| Popis: |
Background: Lack of understanding of factors that contribute to an individual woman's risk of partner \ud violence as well as factors that collectively account for the distribution of violence across settings, \ud continues to compromise efforts to design effective prevention programs. Likewise, key \ud methodological questions remain unanswered, most notably how best to conceptualize, capture, \ud and measure partner violence for the purposes of research. This thesis attempts to bridge these \ud gaps by analyzing the risk and protective factors of partner violence across a variety of low and \ud middle-income settings, with an emphasis on Brazil and Peru. \ud Methods: The analysis herein draws on data from the WHO Multi-Country Study of Domestic \ud Violence and Women's Health, a population-based survey that interviewed over 24,000 \ud reproductive age women, in 15 sites about their experiences of violence. The thesis examines the \ud patterning of partner violence in Brazil and Peru and explores the relative utility of using Latent Class \ud analysis (LCA) compared with traditional WHO case definitions, to identify and classify cases of \ud partner violence. It then uses generalized estimating equations to develop an explanatory model of \ud the factors that best predict an individual woman's risk of experiencing severe partner violence, as \ud identified by LCA. Later chapters present two ecological analyses: one that identifies the cluster- \ud level factors in Brazil and Peru that emerge as most predictive of cluster-level prevalences of \ud domestic violence; and a second analysis that uses the full WHO data set, 18 Demographic and \ud Health Surveys, and a variety of United Nations and independent data bases to test various theories \ud on how macro-level factors work to influence a country's overall level of partner violence. \ud Results: LCA categorizes cases of partner violence differently than the WHO case definition, \ud although both tend to identify similar risk factors. The WHO approach, however, seriously \ud underestimates the effect size for cases of serious violence. Without further research it remains \ud unclear whether the categories identified through LCA represent fundamentally different "types" of \ud partner violence as suggested by some research in high income countries, or merely differential \ud groupings by serverity. \ud At an individual level, partner-related factors emerge as the most predictive of a woman's lifetime \ud risk of partner violence, including exposure to violence as a child, level of controlling behavior, \ud frequency of drunkenness, history of fights with other men and having outside sexual partners. \ud Marital conflict, having more than two children, living together versus being married, not completing \ud secondary school, and poor communication between the couple are also strongly associated with \ud partner violence in both Brazil and Peru. \ud At a cluster level, the proportion of women completing secondary school, norms around male \ud dominance, and the proportion of households in which a partner routinely comes home drunk are \ud among the strongest variables predicting the cluster-level mean of partner violence. At a macro \ud level, a range of variables related to women's status, gender inequality, social norms and overall \ud level of socio-economic development predict a country's prevalence of partner violence. in \ud multivariate analysis, norms related to the acceptability of wife beating and male control of female \ud behavior, as well as women's access to formal wage employment appear the most strongly linked to \ud the distribution of past year partner violence. A country's level of male drinking or male binge \ud 2 drinking does not predict levels of abuse, illustrating that the factors that predict individual level risk \ud can be different from those that predict population-level risk. \ud Conclusion: The next generation of research should focus on longitudinal and mixed method studies \ud to help clarify the temporal associations among variables and identify how and why certain factors \ud emerge as markers for risk. |
