Measurement Structure and Regional Invariance of the Demographic and Health Survey Intimate Partner Violence Items: A Comparative Confirmatory Factor Analysis

Abstract

Intimate partner violence (IPV) is a pressing international issue affecting more than a third of women in lower- and middle-income countries (LMICs). However, accurate global assessment of IPV prevalence among women in LMICs is limited by a lack of consensus around the domains of IPV and sparse evidence on cross-country comparability. We assessed the measurement structure and regional invariance of scales used in population-based surveys to measure IPV (physical, sexual, and emotional) and controlling behaviors. Using confirmatory factor analysis (CFA), we tested unidimensional, multifactorial, hierarchical, bifactor, and bifactor S-1 models for lifetime and past-year IPV across 46 LMICs. We then assessed the invariance of the best-fitting models across countries within world regions using multiple-group CFA. Although other models also showed good fit in most countries, bifactor/bifactor S-1 models had the best fit across all countries and were invariant within most regions. Most bifactor models, especially without controlling behaviors, were primarily unidimensional; IPV can therefore be conceptualized as a single construct with nuanced facets. Researchers seeking to model IPV should consider the bifactor/bifactor S-1 model, unidimensional model, or simple summative measures incorporating physical, sexual, and emotional domains.