Background Suicide mortality is high among the unemployed, but the role of causation and selection models in producing employment status differences remains to be understood. This study analyses the association between unemployment and suicide during different levels of national unemployment adjusting for several factors that might explain or mediate the relationship.
Methods The data comprised annual population-register and death-register information on 25–64-year-old Finns at the beginning of each year in the period 1988–2003; thus, forming 16 separate follow-up cohorts. Experience of unemployment was measured at baseline and during the previous year for each cohort. Suicide was followed for 12 months after each baseline giving a total of 7388 suicides.
Results Overall, age-adjusted suicide mortality was two to three times higher among the unstably employed and almost fourfold among the long-term unemployed. Adjustment for social class and living arrangements had small effect on the HRs, but adjustment for household income per consumption unit decreased the differences by 13% and 31% among the long-term unemployed women and men, respectively. When the national unemployment level was high, excess suicide mortality among the unstably employed was lower than during low unemployment when those becoming unemployed might be more selected. No such differences were found among the long-term unemployed.
Conclusion Long-term unemployment seems to have causal effects on suicide, which may be partly mediated by low income. As the effect of unstable employment is lower during the recessionary stage of the economic cycle some part of the excess suicide among the unstably employed is likely to be attributable to selection into unemployment.
- suicide SI
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Several studies have indicated that unemployment and a lack of full-time employment are related to a twofold to over a threefold excess suicide or attempted suicide mortality among men.1–10 The few studies that have included unemployed women have found at least as high a relative suicide risk as among unemployed men,1 10–13 with the exception of a time-series study from Italy14 that reported a twofold increase among women while it remained threefold among men.
Attempts to explain the association between unemployment and both suicide and attempted suicide have focused on causation and selection. In the former, unemployment may directly or indirectly elevate the suicide risk by increasing stress or mental health problems, or the likelihood of other stressful life-events, such as psychological, social or financial problems, that predispose to suicide. However, the association may be non-causal, stemming from pre-existing social, behavioural (eg, drinking problems) or psychological (eg, an unstable personality12) factors that predict both employment status and suicide risk, or from direct health-related factors (for instance, psychiatric morbidity1 15) that increase the risk for both unemployment and suicide.
The testing of the causal versus the selection hypothesis has given support to both contentions. Claims of a causal effect have been based on results from the UK4 and Denmark.5 In addition, a New Zealand workplace closure study found that exposure to involuntary job loss increased the risk of serious self-harm leading to hospitalisation or death.16 As far as Finland is concerned, living arrangements and income9 13 have been found to account for some of the employment status differences, but excess mortality of 1.6 in non-alcohol-associated and 2.3 in alcohol-associated suicide among men, and almost 2.3 and 3.8, respectively, among women, remained in a 11-year follow-up.
Nevertheless, some argue that much of the association between unemployment and suicide or attempted suicide is non-causal, and is confounded partly by social and socioeconomic factors and partly by pre-existing psychological factors12 or psychiatric morbidity.1 5 15 Moreover, according to previous studies in the Finnish context,7 9 13 education and social class explain about 30% of the suicide differences between the unemployed and the employed among men, and about 15% among women, while a Swedish longitudinal study17 found that controlling for several life-course confounders, including psychiatric diagnosis, explained 65% of the excess suicide risk associated with unemployment.
More research attention has been given to the association between unemployment and total mortality than to its association with suicide, but no consensus about the mechanisms that produce mortality differences has been reached. Study designs that are comparable to ‘natural experiments’ can further our understanding of this issue. Some studies have incorporated workplace downsizing and closure into design: no plausible selection processes are involved because in such context the chances of redundancy increase similarly for all employees regardless of their individual characteristics.18 Furthermore, the association between unemployment and total mortality has also been studied in the context of different general unemployment levels.3 10 19 20 The results suggest that during periods of favourable economic circumstances, when unemployment is likely to be highly selective, its association with total mortality is very strong, while during periods of high unemployment, when the risk of becoming unemployed is considerable, the excess mortality is much smaller or non-existent.19 20 It appears from these findings that the association between unemployment and total mortality established in observational studies may be strongly influenced by selection. However, as far as we know, no similar study design has been used to examine suicide mortality.
The objective of the present study was to analyse the association between the experience of unemployment and suicide mortality among Finnish men and women aged 25–64 years in the period 1988–2003. We distinguished between the stably and unstably employed and the long-term unemployed (see below Data and Methods). The specific aims were to:
Evaluate the strength of the association between unemployment status and suicide before and after adjusting for socioeconomic status, living arrangements and income.
Analyse whether these associations varied during different levels of national unemployment.
Data and methods
The target population consisted of men and women born between 1924 and 1978 and living in Finland between 1987 and 2003. Because of the data protection regulations of Statistics Finland, we did not have access to the total population, which would have violated the rules of data provision. Therefore, the data were based on a two-part sample: (1) an 11% random sample obtained from a Statistics Finland population-register data file covering all Finns with (2) a random oversample of death certificates covering 80% of all suicides (ethics approval TK 53-601-06). Statistics Finland used personal identification codes to link the information on the cause and date of death from the death certificate data to the information in the annual population-register data. All other information was obtained from the population database. We used sample design weights in order to take account of the oversample of deaths and, thus, to obtain results that are representative for the population.
For the purposes of the analyses in this study, we first extracted the men and women who were 25–64 years old on the 31 December for each year in the period 1987–2002 (thus forming 16 separate study baselines), non-institutionalised and in the labour force (either employed or unemployed), and we followed them for suicide mortality for 12 months beginning from the 1 January 1988, 1989 and so forth to 2003. We then aggregated these 16 follow-up cohorts, but in addressing the second research question (see above) we split the data to cover the period of low general unemployment (about 4%) in 1988–1991, very high unemployment (about 15%) in 1992–1995, and the two most recent 4-year periods 1996–1999 and 2000–2003 (with unemployment rates of 12 and 9%, respectively).
For each study cohort the study subjects were classified according to unemployment experience. The unemployed comprised those who were not employed but who had registered as a jobseeker with an employment office. Three categories were formed: (1) the stably employed were those who were employed at the baseline and who had been employed for the whole of the previous year; (2) the unstably employed had experiences of both employment and unemployment; (3) the long-term unemployed were jobless at the beginning of the follow-up year and for the whole of the previous year. Table 1 shows the proportions of person-years in each category.
Occupation-based social class, living arrangements covering both the partnership situation and the household composition, and income were adjusted for in the analyses. We also adjusted for education initially, but due to a zero effect on employment status differences in suicide it was left out of the final models presented here. Social class comprised 10 categories (including upper non-manual, lower non-manual in autonomous or non-autonomous work, and skilled and non-skilled manual21) with only 0.35% of both men and women falling within the category unknown social class. Living arrangements were categorised in eight groups of which married with children (46% of person-years among the men and 44% among the women), single (21 and 16%, respectively) and cohabitors (15 and 17%, respectively) were the largest. Household income per consumption unit during the year in which employment status was measured was used as a continuous variable. Consumption units were calculated as follows: the first adult in the household was given the value of 1.0, any additional adults and children aged 14 years and over the value of 0.5, and each child younger than 14 years the value of 0.3 (the Organisation for economic co-operation and development (OECD) adjusted consumption-unit scale).
The outcome measure used in this study was suicide mortality (in 1988–1995 International Classification of Disease (ICD) 9 codes E950–E959 and in 1996–2003 ICD10 codes X60–X84 and Y870 on the death certificate). We calculated the HRs for suicide from the Cox regression models in Stata22 using robust estimators to account for repeated measures on the same individuals at different baselines. There was an interaction (p<0.001) between sex and employment status, and the data were thus analysed separately for men and women. We also adjusted for the possible confounding effect of the calendar year.
Both men and women who had experienced any length of unemployment had a higher suicide mortality risk than the stably employed (table 1). The unstably employed men had about 2.8-fold higher age-adjusted suicide HR while the respective figure was about 2.3 among the women, and among the long-term unemployed the risk was 3.7-fold. Adjustment for social class and living arrangements had very little effect on the HRs—for example, only 3% among the women (eg, (3.69–3.60)/(3.69–1.00)*100 among the long-term unemployed)—but adjustment for household income per consumption unit decreased the differences substantially: a further 20–31% among the men and 10–13% among the women (model 4 compared to model 3) with the largest adjustments being among the long-term unemployed.
The main effect of the continuous national unemployment rate on individual suicide risk was 1.01 (p<0.001) among the men and 1.02 (p<0.01) among the women. Because there was an interaction (p<0.001) between individual unemployment status and the national unemployment rate among the men, the association between employment status and suicide mortality was also analysed during different levels of national unemployment operationalised here as four periods representing different stages of economic cycle.
Among the unstably employed men the HR was highest (2.7) when the national unemployment rate was low and lowest (2.08) during the period of high general unemployment (table 2). The ratios were again rather high during the two most recent time periods with intermediate levels of unemployment: about 2.2 and 2.5, respectively. A different pattern occurred among the long-term unemployed: the men's HRs were much more similar irrespective of the time period with the lowest ratio (2.56) occurring during the period of 12% unemployment and the highest (2.99) during those of the lowest and highest unemployment.
As with the men, the highest suicide ratio among the unstably employed women (2.38) was during the period of lowest unemployment. The risk was lowest during the two periods with the highest unemployment level (1.99 in 1992–1995 and 1.83 in 1996–1999), but increased to about 2.3 again during the latest time period when unemployment was lower than in the 1990s. The HR fluctuated among the long-term unemployed women being at the highest (4.02) during the most recent time period. Some of these more inconsistent results among the women may be attributable to the small number of suicides.
For this study we had access to comprehensive population registration data with a large random sample of suicides. There were practically no loss of follow-up and because the variables used in this study were obtained from registers there was no self-report bias. Suicide statistics are regarded very reliable in Finland.23 The decision on categorising a death as a suicide is made by the forensic examiner; the medico-legal autopsies are carried out by a small group of well-trained pathologists24 and 97.9% of all violent and accidental deaths are subject to this kind of autopsy. It is rare in Finland for fatal injuries to be undetermined in terms of whether they are intentional or unintentional: they constituted only 4.6% of all accidents and violence in the 25–64 year age group in 1995.25
Our analyses were based on the experience of unemployment in the year before baseline. We used a follow-up period of 12 months in order to update the unemployment status on a regular basis and, thus, avoid bias related to misclassification of exposure. Some previous studies in the Finnish context9 13 used a follow-up time of 11 years and the observed HRs were slightly smaller than observed here. A limitation in the current study was that the data did not enable us to determine the employment status at the time of death only at the turn of each year. It is possible that redundancy could affect health in a manner similar to other stressful life events producing higher excess mortality soon after the event and a weakening effect over time. Further research is needed to confirm such assumptions.
We observed high excess suicide mortality among those who had experienced any length of unemployment. Adjustment for social class and living arrangements decreased the age-adjusted relative differences only a little, but income attenuated the association notably especially among the men. For a more detailed understanding of the pathways between unemployment and suicide and of the mediating effect of income further work should, if possible, include additional confounders such as home ownership or other wealth indicators.26
Workers in a precarious situation—that is, unstably employed—had a lower suicide risk during times of high general unemployment, but among the long-term unemployed the risk either remained quite similar between periods (among the men) or was related inconsistently to general unemployment (among the women). This alteration in relative suicide risk by national unemployment level is in agreement with a comparative study for 26 European Union countries27 that found substantial variation between populations in how sensitive suicide mortality was to economic crises. According to the authors, the explanation for the inverse association between national unemployment level and suicide mortality rate appearing in Finland and Sweden includes governmental commitment to social support during times of economic stagnation. An alternative explanation that we find more compelling is the countercyclical variation in one of the more proximate determinants of suicide namely alcohol consumption.28
The male to female suicide ratio is large in Finland and it is also evident in this study in the absolute rates by employment status. However, in accordance with most previous studies,1 10–13 we found that the relative suicide risk among unemployed women was at least as high as among unemployed men. The one exception is a time-series study from Italy14 reporting a lower RR among unemployed women than among unemployed men. The authors suggest that not having a job has a different influence on the two sexes and that the possibility of choosing a housewife role offers women an alternative to unemployment and an accepted social status. However, feelings of social-identity deprivation most probably vary between societies. For example, in Finland the employment rate among mothers whose youngest child is aged 3 years or more is one of the highest in the OECD countries29 and the very high relative suicide risk among the women in this study sample possibly indicates that in such circumstances staying at home is hardly an option.
The causal-model hypothesis
According to Bartley,30 the causal pathways between unemployment and mortality may include financial strain, stress related to job loss, changes in health-related behaviour and the accumulation of unemployment experiences among certain individuals. In this study, low income accounted for 31% of the excess suicide mortality among the long-term unemployed men but less among the unstably employed. This may reflect the fact that the social welfare system in Finland provides recently redundant workers with more financial resources than it gives the long-term unemployed. There were some legislative changes in the unemployment-benefit system during the period examined in this study but they were mostly only of technical significance. What was probably more relevant was the non-indexation of benefit levels in several years since 1991, which may have led to decreased relative incomes among the unemployed. Excess suicide mortality among the unemployed was, indeed, mediated by household income per consumption unit to a larger extent during the latter two periods than during the first one (results not shown). However, understanding the reasons for these findings requires further research.
Congruent with these results, previous studies on employment status and suicide have shown that low income in Denmark,5 and the lack of access to a car in England and Wales,4 both of which could be consequences of unemployment, have been associated with elevated suicide. We also found that income seems to attenuate employment status differences less among women: this suggests that for them it is perhaps not financial hardship as such, but possibly the fact of not having a job and the latent consequences of employment (giving a time structure to the day, self esteem, the respect of others31) that are significant. Furthermore, unemployment among men in particular may be associated with other negative life events causing problems to escalate from one sphere of life to another.
Studies on workplace closure have found support for the long-term stress-related model: a follow-up study conducted in New Zealand revealed that involuntary job loss increased the risk of serious self-harm leading to hospitalisation or death,16 and a longitudinal study carried out in Denmark showed that unemployed people were significantly lower in psychological well-being than the employed;32 thereby, indicating a possible pathway from unemployment to suicide. Furthermore, healthy men who became unemployed during prospective studies conducted in the USA and Finland showed more symptoms of depression and anxiety, and were more likely to visit the physician33 and to suffer from mental ill-health,34 respectively. Furthermore, according to a meta-analysis on the effect of unemployment on mental health, Karsten and Moser35 found that the effect size was larger for the long-tem unemployed.
Marital status5 and living arrangements9 13 have also been found to be associated with employment status differences in suicide mortality. In this study their effect turned out to be small, which may stem from the study design with mortality follow-up of only 12 months. It is possible that the supportive effects of the family, or alternatively the lack of comfort for those living alone, have impacts that show only in the long term. Furthermore, the causal order between unemployment and living arrangements is unclear in these data with any changes conceivably either preceding or following unemployment. However, these alternatives do not strongly influence the interpretation of our results because adjusting for living arrangements had relatively little impact on employment status differences in suicide.
The non-causal-model hypothesis
It was found in a previous Finnish study7 that occupation-based social class differences accounted for about 30% of the employment status differences in suicide, but in this study the adjustment had very little effect on the risks. The discrepancy in the results of these two Finnish studies may stem from the different age groups (30–54 vs 25–64 years old) and study periods (1981–1985 vs 1988–2003) covered, and also from the differences in the length of mortality follow-up—5 years in the earlier study but only 12 months in this one. In the longer run, the effects of education and social class could, for example, be reflected in enhanced re-employment36 opportunities, which might alleviate the suicide risk.
The data used in this study gave no information on illnesses, but Martikainen7 showed that adjustment for the use of reimbursable medicines or the number of allowed days of sickness absence had little effect on employment status differences in suicide. Such results indicate that direct health-related selection is unlikely to be a major determinant of differences in suicide mortality.
We further assessed the possible selection effect by comparing the suicide mortality between times of very low and high general unemployment. If selection was based on pre-existing social, behavioural or psychological factors, or poor health, its effect should be stronger during times of low unemployment and the relative suicide rates should, accordingly, be higher. A Finnish study on total mortality during the same years20 confirmed these expectations. As unemployment rose during the recession presumably more healthy people became jobless, and the unemployed and the employed also shared other characteristics causing lower excess mortality among the jobless during the period of high unemployment. In the present study the unstably employed men and women had a somewhat reduced suicide risk during times of high general unemployment. This finding is consistent with non-causal explanations. For example, Blakely et al1 indirectly estimated that about half of the excess suicide mortality among the unemployed was due to the confounding effect of mental illness, while Lundin et al,17 using longitudinal data, found that controlling for several life-course confounders, including psychiatric diagnosis, explained 65% of the excess suicide risk associated with unemployment. However, the current study, which adopted a study design that resembled ‘a natural experiment’, did not produce such large selection effects and, in particular, the relative suicide risk among the long-term unemployed did not decline in line with the increasing national unemployment level.
Overall, our analyses provide evidence of both causal and non-causal explanations of the association between unemployment and suicide. As far as the unstably employed are concerned, the stage of the economic cycle at which unemployment occurs seems to play a role; thereby indicating selection into unemployment. However, long-term unemployment seems to have a causal effect on the risk of suicide, which among men in particular may be partly mediated by having a low income.
What is already known on this subject
Unemployment is strongly related to increased suicide mortality. However, the reasons for this are not fully established. Study design examining this association across time periods of different national unemployment rate resembles ‘natural experiment’ and can help to assess whether it is caused by selection or causation.
What this study adds
This register-based study found two to three times higher suicide mortality among the unstably employed and almost fourfold among the long-term unemployed.
Adjustment for social class and living arrangements had small effect on the HRs, but adjustment for household income per consumption unit decreased the differences more notably.
Long-term unemployment seems to have causal effects on suicide, which may be partly mediated by low income.
Because the effect of unstable employment is lower during the recessionary stage of the economic cycle, some part of the excess suicide among the unstably employed is likely to be attributable to selection into unemployment.
Funding This work was supported by the Academy of Finland.
Competing interests None.
Ethics approval This study was conducted with the approval of the Statistics Finland (TK 53-601-06).
Provenance and peer review Not commissioned; externally peer reviewed.