Sample attrition in SHARE and SILC surveys and its effect on cross-sectional estimates of population health

Magdalena Muszynska-Spielauer , Wittgenstein Centre (IIASA, OeAW, Univ. Vienna)
Martin Spielauer, Austrian Institute of Economic Research (WIFO)

The Survey of Health, Ageing and Retirement in Europe (SHARE) and the European Union Statistics on Income and Living Conditions (SILC) are the most commonly used sources of cross-sectional information on the health status of the Europeans. The samples in both are predominantly longitudinal and hence subject to attrition. This paper investigates if attrition in the two surveys is related to health status and to what extent it biases cross-sectional estimates of population health. Health status is measured across the Global Activity Limitations Indicator. Population health statistics are health expectancy and average lifespan with health limitations. The cross-sectional sample in SHARE comes from wave 7 and longitudinal from all except the 3rd wave. In SILC, longitudinal samples are rotational samples of 2015-2017 and the cross-sectional sample refers to 2018. Attrition due to mortality and of the unknown character are stiudied separately. First, mortality is compared to the official statistics. Next, we estimate multinomial logistic regression models with status at the interview as dependent variable. Effect of attrition on cross-sectional population health statistics is evaluated in SILC by comparing results based on the partly attrited entire sample and the new rotational sample only. In SHARE we use an analogous, but an empirical approach, and simulate health status at wave 7 of those who attrited. In both surveys, mortality is underreported, and attrition is health-related. Respondents with limitations are more likely to have died. The effect of health on attrition differs between the age groups, countries and severity of health limitations.

See extended abstract

 Presented in Session 73. Issues in health data analysis