Model Selection for Parsimonious Models via Conceptual Predictive Statistic

• Activity Number: 75
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Zugui Zhang*
• Companies: Christiana Care Health System
• Keywords: Model Selection ; Parsimonious Models ; Conceptual Predictive Statistic ; Backward Selection Strategy ; Model Prediction ; Variance of the Full Model

Abstract

To better support the use of the models in clinical care, we sought to create more parsimonious models that would require fewer predictor variables, avoid potential over-fitting, and could be used more easily in routine clinical care. This was accomplished using Harrell's backward selection strategy to identify those variables providing the greatest prognostic value. To do this, the contribution of each covariate in the multivariable model was ranked by F-value. Variables with the smallest contribution to the model were sequentially eliminated until further variable elimination led to a greater than 10% loss in model prediction, as compared with the initial model. This insured that the remaining covariates explained over 90% of the variance of the full model. The approach of conceptual predictive statistic (Cp) suggests an objective standard for the selection of an appropriate model from a potentially large class of candidate models, by examining any candidate model whether the values of Cp are around the number of parameters in the model. In this study, I proposed Cp with parsimonious models and compared its performance to results from the Harrell's backward selection strategy