Procedure and Participants
The present study used data from two school-based cross-sectional surveys of adolescents in Oslo, Norway in 2015 and 2018 (Young in Oslo 2015 and Young in Oslo 2018). All junior and senior high schools in Oslo were asked to participate in the surveys. Except for schools for students with special needs or difficulties with the Norwegian language and a few private senior high schools, all Oslo schools accepted the invitation. Students at the participating junior or senior high schools were invited to complete an electronic questionnaire in class, containing questions about their social lives, health, leisure activities, drug use, and misbehavior. In the 2015 survey, 23,381 students participated, yielding a response rate of 79%. The response rates at junior and senior high schools were 86% and 72%, respectively. In 2018, 25,287 students participated, with an overall response rate of 74%. Response rates were 83% and 65% at junior and senior high schools, respectively. As the surveys sampled almost a complete population of students attending high school in Oslo, only adolescents that did not attend high school or were absent from school at the time of the survey did not receive an invitation to participate. In total, approximately two out of three adolescents in the age group 13–18 residing in Oslo participated in the surveys. Students consented to participation by filling out the survey; the parents of students younger than age 18 were given the option to decline their children’s participation. The Norwegian Center for Research Data approved all ethical aspects of the senior high school survey, and the survey was conducted anonymously for the students at junior high schools.
Physical fighting was assessed in 2015 by two items from an instrument measuring the frequency of different conduct problems: “How many times have you done any of the following things over the past year (the past 12 months)?” with the items “have been in a fight (without weapons)” and “have been in a fight where you used a weapon (e.g., a knife).” In the 2018 survey, one item assessed participation in physical fighting: “have been in a fight.” At both time points, response options were never (0), once (1), 2–5 times (2), 6–10 times (3), and 11 times or more (4). To generate comparable variables for the two time points, a single variable was computed for 2015, retaining the maximum score for the two items measuring physical fighting. For all regression analyses, physical fighting was dichotomized at both time points into no fights versus at least one fight. All other study measures were assessed identically in the two surveys.
Five instruments assessed activities and situations with varying degree of adult supervision. First, parental supervision (Olweus 1989) was measured using three items on parents’ knowledge of their children’s social life: “My parents usually know where I am, and who I’m with, in my free time,” “My parents know most of the friends I hang out with in my free time,” and “My parents know my friends’ parents.” The response options were not true at all (1), not very true (2), quite true (3), and very true (4). Mean scores were computed, ranging from 1 to 4 (α = 0.74). Second, a mean score was generated by averaging six items measuring participation in the following organized leisure activities in the previous month: “sports club,” “youth club,” “religious organization,” “band, choir, orchestra,” “cultural school/music school,” and “other organization, team, association.” The response options were never (0), 1–2 times (1), 3–4 times (2), and 5 times or more (3), returning a variable with a range from 0 to 3. Third, to assess the amount of time spent at home, the respondents’ indicated in a single item how many times in the previous week they had “been at home the whole evening,” with response options never (0), once (1), 2–5 times (2), and 6 times or more (3). Fourth, a single item with the same response options assessed how many times in the previous week the respondents had “spent the majority of the evening out with friends.” Finally, school truancy in the last 12 months was measured by a single question indicating the frequency of truancy, with response options never (0), once (1), 2–5 times (2), 6–10 times (3), and 11 times or more (4).
Digital media use
A single item was used to assess how much time the respondents normally used outside of school on “activities in front of a screen (TV, computer, tablet, smartphone),” with response options no time (0), less than 1 h (1), 1–2 h (2), 2–3 h (3), 3–4 h (4), 4–6 h (5), and more than 6 h (6). A second single item assessed how much time the respondents spent daily on “social media (e.g., Facebook, Instagram, etc.),” with response options no time (0), under 30 min (1), 30 min to 1 h (2), 1–2 h (3), 2–3 h (4), and more than 3 h (5).
Alcohol intoxication and cannabis use were assessed by two single items from an instrument measuring the frequency of alcohol intoxication and illicit drug use in the previous 12 months, with response options never (0), once (1), 2–5 times (2), 6–10 times (3), and 11 times or more (4).
Years of schooling
A single item measured the respondents’ years of schooling (range 8–13).
The respondents’ socioeconomic background was measured by a composite score, averaging the score of three variables ranging from 0 to 3: (a) the number of parents having a university degree, (b) the number of books in the home of the respondents, and (c) the average score on the four-item Family Affluence Scale II (Currie et al. 2008). The instrument has been presented in detail in previous publications (Pedersen et al. 2018).
Migration background was assessed using a single item separating those with two parents born outside of Norway from the remaining participants.
School grades in the subjects written Norwegian, English, and mathematics were assessed, and a mean score was computed (range 1–6).
The participants’ gender was assessed.
Changes in the prevalence of physical fighting from 2015 to 2018 were analyzed by means of cross tabulations and χ2 tests. Next, several analyses were conducted to investigate whether co-occurring changes in leisure time activities contributed to statistically account for the change in physical fighting. To be able to account for changes in physical fighting, the included variables had to fulfill three criteria: (1) changes had correspond to changes in physical fighting, (2) the variable had to correlate with physical fighting, and (3) the variable had to show a significant indirect effect in mediation analyses (Frøyland and von Soest 2018). To account for the possibility of gender-specific associations between physical fighting and the different leisure time activities, moderation analyses including a dummy variable for survey year, each of the leisure time activities, gender, and interaction terms between gender and the leisure time activities were conducted. The identification of significant interactions would imply a need for conducting gender separated analyses. Second, a series of linear regression analyses identified changes over time in the included leisure time activities. Third, all variables were correlated with physical fighting. Fourth, all leisure time activities that correlated with physical fighting and showed appropriate co-occurring changes were included one by one in separate probit regression analyses, together with a dummy variable for survey year. Whether the change in the leisure activity was significantly related to the change in physical fighting was assessed by means of mediation analyses under the counterfactual framework (VanderWeele 2015), which is the recommended framework for conducting mediation analyses with binary outcomes. Potential outcome probabilities were calculated based on the parameter estimates from the probit analyses, and results were presented as risk differences of these probabilities for the total effect (TE), the natural direct effect (NDE), and the natural indirect effect (NIE). The TE shows the increase in risk for physical fighting between the counterfactual outcomes of letting the total sample be from 2018 and allowing the mediator to change to the value from 2018 compared to letting the total sample be from 2015 and keeping the mediator value to the level from 2015. The NDE shows the increase in risk for physical fighting between the counterfactual outcomes of letting the total sample be from 2018, but keeping the mediator as it was in 2015, compared to letting the total sample be from 2015 and keeping the mediator as it was in 2015. The NIE shows the increase in risk for physical fighting between the counterfactual outcomes of letting the total sample be from 2018 and allowing the mediator to change to the value from 2018 compared to letting the total sample be from 2018 and keeping the mediator value to the level from 2015. Standard errors were estimated using the Delta method. The Delta method generally returns valid estimates in large samples (Muthén et al. 2016). Finally, all leisure time variables with a significant NIE in the bivariate analyses were included in multivariate analyses where the combined NIE of change in all the leisure time variables was calculated. The analyses were conducted based on recommended methods for analyzing the combined impact of multiple mediators on a binary outcome using structural equation models (Nguyen et al. 2016). To account for possible confounding, years of schooling, socioeconomic background, migration background, and school grades were included as control variables in all analyses. Mediation analyses were additionally conducted using the product-of-coefficients method (Hayes 2018), tables are included in the appendices.
The analyses were conducted using Mplus Version 8.3. Missing data were handled by the full information maximum likelihood procedure, thereby providing missing data routines that are considered to be state of the art (Schafer and Graham 2002). All analyses were also conducted using listwise deletion, yielding similar results. Due to the large sample size, only findings with p-values less than 0.01 were considered statistically significant.