OUP user menu

Unemployment and mental health—who is (not) affected?

Mona C. Backhans, Tomas Hemmingsson
DOI: http://dx.doi.org/10.1093/eurpub/ckr059 429-433 First published online: 20 May 2011


Background: The aim of this study is first, to investigate the association between periods of unemployment and mental distress, adjusting for previous health status, and second, to study differences and similarities between groups defined by age, sex, family situation, socioeconomic position and work environment. Methods: The analyses are based on a cohort of participants in Stockholm county council’s Public Health Survey 2002 with a follow-up in 2007. Selected from the initial cohort are respondents 20–59 years who were employed at T1 and had no unemployment in 2001–02. Logistic regression is used and differences between groups are expressed as odds ratios. Interaction analyses are also performed. Results: Initial odds ratios of 1.84 in the group with 1 year of unemployment or more compared to the reference group with no unemployment is reduced to 1.52 after adjustment for prior mental and somatic health. Analyses show that the impact of unemployment in this sample is stronger for men, those working overtime, those with high social support or low control at their previous job, self-employed and those with low occupational class or low previous wage. Regarding family situation, unemployment is least associated with mental distress among individuals living in couples without children. Conclusion: Results show an independent effect of unemployment on mental distress, but this effect varies between groups. Both proposed theories: role loss and differential susceptibility, receive some support. Since all interaction analyses are insignificant, results should be interpreted with caution.


While there is abundant research investigating the potential effect of unemployment on mental health,1–6 few studies focus on variations in the association between demographic and/or socioeconomic groups. Among the individual factors that have been shown to mediate the effect of unemployment on mental health are work-role centrality,7–8 job quality,9 financial strain,10,11 social support, coping strategies, reemployment expectations and unemployment duration.12 These results support not only theories that define unemployment as role loss,13 but also those focussing on the financial aspect of unemployment14 and the importance of time-structure for the mental health of unemployed individuals.15

This study takes as one departure Ezzy’s theory of job loss as a status passage, a divestment passage associated with the failure to successfully maintain or continue in a role.13 If there is a major disruption to an individual’s life this threatens the whole enterprise of identity legitimation; psychological distress after job loss is seen as a product of the failure to find meaning. In a similar vein, Ashforth writes about job loss as ‘role exit’.16 A role exit which is most often characterized by three critical attributes: it is socially undesirable, of great magnitude, and involuntary. The individual’s identification with the role and cohesion with co-workers will determine how difficult the transition is and how negative the consequences. Although our data does not allow an investigation of work role centrality, the theory gives us an idea of how effects of unemployment may differ between groups defined by age and sex, family situation, socioeconomic position and work environment. Hypothetically, those most to lose from job loss are those more privileged in terms of job content and salary.

Contrary to this, according to theories of general/differential susceptibility those with few social and economic resources are more likely to suffer as a consequence of job loss.17–19 In this case, being in a disadvantaged position, in itself a risk factor for mental distress, will exacerbate the negative effect of unemployment on mental health. This contention is not the same as saying that financial difficulties mediate the effect of unemployment on health, but instead points to pre-existing circumstances, like low income, low education, belonging to an ethnic minority etc., explaining differential effects of unemployment.20,21

The aim of this study is first, to investigate the association between unemployment and mental distress, adjusting for previous health status, and second, to study differences and similarities between groups defined by age and sex, family situation, socioeconomic position and psychosocial work environment. To limit selection effects,22–25 previously unemployed or non-active individuals were excluded. The two opposing hypotheses are: (i) those with better jobs (in terms of psychosocial work environment and wage) or stronger attachment to the labour market (men, permanent and full time job, middle-aged) will be more negatively affected by unemployment; (ii) those in disadvantaged positions (lower occupational class, temporary contract, previous mental distress, low wage, women) or with strong breadwinning obligations (men, single income, living with children) will be more negatively affected by unemployment.


The Stockholm Public Health Cohort

Analyses are based on a cohort (n = 23 794) of respondents to the Stockholm county council’s Public Health Survey 2002 (SPHS) with a follow-up in 2007. The population consists of a random sample of residents in the age span 18–84 years. The SPHS is a postal survey with a response rate of 62% in 2002 and 80% in 2007. The questionnaire contains around 100 questions regarding somatic and mental health, health behaviours, financial and family situation, housing, employment and work environment. Linked to the survey is register data regarding remunerated days from unemployment (2001–06), sickness insurance (2001–02), and wage (2002). The use of this data for the study question was approved by the local ethics committee (dnr 2009/1571-31).

From the initial cohort, a selection is made based on the following criteria; the respondents were 20–59 years at T1, they had no unemployment in 2001–02 and they had not become disability pensioners during the follow-up period 2003–06. Furthermore, they were in employment or on leave (most often parental leave) in 2002; students and other groups outside the labour force were excluded (table 1). The sample for analyses consists of 12 605 individuals.

View this table:
Table 1

Stepwise selection of participants for the study

SPHS20–59 years in 2002+ No unemployment 0102, no disability pension 0306+ Employed or on leave 2002Of which unemployed
23 79417 10414 00712 6051271 (10.1%)
43.8% men42.7% men44.1% men45.2% men40.4% men

Variables in the model


Unemployment in 2003–06 is measured as the total number of registered days with unemployment. Total unemployment duration is categorized into four groups: 0 days, 1–130 days (5 remunerated days/week equals <6 months), 131–260 days (6–12 months), ≥261 days (>12 months). This categorization was based on first, the official definition of long-term unemployment in Sweden (6 months) and second, on the size of exposed groups. In the stratified analyses, these groups are collapsed into three: 0 days, 1–130 days and ≥131 days. To calculate the synergy index, the exposure is dichotomised into no unemployment or any unemployment.


GHQ12 in 2007 is used as a measure of mental distress with a cut-off of 3 points or more. GHQ12 is one of the most frequently used psychiatric screening instruments, with the aim of discriminating between those with and without minor psychiatric disorders26 and it has often been used in studies of health effects of unemployment.12,27,28 Although a cut-off score of 3/4 is sometimes found to be optimal, we chose 2/3 as the mean score was so low in our population (1.83 for the full cohort in 2002).29,30


The association between unemployment and mental distress is adjusted for the following possible confounders: sex and age (classified into 20–29, 30–39, 40–49 and 50–59 years), mental health status in 2002 (GHQ12 dichotomized), any sickness absence in 2001–02, self-rated health less than good, long-standing limiting illness in 2002.


Stratified analyses are performed for men and women; age dichotomized into 20–39 and 40–59 years; according to family situation (cohabiting or not, living with children or not), mental distress in 2002 and occupational class (measured using the EGP-schema31 and collapsed into three categories: manual workers and lower non-manual, middle and higher non-manual employees and self-employed). Psychosocial work environment is assessed using questions on social support, job demand and job control.32 All three dimensions are dichotomized. Analyses are also performed based on type of labour contract (permanent or temporary); working hours (1–35, 36–45, ≥45 h) and wage (divided into tertiles) in 2002.

In the interaction analyses, all exposures are first dichotomised and combined into four groups, with the reference group being unexposed to both risk factors. The following categories are assessed as potentially ‘vulnerable’ to the effects of unemployment: younger people, women, people with children, people with prior mental distress, manual workers, people with low social support, high demands or low control at work, those with temporary work contracts, on part time, or low income.


GHQ12 was initially analysed as a categorical outcome ranging from 0 to 12, using ordinal logistic regression, but with inclusion of confounders the number of empty cells soon became a problem. This persisted even when small categories (high scores) were collapsed. Instead, GHQ12 is dichotomized. Zero days of unemployment is the reference category and differences between the reference category and exposed groups are assessed using odds ratios. The base model contains only the exposure unemployment. Confounders are added in blocks and discarded if unrelated to the outcome or if they do not change estimates regarding unemployment and mental distress. Long-standing limiting illness was unrelated to mental distress, and is excluded from the full model.

Stratified analyses are performed for the base model and the full model as established previously. As a more stringent test of differences in effect between groups, interaction analyses are also performed. Interaction is defined as effect-measure modification or departure from additivity.33 As a first condition for interaction, both exposures should be independent risk factors for the studied outcome. According to analyses, being a woman, being younger, having a prior history of mental distress, being single (regardless of children), lacking social support at work, having high job demands or low job control, being in temporary employment, and having low income are risk factors for subsequent mental distress. Being a manual worker or working either part time or overtime are not risk factors in themselves. However, as these factors in previous analyses seem to have a role to play in the association between unemployment and mental distress, the variables are included in the interaction analyses. For the calculation of the synergy index, a SAS program created by Lundberg et al.34 is used. In cases where synergy is present (and the risk increase is more than additive), the synergy index is significantly above one and in cases of antagonism, it is significantly below one.


Descriptive results show that 20.2% are defined as mentally distressed in 2007 among those with no unemployment, and 31.9% among those with >12 months unemployment. Regression analyses show a high crude odds ratio (OR) in the group with 260 days (1 year) of unemployment or more compared to the reference group with no unemployment (table 2). Also, for those with 1–130 days (<6 months) and 131–260 days (6–12 months) ORs are elevated. For the group with the longest unemployment duration age and sex is of no importance while much is explained by previous mental and somatic health. In the full model, OR drops by 40%. For the next exposure group, age and sex explains more than previous mental health (the addition of which even strengthens associations). For the group with <130 days unemployment much is explained by the age and sex composition. After addition of previous mental health the confidence interval (CI) includes 1.

View this table:
Table 2

Logistic regression between unemployment and mental health for the full sample. Odds ratios and confidence intervals (CI)

Base modelCI+ age and sexCI+ GHQ12 2002CI+ SRH, sickness absence 01–02CI
1–130 days unemployment1.481.23–1.781.251.04–1.521.150.94–1.401.120.92–1.37
131–260 days unemployment1.571.20–2.051.391.06–1.821.471.11–1.941.371.04–1.82
261+ days unemployment1.841.46–2.331.851.46–2.351.611.26–2.061.521.18–1.95
30–39 years0.750.66–0.860.760.66–0.870.750.65–0.86
40–49 years0.640.55–0.730.670.58–0.780.650.56–0.75
50–59 years0.410.36–0.480.460.40–0.540.430.37–0.51
GHQ12 3+2.862.59–3.162.462.21–2.74
Any sickness absence1.191.06–1.34
SRH less than good1.661.48–1.87
n11 98411 98411 62911 524
  • Reference group omitted from table.

Stratified analyses show that ORs are higher in the male subsample (table 3). Associations are weak in younger age groups. Regarding family situation, unemployment is least associated with mental distress among individuals living in couples without children. Among single people with children estimates are contradictory and far from significant, probably due to the size of the exposed groups (n = 43 and n = 46, data not shown). The effect of unemployment exposure does not differ between those with and without prior mental distress (data not shown).

View this table:
Table 3

Logistic regression between unemployment and mental health. Odds ratios and confidence intervals (CI)

Stratified bynExpBase modelCIFull modelCI
Age (years)
Family situation
    Single no1071–1301.340.85––1.78
    Couple w2331–1301.381.02–1.891.070.76–1.49
    Couple no1971–1301.531.09––1.56
    children195131+1.491.06– 2.101.360.95–1.95
Occupational class
    Manual & low3111–1301.441.10–1.881.100.82–1.47
    Mid & high2531–1301.310.97–1.770.990.71–1.37
Employment contract
  • Stratified analyses part 1. Reference group omitted.

Among middle and higher non-manual employees, there is no association between unemployment and mental distress at T2 after adjustment. The association is especially strong for the self-employed. There is little association between unemployment and mental distress for those with temporary employment. For those who experience low social support at work, there is no association between unemployment and mental distress (table 4). Results regarding job demands are contradictory. Those with low levels of job control display a stronger association between unemployment and mental distress, especially among those with short-term unemployment. People working overtime have higher odds ratios than those working either part time or full time. Regarding wage at T1, ORs for both long- and short-term unemployed remain significant in the full model only for those in the lowest tertile.

View this table:
Table 4

Logistic regression between unemployment and mental health. Odds ratios and confidence intervals (CI)

Stratified bynExpBase modelCIFull modelCI
Job support
Job demands
Job control
Working Time
    Part time991–1301.470.92–2.331.050.63–1.76
    Full time3671–1301.491.17–1.911.140.87–1.48
    Tertile 13331–1301.591.23–2.041.311.00–1.72
    Tertile 21831–1301.040.72–1.510.870.59–1.30
    Tertile 31081–1301.450.90–2.341.050.62–1.76
  • Stratified analyses part 2. Reference group omitted.

In order to perform interaction analyses, all strata are dichotomized. Based on results from previous analyses working time is dichotomised as overtime vs. others, and wage into 1st tertile vs. others. Occupational class is categorized into middle or higher non-manual employees vs. all others. There is no ‘natural’ dichotomization of family situation. Based on the established risk group, family situation could be dichotomized as all single vs. all cohabiting regardless of children. According to the results of the stratified analysis the variable is also dichotomized as cohabiting without children vs. all others.

The only significant result is found for income (SI 1.89; 1.04–3.41), but only in the base model (see Supplementary data). The effect of family situation is assessed in several ways. When dichotomization is based on the risk group (single with or without children vs. others), SI is 1.21 after adjustment (data not shown), and when it is based on the stratified analysis (cohabiting no children vs. others), SI is 1.53, both insignificant. As one of the reference groups has an OR below 1 in the full model, estimates of the SI are imprecise. When reference groups are switched, the SI is however still insignificant (not shown). Also for occupational class there are problems with risks below one. When the reference group is switched, the SI is significant in the base model (0.30; 0.12–0.78), but not after adjustment. Job control borders at significance. For wage and employment contract, the OR is well below 1 in one exposure group after adjustment. When the reference group is changed to the opposite, the SI is however insignificant (data not shown).

Interaction analyses, although insignificant, support hypothesis 1 (role loss) regarding sex, social support at work, and working time. Analyses support hypothesis 2 (susceptibility) regarding family situation, occupational class, job control and wage. After adjustment, there are no signs of interaction with age, prior mental distress, job demands or employment contract.


In accordance with results of previous studies, the analysis for the full sample shows that associations between unemployment and subsequent mental distress, although attenuated, persist after adjustment. There is also a clear effect of unemployment duration. The two opposing hypotheses both receive some support.

Contrary to previous Swedish research,35–37 there is a clear difference between men and women, which could be interpreted as an effect of men’s higher identification with the job role, stronger breadwinning obligations and/or better jobs and thus more to lose from unemployment. Even after adjustment for employment contract, psychosocial work environment and wage, the SI remains virtually unchanged. Having high social support at work is associated with mental distress after unemployment, showing how attachment to and identification with a particular work place may create difficulties in situations of job loss. Those working overtime are also more affected by unemployment. It is likely that these individuals have especially strong work commitment, alternatively that their way of dealing with anticipated lay-offs has been to take on more work; a strategy that failed. That self-employed are more affected than others could partly be due to a sense of personal failure, coupled with strong economic repercussions. The self-employed are thus likely to be especially vulnerable to both role loss and economic consequences.

There is no effect of having children per se, but the combination of double-income, no kids is advantageous in situations of job loss. This finding could be due to having low breadwinning obligations and high social support. The results further show a protective effect of having a high occupational class, and a stronger effect of unemployment among those with low previous wage and low job control. These results could be explained by a lack of resources (social, educational, financial) and difficulties in finding a new job in less advantaged groups. Among those with low previous wage we might also find a larger proportion of people with only the basic flat rate benefit (who have not qualified for regular unemployment benefits) for whom unemployment may lead to serious financial strain.35

Regarding socioeconomic position, the hypothesis of general susceptibility thus is supported, whereas social and behavioural job-related aspects, as well as gender differences, support the hypothesis of role loss. It is possible that we would have found stronger evidence for general susceptibility in a different policy context—most in this sample would have had access to the rather generous unemployment insurance (80% of previous income during this time period, although with a fairly low ceiling). That social insurance may modify the effect of unemployment on health is indicated by Stuckler,38 while Nordenmark et al. show that income-related, in contrast to flat-rate benefits, while buffering, also preserve pre-unemployment differences in mental distress.28

Data limitations

While the use of registered unemployment means that all classified as unemployed also are unemployed (unless working in the black economy), we may miss some who consider themselves unemployed but have not registered. Excluding this group could lead to an attenuation of the associations between unemployment and mental distress as they are further disattached from the labour market. Even though an attempt was made to account for selection effects, these cannot be ruled out as adjustment was made only regarding one year; hence individuals with mental distress prior to 2002 could still be selected into unemployment. As only a minority of those with mental distress in 2002 also score above 3 in 2007, the group is highly volatile.

This study is naturally based only on those that responded at both time points. This means that the cohort is limited to 50% of those that were first sampled in 2002. There was an overrepresentation of men, immigrants, and single people among non-responders to the 2002 questionnaire. It is probable that there is also a selection of those less likely to be mentally distressed and less likely to become unemployed to the cohort. By the same logic, there should be a higher response rate and thus less response bias in our subsample (employed at T1 and without recent unemployment experience). A follow-up of a previous panel (1998–2002) also showed no selection to non-response from the first to the second wave. The problem of low response rates plagues most health surveys and the large sample cannot compensate for this. However, it is not clear that associations should differ between responders and non-responders.

That interaction analyses are never significant in the full model is disconcerting, and results should be interpreted with caution. Confidence intervals of the synergy index reflect the combined effect of all groups’ CIs, which often overlap due to the small size of some combinations of exposures.


Results show an independent effect of unemployment on mental distress, but the association is different in different groups. Both theories: role loss and differential susceptibility, receive some support. Mental distress after unemployment could result both from status change relative to a previously advantaged situation and from lack of resources due to an already disadvantaged situation.

Supplementary data

Supplementary data are available at EURPUB online.


This research was funded by Stockholm County Council

Conflicts of interest: None declared.

Key points

  • Unemployment is shown to have a moderate impact on mental distress even after adjustment for previous mental health status.

  • The consequences of unemployment vary between groups.

  • Circumstances under which unemployment is more likely to lead to impaired mental health are: having strong attachment to the labour market or to the prior work place, having high breadwinning obligations, or being disadvantaged in terms of occupational class, wage or job quality.


The authors would like to thank Michael Lundberg for valuable help concerning the interaction analyses.


View Abstract