Item Response Modeling

Abstract

Item response models are applied for analyzing item scores of psychological and intelligence tests, and they are based on exponential relationships between the psychological traits and the item responses (Baker and Kim 2004; De Boeck and Wilson 2004). Items are usually questions with “yes” or “no” answers. Item response models were invented by Georg Rasch, a mathematician from Copenhagen who was unable to find work in his discipline in the 30ths and turned to work as a psychometrician (Rasch 1980). These models are, currently, the basis for modern psychological testing including computer-assisted adaptive testing (Van der Linden and Veldkamp 2004). Advantages compared to classical linear testing include first that item response models do not use reliability as a measure of their applicability, but instead use formal goodness of fit tests (Zwinderman 1991). Second, the scale does not need to be of an interval nature. As a consequence the effects of covariates can be analyzed and reported with odds ratios, independently of the item format and population averages. Ceiling effects are, therefore, much less of a problem than they are with classical linear methods (Fischer 1974).