171 – Design and Adjustment Challenges in Modern Surveys
Using Proportional Hazard Regression for Building Imputation Models
Christele Augard
Sanofi Pasteur (France)
Robert D. Small
Sanofi Pasteur
In this paper we introduce a semi parametric method for imputing missing values when the distribution of the data is not known. The method is a version of the partial likelihood used in proportional hazard modeling in survival analyses. The various extensions (to time varying covariates for example) carry over even when the missing data are not time or positive data. We compare the method with methods assuming normality and evaluate it when the dependent variable is far from normal