Maartje Boer

CHAPTER 4 112 Results Bivariate Correlations Table 4.1 shows the descriptive statistics and correlations of the individual- level and country-level variables. On the individual-level, intense and problematic SMU were correlated, with small to moderate effect size ( r = 0.269). Intense SMU was associated with lower mental and school wellbeing, although effect sizes were small ( r < 0.119). Intense SMU was associated with higher levels of friend support, with a small effect size ( r = 0.117), but not with family support. Problematic SMU was correlated with lower mental, school, and social wellbeing, with effect sizes ranging from small (friend support: r = -0.068) to moderate (psychological complaints: r = 0.290). At the country level, a higher prevalence of intense SMU was strongly associated with a higher prevalence of problematic SMU ( r = 0.476). The cost of mobile broadband and internet speed were not correlated with countries’ intense and problematic SMU prevalence. Model Selection Table 4.2 shows the model fits of models (M) following a stepwise procedure. Results showed that in all models, adding random slopes for intense and problematic SMU improved model fit in terms of AIC and/or BIC (M1-6 b and M1-6 c ), which suggests that associations between both types of SMU and all six wellbeing indicators varied across countries. For life satisfaction, only the cross-level interaction between intense SMU and country-level intense SMU prevalence further improvedmodel fit (M 1d ). The same applied to psychological complaints, although the respective cross-level interaction improved AIC, but not BIC (M 2d ). For both school wellbeing outcomes, models without any cross-level interaction showed the best model fit (M 3c and M 4c ). For both social wellbeing outcomes, the model with all four cross-level interactions showed the best model fit in terms of AIC, but not BIC (M 5g and M 6g ). For each wellbeing outcome, we selected the models with the best model fit for further interpretation. When AIC and BIC were inconclusive, we selected the models with the lowest AIC, because these models included cross-level interactions that reduced the country-variance in the investigated associations, suggesting that the respective cross-level interactions were present.