Testing for predictive validity in health care education research: a critical review

The assessment of predictive validity is the essential core from which a sound model of prediction is built.

Three methods for assessing predictive validity in health care education research were reviewed: longitudinal profile development, cross-validation, and inspection of the adjusted R2. A total of 47 articles published between 1973 and 1993 in nine health care disciplines were critically reviewed to determine whether the studies tested for predictive validity by using these methods.

Very few of the 47 studies used at least one of the three methods for assessing predictive validity. Furthermore, the proportion of variance explained that is reported in the articles is typically small even before assessment of predictive validity. It is sobering to note that these small values may be inflated, since shrinkage is likely to occur when assessing predictive validity on a second, or cross-validation, sample.

The scarcity of testing of predictive validity in the studies reviewed highlights the necessity of future research to establish the degree of predictive validity, if improvements in predicting success in health care education research are to be realized.