Peterman et al. Respond

Abstract

We thank Linos and Kawachi for drawing attention to the importance of community-level and other contextual factors in predicting violence against women. The variation by province in our study results do suggest upstream social determinants of violence,1 including but not limited to exposure to conflict, level of economic development, and access to services. Unfortunately, the Demographic and Health Survey (DHS) data did not collect cluster-level data to match community-level characteristics to our sample, and therefore we are unable to operationalize indicators beyond a regional fixed-effect. However, we would add a word of caution to the use of attitudes and social norms around spousal violence as a predictor of an individual woman's experience of violence because of the simultaneity bias in modeling this relationship. While household-level attitudes toward spousal violence may determine an individual's experience of violence or an intimate partner's perpetration of violence,2,3 experience of violence will most certainly also have an effect on attitudes toward spousal violence.4,5 In this instance, both causation and reverse causation occur simultaneously, biasing the estimated coefficients. As Hindin et al. note, it is particularly difficult to sort out the causal ordering of these two outcomes using cross-sectional surveys.6 The use of community-level averages (or non-self community averages) of attitudes on violence may lessen the simultaneity bias, but it is still present. Furthermore, at least two of the articles cited by Linos and Kawachi model violence against women as a function of household-level attitudes7,8 and therefore suffer from the aforementioned bias—as do four of the studies we cited.2–5 Therefore, although we support Linos and Kawachi's call to study community-level factors associated with violence against women, we believe further research must be conducted to account for simultaneity bias and improve upon measurement of norms versus concrete environmental influences rather than simply looking at associations, which may paint an incomplete or incorrect picture.