Abstract:
|
Nonresponse occurs when a sampled subject fails to complete either part or all of a survey. It can be classified as either item nonresponse, if the subject does not respond to certain survey questions, or as unit nonresponse, if the subject fails to participate in the entire survey. Nonresponse reduces sample size and study power and can cause biased estimates if significant differences exist between those who respond and those who do not. The usual approach to unit nonresponse bias detection and adjustment is to utilize weighting adjustment methods that are based off auxiliary information, or more recently, use propensity score weighting. In this study, we examine the impact of link misspecification on propensity score weighting when used in unit nonresponse adjustments through various simulations. Researchers should investigate the link that best fits their data before using propensity score weighting.
|