This study uses the “Northern Swedish Cohort” (NoSCo), a prospective cohort study of all pupils in a medium-sized industrial town in northern Sweden (n = 1083) who completed or should have completed their final year of compulsory school in 1981 at age 16. This cohort was investigated using a comprehensive questionnaire containing more than 90 questions covering areas such as somatic and mental health, health behavior and labor market experiences. The participants were revisited with the same questionnaire at ages 18, 21, and 30, and most recently at age 43. The initial response rate was 99.7 % and the 27-year follow-up response rate was 94.3 % (n = 1010) of those still alive (n = 1071). The population is known to be representative of their age cohort on the national level in terms of demography, socioeconomic position, health and health behaviors although the local unemployment levels in their home town were initially higher than the national average [19, 20].
The study, including consent methodology, has ethics approval from the Ethics Committees of Uppsala University, Umeå University and Statistics Sweden as well as by the Regional Ethics Vetting Board in Umeå. Written consent has not been requested from these committees. The respondent is regarded as giving written consent when answering the questionnaire. Participants were/are able to opt out at any time simply by not completing any of the waves of the survey.
Exposure to open unemployment and youth programs
In order to measure exposure to open unemployment (here defined as actively seeking work while not being in employment, education or training) and participation in youth programs we use a battery of questions from the NoSCo that asked the 21 year-old respondents how many weeks they had spent in employment, studying, in open unemployment, and participating in youth programs since their previous interview at age 18. From these responses we constructed two variables measuring cumulative exposure (in months) to open unemployment and participation in youth programs between ages 18 and 21. In order to exclude frictional exposure (many individuals will for example be registering short periods of unemployment exposure when simply switching jobs) and focus on substantial exposure these variables were then dichotomized into more or less than 6 months of open unemployment between 18 and 21, and more or less than 6 months’ participation in youth programs between 18 and 21 (where less than six months is treated as no exposure). Six months exposure here represent the Swedish administrative definition of long term unemployment for youth. Additionally analyses were also performed, and are presented, using a combination variable of exposure 18–21 (no exposure, open unemployment, youth programs, both youth programs and open unemployment), where the cut off for exposure was lowered to five months in order to ensure a large enough group only exposed to open unemployment (n = 39). This is done to test the sensitivity of results to combinations of exposure. Sensitivity analyses further show results to be similar when using exposure to open unemployment and exposure to youth programs measured in number of months.
The dependent variables
The dependent variables in the study were identical indexes focusing on depressive, anxiety, and panic related symptoms, which can be conceptualized as internalized mental health symptoms (IMHS), at ages 21 and 43. Three questions based on measures of internalized mental health symptoms as defined at the time of the original survey [19, 21] and taken from well-known and validated surveys [22, 23] were used to create a composite index based on a theoretical and clinical evaluation of the seriousness of individual symptoms and combinations of different symptoms [24]. Respondents were asked whether they had or had not (no = 0 and yes = 1) experienced worry/anxiousness and anxiety/panic during the last 12 months. In addition, they were asked how often during the past 12 months they had experienced sadness or felt low on a 4-grade ordinal scale variable coded as never = 0, sometimes = 1, often = 2, always = 3. An ordinal scale index was created based on these questions ranging from 0 = not experiencing any symptoms to 8 = often feeling both worry/anxiousness and anxiety/panic while also often or always feeling sad and low. The order of the values in the scale was based on clinical judgments of the seriousness of these mental health symptoms as well as the frequency of symptoms. Experiencing anxiety/panic was for instance regarded as more serious in as compared to being worried/anxious. A detailed description of the order of the values in the scale is presented in Additional file 1: Table S1.
The mean IMHS score for all ages was low (1.13 at age 16, 1.22 at age 21 and 1.50 at age 43). The distribution of respondents on the index was skewed in a way that could be expected in relation to mental health problems with the most common value being 0 at all ages, which represented about half of the respondents, followed by 1 and so on.
Confounders
Because previous studies have demonstrated reverse causality between mental health and unemployment [25, 26] it is essential to control for baseline mental health. This was done by using the respondents’ IMHS at age 16 (which was created in the same way as the dependent measures for ages 21 and 43) as a measure of their mental health before potential exposure to youth unemployment. We also controlled for semesters in education at ages 18–21 to reflect the fact that not all youths face unemployment risks. Many youths in the cohort spent at least some of the period between ages 18 and 21 in regular education. We controlled for this by creating a variable that counted the number of spring or autumn semesters spent in education during the period of interest. This variable took values of 0–8, with observations evenly distributed over the range. Youths with higher scores had correspondingly lower risks of exposure.
We also controlled for seven pre-exposure confounders relating to youth unemployment and IMHS at 21 and/or 43: gender, long-term somatic health problems at 16, truancy at 16, parental social class at 16, parental employment at 16, living with both parents at 16, and paternal health problems at 16. These confounders were chosen among those available on the basis of them having either a relationship with either with the probability of exposure to open unemployment and youth programs and/or IMHS. Respondents were identified as having long-term somatic health problems if they reported suffering from diabetes mellitus, hypothyroidism, neurological disorder, impaired hearing, impaired vision, asthma or eczema. Truancy was scored on a scale ranging from 1 to 5 (1 = truancy at least once a week; 5 = no truancy ever). Three parental social class categories based on the parents being manual workers (blue collar workers) or professional workers (white collar workers) were defined: both parents blue collar, one blue collar and one white collar, and both white collar. Two parental employment categories were defined: both parents employed and one or both parents unemployed. Respondents were identified as having paternal health problems if their father suffered from alcoholism or physical/mental ill health.
Statistical analysis
The dependent variables, IMHS at ages 21 and 43, were created from a theoretical clinical perspective and can be considered to be functional on an ordinal level. To accommodate this, ordinal regressions were performed using the Polytomous Universal Model (PLUM) in SPSS with the logit link function. This procedure reports odds ratios in a way similar to that used for binary logistic regression models, but the possible outcomes are expanded. The difference is that ordinal regression relates to the odds of one group having a higher or lower score on the dependent variable than another [27]. The regression coefficients are not dependent on the steps of the ordinal dependent variable. Different equations (the number of categories of the ordinal dependent variable −1) are calculated, each with a different intercept, but with the same slopes [28]. Because these intercepts are not used to interpret the results they are not reported in the tables. To assess the results’ robustness, all of the analyses were tested using alternative approaches with different data requirements. The substantive results were not changed by using binary logistic regressions (dichotomizing IMHS as no symptoms = 0, one or more symptoms = 1) or by treating IMHS as a continuous variable (ranging from 0 to 8) and using a repeated measures linear mixed models approach with random intercepts.
To facilitate the interpretation of the odds ratios for the two different types of unemployment exposure in relation to each other and the confounders, the analyses in Tables 2 and 3 are built up stepwise. In both tables, Model 1 includes only exposure to open unemployment together with baseline IMHS at age 16; Model 2 includes only exposure to youth programs together with baseline IMHS at age 16; Model 3 includes exposure to both open unemployment and youth programs together with baseline IMHS at age 16; and Model 4 adds all measured pre-unemployment exposure confounders. No significant interactions between exposure to open unemployment and exposure to youth programs or between the exposure variables and any other variable were identified. Therefore the models do not include interaction terms.
In addition analyses are performed in Table 4 using only one fully controller model for IMHS at age 21 and age 43 respectively. This is done to test the sensitivity of results to combinations of exposure.
All statistical analysis was performed using SPSS v.22.0 (IBM Corp., Armonk, NY, USA).