Context or composition? Spatial Bayesian analysis of sexual violence against adolescent girls and young women in Namibia

There is considerable spatial variation in sexual violence (SV) against adolescent girls and young women (AGYW) within and across countries; however, the reasons for this remain poorly understood. This study investigated the relative contributions of compositional (individual- and household-level) and contextual (community-level) factors to spatial variations in the risk of SV against AGYW in Namibia. Understanding these dynamics is vital for designing effective prevention strategies. We analysed data from the 2019 Namibia Violence Against Children and Youth Survey, comprising 4211 AGYW aged 13-24 years, across all 14 regions. Bayesian hierarchical logistic regression was employed to model the risk of SV. Spatially structured and unstructured random effects were included, and variance decomposition techniques were applied to quantify the contributions of contextual and compositional influences. Significant spatial clustering of SV was observed. Most of the variance in SV risk was attributable to compositional factors, including perceived community safety (OR = 2.07; 95% CrI: 1.58, 2.72), romantic relationships (OR = 1.39; 95% CrI: 1.11, 1.74), and witnessing domestic violence (OR = 1.37; 95% CrI: 1.10, 1.71). None of the contextual factors examined were statistically significant. However, residual regional heterogeneity suggests possible unmeasured contextual influence. These findings highlight the importance of integrating spatial and multilevel perspectives into violence research. Given that the risk factors were primarily compositional, individual- and household-level interventions may be effective across settings. However, persistent spatial clustering suggests that certain geographic areas may require intensified interventions or additional resources and investigation of potential unmeasured contextual factors.