Understanding SGM Well-Being: Do unique determinants outperform traditional measures?

Sexual and gender minorities (SGMs) generally report lower levels of subjective well-being. The frequency of minority stress and feelings of social safety are thought to explain differences in SGM well-being (Diamond & Alley, 2022; Meyer, 2003). However, two core issues limit the advancement and generalizability of research on SGM well-being. First, theories of well-being emphasize how individuals uniquely construct joy in their lives. Yet, measures of well-being determinants capture shared determinants across people (a nomothetic approach) at the expense of unique determinants (a person-specific approach) (Nissen & Beck, 2025; van Zyl & Dik, 2025). An overemphasis on nomothetic approaches masks the unique construction of determinants that inform individual experiences of well-being. Second, person-specificity is compatible with a social justice orientation to SGM well-being because it provides a framework and toolbox for recentering the voices of marginalized communities, like SGMs, in understanding what factors are most important to their well-being. To address these issues, the present study compares the performance of various machine learning models and feature sets to see which work best for whom. Such an approach provides a more nuanced understanding of (1) the unique indicators of well-being for a person and (2) allows for commonalities to be assessed across SGMs that better highlight core features of SGM well-being as a collective. We will use data from a sample of SGM college students who responded to items of subjective well-being, unique queer joy indicators, minority stress, and social safety (n = 158) via experience sampling method (ESM) surveys over the course of three weeks. We will specifically use three machine learning models (elastic net, BISCWIT, random forest) and several feature sets, which include between-person measures of minority stress (i.e. discrimination and microaggressions) and social safety, as well as unique, participant-generated items of queer joy to predict future (t + 1) well-being. Separate models will be run for each of our subjective well-being indicators (positive affect, negative affect, life satisfaction).