ABSTRACT
This chapter demonstrates that cross validation can be performed in different ways. It suggests that the cross model validation method as an instrument in model selection problems. The chapter shows how this works in a very elementary situation, which is later developed into a more complex analysis. Cross validation is perhaps most valuable when the data are highly multivariate, or complex in some other way, and many models are conceivable. However, larger analyses are best handled directly on a computer. In the statistical literature several different procedures have been suggested for variable selection in regression. The most well known are based on best subset regression together with some decision rule for the model size. If the best model is not there to begin with the true predictability of the selected model may improve when the number of models increases.
