Main content



Loading wiki pages...

Wiki Version:
## Data ## - The data generated and analysed for the study can be found at [Figshare][1]. - Please cite the data as Beyens, I., Pouwels, J. L., van Driel, I. I., Keijsers, L., & Valkenburg, P. M. (2021). *Data set belonging to Beyens et al. (2021). Social media use and adolescents’ well-being: Developing a typology of person-specific effect patterns.* [Data set]. doi:10.21942/uva.14557902 ## Analyses ## The Analyses folder contains all syntax files that were used for (1) the data preparation, descriptive statistics, and graphical presentations, (2) the assumption checks, (3) the main analyses, and (4) the sensitivity analyses. 1. **Data Preparation, Descriptive Statistics, and Graphical Presentations** The data preparation, descriptive statistics, and graphical presenetations were performed using R. 2. **Assumption Checks** We tested the required assumption of stationarity, by investigating whether the mean of well-being did not systematically change over the course of the study (McNeish & Hamaker, 2019). To that end, we compared a two-level fixed effect model including day of the study as predictor of well-being with an intercept-only model (i.e., a model with well-being, but without predictors). The analyses were performed in Mplus Version 8.4. 3. **Main Analyses** We tested two-level autoregressive lag-1 models with well-being as the outcome, lag-1 well-being as the predictor, and time spent with active private (model 1), passive private (model 2), and passive public (model 3) social media as time-varying covariate, following the procedure of McNeish and Hamaker (2019). The analyses were performed using Dynamic Structural Equation Modeling in Mplus Version 8.4. 4. **Sensitivity Analyses** We conducted three sets of sensitivity analyses to examine the robustness of the results: (1) We examined each of the three models without autoregressive effects; (2) we examined an AR(1) model with all three types of social media use included as predictors; and (3) we tested each of the three AR(1) models while excluding participants who provided potentially untrustworthy responses to the ESM questions (n=8) from the analyses. The analyses were performed in Mplus Version 8.4. [1]: