Evaluación de la adecuación de modelos psicométricos para el análisis de pruebas conformadas por testlets

Abstract

The aim of this doctoral thesis was to compare the most common psychometric models applied to testlet-based tests. The models were compared based on their fit indices, the magnitude and precision of the item parameters, the precision and fairness of the scores, and their impact on the classification of the examinees. Based on a systematic review of the literature, ten psychometric models were selected from the item response theory, the testlet response theory, and Bi-factor approach; these models were fitted to a data set of a reading comprehension test used to select students (n = 30,853) to enter a Mexican public university (see UABC, 2017). The model fit was compared based on the goodness of fit indices and the information indices (AIC, BIC and DIC). The magnitude and precision of the parameters were compared, as well as the reliability of the scores obtained by each model. To assess the impact of the testlets on the fairness of the test, the detection differential item functioning was compared via the logistic regression procedure. Finally, the capacity of each of the ten models to select the examinees based on their skill level was compared...