Background
Understanding variation in mortality through time and across space is critical to quantifying and monitoring population health. In settings where ongoing, routine systems to monitor mortality - e.g. vital statistics - do not exist or function poorly, there are very few options available to obtain measures of mortality that cover both time and space, particularly space. In large part due to the innovative work of William Brass - and many after him - formal demography has a suite of 'indirect estimation' methods designed to obtain reasonable estimates of key mortality indicators from very few data that can occasionally be elicited from enough people to form a comprehensive picture of mortality throughout a given place, e.g. a country.
The 'Brass Method' to estimate child mortality utilizes simple data obtained from a population of women of reproductive age; from each woman at the time of the interview:
- her age,
- how many children she has had, and
- how many of those children are still alive.
Obtaining these data in a survey or census typically requires only two additional questions, the second two relating to children ever born. This parsimony has resulted in these questions being added to whole censuses and many surveys over the past several decades. In a number of countries in Africa this is the only country-wide source of mortality information, and consequently, it is very valuable.
Traditional approaches to indirect estimation methods are deterministic in the sense that they produce specific numbers without a sense of how certain the method is that those are the the numbers most consistent with the data. Situations in which indirect estimation is necessary are inherently low-information, and the methods themselves require a set of assumptions and some additional information, and as a result, there are many ways in which a slightly different estimate could be obtained → there must be some degree of uncertainty in the estimates.
Project
OSU Sociology graduate student Yue Chu and I are exploring alternative approaches to the traditional Brass method to estimate child mortality from the three questions listed above. Yue has complete a master's degree based on this work, and we will present some preliminary results at the Population Association of America (PAA) meeting in 2021.
We stay close to the spirit of the method, but we have developed a new Bayesian estimation approach that simultaneously estimates age-specific fertility for the mothers and age-specific mortality for their children, with uncertainty in both. Initial work is based on data from the demographic and health surveys (DHS)