Inventaire d'enquêtes Demostaf

This paper considers the impact of sample-attrition through dropouts on mortality analyses, using the pioneering IFORD survey of Yaounde (Cameroon). The essential issue in the IFORD surveys is the possibility that mortality of members of the cohort may differentially select some children, with specific underlying characteristics. The paper implements a method to assess the following three distinct concerns that may arise in the analysis of the IFORD data: (a) the estimation of the relationship between the covariates and the rate of occurrence of mortality or attrition over time; (b) the study of the interrelation between processes under a specific set of conditions during intervals between rounds for those children still alive and in the survey at the beginning of each interval; and (c) the issue of whether mortality and mortality differentials are affected when attrition is ignored. This approach accounts for sample-selection bias that may have resulted in the attrition process. The analyses provide insights into the debate, which has been ongoing since the late 1970s among students of African demography, regarding the selection problem in the IFORD surveys. Based on a multinomial survival modelling and bivariate probit with sample-selection framework, the results substantiate the belief that average levels and differentials of mortality would not have been different for children who dropped out from the survey than for those who remained in the survey. This is evidence that mortality estimates are virtually unaffected if attrition is ignored.