Extensions of the Penalized Spline Propensity Prediction Method for Monotone Missing Data

• Activity Number: 207
• Type: Contributed
• Date/Time: Monday, August 7, 2006 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Guangyu Zhang*+ and Roderick J. Little
• Companies: University of Michigan and University of Michigan
• Keywords: monotone missing data ; penalized spline propensity prediction

Abstract

Little and An (2004) proposed a Penalized Spline Propensity Prediction (PSPP) method, which yields an estimate of the marginal mean of the missing variable with a double robustness property. We study extensions of the PSPP method to monotone missing data, where the variables can be ranged so that all subsequent variables, say Yj+1,.,Yp , are missing for cases where Yj is missing, for all j = 1,.,p-1. This pattern arises frequently in longitudinal studies due to attrition. Let (X1,.,Xq,Y1,.,Yp) denote a p+q-dimensional vector of variables with Y1,.,Yp represent the monotone missing data part and X1,.,Xq fully observed covariates. Our research objective is to estimate the marginal and conditional means of Y1,.,Yp . Results of a simulation study illustrate the efficiency and robustness property of the stepwise PSPP method. We also apply the method to an online weigh loss study.