Appropriately Assessing the Effectiveness of Pregnancy Care in Reducing Infant Mortality: Evidence from China, India and South Africa

Sarah A. Burgard, University of Michigan

Statistical efforts to assess the effectiveness of medical pregnancy care for reducing infant mortality have resulted in mixed findings. We focus on the way that adverse or positive selection can obfuscate the true association between medical pregnancy care use and infant mortality risk. We focus on identifying these biases and exploring the relative effectiveness of different statistical methodologies in reducing the bias. Using data on recent births from China, India and South Africa in the 1990s, we estimate and compare mother-level random effects models, mother-level fixed effects models, joint mother-level random effects models (the multilevel multiprocess model) and propensity score matching. Selection bias is a substantive issue without a single best statistical solution; using the specific example of medical pregnancy care and infant mortality, we compare the performance of these statistical corrections across very different societal contexts and examine the extent to which pregnancy care effectively promotes infant survival.