The IRT Procedure (Experimental)

The item response theory (IRT) model was first proposed in the field of psychometrics for the purpose of ability assessment. It is most widely used in education to calibrate and evaluate items in tests, questionnaires, and other instruments and to score subjects on their abilities, attitudes, or other latent traits. Today, all major psychological and educational tests are built using IRT, because the methodology can significantly improve measurement accuracy and reliability while providing potential significant reductions in assessment time and effort, especially via computerized adaptive testing. In a computerized adaptive test, items are optimally selected for each subject. Different subjects might receive entirely different items during the test. IRT plays an essential role in selecting the most appropriate items for each subject and equating scores for subjects who receive different subsets of items. Notable examples of these tests include the Scholastic Aptitude Test (SAT), Graduate Record Examination (GRE), and Graduate Management Admission Test (GAMT). In recent years, IRT models have also become increasingly popular in health behavior, quality of life, and clinical research. The Patient Reported Outcomes Measurement Information System (PROMIS) project, funded by the US National Institutes of Health, is an excellent example. By using IRT, it aims to develop item banks that clinicians and researchers can use to collect important information about therapeutic effects that is not available from traditional clinical measures.

Early IRT models (such as the Rasch model and two-parameter model) concentrate mainly on dichotomous responses. These models were later extended to incorporate other formats, such as ordinal responses, rating scales, partial credit scoring, and multiple category scoring. Early applications of IRT focused primarily on the unidimensional model, which assumes that subject responses are affected only by a single latent trait. Multidimensional IRT models have been developed, but because of their greater complexity, the majority of IRT applications still rely on unidimensional models.

For an introduction to IRT models, see De Ayala (2009) and Embretson and Reise (2000).