Identifying Person-Fit Latent Classes, and Explanation of Categorical and Continuous Person Misfit.

Abstract

Latent class (LC) cluster analysis of a set of subscale lz person-fit statistics was proposed to explain person misfit on multiscale measures. The proposed explanatory LC person-fit analysis was used to analyze data of students (N = 91,648) on the nine-subscale School Attitude Questionnaire Internet (SAQI). Inspection of the class-specific lz mean and variance structure combined with explanatory analysis of class membership showed that the data included a poor-fit class, a class showing good fit combined with social desirability bias, a good-fit class, and two classes that were more difficult to interpret. A comparison of multinomial logistic regression predicting class membership and multiple regression predicting continuous person fit showed that LC cluster analysis provided information about aberrant responding unattainable by means of linear multiple regression. It was concluded that LC person-fit analysis has added value to common approaches to explaining aberrant responding to multiscale measures.