Department of Statistics, Iowa State University
Semi-Parametric Quantile Regression Imputation for Missing Response and Covariates Subject to NMAR Nonresponse
Development of imputation procedures for data that has extreme values or nonlinear relationships, and has both of the response and the covariate subject to missing, is a challenge task in surveys. Further, allowing for not missing at random (NMAR) nonresponse, where the response probability depends functionally on the value to be imputed, adds a significant complexity. In this project, we develop an imputation method based on semi-parametric quantile regression to deal with those situations. We validate our procedure and a linearization based variance estimator through simulation. An application to impute veterinary expenditures using data from a 2011 survey of pet owners will be presented.
Refreshments at 3:45pm in Snedecor 2101.