Validation of census and survey data used to estimate adult mortality in sub-Saharan Africa – MADAS

Information on adult mortality is essential to assess the extent of socioeconomic change in sub-Saharan Africa and in particular to assess the attainment of several millennium development goals. In settings where civil registration is far from complete, data on the survival of close relatives (parents, siblings, household members) collected during censuses or surveys are used to measure mortality at adult ages, including maternal mortality. Such approaches to mortality measurement are particularly common in sub-Saharan countries.

Even though these retrospective methods present numerous advantages (ease of data collection, low costs, potential use in samples), they are however affected by large biases, due either to sampling errors, or to misreporting of a relative’s current age or a deceased relative’s age at/date of death, and misreporting of the cause(s) of a relative’s death. Whereas numerous studies have sought to address and correct sampling errors, misreporting and recall errors have garnered much less attention. Such errors have only been assessed at the aggregate level by comparing mortality estimates obtained using data on close relatives to their expected levels. While these comparisons highlight the scale of misreporting in retrospective data on close relatives, they do not allow identifying the sources of these errors. As a result, it has been impossible to correct mortality estimates obtained from these data.

The MADAS project (Mortalité des adultes en Afrique sub-saharienne - Adult mortality in sub-Saharan Africa) seeks to undertake a detailed study of the prevalence and the consequences of misreporting in retrospective data, in order to improve measurements of adult mortality in sub-Saharan Africa. The key innovation of the proposed project is the collection of retrospective information on the survival of close relatives in population observatories. These demographic surveillance sites (DSS), as they are also referred to, provide independent and validated data on the survival of a respondent’s close relative. By linking at the individual level the retrospective reports (collected in conditions similar to those in place during censuses or surveys, and using the same data collection tools) with data collected prospectively, we will be able to validate the retrospective data so commonly used by demographers to estimate adult mortality.

A pilot study of the proposed project was conducted in 2010 in the Bandafassi DSS in Senegal. It allowed testing the linking approach in the case of survey data on sibling survival, information that is commonly used to estimate maternal mortality rates in African countries. In this application, we propose to extend the study to the validation of data on household deaths in the past 12 months and on orphanhood. We also propose to extend the study to two other DSS in Senegal in order to assess the impact of context on the extent of reporting biases.

This research program ought to provide a better understanding of the misreporting biases affecting retrospective mortality data in order to correct them. It will rely on micro-simulations of kinship structures to estimate the effect of misreporting on estimates of mortality rates. It will also identify adjustment factors, and will ultimately lead to measurement innovations allowing a more accurate estimation of the level and causes of adult mortality. In particular, it will seek to develop new tools for the measurement of maternal mortality and violent deaths. Through this unique focus on explaining differences between observed and reported mortality events, it will eventually lead to recommendations aiming to improve the measurement of adult mortality in sub-Saharan countries.