Research Seminar
On the analysis of Bayesian semiparametric IRT-type modelsErnesto San MartínDepartment of Statistics PUC and Measurement Center MIDE UC | |
| Abstract: | Normality of the abilities' distribution is a routine assumption in IRT modeling. However, such an assumption may be unrealistic, obscuring important features of between-subject variation. This paper describes Bayesian semiparametric IRT-type models, where the distribution generating the individual abilities is specified using a prior distribution on the space of probability measures, particularly Dirichlet and Polya Trees processes. The paper establish sufficient conditions for identification and consistency of Bayesian semiparametric IRT-type models. The identification conditions are much stronger for binary responses than for unbounded count responses. These findings are corroborated by a set of simulations. The appropriate posterior simulation schemes are described; the nonparametric priors are compared through a simulation study, and the estimation methods are illustrated by applying them to two practical examples widely used in educational measurement. |
| Date: | Wed Jun 20, 9:45 am - 10:45 am |
| Place: | room 00.14 (Department of Psychology, Tiensestraat 102, 3000 Leuven) |