Abstract
Effects of the scale number of categories on the Muraki Graded Response Model. Effects that variation in the number of categories in ordinal scales has on the Graded Response Model (as proposed by Muraki (1990)) are evaluated by means of Parscale 3.2 (Muraki and Bock, 1998). Two sets of simulated data (unidimensional and bidimensional) and an empirical data set are analyzed. General results show that when data are unidimensional, the best goodness of fit indices are achieved between 4 and 6 categories. Satisfactory estimates of parameter Θ are also obtained. When data are bidimensional 6 categories are needed as a minimum to achieve satisfactory fit indices. In this case, Θ estimates represent the average between the two latent factors.
