Death distribution methods for estimating adult mortality

Abstract

Abstract The General Growth Balance (GGB) and Synthetic Extinct Generations (SEG) methods have been widely used to evaluate the coverage of registered deaths in developing countries. However, relatively little is known about how the methods behave in the presence of different data errors. This paper applies the methods (both singly and in combination) using non-stable populations of known mortality to which various data distortions in a variety of combinations have been applied. Results show that the methods work very well when the only errors in the data are those for which the methods were developed. For other types of error, performance is more variable, but on average, adjusted mortality estimates using the methods are closer to the true values than the unadjusted. The methods do surprisingly well in the presence of typical patterns of age misreporting, though GGB is more sensitive to coverage errors that change with age. The Basic SEG method (that is, making no adjustments for possible change in census coverage) is very sensitive to such coverage change, but the Extended SEG method (that is, adjusting census coverage to obtain a set of completeness estimates that show no trend with age) is little affected. Fitting to the age range 5+ to 65+ is clearly preferable to fitting to 15+ to 55+. Both GGB and SEG are very sensitive to net migration, which is an Achilles heel for all of the methodologies in this paper. In populations not greatly affected by migration, our results suggest that an optimal strategy would be to apply GGB to estimate census coverage change, adjust for it and then apply SEG; in populations affected by migration, applying both GGB and SEG, fitting both to the age range 30+ to 65+, and averaging the results appears best. (ProQuest: ... denotes formulae omitted.) 1. Introduction The study of adult mortality in less developed countries is problematic due to data quality issues. Incomplete vital registration, inaccurate censuses, and misreporting of age at death or age of the living are among the problems often encountered by researchers wishing to use these data-sets (United Nations 1983, 2002; Bhat 1990). Considerable ingenuity has been shown in the development of methods to estimate adult mortality despite these data challenges. There are three broad groups of methods for evaluating data quality or otherwise estimating adult mortality: (1) death distribution methods that assess the completeness of death recording relative to census recording, (2) methods based on intercensal survival, and (3) methods that convert indicators of mortality levels based on survival of close relatives into standard life table functions. Where the necessary data exist, death distribution methods are the method of choice because they provide age-period specific estimates of mortality rates (Hill 2001). These methods compare the distribution of deaths by age with the age distribution of the living and provide the age pattern of mortality in a defined reference period. Standard methods require two population censuses (or large sample surveys) to provide age distributions of the living and the changes of such distributions over time, plus information to calculate an age pattern of deaths for the intercensal period. If the completeness of death recording relative to population recording can be estimated, and there are no other data errors, any differential in completeness can be adjusted for, and unbiased death rates and standard life table functions calculated. However, the methods require numerous assumptions about the population they are applied to and about the nature of typical data errors. Standard methods assume the population to experience no net migration. Strong simplifying assumptions are made about data errors: no age misreporting (of either population or deaths), proportionately constant omission of deaths by age (an assumption that also implies no selectivity bias in deaths that are reported) and that any change in census coverage has been proportionately constant by age. …