Choosing between alternative statistical models

Background

Any scientific model will have to make simplifying assumptions about reality. Nevertheless, statistical models are important tools to summarize patterns from underlying data. Statistical models can well be used to make predictions for future subjects. We consider some general issues in choosing a type of statistical model in a prediction context, with illustration in a case study on modelling age–outcome relationships in medicine. We also summarize results from some empirical comparisons of alternative statistical models.