3. The potential protective or compensatory role of SOC on the association between stress and health.

#### **2. Method**

#### *2.1. Participants*

This cross-sectional study was based on data from adolescents in public lower- and upper-secondary schools, in five municipalities from inland and coastal rural areas in the county of Trøndelag, located in Central Norway. The schools offer vocational and academic study tracks that are representative of Norwegian upper secondary schools. In the data collection from 2016, 1906 students were invited to participate in the study, with N = 1282 responding on a questionnaire (a response rate of 67%). Non-responses were caused by students not being at school at the time of data collection, non-willingness to participate or because some classes did not have the chance to participate as the teachers could not administer the questionnaire. No detailed information was available on non-responders. Adolescents <13 or >19 years (n = 49) were excluded, resulting in n = 1233 (64%) being included in the study sample (Table 1).


**Table 1.** Demographic characteristics of the sample.

#### *2.2. Procedure*

Data collection was approved by the Regional Committee for Medical Research Ethics (approval number 2016/1165). Prior to data collection, a written information letter was sent to all students and to parents of those ≤15 years, underscoring that participation was voluntary and anonymous, that participants were free to withdraw from the study, and that the collected information was treated with confidentiality. According to research ethical guidelines, written consent was required from adolescents and their parents when adolescents were ≤15 years. Adolescents ≥16 years gave consent by answering the questionnaire. Questionnaire administration was completed with help from teachers in whole class groups during one regular school session (of the teachers' choice) of 45 min, in 2016.

#### *2.3. Measures*

Self-rated health (SRH) was assessed by one item, "How is your health now?" The response options were: 1–'bad', 2–'not so good', 3–'good', 4–'very good', and 5–'extremely good'. Assessment of health using one item was previously found to be satisfactory for use in other studies on adolescents' health [13,14].

Subjective health complaints (SHC) was measured by 12 items comprising physical symptoms (e.g., stomachache, headache, pain in the back/arms/legs, and cold) and mental symptoms (e.g., bad mood, felt lonely, nervous, sad, or irritable). Participants responded on a four-point scale ranging from 1–'not bothered' to 4–'very much bothered', where higher sum scores indicated higher symptom load. Cronbach's α for the instrument was 0.86.

Sense of coherence (SOC) was assessed with the 13-item Orientation to Life Questionnaire consisting of 13-items rated on a seven-point scale; higher sum scores indicated stronger SOC. The questionnaire has been extensively validated and used cross-culturally, both in adult and adolescent samples [38,39]. In the present study, Cronbach's α was 0.82.

Adolescent stress was measured by use of the Norwegian 30-item version of the Adolescent Stress Questionnaire (ASQ-N). Each item was rated on a five-point Likert scale ranging from 1–'not at all stressful' or 'irrelevant to me' to 5–'very stressful', where a higher sum score indicated higher stress level. The scale was validated for use in Norwegian adolescents [40] and adolescents in other European countries [41–43]. Cronbach's α for the instrument in the present study was 0.94.

Mental well-being (MWB) was assessed with the 14-item version of Warwick–Edinburgh Mental Well-Being Scale (WEMWBS) [44]. The respondents were asked how they had felt about seven positively worded statements over the past two weeks. The values ranged from 1–'None of the time' to 5–'All of the time', where higher sum scores indicated higher levels of mental well-being (range 14–70). The WEMWBS was validated in the general population [44,45], clinical samples [46], and in adolescents [47–49]. Cronbach's α for the scale in the present study was 0.91.

Symptoms of depression was measured using a non-clinical depression scale appropriate for measuring non-clinical depressive attributes [3]. The scale consisted of a 15-item questionnaire measuring respondents' levels of current depressive moods. Item choice was informed by reference to commonly experienced depressive features outlined in the Diagnostic and Statistical Manual–Fourth Edition [50], and to the Zung Self Rating Depression Scale [51]. The items were rated on a 5-point Likert scale ranging from 1–'never' to 5–'always', where higher scores indicated a higher symptom load. The scale was used in previous studies in the adolescent population [26,40] Cronbach's α for the instrument in the present study was 0.94.

Socioeconomic status (SES) was measured in terms of mother's and father's education, employment status, and adolescents' perception of their family's economic situation. Mother's and father's education were assessed separately using one item: "'What is your parents' highest education?"; 1–'Primary and lower secondary school', 2–'Upper secondary school', 3–'University up to 4 years', 4–'University, more than 4 years', 5–'Don't know'. Mother's and father's employment status was assessed separately with the item "'What is your parents' employment status?"; 1–'stay at home', 2–'unemployed', 3–'part time job', 4–'full time job', 5–'other'. Adolescents' perception of family economy was assessed by one item: "How has the family economy been during the last two years?"; 1–'We have had bad economy the whole time', to 5–'We have had good economy the whole time'.

#### *2.4. Statistical Analyses*

Statistical analyses were conducted using SPSS, version 22.0 BM SPSS, Armonk, NY, USA. Descriptive statistics included frequencies, means, and standard deviations. T-tests were calculated to test sex mean differences on the scales in the study. To evaluate the strength of the sex mean differences, effect sizes were calculated following Cohen's [52] guidelines for small (0.20), medium (0.50), and large (0.80+) effect sizes. Bivariate correlations between the continuous variables of age, SES, stress, SOC, and health (MWB, depression, SRH, SHC) was tested using Pearson's product-moment correlation. Multiple linear regression analysis was applied to investigate associations between sex, age, SES, stress, SOC, and the outcome of each of SRH, SHC, MWB, and depressive symptoms. The interaction effects including combinations of sex, stress, and SOC were also tested. An assumption for conducting linear regression analysis is to have continuous variables. As stated by Wu and Leung [53], Likert scales are often treated as interval scales when included in regression analyses, when strictly speaking, it is an ordinal scale. Meanwhile, a study by Tacoby [54] also showed that the decisions used in measurement levels depended on the researcher's interpretation of the differences among the observational categories into which the empirical objects are divided. When considering the dependent and independent variables of stress and SOC for use in the present study, the assumption of continuous variables was met as the variables were constructed as sum scores. The SES variables including mother's and father's education level and employment status were originally scaled at the ordinal level. In the analyses, the variables were therefore constructed as summed scores representing parents' education and parents' employment status. In the survey, the values 'I don't know' and 'other' were included in the assessment of SES variables to ensure valid responses from the participants. In the regression analyses, these values were excluded, due to the assumption of including only continuous variables. Model assumptions for linear regression analysis were tested, and no indications of multicollinearity (VIF < 0.10 and tolerance > 0.02, correlations < 0.80) were found. The assumptions of linearity, homoscedasticity, and independent residuals were also met, where the Durbin Watson test were close to 2 for all models and the residuals were normally distributed through an inspection of the scatterplot [52]. The independent variables were included in the following order: (1) sex and age; (2) SES, (3) stress; (4) SOC; (5) sex × stress, and sex × SOC, and SOC × stress. The last step of the four regression models is presented in the results section; statistical significance was set to *p* ≤ 0.05.

#### **3. Results**

#### *3.1. Mean Scores and Correlations of the Included Scales*

The distribution of sex, age, and socio-economic status (SES) is presented in Table 1. When looking at sex, 580 (47%) were girls and 644 (52.2%) were boys; 9 did not report sex. Mean age was 16.62 years (SD = 1.61 years) for the total sample; for boys it was 16.68 years (SD = 1.60 years), and for girls it was 16.55 years (SD = 1.61 years). Table 2 presents an overview of the sex mean differences on the included scales. Boys scored significantly higher on SOC, MWB, and SRH, whereas girls scored significantly higher on SHC and depressive symptoms, showing weak-to-moderate strong mean differences. The correlation analysis is displayed in Table 3. The main variables of MWB, depressive symptoms, SHC, SRH, stress, and SOC showed moderate-to-strong correlations in expected directions; the strongest correlations were between SOC, depression, and MWB. The SES variables moreover showed weak to moderate strong correlations with the other variables.


**Table 2.** Sex mean differences on stress, sense of coherence, mental wellbeing, symptoms of depression, self-rated health, and subjective health complaints.

Note. \* *p* ≤ 0.05; \*\* *p* ≤ 0.01; \*\*\* *p* ≤ 0.001.

**Table 3.** Correlations between the study variables.


Note. \* *p* ≤ 0.05; \*\* *p* ≤ 0.01.

### *3.2. Regression Analyses for Variables Associated with Mental Wellbeing (MWB) and Depressive Symptoms*

Table 4 presents the results of the multiple linear regression analyses investigating the associations between sex, age, SES, stress, SOC, and the dependent variables depressive symptoms and MWB. When looking at the two models, sex was significantly related with depressive symptoms, where girls reported higher scores than boys; no significant sex differences were found on MWB. Age showed a non-significant association with MWB and a weak positive and significant association with depressive symptoms, indicating that adolescents seem to have a stable level of MWB and a weak increase in symptoms of depression across age groups. Of the SES variables, perception of stronger family economy showed a significant positive and weak association with MWB. Parents' employment status also showed a significant and positive association with depressive symptoms. Stress was significantly positively associated with depressive symptoms (22% explained variance), but not with MWB, after being controlled for the other variables. A strong positive relation was found between SOC and MWB (20% explained variance), whereas a significant strong and inverse relation was found between SOC and depressive symptoms (24% explained variance), controlled for the other variables. Significant interaction effects were found between sex × stress on MWB, and of sex × SOC on depressive symptoms, where the associations were strongest for girls. A significant interaction effect was also found between stress × SOC on depression, indicating that the strength of the relation between stress and depressive symptoms depended on the level of SOC. The total explained variance in the two regression models was 41% in the model with MWB and 68% in the model with depressive symptoms.


**Table 4.** Summary of the hierarchical regression analysis for variables associated with mental wellbeing and depressive symptoms.

Note. \* *<sup>p</sup>* ≤ 0.05; \*\* *<sup>p</sup>* ≤ 0.01; \*\*\* *<sup>p</sup>* ≤ 0.001. Sex: value 0—girls; value 1—boys. Cases deleted listwise. Adjusted R2 <sup>=</sup> 0.41 for model with mental wellbeing and R<sup>2</sup> = 0.68 for model with depression.
