Old age mortality in South Africa

Includes bibliographical references (leaves 71-74).This study estimates the mortality of the South African oldest old age population (in five year age groups from age 75 up to the open age interval 100 and above) and in the process re-estimates the numbers of people in the population at these ages at the time of the 1996 and 2001 censuses, and the 2007 Community Survey. In countries where the data on the old age population have been verified, it has been observed that the data are marred by errors in the form of age exaggeration, age digit preference, relative under/over count of the population and under-registration of deaths. These errors have been observed to have the net effect of underestimating mortality of the oldest old age groups. The current research applies the method of extinct generations to estimate indirectly the population numbers at the oldest old age groups (75 up to 100 and above) using data on reported deaths alone. Age heaping and year of birth preference in the reported deaths are assessed using ratios of the probability of death estimated from the data. Age exaggeration in the data on reported deaths is assessed using ratios of deaths compared with same ratios from a standard population. Age heaping and year of birth preference in the census/survey population is assessed using the modified Whipple's Index of age accuracy. The Generalized Growth Balance (GGB) and Synthetic Extinct Generations (SEG+delta) methods are applied to adjust for under reporting of deaths and to assess patterns of age exaggeration in the census/survey population. The difference between the estimates of the completeness of reporting of deaths from the two methods is small (less than 1 per cent) and has been observed to have little impact on the mortality estimates. Final estimates of the completeness of reporting of deaths used are those derived using the SEG+delta method. After re-estimating the population numbers and adjusting for completeness of reporting of deaths, mortality rates were then estimated. Results obtained from the method of extinct generations suggest that there is no systematic difference between the census/ survey population and the population numbers estimated from deaths except at ages 95 and above. Measures of age accuracy show that there are patterns of preferring 1910, 1914, 1918, 1920 and 1930 as the years of birth in the census/survey population and these patterns are also found in the registered deaths. The impact of these errors was investigated and the results show that preference of certain years of birth cause fluctuations in the mortality rates. Patterns observed after applying the SEG+delta method suggest that the completeness of reporting of deaths falls with age at the advanced ages (from age 90 and above) and as a result, the estimated mortality rates above this age are lower than those estimated from the United Nations Population Division (UNPD) and US Census Bureau (USCB) population projections, and Dorrington, Moultrie and Timaeus (2004). Conclusions reached are that the mortality rates for the age groups 75 to 89 derived after re-estimating the population numbers and after allowing for the fall in the completeness of reporting of deaths are lower but not significantly different from those inferred from the UNPD and USCB population projections, and estimates derived by Dorrington, Moultrie and Timaeus (2004). The research recommends mortality estimates from the UNPD since they are the closest to the estimates derived using the published census population numbers for the whole period between the nights of 9-10 October 1996 and 9-10 October 2001. However, the research produced better estimates of the oldest old age population numbers relative to the census/survey numbers