Cross Validation

Cross validation refers to a procedure in which an analysis is performed on one dataset and the parameters used in a second dataset. At the same time, an analysis is performed on a second dataset and then the parameters are used on the first dataset. A successful cross validation would occur when the overall estimation of the process has equal accuracy for both datasets. The argument is that the resulting equation or estimation process generated using one set of data will cross validate or continue to be accurate when applied to an entirely new set of data. Similarly, going from the second set of data to the first set retains the same level of accuracy. The procedure is useful when the desire is to provide a means of predicting a score using a combination of predictors. The goal is the generation of an equation that will maintain a high level of prediction when used on different samples. If successful, additional future samples would have the same level of accuracy when using the generated equation. The technique has particularly strong application when the original sample is very large and the future samples are much smaller. The larger sample for the original estimation should provide a great deal of accuracy in the estimation of the parameters that may not be as accurate when considering smaller samples. This entry provides a detailed example that offers further explanations of cross validation as well as [Page 306]justification for its use. Next, the aspect of cross validation that examines whether an equation predicted by a theory can continue to be equally predictive in all contexts is examined. Finally, the value of cross validation is reviewed.