The importance of correcting for health-related survey non-response when estimating health expectancies: Evidence from The HUNT Study

BACKGROUND: Most studies on health expectancies rely on self-reported health from surveys to measure the prevalence of disabilities or ill health in a population. At best, such studies only correct for sample selection based on a limited number of characteristics observed on the invitees. OBJECTIVE:...

Full description

Saved in:
Bibliographic Details
Main Author: Fred Schroyen
Format: Article
Language:English
Published: Max Planck Institute for Demographic Research 2024-04-01
Series:Demographic Research
Subjects:
Online Access:https://www.demographic-research.org/articles/volume/50/25
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:BACKGROUND: Most studies on health expectancies rely on self-reported health from surveys to measure the prevalence of disabilities or ill health in a population. At best, such studies only correct for sample selection based on a limited number of characteristics observed on the invitees. OBJECTIVE: Using longitudinal data from the Trøndelag Health Study (HUNT), I investigate the extent to which adjustments for a health-related sample selection affect the age profiles for the prevalence of functional impairment (FI) and the associated disability-free life expectancy (DFLE). METHODS: I estimate a probit model with sample selection under the identifying restriction that the strength of the health-related selection is of similar order to the strength of the selection on observable characteristics. I then compute the selection-adjusted FI prevalence rates and trace out the implications for DFLE using the Sullivan method. RESULTS: The analysis confirms that poor health measured at younger ages correlates with nonresponse behaviour in later waves of the survey, and that even for a conservative lower bound for the assumed degree of health-related selection, the estimated age profiles for DFLE lie systematically below the corresponding profiles when controlling only for selection on observable characteristics. CONCLUSIONS: Health related non-response downwardly biases the raw sample prevalence rates for FI obtained from survey data and contributes to overestimating the expansion in DFLE. CONTRIBUTION: I present a statistical framework for taking health-related survey non-responses into account when estimating the prevalence rate of FI. The framework can be used to gauge the sensitivity of estimated (changes in) DFLE to health-related sample selection.
ISSN:1435-9871