ABSTRACT
Item response theory is a measurement theory and advanced modeling approach that allows estimating latent variables as the common variance from multiple items, and how the items relate to the construct (latent variable). IRT holds promise to enable the development of briefer assessments, including short forms and adaptive assessments, that have strong reliability and validity. In IRT, a person's construct score is estimated based on their item responses. The construct is estimated as a latent factor that represents the common variance among all items as in structural equation modeling or confirmatory factor analysis (CFA). IRT is similar to an ordinal or categorical approach to CFA. IRT models can estimate up to four parameters, including difficulty (severity), discrimination, guessing, and inattention/careless errors.
IRT conceptualizes reliability in a different way than classical test theory does. In classical test theory, the same standard error of measurement applies to all construct levels. However, IRT estimates how much measurement precision (information) or imprecision (standard error of measurement) each item, and the test as a whole, have at different construct levels. Thus, IRT conceptualizes reliability in such a way that precision/reliability can differ at different construct levels, unlike in classical test theory.
