Model Selection and Statistical Inference

Research project P 13868 Model Selection and Statistical Inference Benedikt M. PÖTSCHER 11.10.1999 In many statistical applications a (formal or informal) model selection stage making use of the data precedes the model fitting and inference stage. Although statistical properties of inference procedures (like estimators or confidence intervals) are obviously affected by such ,model uncertainty", the vast majority of theoretical analyses of inference procedures ignores the problem and procedes as if the statistical model had been selected without recourse to the data. Breiman (1992) calls this ,a quiet scandal in the statistical community". The proposed project is aimed at a better understanding of the effects of model selection on subsequent statistical inference.