Propensity score adjustment under non-ignorable non-response using information from paradata
Propensity score adjustment under non-ignorable non-response using information from paradata
Date: | Friday, March 08 |
Time: | 12:00 pm -- 1:00 pm |
Place: | Snedecor 2113 |
Speaker: | Jongho Im, Department of Statistics, ISU |
Abstract:
Paradata are data about survey processes, such as call records, interviewee attitudes and audit trails. The paradata are not main questions in a survey, but they offer additional information for each observation unit that can be useful in understanding the level of non-sampling errors such as nonresponse error. For example the number of attempts to contact interviewee can be used to adjust unit non-response when the response mechanism is non-ignorable. By assuming a parametric model for the conditional response probability that directly uses the study variable y as the covariate, the model parameters can be estimated from the conditional likelihood.
In this study, we consider propensity score weighting adjustment with non-ignorable non-response when there are such paradata from a followup sample data form. The proposed method provides a consistent estimator of the model parameters specified in the response model. We take some simulation studies and a real data example from a Korean household survey of employment to demonstrate the performance of the proposed method in this setting.