Sample Design and Cross-sectional Weights for the Brazilian PCSVDF-Mulher Study: Integrating a Refreshment Sample with an Ongoing Longitudinal Wave to Calculate IPV Prevalence

Abstract

Addressing unit nonresponse between waves of longitudinal studies using sampling design in weighting has recently incorporated a new strategy based on the availability of supplemental samples, collecting either refreshment or replacement samples on an ongoing larger sample. We implement an approach for calculating individual cross-sectional weights and apply them to the 2016 and 2017 waves of the PCSVDF-Mulher study, a large ($\approxeq 10,000$), complex, multipurpose, population-representative and interdisciplinary longitudinal household dataset in Brazil to study Intimate Partner Violence (IPV). We developed a set of weights that combines a refreshment sample collected in 2017 with the ongoing sample started in 2016. With this set of individual weights, we calculated IPV prevalence for nine capital cities in Brazil. As far as we know, this is the first attempt to calculate cross-sectional weights with supplemental samples applied to a representative sample in a developing country. Our analysis produced a set of weights that shows neglected survey design methodological improvements on IPV measurement. One of our findings pointed out that, even in well-designed longitudinal surveys, the indiscriminate use of unweighted designs to calculate IPV prevalence might inadvertently inflate values, bringing distortions with scientific implications. Also, by not incorporating weights, the analysis neglects variance reduction gains.