In medicine and health sciences we often measure constructs that are not directly observable, such as quality of life, pain, or depression. Measurement of such constructs requires use of questionnaires, which validity and reliability is assessed using statistical measurement models. The Item Response Theory (IRT) provides statistical models which model the relationship between the construct and the questionnaire responses. IRT models are increasingly being used in addition to the Classical Test Theory (CTT) model. Both models have different assumptions and the analyses can complement each other. CTT, for example, focuses more on the validity and reliability of sum scores, while IRT focuses more on the validity and reliability of individual items in a questionnaire. In addition, IRT has several advantages over CTT.
An important practical advantage of the IRT based measurement instruments is the flexibility for use in research and clinical practice. For instance, IRT models can be used to create short form questionnaires tailored for specific target groups. Furthermore, a computerized adaptive test (CAT) can be developed which selects the most informative questions for each individual during the administration, based on the previous responses of the individual.
This course is an introductory course on IRT with a focus on its use in the development and validation of questionnaires used in medical and health sciences. We focus on the conceptual understanding of IRT and less on the statistical details. We will cover the underlying principles of IRT and the conceptual differences between IRT and CTT. Then, we will study the theoretical pillars of IRT (e.g. the concept of latent trait, statistical modeling of items and responses, assumptions of IRT). Once the ground is set, we will move on to the practical issues in applications of IRT, such as checking the model assumptions, evaluating the model fit and estimating reliability of measurements. Finally, we will focus on more advanced issues such as differential item functioning, the principles behind item banking and computerized adaptive testing. As an example we will be using the Patient-Reported Outcomes Measurement Information System (PROMIS), as the items included in these questionnaires were selected under the IRT model and are widely used in clinical settings.
