Measuring Physical and Sexual Intimate Partner Violence Experiences: Comparison of Indirect and Direct Methods in...

Intimate partner violence (IPV) threatens women's health; hence, accurate measurement is critical. Indirect questioning via list experiments or item-count methods offers added privacy and may minimize underreporting. However, less is known about participants' ability to answer item counts and, critically, how indirect assessment compares with direct methods. To advance these questions, our list experiment adaptation entails indirect lists (with and without the IPV question for item-count comparison) and direct questioning, for sexual and physical IPV experience separately. By not constraining item-count responses, we assess list experiment numerical comprehension. The study aims to examine (a) numerical comprehension for indirect assessment and (b) the difference in past-year IPV prevalence estimation generated via indirect versus direct measures. A 2024 cross-sectional survey with partnered women was conducted across four Kenyan counties (n = 3,323). Numerical comprehension was indicated by a non-plausible difference in the five-item and four-item lists (>1 or <0). Among comprehenders, IPV prevalence was estimated using direct and indirect methods, with the difference in prevalence estimation assessed overall and by sociodemographic sub-groups. List experiment comprehension was compromised for 12.8% and 10.1% of women, for physical and sexual IPV, respectively. The prevalence of past-year physical and/or sexual IPV was 22.2% as assessed by direct and 36.7% as assessed by indirect methods, resulting in a difference of 14.5 percentage points. Indirect methods produced higher IPV prevalence estimates across all sociodemographic sub-groups, suggesting that direct methods may risk underestimation. Results highlight numeracy as a limitation in list experiments, particularly for vulnerable sub-populations. List experiments can also be cumbersome and limited in their breadth. Ultimately, direct IPV estimation remains a best practice for population surveys, acknowledging risk for underestimation. Indirect methods can validate direct IPV estimation on a limited set of experiences and add value where privacy issues constrain direct methods.