Abstract:
|
Nonresponse weighting adjustment using propensity score (PS) is a popular tool for handling unit nonresponse. However, including all the auxiliary variables into the propensity model can lead to inefficient estimation and the consistency is not guaranteed if the dimension of the covariates is large. In this paper, a new Bayesian method using the Spike-and-Slab prior is proposed to handle the sparse propensity score estimation under the parametric model assumption on the response probability. The proposed method does not assume any model for the outcome variable and is computationally efficient. Instead of doing model selection and parameter estimation separately as in most frequentist methods, the proposed method simultaneously selects the true sparse response probability model and provides consistent parameter estimation and corresponding inference, which can be quite involved in the frequentist methods. The finite-sample performance of the proposed method is investigated in limited simulation studies, including a partially simulated real data example from the Korean Labor and Income Panel Survey.
|