Study design and participants
“Ungdata” is a cross-sectional, large, national survey, designed for adolescents and conducted at the municipal level in Norway (33). “Ungdata” covers various aspects of young people’s lives, such as relationships with parents and friends, leisure activities, health issues, local environment, well-being, and school issues. The survey also includes questions about tobacco and drug use, and participation in violence, bullying and self-harm. We were granted access to the “Ungdata” from The Norwegian Social Research Institute (NOVA). NOVA is responsible for the national coordination of the project, while the regional Drug and Alcohol Competence Centres are responsible for conducting the municipal surveys.
Participants in the “Ungdata” study were high-school pupils and students (grades 8th–13th, 50% girls) from 85 different municipalities in Norway, conducted in “the Ungdata” survey in 2014 (N = 47 450) (33, 34). The response rate was 84% among the younger high school pupils (grades 8th-10th) and 66% among the older high school students (grades 11th-13th) (34). All participants filled in an online questionnaire during school hours. The questionnaire consisted of both a core part, which is identical for all municipalities, and an elective part, with questions that municipalities could choose, based on interest and need. Questions on self-harm were in the elective part of the questionnaire. Thus, only adolescents living in the 23 municipalities that had chosen to include this part of the questionnaire could respond (N=16 182). A total of 14 093 adolescents answered the questions relating to self-harm, yielding a response rate of 87%.
A public health coordinator in each municipality administered the survey, and local contacts in each school approached the adolescents together with teachers. The local contacts and teachers ensured that survey procedures were followed and that adolescents did not collaborate while responding to the survey. All of the municipal surveys were conducted anonymously. Participation was voluntarily. Parents were informed about the surveys and given the opportunity to withdraw their children from participation (35). The data were cleaned by NOVA Ungdata according to their criteria to remove non-serious responses prior to issuing the data. Details of the criteria can be found in the UngData report (33). Because the dataset was conducted without personal identifiable information, specific ethical approval was not required for this study and was waved by the Norwegian Centre for Research data (NSD). We conducted the analyses in accordance with the NSD’s data protection regulations.
Self-harm was the dependent variable. To measure self-harm, participants were asked: (1) “Have you ever tried to harm yourself?” Those who responded yes, were then asked: (2) “Have you tried to harm yourself in the past 12 months?”. We wanted specifically to study those who had been self-harming the last year. This was because those who have self-harmed in the last twelve months are most at risk in the future, and also because the time line was more aligned with the bullying measure (see below for the bullying measure). Thus, we grouped adolescents into two categories: Those who had self-harmed in the past 12 months (value 1) and those who had not self-harmed in the last 12 months (regardless of if they had harmed themselves prior to this) (value 0).
Bullying behaviour was a categorical independent variable. Bullied by peers was measured by combining two questions: “Are you sometimes teased, threaten, or frozen out by other young people in school or in your free time?” and “Are you sometimes teased, or threaten, by other young people online or on your mobile phone?” Response options for both questions were: (1) Yes, several times a week, (2) Yes, around once a week, (3) Yes, around once a fortnight, (4) Yes, once a month, (5) Almost never (6) Never. Those who responded with 1 to 4, on both, or one of the questions, were categorised as: “Bullied by peers” and valued 1, and those who responded with 5 or 6 on both questions were categorised as “Not bullied by peers” and valued 0. Bullying other peers was measured by combining the following two questions: “Do you sometimes take part in teasing, threatening or freezing out other young people at school or in your free time?” and “Do you sometimes take part in teasing, and/or threatening other young people online or by mobile phone?” Response options for both questions were: (1) Yes, several times a week, (2) Yes, around once a week, (3) Yes, around once a fortnight, (4) Yes, once a month, (5) Almost never (6) Never. Adolescents who responded with 1 to 4, on both, or one of the questions, were categorised as “Bullying other peers” and valued 1, and adolescents who responded with 5 or 6 on both questions were categorised as “Have not bullied other peers” and valued 0. We then created a new variable combining the variables “Bullied by peers” and “Bullying other peers”. First, those who were neither “Bullied by peers” nor “Bullying other peers” were coded with 0. Those who had been “Bullying other peers” but were not “Bullied by peers” were coded with 1 (“bullies”). Those who were both “Bullied by peers” and “Bullying other peers” were coded with 2 (“bully-victims”). Finally, those who were “Bullied by peers”, but not “Bullying other peers” were coded with 3 (“bullied”).
Gender was a dichotomous variable with Boys (0) and Girls (1).
Social Support of Parents and Social Support of Friends
The measure “willingness to seek social support” was based on Sarason´s social support measure (36). We used two separate questions, one referring to parents and one to friends. Participants were asked: “Imagine that you have a personal problem. You feel sad and need someone to talk to. Who would you talk to, or ask for help? A) Parents? B) Friends? Response options were: (1) Definitively, (2) Maybe and (3) No. For each support person, we combined response options (1) “Definitively” and (2) “Maybe”, with “Yes” (coded with two), while No was coded with 1.
Best friend/friendship was measured with the question: Do you have at least one friend who you trust completely and can tell absolutely anything to?
1) Yes, I am sure, 2) Yes, I think so, 3) I do not think so, and 4) I have no one I could call a friend now a days”.
Father- and Mother Education
Father/Mother education was measured by responses to the questions: a) “Does your father have university or college level education?” (1) Yes (2) No and b) “Does your mother have university or college level education?” (1) Yes (2) No.
Psychological Distress: Symptoms of Depression and Anxiety
The measures of symptoms of depression and anxiety were short versions of the scales: “Hopkins Symptom Checklist” (HSCL) (37, 38) and “Depressive Mood Inventory” (39). Earlier studies have shown that short version of HSCL has good validity (40, 41). Participants were asked: «During the past week, have you been affected by any of the following issues: 1) ”Felt that everything is a struggle”; 2) ”Had sleep problems”; 3) ”Felt unhappy, sad or depressed”; 4) ”Felt hopelessness about the future”; 5) ”Felt stiff or tense”; 6) ”Worried too much about things”; 7) ”Suddenly felt scared for no reason”; 8) ”Felt constant fear or anxiety”; 9) ”Felt exhausted or dizzy”; 10) ”Been nervous or felt uneasy”; 11) ”Been easily moved to tears” and 12) ”Tended to blame yourself for things”. Each question had four response categories: (1) Not at all, (2) Not much; (3) Quite a lot; and (4) A great deal. We then conducted a Principal components factors analysis (PCA) with oblique rotation. Based on eigenvalues and scree-plots, we found two factors: questions 1-6 (Depression) and questions 7-12 (Anxiety). A mean score index was then calculated for depression and anxiety. Cronbach´s alpha was 0.9 for each of the factors, indicating a satisfactory reliability of each measure.
Family’s Financial Situation
Family’s financial situation was based on responses to the question: “Financially, has your family been well off or badly off, over the past two years? The response categories were: (1) ”We have been well off the whole time”, (2) ”We have generally been well off”, (3) ”We have neither been well off or badly off” (4) ”We have generally been badly off” and (5) ”We have been badly off the whole time”.
School Behavioural Problems
The School behavioural problems (SBP) was measured by four items from the conceptual domain of the “School-related problem behaviour” in the “Bergen Questionnaire on Antisocial behaviour” (42, 43). This measure has been used in earlier UngData-studies (Young in Norway studies) in 1992, 2002, and 2010, and the scale was shortened down when the questionnaire was revised in 2013 (33). The measure included: “Have you done or experienced any of the following things during the past 12 months?”: ”Had a big argument with a teacher”; ”The school have contacted your parents for something bad you did”, ”Skipped school” and “Been in a fight (without weapon)”. The response options were: (1) “Never”; (2) ”Once” (3) ”2 – 5 times”; (4) ”6 – 10 times” and 5)”11 times or more”. PCA with oblique rotation was conducted for the four questions. Based on eigenvalues and scree-plots, the items loaded on one factor. A mean score was calculated for the four items. Cronbach´s alpha was 0.7, indicating a satisfactory level of reliability for school problems.
Relationship with Parents: Parental Monitoring and Parental Conflict
This measure is based on elements of the concept of parenting style (45), particularly authoritative parenting (46, 47), from the “strictness-supervision scale”, which assesses parental monitoring and limit setting. In addition, some of the items measured the concept “conflicts with or between parents” (33). The adolescents responded to the following questions: ”Below are some statements that may describe your relationship with your parents”: a) ”My parents usually know where I am, and who I am with, in my free time”; b) ”My parents know most of the friends I hang out with in my free time”, c) ”I try to hide what I do in my free time from my parents”; d)”My parents know my friends’ parents”; e) ”I often argue with my parents”, f)”The adults in my family often argue” and g) ”My parents know who I am in touch with on the internet”. Response were given on a 4-point scale from: (1) Very true; (2) Quite true; (3) Not very true, and (4) Not at all true. We conducted a PCA with oblique rotation on the seven statements. Based on eigenvalues and scree plots, two factors were extracted: “Parental monitoring” (items a, b, d and g) and “Parental conflict” (items c, e and f). A mean score index was calculated for each and Cronbach’s alpha showed a satisfactory level of reliability (0.7) for both concepts; (“Parental monitoring”, α = 0.7 and “Parental conflict”, α = 0.7) (48).
The questions measuring school well-being have previously been used in studies of young people in Norway (33), and include the following: ”Do you agree or disagree with the following statements about your situation at school:, ”I enjoy school”, ”I feel that I fit in with the students at my school”, “I often do not want to go to school”, ”My teachers care about me”, and ”I am bored at school”. Evaluation of the statements was on a 4-point scale from: (1) “Totally agree”, (2) ”Somewhat agree”, (3) ”Somewhat disagree” to (4) ”Totally disagree”. To measure how the items correlated with each other and if they fitted together, a PCA with oblique rotation of the five statements was conducted. Based on eigen-values and screeplots, only one factor was extracted. A mean index score was calculated for the five items after reversing responses in the first, second and forth items so a high score represented good school well-being. Cronbach´s alpha was 0.72, indicating a satisfactory level of reliability.
To test significant differences between those who had self-harmed during the last 12 months and those who had not self-harmed during the last 12 months, we used chi-square tests for all categorical va riables. Independent-samples t-tests were conducted to compare the continuous variables. To assess the relationship between the dichotomous dependent variable/the outcome variable: “self-harmed last 12 months”, (yes/no), and the independent variable (bulling behaviour) and the covariates (gender, depression, anxiety, family’s financial situation, mother/father education, school problems, parental conflict, parental monitoring, school well-being, parental support, friend support and friendship), we conducted logistic regression analyses. Gender, bulling behaviour, parental support, friend support and friendship were treated as categorical variables. We conducted bivariate logistic regression analyses for the dichotomous dependent variable “self-harmed last 12 months”, and the independent variable (bullying behaviour) and each covariate separately. Then we ran various logistic regression models, with the addition of the covariates in order to evaluate which factors would reduce the strength of the association between bullying behaviour and self-harm. We did this first for all potential risk factors and then for all potential protective factors. To determine whether the potential protective factors moderated the relationship between bullying behaviour and self-harm, we performed separate regression analyses, adding one interaction term at a time. All continuous variables in the interaction terms were first standardised (49). We then plotted the values of the unstandardised regression coefficients (including intercept) and means and standard deviations of the independent variables, moderators and the interaction terms in the cells indicated in the Jeremy Dawson’s excel sheet, http://www.jeremydawson.com/slopes.htm (49-51). This was to aid interpretation and to visualise the significant interaction effects.