Online Program Home
My Program

Abstract Details

Activity Number: 260 - SPEED: Topics in Bayesian Analysis
Type: Contributed
Date/Time: Monday, July 30, 2018 : 3:05 PM to 3:50 PM
Sponsor: Section on Bayesian Statistical Science
Abstract #332995
Title: Combining Predictive Mean Matching with the Penalized Spline of Propensity Prediction Method When Performing Multiple Imputation
Author(s): Jay Xu* and Roee Gutman
Companies: and Brown University
Keywords: Bayesian Statistics; Integrated Nested Laplace Approximation; Markov chain Monte Carlo; Multiple Imputation; Penalized Spline of Propensity Prediction Method; Predictive Mean Matching

Predictive mean matching (PMM) is a commonly used method to impute missing data, combining parametric modelling with hot-deck imputation, making it more robust to model misspecification than purely parametric methods. PMM imputes missing values using a random hot deck procedure with a specified distance metric, and the distances are computed by fitting a linear regression model to the data. While the actual imputations do not follow the linear model prediction, the method's success still largely hinges on some degree of linearity in the relationship between the variables and the linear regression model accurately capturing the ranking of the observed units to form the donor pool. The Penalized Spline of Propensity Prediction method (PSPP) is a semi-parametric method for imputation that has been shown to provide doubly-robust estimates of the mean in a variety of settings. However, it makes the assumption that the outcome variable is Normally distributed conditional on the propensity score and the other covariates, which can lead to inaccurate results when the data are non-continuous or non-Normal. We propose a method that combines PSPP with PMM to impute univariate missing data.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program