Grant Morgan, an assistant professor of education psychology at Baylor University, will present “Finite Mixture Modeling With Nonnormal or Categorical Indicators: A Monte Carlo Examination of Fit Indices for Model Selection” from noon to 1 p.m. on March 20 in SZB 324.
Morgan will discuss the findings from three simulation studies that examined the performance of fit indices that applied researchers commonly use in model selection. Each of the studies included design factors that reflect conditions found in applied research as well as categorical or nonnormally distributed mixture indicators. Morgan will explore the accuracy of the fit indices in the studies and present recommendations for fit index use and potential solutions for nonnormality.
Finite mixture modeling for the purposes of classification is a model-based approach that can be used to identify groups of cases underlying a multivariate dataset and offers significant improvement over traditional, distance-based clustering procedures (i.e., cluster analysis, multidimensional scaling). Among the major advantages of finite mixture modeling is the inclusion of a variety of fit measures, each of which reflects a slightly different aspect of model fit, and the ability to simultaneously use variables measured at different metric levels and distribution shapes. Despite these advantages, model selection aided by fit indices remains a challenge to researchers.