Social media as a source for understanding differences in wellbeing

This dissertation investigates the complex interplay between social media use (SMU) and mental health and wellbeing (WB), with a particular focus on the genetic, psychological, and computational underpinnings of this relationship. It addresses three core research questions: (1) Is social media use associated with wellbeing and mental health outcomes? (2) What are the underlying genetic and environmental sources contributing to this overlap? (3) Can social media language be leveraged to validly measure wellbeing? To explore these questions, I employed methods from diverse fields, including behavioural genetics, psychometrics, network science, and computational social media text mining (SMTM). This integrative methodology allowed for a nuanced understanding of individual differences in wellbeing and the mechanisms through which social media behaviours intersect with psychological outcomes. In this thesis, Chapter 1 serves as a introduction to the topics covered and lays out the main motivations to conduct the work presented in this thesis. In Chapter 2, I used multivariate twin models with data from the Netherlands Twin Register (NTR) to assess the heritability of SMU and its genetic correlation with wellbeing and anxious-depressive symptoms. Findings revealed that SMU and WB share a small but significant genetic overlap, with SMU behaviours showing moderate-to-high heritability (32%–72%). Notably, time spent on social media exhibited the highest heritability, and the (small) observed associations between SMU and WB were largely driven by shared genetic influences if a statistically significant association was observed. Chapter 3 presents a systematic review and meta-analysis of SMTM methods for assessing wellbeing, synthesizing evidence from 45 studies. The results demonstrated moderate validity for SMTM-derived wellbeing estimates (meta-analytic r = 0.33 for individuals, r = 0.54 for regional measures), while also highlighting methodological limitations—especially the dominance of English-language studies and the underuse of eudaimonic wellbeing constructs. Chapter 4 addressed some of these limitations by comparing SMTM-based and survey-based wellbeing structures using network analysis. Using both hedonic and eudaimonic measures, and applying Latent Dirichlet Allocation (LDA) to language data, I found structurally similar network dimensions of wellbeing across survey and language data, indicating that SMTM can meaningfully approximate psychological constructs of wellbeing. Chapter 5 extended SMTM to Dutch-language data, integrating survey responses, Facebook likes, and social media language from the NTR. While associations between language features and wellbeing replicated patterns seen in English-language studies, predictive power of machine learning models remained modest (average r = 0.11), largely due to small sample sizes. Nevertheless, language use and Facebook activity provided valuable insights, especially when combined with demographic and behavioural features. Chapter 6, summarises the dissertation's findings, discuss their implications, provide recommendations for researchers, and outline questions for future research. Appendix 1 and Appendix 2 offer a guideline for social media data collection in the Netherlands Twin Register and include some insightful statistics as well as materials used in these data collection efforts. This thesis contributes to wellbeing science by showing that SMU and WB are influenced by overlapping genetic factors, and that SMTM—especially when incorporating both hedonic and eudaimonic dimensions—can offer a valid, complementary method for assessing wellbeing. It also highlights critical methodological and ethical challenges, including data access limitations and demographic biases in social media data. Future directions include developing multilingual, inclusive SMTM tools, improving interpretability of data-driven methods, and integrating diverse data types across platforms. Ultimately, this dissertation supports the use of multimethod approaches—combining traditional surveys with computational analyses of digital traces—to build a more comprehensive and dynamic understanding of wellbeing in the digital age.