Does the selective attrition of a panel survey of older people affect the multivariate estimations of subjective well-being?

The increased population aging has resulted in a growing need for longitudinal studies about the quality of life among older people. Nevertheless, the results of these investigations could be biased because more disadvantaged people leave the original sample. The purpose of this study is to examine how the selective attrition observed in a panel survey affect multivariate models of subjective well-being (SWB). The question is if we could do reliable longitudinal investigations concerning the predictors of SWB in old age.

This paper examines attrition in a panel of older people in Chile. Attrition was evaluated in the variables that affect elderly SWB. Probit models were fitted to compare dropouts with nondropouts. Then, multivariate probit models were estimated on satisfaction and depressive symptoms, comparing dropouts and nondropouts. Finally, we compared weighted and unweighted multivariate probit models on SWB.

The attrition rate in 2 years was 38.8%, including deaths and 32.9%, excluding them. Survey dropouts had lower satisfaction but not higher depressive symptoms. Among SWB predictors, people without a partner and with lower self-efficacy abandoned more the study. When applying the Becketti, Gould, Lillard, and Welch test, the probit coefficients of the predictor variables on SWB outcome variables were similar for dropouts and nondropouts. Finally, the comparison of multivariate models on SWB with weighting methods did not find substantial differences in the explanatory coefficients.

Although some predictors of attrition were associated with SWB, attrition did not produce biased estimates in multivariate models of life satisfaction life or depressive symptoms in old age.