Is unemployment in young adulthood related to self-rated health later in life? Results from the Northern Swedish cohort

It is generally agreed that unemployment is related to poor health [1–3]. It has been debated whether unemployment causes poorer health or if poorer health among unemployed individuals can be solely explained by poor health increasing the risk of getting unemployed. The most common view is that unemployment in fact causes poorer health [3], but there are also a few studies arguing against this [4]. The study context has been shown to have a major role in explaining how unemployment is related to poorer health [2], so it is possible, therefore, that both those who argue for and against the causality link might be partly correct.

Less is known about the long-term effects of unemployment on health. The few studies that have examined this are well in agreement that unemployment is related to poorer long-term health [5–9] as well as other social adversities such as lower income [10]. A long-term negative health effect from youth unemployment has been shown in studies using the Northern Swedish Cohort [11] in the form of psychological symptoms [6, 8, 9], somatic symptoms [5, 6], and high blood pressure [7] in follow-ups of 16 year olds at 30 and 42 years of age. It is rare for studies to look at long-term follow-ups of health effects from unemployment at a later age. Strandh et al., using data from the Northern Swedish cohort, could not confirm a long-term effect on psychological symptoms from unemployment at the age of 30 years in a 12-year follow-up [8].

Many variables have been included as part of the statistical analysis for studies of the relationship between unemployment and health, the most common being gender, age, education level, marital status, household income, geographic location, and social network/support [2]. In studies of the relationship between unemployment and health, most of these variables are commonly only of interest as potentially confounding variables and are not presented with stratified estimates for each outcome of the factors. Gender, age, and geographic location were actually the only factors where results were reported for each outcome in at least 5 of the 41 studies in a recent review [2].

In previous studies with results presented on the group level, the effect on health from becoming unemployed has usually differed between groups [2]. The groups most disfavored by unemployment often vary depending on when the study was performed and the target population that was studied. Only for manual workers (compared with non-manual workers) [12–14], those unemployed due to health reasons (compared with those unemployed for other reasons) [15], and those with a weak social network (compared with those with a strong social network) [16, 17] has a greater risk for poor health been unequivocally demonstrated. However, these conclusions are based on only a few studies, so it can still be questioned whether similar conclusions can be drawn for any context or population.

There is a need for gendered and contextualized analyses as studies in the field have shown that unemployment could have various impacts on the health status of men and women [12, 18, 19]. Here, Raewyn Connells theories about how the patterned relations between men and women that form gender as a social structure, could be useful [20]. According to her theory, gender relations are on a structural level integrated into the labor market and in this way, different labor market conditions are constructed for men and women. Also for other groups, such as age and education, stratified results have been inconclusive, and they seem to depend on the study context. For some groups, there have even been results indicating no health effect or even potentially positive health effects from unemployment, e.g. for Spanish women [14] and for Swedes with only a primary-school education [21]. Thus it is not usually possible to draw general conclusions about how unemployment affects different groups [2].

Potentially confounding variables must be considered in the analysis of the effect on health from unemployment, and they need to be handled well in the statistical model. One of the keys to having good estimates of the health effect from unemployment is being able to overcome the problem arising from health selection, which appears due to people who become unemployed being more likely to have previous health problems than those who remain employed. For long-term effects due to unemployment, caution about unemployment during the follow-up period is needed to avoid interpretations related to a more recent unemployment experience. In previous studies this has been handled in the statistical analysis model [5–9], while a novel approach used in our study is to only include those without unemployment in the follow-up period. The health effect from being re-employed has been studied many times [1], but such research has had a different focus than our study, which is on the lasting health effect later in life due to unemployment.

Thus, our aim was to examine how unemployment is related to long-term self-rated health among 30 year olds, with an emphasis on how health differs in relation to education level, marital status, previous health, occupation, and gender.

In our study, we show in a 12-year follow-up of 30 year olds that there is a negative health effect from being unemployed at the age of 30 despite having had steady employment from the ages of 30 to 42 years. This provides evidence for a long-term effect from being unemployed at an older age than has previously been shown. Despite rather small samples (100–300 individuals), it is evident from our stratified results that the effect from unemployment differs between groups of individuals. Most pronounced is the long-term negative health effect for those with upper secondary education, those living alone, medium- and high-level white-collar workers, and women, while there was at least an indication of a negative long-term health effect of unemployment for all other groups.

Strandh et al. studied the long-term health effect from unemployment at the age of 30 from the same cohort for psychological symptoms, but could not confirm a long-term negative health effect [8], which is in contrast to our results. Their study used a different health measure, which probably explains the different results. Our findings of a larger negative health effect for women than men is in agreement with two previous studies from the Northern Swedish Cohort [19, 32], as well as with other Swedish studies [21, 33]. Contrary to these studies that had a short-term perspective on the health consequences from unemployment, our study investigated the long-term effect. It is an interesting finding, therefore, that women seem to be more disfavored from being unemployed both in the long- and short-term than men. There are still not any commonly agreed upon reasons as to why Swedish women seem to suffer more from unemployment than men. In applying Connell’s theory of gender relations on the Swedish labor market [20], it turns out that it is strongly gender segregated with women working in less paid occupations within the public sector and in worse work-environment [34, 35]. In addition, women work in more unsecure labor market positions and much more often than men they have part-time work. These could potentially be part of the explanation for our findings, and because it is important to find explanations behind the differences between unemployed men and women we recommend that more research is conducted on this topic from a gender theoretical perspective.

Few studies have presented results for health effects of unemployment for those with different education levels, and those that have looked at this have shown inconclusive results [2]. In our study, we showed a large negative health effect due to unemployment for those with upper secondary education, while there was a lower effect on health for those with secondary education and university education. Despite estimates with a potentially large bias due to few unemployed with an upper secondary education, this group’s health status seems likely to be more affected from unemployment in our study population. Also for occupation, only a few studies have presented stratified results [2]. Our study shows opposite results compared to these studies, indicating more health problems for the unemployed from a high-level occupation class than others. Stratified results for marital status have only been presented in two studies, and they showed no apparent difference between married and single individuals [12, 21]. Our study is, therefore, the first to show results indicating that single individuals might be more affected by unemployment than married individuals. Lack of a strong social network has been shown to be related to poorer health in two previous studies [16, 17], and perhaps our results could be interpreted as there being weaker social structure among single persons and that this has a negative effect on their health in relation to married people if they become unemployed. Differences in results for occupation and marital status between our study and others might be explained by us having a focus on the long-term effect from unemployment, which was not the case for previous studies. Although our study provides new and valuable information, more research is needed to be better informed about how education, occupation, and marital status are related to poor long-term health due to unemployment.

A strength with our study is the very low attrition rate (6%). Still, despite a large sample of 1010 participants at the latest follow-up, not all stratified results fulfilled the recommendation for logistic regression of at least 10 events per variable for the less-common outcome [29]. Thus, the stratified results might be unreliable and give biased results, both for these and other estimates where the sample size was only slightly above the recommended threshold. Nevertheless, our results on the group level are valuable from a descriptive point of view, even if they cannot give very strong statistical evidence. We restricted our selection of individuals to those without unemployment during the follow-up because we did not want a prolonged unemployment experience to affect our effect estimates. In our sensitivity analyses, results were similar when a variable for unemployment during follow-up was used, while not excluding individuals with unemployment during the follow-up gave a slightly stronger negative health effect. Thus, this selection criterion is likely to have little bearing on the estimates.

A negative long-term health effect due to unemployment might be related to precarious employment during the follow-up period. Of the unemployed with no unemployment during the follow-up in our study, 42 experienced precarious employment during the follow-up with only 15 of these experiencing it during the majority of the follow-up period. To avoid effect estimates that are mainly reflecting effects due to precarious employment, a further limitation of our study sample to only those without precarious employment during the follow-up period was an alternative. However, analysis based on such restriction would also require that precarious employment during 1992–1995 was used to define the employed group and a larger focus on precarious employment which was not the scope of our study. It would also result in a too small sample to have reliable estimates to restrict ourselves in such a way. We therefore considered our definition of the study sample and the labor market status groups to be the most feasible. If there is a non-negligible bias due to precarious employment during the follow-up period for our effect estimates then our conclusions are still likely to be valid, although they would then be indicative of unemployment being related to poor health due to future unstable labor market positions rather than due to the unemployment itself. It has been shown in studies from our study cohort that those with precarious employment has a poorer health than those with a stable employment [36, 37]. However, these studies have not evaluated the long-term consequences from having had a precarious employment, which is an angle that would be valuable to study and could be recommended for future research.

The idea with propensity scores is to create two groups in the same manner as an RCT and thus to avoid problems from confounding. For the overall long-term health effect due to unemployment, our results were similar to those from the G-computation that were derived using logistic regression. Similar results are an indication that both methods work well, and are also in line with previous comparisons between propensity score techniques and logistic regression [38], but unmeasured confounders can still cause problems for the estimates. Also for our third method, logistic regression, the conclusions were generally similar. The confidence intervals estimated from the non-parametric Bootstrap, which was used for the G-computation and the IPW estimators, were derived from small samples. The G-computation estimator had small confidence intervals for stratified estimates and might therefore give more significant results than there really are. On the population level, the sample sizes are large enough so the bias for the confidence intervals based on a non-parametric Bootstrap should be very small.

Health prior to unemployment might have been an important confounding factor to include in our analyses to limit the bias. We could have used information from earlier questionnaires (1981, 1983, and/or 1986) in the Northern Swedish Cohort. However, these questionnaires did not include self-rated health and were not informative about health close to the unemployment period (1992–1994) of our study. We did use self-rated health at the time of recent/current unemployment in order to explain how health has changed between the two occasions (1995 and 2007). Interestingly, our stratifications on self-rated health in 1995 showed very similar results. This indicates that health status at the time of becoming unemployed at most might play a small role in the long-term health experience from unemployment. This also gives good support for our results being highly reliable and not confounded by previous health. Avoiding unemployment therefore seems valuable from a public health perspective not only for those that suffer from poorer health at the time of the unemployment, but also for those whose health is at most marginally affected at the time of the unemployment.

In a previous review, it was shown that only gender, age, and geographical location are commonly reported on the group level despite results on the group level showing that the context matters for the extent to which unemployment leads to poorer health [2]. Our study supports the recommendation from this earlier review that there should be greater focus on results on the group level. In the review, it was also reported that it is common to include many factors in the statistical model for the analysis of the health effect due to unemployment in order to control for confounding effects, as is also the case with other social epidemiological research questions. Still, it is apparent that only a few, if any, of these factors are commonly analyzed on the group level. We therefore support the recommendation to increase the focus on the group level for our and other research questions within the social epidemiological research field.

We limited ourselves to self-rated health as the health outcome. It would be valuable to investigate the long-term health effect of unemployment on other health outcomes, e.g. somatic symptoms, especially because the study by Strandh et al. showed different results than ours. In future studies with similar research questions as ours, we propose that the G-computation and IPW estimators should be used more frequently because they present the marginal effect and not the relative effect like logistic regression. An article with a more thorough evaluation of the sensitivity to the model set-up, the impact from the definition of labor market status, and the similarity of estimates from G-computation and propensity scores with IPW is planned from the same material as in this article. These issues are highly important for the interpretation of the causal effect from unemployment on health.

Because there is a limited number of studies presenting results for groups of individuals, more research is needed to better understand for whom and to what extent unemployment is related to poor health. Still, studies based on small samples such as ours can provide valuable evidence for policy makers, and such studies from a public health perspective can help guide decisions on which labor market measures to prioritize and to whom they should be directed. It might, for instance, be recommendable to focus more on understanding why women seem to suffer more from unemployment than men and to potentially prioritize efforts that can improve their chances of being re-employed and thereby improving public health. Our results are also important to provide support for future studies that can confirm the relationships observed in this study. The study context has been shown to be highly relevant for research within unemployment and health, but we still think that our study will provide highly relevant information both for other areas within Sweden and in other countries.

In comparison to young adults with employment, those who are unemployed suffer from poorer health not only shortly after their unemployment, but also later in life. Our study therefore implies that it is important to lower the unemployment rate during young adulthood from not only an economical, but also from a long-term public health perspective.

The health effect due to unemployment varies between groups. For future research, it is important to put more emphasis on identifying groups of individuals for whom unemployment is most related to ill health so that efforts can be put towards the groups with the greatest need.