A Machine Learning Study of Predictors of Homophobia Across the Globe.

Much of the existing research on homophobia has focused on a limited range of correlates, typically within single-country contexts. The present study employed a machine learning approach (Random Forest) to analyze more than 350 potential predictors of homophobia using data from the World Values Survey across 62 countries. The analysis identified eight broad clusters of variables, comprising 78 top predictors associated with homophobic attitudes: (1) traditional moral norms and sexual conservatism, (2) religious beliefs and practices, (3) authoritarianism and social conformity, (4) xenophobia and outgroup distrust, (5) low civic engagement, (6) traditional gender, family, and work values, (7) limited digital engagement, and (8) perceived societal threat and insecurity. Sexual-moral traditionalism, gender and family conservatism, exclusivist, ethnocentric beliefs, and being a Muslim emerged as the top predictors (i.e., they showed the strongest associations with homophobia after accounting for the other variables in the model). These findings suggest that homophobia is not an isolated attitude but is embedded within a broader ideological framework characterized by rigid social norms, group-based distrust, adherence to traditional values, and a sense of perceived threat. Identifying these clusters and ranking predictors by their predictive strength improves our understanding of the psychological and sociocultural bases of homophobia and informs priorities for future research and intervention design.