A Note on the Sisterhood Estimator of Maternal Mortality

Abstract

This paper comments on the Graham et al. extension of the sibling survival technique of Hill and Trussel to estimate maternal mortality. The basic ideas behind the sisterhood method are described. The Graham et al. thesis is found to be confusing and misleading in that multiple reporting or restricting the sample are viewed erroneously as equally valid alternatives. The sampling frame is confusing. All sisters drawn independently from the same distribution who belong in the sample assures that the 3 sources of error cancel each other exactly. The key assumption of the sisterhood method is independence of mortality experience of adult sisters. A probability model of dying of maternal causes is established and then applied to a numerical example of illustrate the point. An important point is that there should be the possibility that all sisters could be interviewed. When the method uses census rather than survey data, the bias is dependent on sibship size and value of p. The bias for surveys depends on the size of clusters and the extent to which sisters live geographically close by. Multiple reporting is essential for the technique.