Attrition Bias in Longitudinal Survey: A Validation Study with Interviewer's Record of Respondent Mobility
This paper aims to examine respondent's mobility-related non-response in the
longitudinal survey. We use the interviewer's record of respondent mobility, which can
be observed even if the respondent does not participate in the respective wave, as a
source of validation data. Using the Keio Household Panel Survey (KHPS) 2004–2007
as a primary dataset, household mobility equations are estimated for selected
subsample of non-attritors by two competing methods—an inverse probability weighted
(IPW) estimator and a sample selection (SS) estimator. These two estimators are
compared with a probit estimates using complete sample including both attritor and
non-attritor. It is found that SS generally outperforms IPW in terms of coefficient
estimates, suggesting that the mobility-related non-responses in the KHPS are
non-ignorable. However, the results of Hausman test cannot find any significant bias for
either IPW or SS estimator.
