Predicting Intimate Partner Violence in South Asia Using DHS Data

Intimate partner violence (IPV) is a common form of violence against women and violation of human rights. It has become an increasingly salient social problem during the Covid-19 because potential victims of IPV now have to spend more time with their perpetrators at home due to the lockdown (Evans et al, 2020). This “pandemic within a pandemic” (Evans et al, 2020) merits serious scholarly attention. For governments, development agencies and non-profit organizations that promote human rights, a key question is how to accurately identify potential victims. Better targeting has important policy implications since it can inform the allocation of scarce financial and human resources to provide social support and protection. In addition to an improvement on aid targeting, these organizations can also benefit from the research insights into what socioeconomic and demographic features distinguish victims from non-victims. These insights can be used to facilitate policy interventions that seek to address this issue along a certain dimension such as increasing women’s employment or shifting discriminatory social norms against women. In this report, I used nationally representative survey data from three South Asian countries (India, Pakistan and Nepal) to train machine learning models able to accurately identify women vulnerable to IPV and to examine socioeconomic and demographic features predictive of IPV.

Learn more about this on the Github.