Estimating the undercount of intimate partner violence using multiple systems data

Abstract

Abstract Reliable estimation of the underreported incidence of intimate partner violence (IPV) is essential to understand the true magnitude and public health burden. An integrated data system that combines multiple sources is commonly used to estimate such a hard-to-count population. Motivated by a real-life IPV dataset, we develop a novel Bayesian estimation strategy based on a trivariate Bernoulli model (TBM) under the multiple systems estimation (MSE) framework. The model accounts for the inherent dependencies commonly found between IPV data sources and provides practical interpretations. Simulation studies suggest that the proposed method outperforms existing estimators in terms of absolute error, coverage and robustness under varying dependence structures. The model is applied to analyse a multiple systems IPV dataset from a United States (U.S.) county and reveals that approximately 23% of IPV incidents are not captured by official records. These findings underscore the importance of statistical model-based estimation of the true size of IPV incidents to improve surveillance and evidence-based policy responses.