Possibilities to deal with unknown vital status in the Survey of Health, Ageing and Retirement in Europe (SHARE)

Longitudinal surveys aim to correctly represent the population of interest over time. In this respect, panel attrition, i.e. the systematic drop-out of sample members, is a major challenge for maintaining long-running panel surveys. A second problem might arise when some sample members die during the life of the panel. This holds in particular for panel surveys that consider (mainly) older people, because here the overall mortality rate is higher than in studies including all age groups. Distinguishing between mortality and other forms of attrition hence is crucial as the death of respondents in a longitudinal survey is a natural process that needs to be considered in order to maintain representativeness of the panel sample. If mortality is not taken into account properly, attrition analyses might overestimate the effect of systematic drop-outs for variables that are highly correlated with mortality, such as age or health of the respondents. Therefore, lacking information on the reason why a former respondent cannot be contacted anymore and thus on the vital status is a huge problem in many longitudinal studies that further increases from wave to wave. Using the Survey of Health, Ageing and Retirement in Europe (SHARE), three methods are implemented in this paper to examine the extent of missing death reports. The first method randomly assigns people with unknown vital status to death. The second method uses mortality rates form life-expectancy tables to extrapolate the expected number of deaths among the panel members with unknown vital status. The third method models deaths from data internal to the survey. The correction methods are compared to the original, uncorrected sample and the implications for analyses of died sample members as well as attrition analyses are explored.