Testing Predicted Clusters

Testing predicted clusters of questionnaire items can be a source of abundant feedback on the theory that is behind the prediction. Such testing is often performed by means of confirmatory factor analysis. However, that method offers insufficient feedback at the level of items, while its goodness-of-fit indices are notoriously unreliable. Richer and more precise feedback would be generated by a data-driven optimization of the predicted clusters (factors), allowing a comparison between predicted and empirical clusters (factors) at the level of items. Contrasting these two provides a basis for classifying the items into hits, false positives, and false negatives. This division greatly facilitates reinterpretation of the clusters (factors) and evaluation of the items. In addition, it offers a basis for two new measures of goodness of fit with respect to the correct assignment of items to clusters (indicators to factors). Application of this new approach to a questionnaire on Obsessive–Compulsive Disorder will serve as an illustration of its merits for both a qualitative and quantitative evaluation of the predicted clusters.