Dead or Alive? Dealing with unknown eligibility in longitudinal surveys

Abstract

Longitudinal surveys follow people over time and some of these people will die during the life of the panel. Through fieldwork effort, some deaths will be reported or known, but others will be unobserved due to sample members no longer being issued to field or having inconclusive fieldwork outcomes (such as a noncontact that is not followed by a contact at a later wave). The coverage of deaths identified among sample members has flow-on implications to nonresponse correction. Using the Household, Income and Labour Dynamics in Australia (HILDA) Survey, four methods are used to examine the extent of missing death reports. The first method matches the sample to the national death register. The second method uses life-expectancy tables to extrapolate the expected number of deaths among the sample with unknown eligibility. The third method is similar but models deaths from data internal to the survey. The fourth method models deaths as part of the attrition process of a longitudinal survey. The last three methods are compared to the first method and the implications for the construction of balanced panel weights and subsequent population inference are explored.