A model-based boosting approach to risk factors for physical intimate partner violence against women and girls in Mexico

The goal of this study was to identify and describe the extent to which a comprehensive set of risk factors from the ecological model are associated with physical intimate partner violence (IPV) victimization in Mexico. To achieve this goal, a structured additive probit model is applied to a dataset of 35,000 observations and 42 theoretical correlates from 10 data sources. Due to the model's high dimensionality, the boosting algorithm is used for estimating and simultaneously performing variable selection and model choice. The findings indicate that age at sexual initiation and marriage, sexual and professional autonomy, social connectedness, household overcrowding, housework division, women's political participation, and geographical space are associated with physical IPV. The findings provide evidence of risk factors that were previously unknown in Mexico or were solely based on theoretical grounds without empirical testing. Specifically, this paper makes three key contributions. First, by examining the individual and relationship levels, it was possible to identify high-risk population subgroups that are often overlooked, such as women who experienced sexual initiation during childhood and women living in overcrowded families. Second, the inclusion of community factors enabled the identification of the importance of promoting women's political participation. Finally, the introduction of several emerging indicators allowed to examine the experiences faced by women in various aspects of life, such as decision-making power, social networks, and the division of housework.