Computerized Classification Testing to Predict an Observable Outcome
Matthew Finkelman, Tufts University School of Dental Medicine
Yulei He, Harvard School of Public Health
Giles Hooker, Cornell University
Wonsuk Kim, Measured Progress
Robert Keller, Measured Progress
Barbara Gandek, Tufts University School of Medicine
Computerized classification testing (CCT) is a well-known psychometric approach to categorizing examinees into groups. Most existing CCT methods rely on the latent variable models of item response theory (IRT), which were originally developed for assessments where there is no external outcome variable. This research examines the use of CCT in a different context: the prediction of a medical event from responses to a health questionnaire. Because the outcome is observable rather than latent, new item selection procedures and variable-length stopping rules are needed. Methods are illustrated using data from the Medicare Health Outcomes Survey.