Discretization of measures: an IRT approach

The paper describes a new method that can be used to discretize a continuous variable obtained applying the Rasch model to the data coming from a questionnaire. The idea underlying this proposal can be applied to every model belonging to the class of the IRT models. The estimated parameters of the Rasch model are used to estimate the most probable response for a subject with an estimated level of latent trait to a representative item; the obtained response is the category assigned to the categorized version of his/her latent trait. The performance of this method is compared to the performance of other common unsupervised discretization methods.