-age population estimated from the Current Population Survey.²⁵

Statistical Analysis

We used both ordinary least-squares linear probability and logistic state fixed-effect models that exploited exogenous variation in the generosity of state unemployment benefit programs over time across states. State fixed effects exploit the longitudinal nature of the data by assessing the association between changes in benefit generosity and changes in self-rated health, controlling for permanent characteristics that vary across states. We used individual random effects in our primary analysis but also report results from models that incorporated individual fixed effects. Individual fixed-effects estimators are attractive because they control for unobserved individual-level heterogeneity that may be correlated with the explanatory variables. On the other hand, individual fixed effects may be overly restrictive because they identify effects of unemployment benefits only for individuals who experience more than 1 spell of joblessness. In addition, unemployment benefit generosity varies at the state level, not at the individual level. As a result, we focused primarily on the results from individual random-effect models that incorporate state fixed effects to estimate the health effect of changes in state unemployment benefit generosity.

Our main ordinary least squares model specification is as follows:

where H is a binary indicator of poor self-reported health for individual i in state j and year t, is the intercept, U is whether an individual experienced joblessness in t-1, UB is the mean-centered natural log of maximum unemployment benefits in state j for year t-1, UR is the mean-centered lagged state unemployment rate for state j each year, X is a vector of individual controls, S represents state fixed effects, T represents year fixed effects, and ɛ is the regression error term. Employment, unemployment rates, and state benefit levels are lagged by 1 year because the PSID questionnaire asks about employment status in the previous year. The natural log of benefit levels captures proportional increases in maximum benefit levels. The model specification is analogous for logistic regression.

The main estimate of interest was U*UB, which assesses the interaction between joblessness and unemployment benefits and indicates whether larger maximum unemployment benefits at the time of unemployment in a worker’s state of residence moderate the impact of unemployment on health. A coefficient indicates that the impact of unemployment on health is weaker if state maximum unemployment benefits are higher. This estimate can also be interpreted as a difference-in-difference estimate, as we expected the impact of the benefits to be larger for heads of household that become unemployed (experimental group) than for employed heads of household (control group), who should not benefit directly from an increase in unemployment benefits.

Also of interest was the term UR*UB, which examines whether larger maximum unemployment benefits offset the impact of aggregate economic downturns. Combining both aggregate and individual-level unemployment measures enabled us to distinguish the effects of benefits on individual unemployed workers from effects that have an impact upon the health of the entire working population. In all models standard errors were robust and clustered at the state–year level and therefore consistent in the presence of correlated errors within state–years.

RESULTS

Sample descriptive statistics disaggregated by employment status and gender are summarized in Table 1. Of the sample, 17.7% experienced at least 1 episode of job loss. Ten percent of individuals who were gainfully employed in the previous year reported poor health, whereas 24.9% of individuals who experienced job loss in the previous year reported poor health. Compared with 2 years before, unemployed individuals were 5.3 percentage points more likely to report poor health, whereas among the employed, the share reporting poor health only increased by 0.9 percentage points over the same period. Men made up nearly 80% of the sample of heads of household. Female heads of household on average lived in state–years with comparatively less-generous unemployment benefit programs than male heads of household. Consistent with previous research,^26,27 we found that women had higher likelihood of reporting poor health than men; this gender pattern was exhibited among both the employed and unemployed.

TABLE 1—

Sample Descriptive Statistics Disaggregated by Employment Status and Gender: United States, 1985–2008

Employment Status	Poor Health, Mean (SD)	Poor Health in t-2, Mean (SD)	State Unemployment Rate in t-1, Mean (SD)	Real Total Unemployment Benefit in t-1, 1999 US $, Mean (SD)	Real Family Income in t-2, 1999 US $, Mean (SD)	Age, y, Mean (SD)	Male	No.
Total
Employed	0.100 (0.300)	0.091 (0.288)	4.601 (1.555)	7877.80 (2253.23)	60 549.07 (65 610.89)	41.2 (10.4)	80.3%	63 852
Unemployed	0.249 (0.433)	0.196 (0.397)	4.743 (1.577)	7631.11 (1879.44)	38 938.81 (43 298.72)	41.5 (12.8)	64.3%	2943
Total	0.106 (0.308)	0.096 (0.295)	4.608 (1.556)	7866.93 (2238.64)	59 596.91 (64 941.23)	41.2 (10.5)	79.6%	66 795
Men
Employed	0.086 (0.280)	0.078 (0.269)	4.605 (1.554)	7931.46 (2294.49)	67 386.88 (70 580.69)	41.0 (10.2)	. . .	51 276
Unemployed	0.241 (0.428)	0.176 (0.381)	4.802 (1.568)	7751.63 (1870.12)	48 361.39 (50 389.28)	42.5 (13.0)	. . .	1891
Total	0.092 (0.288)	0.082 (0.274)	4.612 (1.555)	7925.06 (2280.98)	66 710.20 (70 050.88)	41.1 (10.3)	. . .	53 167
Women
Employed	0.156 (0.363)	0.146 (0.353)	4.585 (1.562)	7659.02 (2062.22)	32 669.33 (24 026.73)	41.9 (11.1)	. . .	12 576
Unemployed	0.264 (0.441)	0.233 (0.423)	4.638 (1.590)	7414.47 (1877.60)	22 001.46 (15 331.90)	39.6 (12.2)	. . .	1052
Total	0.164 (0.371)	0.152 (0.359)	4.589 (1.564)	7640.15 (2049.54)	31 845.83 (23 642.23)	41.7 (11.2)	. . .	13 628

Note. Data are from the Panel Study of Income Dynamics²³ and the US Department of Labor Employment and Training Administration.

To illustrate the generosity of benefits relative to household income, we calculated the maximum household unemployment benefit replacement rate, which reflects the proportion of income that is maintained through unemployment benefit receipt, by dividing the real maximum unemployment benefit level in t-1 by real household income in t-2 (Figure A, available as a supplement to the online version of this article at http://www.ajph.org). On average, maximum allowable benefits corresponded to between one fourth and one third of household income. The mean household income replacement rate was 25.9%, but for half of respondents, the mean replacement rate was less than 18.9%. When we adjusted the replacement rate for household size by dividing household income by the square root of the number of household members,²⁸ the mean replacement rate was 34.4%, but less than 28.6% for half of the respondents.

Figure 1 plots lines of best fit for the probability of reporting poor health along the maximum level of real total unemployment benefits, separately for employed and unemployed workers, by gender. Displaced workers had higher probabilities of poor self-reported health than employed workers at all levels of benefits, but among men (Figure 1a), both employed and unemployed respondents had lower probabilities of poor health as benefit levels increased. Among men, the slope was noticeably steeper for unemployed workers, so that the health gap between employed and unemployed male workers became smaller as benefits increased. Among women (Figure 1b), more generous benefits predicted lower probability of poor health, but the slopes were nearly identical for the employed and unemployed. Based on these clear differences and because of historical disparities in employment patterns by gender, we stratified the sample and primarily examined whether there were effects of unemployment benefit programs for men.

FIGURE 1—

Probability of poor self-reported health relative to maximum allowable real unemployment benefit levels, unemployed and employed US workers, among (a) men and (b) women: 1985–2008.

Note. Data are from the Panel Study of Income Dynamics²³ and US Department of Labor Employment and Training Administration.

For men, model results that used individual random effects with state and year fixed effects are summarized in Table 2. In linear probability models that did not include interactions, unemployment at time t-1 was associated with higher likelihood of reporting poor health in the following year (b = 0.0819; 95% confidence interval [CI] = 0.0644, 0.0995). Higher unemployment rates at time t-1 were associated with slightly lower likelihood of reporting poor health (b = −0.002; 95% CI = −0.00425, −0.000195). The main effect for benefit levels indicates that, as expected, benefit generosity was not associated with health among employed workers. In contrast, after we included all interactions (Table 2), the interaction between joblessness and benefit generosity was negative and significant (b for interaction between joblessness and benefits = −0.124; 95% CI = −0.197, −0.0523), suggesting that a 10% increase in benefits for the unemployed reduces the risk of poor health by 1.3% points (−0.00014 plus −0.0124). This indicates that a 63% increase in state benefit generosity would be required to fully offset the impact of unemployment on self-reported health. The interaction between unemployment rates and benefits was not statistically significant (b = 0.00361; 95% CI = −0.00202, 0.00924).

TABLE 2—

Ordinary Least Squares and Logistic Individual Random-Effects Models of the Probability of Reporting Poor Health in Time t Conditional on State Unemployment Benefit Generosity at t-1: US Men, 1985–2008

	OLS Random Effects, b (95% CI)		Logistic Random Effects, OR (95% CI)
Variables	Main Effects	Interactions	Main Effects	Interactions
Joblessness in t-1	0.0819*** (0.0644, 0.0995)	0.0794*** (0.0623, 0.0965)	2.911*** (2.419, 3.504)	2.790*** (2.305, 3.378)
Natural log real total max benefit in t-1	-0.00668 (-0.0322, 0.0189)	-0.0014 (-0.0268, 0.0240)	0.907 (0.509, 1.617)	0.977 (0.547, 1.746)
Working age state unemployment rate in t-1	-0.00222* (-0.00425, -0.000195)	−0.00218* (-0.00414, -0.000216)	0.947* (0.907, 0.989)	0.951* (0.910, 0.993)
Joblessness × natural log real total max benefit in t-1		-0.124*** (-0.197, -0.0520)		0.379* (0.168, 0.852)
Working age unemployment rate in t-1 × natural log real total max benefit in t-1		0.00361 (-0.00202, 0.00924)		1.059 (0.925, 1.212)
Poor health in t-2	0.232*** (0.214, 0.249)	0.232*** (0.214, 0.250)	4.766*** (4.053, 5.605)	4.773*** (4.058, 5.613)
Natural log real family income in t-2	-0.0426*** (-0.0478, -0.0374)	-0.0427*** (-0.0479, -0.0375)	0.434*** (0.396, 0.475)	0.433*** (0.396, 0.474)
Age	0.00492*** (0.00452, 0.00531)	0.00492*** (0.00452, 0.00532)	1.084*** (1.077, 1.091)	1.084*** (1.077, 1.092)
State fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
Observations	52 892	52 892	52 892	52 892
Number of respondents	9408	9349	9349	9349

Note. CI = confidence interval; OLS = ordinary least squares; OR = odds ratio. Data are from the Panel Study of Income Dynamics²³ and the US Department of Labor Employment and Training Administration. Robust confidence intervals in parentheses; standard errors clustered at state–year level.

*P < .05; ***P < .001.

Results from logistic models are summarized in Table 2 and suggest a similar pattern to that observed for linear probability models. In the full model, joblessness was associated with a significantly increased risk of reporting poor health (odds ratio [OR] = 2.79; 95% CI = 2.305, 3.378), but this effect was offset by the presence of generous unemployment benefits (OR for interaction between benefits and unemployment = 0.379; 95% CI = 0.168, 0.852).

We also ran models for women (Table A, available as a supplement to the online version of this article at http://www.ajph.org). In line with Figure 1, we found no statistically significant difference in the likelihood of poor self-reported health on the basis of the interaction between unemployment benefit generosity and joblessness with linear probability models.

We conducted a number of robustness checks and additional analyses. Table 3 contains the results of individual fixed-effects models for men, in which the interaction term U*UB was identified only for individuals who experienced multiple job losses. Results were similar to those presented in Table 2. In the full linear probability model with individual fixed effects (Table 3), unemployment in t-1 was associated with higher likelihood of poor self-reported health (b = 0.0618; 95% CI = 0.045, 0.0785), but more generous unemployment benefits weakened the effect of unemployment (b = −0.0751; 95% CI = −0.148, −0.00232). The direction and magnitude of effects was similar in logistic models (Table 3), but confidence intervals for the interaction between joblessness and benefits were wide and included the null. This most likely reflects the fact that estimates were only based on individuals with multiple job-loss episodes, leading to a loss in statistical power.

TABLE 3—

Ordinary Least Squares and Logistic Individual Fixed-Effects Models of the Probability of Reporting Poor Health in Time t Conditional on State Unemployment Benefit Generosity at t-1: US Men, 1985–2008

	OLS Fixed Effects, b (95% CI)		Logistic Fixed Effects, OR (95% CI)
Variables	Main Effects	Interactions	Main Effects	Interactions
Joblessness in t-1	0.0632*** (0.0462, 0.0802)	0.0618*** (0.0450, 0.0785)	2.028*** (1.641, 2.507)	1.985*** (1.599, 2.464)
Natural log real total max benefit in t-1	-0.0109 (-0.0386, 0.0168)	-0.00806 (-0.0355, 0.0194)	0.754 (0.388, 1.465)	0.777 (0.399, 1.513)
Working age state unemployment rate in t-1	-0.00148 (-0.00339, 0.000424)	-0.00145 (-0.00329, 0.000396)	0.967 (0.923, 1.013)	0.635 (0.265, 1.523)
Joblessness × natural log real total max benefit in t-1		-0.0751* (-0.148, -0.00232)		0.97 (0.925, 1.016)
Working age unemployment rate in t-1 × natural log real total max benefit in t-1		0.00404 (-0.00118, 0.00927)		1.033 (0.903, 1.182)
Poor health in t-2	0.0344*** (0.0139, 0.0549)	0.0344*** (0.0139, 0.0549)	0.965 (0.863, 1.079)	0.965 (0.863, 1.079)
Natural log real family income in t-2	-0.00961*** (-0.0151, -0.00412)	-0.00972*** (-0.0152, -0.00424)	0.964 (0.850, 1.093)	0.962 (0.848, 1.092)
Age	-0.00412 (-0.0101, 0.00182)	-0.00407 (-0.0100, 0.00187)	0.916 (0.794, 1.057)	0.917 (0.795, 1.057)
State fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
Observations	52 892	52 892	52 892	12 173
Number of respondents	9408	9349	9349	1586

Note. CI = confidence interval; OLS = ordinary least squares; OR = odds ratio. Data are from the Panel Study of Income Dynamics²³ and the US Department of Labor Employment and Training Administration. Robust confidence intervals in parentheses; standard errors clustered at state–year level.

*P < .05; ***P < .001.

In addition, in separate models, we adjusted the household income explanatory variable by dividing by the square root of the household size²⁸; the results did not materially change. We also clustered errors at either the state level or at the individual level in separate analyses and found that this did not affect the results.

DISCUSSION

This study was motivated by the need to understand how unemployment benefit policies influence workers’ health. We found that generous state unemployment benefits were associated with lower likelihood of reporting poor health among unemployed male workers. One might also hypothesize that unemployment benefits could lead to health improvements of the employed population, for example, by reducing the stress from the fear of job loss associated with labor market fluctuations.²⁹ Although the graphical representation in Figure 1 indicates that both the employed and unemployed might enjoy better self-rated health in state–years with more generous unemployment benefits, on the basis of our model results, we found no evidence of an effect of unemployment benefits for the employed or women heads of household or that effects of unemployment benefits differ on the basis of labor market conditions.

Results from our study provide insight into the mechanisms linking job loss to health. Theoretically plausible mechanisms linking job loss to self-reported health include financial distress, stigma, social isolation, or reduced “meaning in life.”^30,31 We found larger maximum cash unemployment benefits have a protective effect on self-reported health during periods of unemployment. This suggests that the relationship between poor self-reported health and unemployment may partially be the result of income loss after job loss, as a policy that relieves income losses is associated with weaker effects of unemployment on health. Although it is likely that income is not the only mechanism through which unemployment influences health, our findings highlight the potential of income support programs to not only smooth consumption during unemployment spells, as has been suggested in the literature,³² but also to influence health after job loss.

Economic theory suggests a potential explanation for how unemployment benefit programs buffer the impact of job loss on health. The canonical Grossman model of the “demand for health” and “health investment” posits that individuals require health so that they can maximize their utility and enjoy life.³³ Time spent working increases income, which allows individuals to purchase health inputs such as healthy food, but at the same time, working reduces time to invest in health-promoting activities such as exercise, or may even harm health as a result of exposure to adverse working conditions. According to this model, the unemployed who do not have access to financial resources may face difficulties maintaining health if they cannot afford to invest in health-enhancing inputs. At the same time, individuals who are not working have more leisure time available that can be used for health-promoting activities. Access to generous unemployment benefits may therefore protect health through 2 pathways: (1) increasing access to financial resources that can be used to invest in health and (2) providing the unemployed with additional time to engage in health-promoting leisure activities during unemployment spells. The notion that access to money weakens the deleterious health effects of unemployment is consistent with research on the effects of unemployment benefits and liquidity on unemployment duration and leisure.^34–36

There are limitations to our analysis. Our results are based on self-reported health, which captures a combination of complex physical and mental health dimensions. It is possible that our results were driven by effects on mental health; consistent with this hypothesis, earlier research suggests that higher unemployment benefits are associated with lower suicide rates.¹⁶ Unfortunately, most waves of the PSID do not contain a wider set of relevant indicators, such as objective health measures or other well-being metrics, which could provide additional insight into the linkages between unemployment benefits and health. Second, our study design enables us to identify the net effect of unemployment benefit policies, but it does not capture the direct effect of receiving benefits. Although the latter is of interest, our approach has 2 main advantages. A first advantage is that we overcome selection bias inherent to the nonrandomized allocation of unemployment benefits. A second advantage is that we are able to provide estimates of the net effect of a policy intervention that would change the generosity of unemployment benefits. This is important because it has been estimated that a non-negligible proportion of eligible unemployed workers do not claim unemployment benefits, so the direct effect of receiving benefits might overestimate the impact of a policy change on the health of unemployed workers.

Although social programs such as Social Security, the Earned Income Tax Credit, US welfare reform, and the food stamp program were not motivated by health concerns, a growing literature suggests that these social policies may bring important health benefits.^37–40 Our results add to this body of literature and suggest that unemployment benefits have the potential to improve health. On the basis of the effect of joblessness on health and the estimated effect of benefits on unemployed males (Table 2), we estimate that a 63% increase in the maximum unemployment benefits a worker is entitled to receive offsets the impact of unemployment on health among men.

The current financial crisis has sparked debates on the costs and benefits of social programs. Our findings suggest that a cost–benefit analysis of unemployment benefit policies should take into account the potential loss in health that would result from diminishing the comprehensiveness of unemployment benefit programs.