Violence against women in Colombia: A Bayesian analysis approach to a wicked policy problem

Violence Against Women (VAW) is commonly understood as a multidimensional and structurally embedded phenomenon, entailing uncertainty not only in its occurrence but also in the collection and processing of the data that inform policy. Yet, public decision-making mechanisms remain–across most social issues–largely confined to descriptive statistics and general indicators. It is, therefore, difficult to translate VAW into data-driven policy. This article addresses these issues by proposing a Bayesian analysis that copes with complexity while supporting informed public decisions. Drawing on the existing normative Colombian framework, which defines VAW as a complex phenomenon, we aim to show that Bayesian models can be effectively used to strengthen informed public decision-making while accounting for complexity and quantifying the uncertainty of the estimates. To do so, we develop and implement a Bayesian multinomial logistic model that simultaneously examines multiple typologies of VAW using the Colombian national administrative dataset of non-lethal injury records (2015–2024). We complement it with a Bayesian ANOVA that compares the magnitude of the effects of each explanatory dimension. Both are estimated by Markov chain Monte Carlo (MCMC). The results reveal differentiated profiles: domestic violence is strongly associated with partners, ex-partners and family members as perpetrators, sexual violence with girls and adolescents and the household setting, and sociopolitical violence with perpetrators outside the intimate sphere. The model discriminates well between typologies on held-out data (area under the ROC curve, AUC, 0.854–0.945), and the Bayesian ANOVA identifies the type of perpetrator and the life-cycle stage as the dimensions whose levels differ most in their effects. Future research should incorporate the spatial and temporal dynamics of VAW and the under-reporting in administrative records. We show that this approach provides a coherent analytical framework to support more informed and context-sensitive policy decisions in the presence of complexity and uncertainty.