A Bayesian person fit evaluation for polytomous response data

Studies about Person-fit are generally produced under a frequentist approach. For example, Meijer & Sijtsma (2001) discussed many parametric and non-parametric indexes in their review on this topic. However, it exists also few papers about the investigation of person-fit in a Bayesian context (e.g. Glas & Meijer, 2003; Van Der Linden & Guo, 2008). In this talk, we present a new method based on the evaluation of informative hypotheses using the Bayes factor. This approach is non-parametric in nature and can be applied to a large variety of situations and many types of data. Here, we focus on the use of Bayesian person-fit methods that can be used with polytomous response data. This presentation is divided in two sections. First, we present the technical aspects of this approach by discussing some hypotheses of interest, the nature of the prior and the nature of the posterior. Second, we present results from a real data matrix. The first analysis shows that Bayesian person-fit evaluation is efficient and can be easily applied to small data matrices.