For the present study, participants were originally recruited from a large representative sample of the Swedish working population in 2003–2005 arising out of the Swedish Work Environment Survey (SWES). SLOSH was conceived as a follow-up to SWES with a more detailed prospective collection of data on work environment and health . The SLOSH population is surveyed biennially since 2006 by self-administered postal questionnaires. Participants fill in one of the two questionnaires: one addressed to workers gainfully employed for at least 30 % of full time, or one designed for “not gainfully employed” respondents working less than that or not at all; the latter category includes mostly persons outside the labour force (homemakers, non-working students, pensioners etc.). From 2008 onwards, the SLOSH questionnaires make it possible to assess the impact of company downsizing and the resulting changes in employment status on the health of the workers. Therefore, we used data collected in the second (2008) and third (2010) waves of SLOSH. Data collection in 2008 occurred before the outbreak of the Great Recession. In this report 2008 is treated as baseline and 2010 as follow-up.
In total, 8771 persons returned a questionnaire in both 2008 and 2010; attrition rate between these waves was 23.3 %. Those with missing or incomplete data on education or depressive symptoms (N = 418) were excluded from the analyses. Given our research focus, we further excluded respondents who were unlikely to experience collective compulsory redundancies during the Great Recession:
1179 self-employed, farmers and workers of microenterprises with less than 10 employees;
1800 workers of larger enterprises with some downsizing but no collective compulsory redundancies (questions on compulsory redundancies not completed);
1529 respondents considered economically inactive by ILO definition  (non-working students, homemakers, retirees) or non-employed for reasons other than downsizing during the Great Recession.
This perspective was adopted for the following reasons: (a) Downsizing in microenterprises (including farms) is not subject to the stringent LIFO rule stipulated in the Employment Protection Act : exemptions of “key workers” from this rule may affect both the downsizing procedure  and mental health outcomes, possibly resulting in artificial inflation of the results due to a higher level of psychological distress. (b) Staff reductions without compulsory redundancies can represent strategic downsizing aimed at promotion of long-term organisational benefits. This approach may be associated with a less detrimental health impact due to support and employment of surplus workforce  (inclusion of persons affected may result in artificial deflation of the results). (c) Being non-employed might be associated with poorer mental health (i.e. artificial deflation of the results due to the inclusion of voluntarily unemployed or disabled persons in the reference group).
Furthermore, given our research focus on people who cannot withdraw from situations of compulsory redundancies – workers who lost their jobs and layoff survivors – we have also excluded 238 employees of downsized organisations who retired, quit or found another job before becoming unemployed. These exposures might be associated with either poorer or better mental health. Finding another job could represent a healthy outcome to impending layoffs, as healthier and better educated workers may find it easier to obtain new employment before the actual job loss. In contrast, older and less healthy workers are overrepresented in jobs that have become obsolete as a result of technological developments . In fact, difficulties obtaining new employment may be influencing the decision to retire early. Small numbers and a highly heterogeneous composition of respondents would prohibit a detailed analysis of relationships in this group.
Finally, because of missing data on either organisation size, permanence of employment, self-employed/farmer status or downsizing exposure, we excluded another 104 persons.
Consequently, the final analytic sample consisted of 3503 persons. The sample comprised three groups of permanent and temporary workers employed in small (10 to 49 workers), medium and large organisations (≥50 workers): (a) 1845 workers in companies with no downsizing, (b) 1462 layoff survivors, who remained at work and (c) 196 displaced workers who lost their jobs through compulsory redundancies.
Dropout analysis and representativeness of the analytic sample
We conducted a dropout analysis to test, whether dropout between the 2008 and 2010 SLOSH waves was related to demographic characteristics (age, gender, marital status and education), employment and depression. Results from a multivariate logistic regression (outcome: loss to follow-up in 2010 versus responding to SLOSH in both 2008 and 2010) indicate that dropout is significantly predicted by male gender, younger age, lower education, non-permanent employment and being single, all p < 0.001. Similar patterns of dropout were earlier reported for non-respondents to SLOSH in relation to SWES participants [27, 28]. Yet our results indicate that depression did not significantly affect the likelihood of loss to follow-up in 2010 (p > 0.05).
Furthermore, using descriptive statistics, we checked whether our analytic sample can be regarded as representative for the sample of all respondents to the 2008 SLOSH wave. Compared to the total sample, the analytic sample included higher proportions of men (46 vs. 45 %), singles (45 vs. 44 %), university educated (39 vs. 36 %) and permanently employed workers (95 vs. 88 %) and had a lower mean age (48 vs. 49 years). However, the distribution by the level of depression at baseline showed that the proportions of non-depressed respondents were equal in both samples (76 %). Thus, the representativeness of the analytic sample is adequate with respect to likely depression prevalence.
The study was approved by the Regional Research Ethics Board in Stockholm (Ref.no: Dnr 2006/158-31, Dnr 2008/240-32 and 2010/0145-32). All subjects gave their written informed consent. The entire data collection was carried out by Statistics Sweden on behalf of the Stress Research Institute at Stockholm University. Statistics Sweden delivered the data to the researchers in such a form that neither individuals nor groups based on, e.g., employment at a certain workplace, could be identified.
Exposure to downsizing during the Great Recession
Downsizing is defined as a process whereby an organisation reduces its personnel, particularly, through redundancy . All participants were asked in 2010 whether they had experienced downsizing in the last two years. The unexposed group answered this question in the negative; job losses for other reasons were precluded. Persons with “yes”-responses were further prompted to specify a proportion of the employees made redundant on a scale including “less than 8 %”, “between 8 and 18 %” and “18 % or more”. People with non-missing responses to this item were further classified based on the question: “In what way were you personally affected by the changes?”. Exposed subjects, who answered that they received a warn notice and became unemployed, were classified as displaced workers. The exposed group of layoff survivors included persons continuously employed in the same downsized organisation, both notified workers who did not have to leave and those never notified. We coded the variable as 1 = employees in non-downsized organisations (reference group), 2 = layoff survivors and 3 = workers displaced due to downsizing.
Depressive symptoms in 2008 and 2010 were assessed by a brief version of the depression subscale of the Hopkins Symptom Checklist 90 . The scale (SCL-CD6) measures one-week prevalence and includes six items covering the core symptoms of depression: lowered mood (“feeling blue”), loss of interest (“feeling no interest in things”), reduced energy and diminished activity (“feeling lethargy or low in energy”), marked tiredness and, possibly, psychomotor retardation (“feeling that everything is an effort”), excessive worries reflecting psychic anxiety, phobic, hypochondriac or obsessional symptoms (“worrying too much about things”), and self-accusation due to feelings of guilt or unworthiness (“blaming yourself for things”). This scale is particularly suitable for assessment in large population surveys because of its brevity, ease of administration and the central role of clinical validity in the selection of items . Respondents rated how much they have been troubled by each symptom on a five-point Likert scale. The total sum score ranges from 0 to 24. The SCL-CD6 with a coefficient of homogeneity of 0.70 by Mokken analysis indicates a meaningful dimensional measure of depression severity. The scale has proven to be valid and had higher uni-dimensionality than longer epidemiologic instruments, thus being more specific for measuring depression as the underlying construct. The standardisation of the SCL-CD6 was based on receiver operating characteristic analysis, using the Major Depression Inventory as an index of validity. A score of 17 or higher was found to be the best cut-point for major depression (sensitivity 0.68, specificity 0.98), with significantly predicted subsequent use of antidepressants and hospitalisations for depressive episodes . The subjects were classified in accordance with their score values as being likely to have major depression (from 17 to 24), less severe depression symptoms (from 10 to 16) and no depression (from zero to nine) [31, 32]. The variables denoting the level of depression at baseline and follow-up include three categories based on this classification: 1 = no depression, 2 = less severe depression symptoms and 3 = major depression.
Demographic factors included age, sex (in the analyses combining both genders) and education, derived from national registers, and self-reported marital status. These factors are regarded as potential confounders as they may influence the experience of unemployment and depression [33, 34]. Education has been the key proxy for socioeconomic status (SES) in earlier studies: it has the advantage of relative stability across the life course in adults. Moreover, education is less prone to bias of reverse causality (i.e. health affecting SES) than measures like income and occupation . We linked the register-based information with the questionnaire data by means of unique ten-digit personal identification numbers. Age was measured in years, educational level included three categories: 1 = mandatory education only, 2 = high school or comparable, 3 = university degree. Marital status was assessed with a direct question and coded as 1 for married/cohabiting and 0 for single.
Employment variables included permanence of employment at baseline and a measure of changes in the employment status at follow-up. We controlled for these variables in order to account for a potentially adverse impact of lost seniority after losing a permanent job  and to consider the effects of reemployment: finding paid employment is known to reduce depression risks in displaced workers . Permanence of employment was assessed with a direct question and coded as 0 for various types of temporary employment, such as project-based or substitute, and 1 for permanent employment. Regarding employment status, respondents were categorised by the type of questionnaire they completed in 2010 as being either “gainfully employed” for at least 30 % of full time (code 1), or “not gainfully employed” i.e. working less than that or not at all during the past 3 months (code 0).
We also adjusted for past redundancies to exclude the possibility that the observed depression risks are due to long-term psychological scarring from earlier layoffs  prior to the recession. This variable is coded as 1 if respondents indicated in 2008 that they had survived layoffs or had been laid off in the previous two years (2006–2008); otherwise it is coded as 0.
A self-reported measure of long-term sickness captures underlying chronic medical and psychiatric conditions, which may be associated with depressive symptoms and influence one’s experience of unemployment [38, 39]. It is based on the information on long-term leaves with sickness benefits, activity or sickness compensation (0 = no long-term sick leaves in both 2008 and 2010; 1 = long-term sick leaves in 2008 or 2010).
For all statistical analyses, we used the STATA software package, version SE 11.2. First, we calculated descriptive statistics (numbers and percentages, means and standard deviations (SD)) and evaluated the bivariate gender-specific associations of socio-demographic and health characteristics with the exposure status using Pearson’s χ2 test and analysis of variance, when appropriate. Significance was considered at p < 0.05.
Second, we performed multivariate analyses of relationships in line with our research questions. The first set of multivariate analyses examined job displacement and surviving a layoff during the Great Recession as key predictors of depressive symptoms at follow-up (social causation). Relative risk ratios (RRR) with 95 % confidence intervals (CI) were estimated from multinomial logistic regression models. While risks of depressive disorders are generally higher in women, job displacement may be more detrimental in men due to demands posed by the traditionally male-gendered responsibility for breadwinning . Therefore, in addition to estimating the strength of relationships in the total analytic sample, we performed analyses stratified by gender, while adjusting for demographic and employment variables, depression at baseline, past redundancies and long-term sickness. Exposure status, education and depression at baseline were treated as factor variables: this procedure creates dummy variables for the levels of categorical regressors .
In the second set of multivariate analyses, we examined whether pre-existing depression increases the risk of being laid off (i.e. becoming unemployed) when organisations downsize (health-related selection). These analyses were restricted to 1658 victims (i.e. displaced workers) and survivors of layoffs during the Great Recession. Unemployment at follow-up was coded as 1 for displaced workers and 0 for layoff survivors. Level of depression at baseline was treated as the key explanatory factor variable. The multinomial logistic regression models for men, women and both sexes combined were adjusted for variables which can affect the probability of losing a job during downsizing, including demographic factors, permanence of employment, past redundancies, long-term sickness and scale of downsizing during the Great Recession (dichotomously coded: large versus minor staff reductions of less than 8 %).