Detecting Bias in DHS Infant and Child Mortality Estimates

Thomas W. Pullum, University of Texas at Austin
Jeremiah M Sullivan, Macro International Inc.

This paper is an assessment of the quality of the DHS estimates of infant and child mortality, particularly with respect to possible biases due to misrecording of children’s birthdates. In some surveys, interviewers appear to have reduced their workloads by recording recent births earlier than they actually occurred, thus avoiding the health questions asked for recent births. This tendency is greater for deceased than for surviving children. Thus, standard DHS estimates for the period immediately prior to the survey are negatively biased, those for the penultimate period are positively biased, and time trends are distorted. Inconsistencies in comparisons of successive surveys in the same country, and apparent recent rapid declines in infant mortality rates in some surveys, are probably due to such displacement. The paper develops procedures to identify such differential displacement and includes adjusted estimates of levels and trends in mortality for DHS surveys that have been most affected.