Structural Signatures of Gender Norms: Cross-National Predictability of Attitudes Justifying Violence Against Women

Violence against women is sustained not only by individual behavior but also by social norms that legitimize coercion and control. While attitudes justifying intimate-partner violence have been extensively documented in large-scale household surveys, they are rarely analyzed as structured, predictable population-level phenomena. Here, we model the continuous prevalence of violence-justifying attitudes across 70 countries and demographic subgroups using country-resolved supervised machine learning with strict out-of-sample evaluation. Drawing on harmonized estimates derived from the Demographic and Health Surveys, we quantify how much cross-subgroup variation in normative acceptance is explainable from survey structure alone. By comparing full models that incorporate attitudinal scenario framing with demographics-only baselines, we show that high predictability arises from fundamentally different sources across countries: in some contexts, demographic stratification--particularly education--structures normative acceptance, whereas in others, conditional justification narratives dominate. Integrating independent country-level indicators of gender inequality, human development, and democratic quality reveals that violence-justifying norms are most predictable in structurally polarized settings rather than within a single cultural regime. Together, these findings demonstrate that normative acceptance of violence is not uniformly diffuse but can form coherent, structurally embedded patterns. This cross-scale framework provides a quantitative basis for identifying where prevention strategies may benefit most from demographic targeting versus direct challenges to context-specific justifications of violence. Significance statementNormative acceptance of intimate-partner violence is a measurable societal risk factor, yet it is rarely analyzed as a structured population-level phenomenon. Most quantitative studies remain descriptive, and machinelearning analyses using large-scale household surveys typically focus on individual-level classification of victimization or vulnerability. Here, we model the continuous prevalence of violence-justifying attitudes across 70 countries and demographic subgroups using country-resolved supervised regression with rigorous out-of-sample evaluation. By contrasting demographics-only models with those incorporating attitudinal scenario framing, we show that cross-national differences in predictability arise from distinct sources--demographic stratification in some contexts and conditional justification narratives in others. Linking these patterns to independent indicators of gender inequality, human development, and democratic quality reveals that highly structured norms emerge in structurally polarized settings, highlighting where targeted prevention strategies are most likely to be effective.