Maartje Boer

CHAPTER 8 244 data were not completely missing at random (ꭓ 2 (2,564) = 3073.68, p < 0.001). To overcome potential bias that is often associated with listwise deletion of respondents when data are not missing completely at random, missing data were imputed using multiple imputation by chained equations using Stata 13.0 (Royston & White, 2011; StataCorp, 2013). Particularly, missing data were imputed based on available data on the study variables in other waves with the data in ‘wide format’ ( n = 1,419). Multiple imputation is considered to reduce potential bias related to missing data even when the percentage of missing data is very high (Madley-Dowd et al., 2019). We conducted five imputations, which were based on predictive mean matching using the five nearest observations. As such, all 1,419 respondents were retained for the analyses. Data Organization After imputation, data were restructured into ‘long format’. That is, observations reflected repeated measures (i.e., level 1, within-person level: n = 5,676), which were nested in adolescents (i.e., level 2, between-person level: n = 1,419). Subsequently, to examine SMU activities and SMU problems and their associations with life satisfaction on both levels, we computed adolescents' person-specific means of SMU activities and SMU problems based on their respective repeated measures. Also, we computed adolescents' person- specific means of upward social comparison to test whether these means explained potential individual differences in the within-person associations between SMU activities and life satisfaction. Subsequently, the repeated measures of adolescents’ SMU activities and SMU problems were centered using their computed person-specific mean (Wang & Maxwell, 2015). Due to this centering, associations on the first level denote, for example, whether changes in SNS viewing intensity relative to one’s average SNS viewing intensity were associated with changes in life satisfaction relative to one’s average life satisfaction. The continuous time-invariant predictors (i.e., average SNS viewing) and the moderator (i.e., upward social comparison) were centered using the grand mean. Associations on the second level reflect, for example, whether adolescents with higher means in SNS viewing intensity reported higher means in life satisfaction than adolescents with lower means in SNS viewing intensity.