Background: Since the 1950s, many indirect or semi-indirect methods have been developed to either adjust mortality estimates or generate complete life tables from mortality indices in countries lacking high quality vital registration data. These methods are underused for estimating older adult mortality.

Objective: I seek to answer the following questions: How to better estimate older adult mortality from imperfect data? Can consistent estimates be derived from indirect-based methods? If not, what could explain the possible differences?

Methods: After adjusting population and intercensal death counts for incompleteness using death distribution methods, data from the last three censuses in Burkina Faso (1985, 1996, 2006) were fitted using Singular Value Decomposition (SVD) and Brass models, and specifically the Makeham model (MKH) for extrapolation to advanced ages where large age errors were suspected. The resulting estimates were then compared in terms of age patterns and risk of death between the ages of 50 and 80.

Results: Estimates from the SVD model are higher than those from both the adjusted data and the Brass model, which are consistent, but only before age 70. Extrapolation by the MKH model reveals obvious underestimations in the adjusted data beyond age 70, but of smaller magnitude than those suggested by the SVD model. When compared with the empirical data from the Human Mortality Database (HMD), all estimates agree with the empirical data before age 70, but only the estimates from the SVD and MKH models remain consistent beyond age 70.

Conclusions: When used to infer mortality in older adults, the estimates from empirical models such as the SVD model should be taken with caution. Further refinements of the model are required to better reflect the observed mortality level at older ages.

Contribution: This study highlights issues with using empirical models indexed by child and adult mortality to infer mortality at older ages from imperfect data.