Investigation of the effect of mixture IRT models under different conditions on parameter recovery and classification accuracy with simulative and real data

The aim of the study is to examine the effect of mixture IRT models on item parameter estimation and classification accuracy for different conditions with simulative and real data. In the simulation study, Mixture IRT models (Rasch, 2PL, 3PL), sample size (600, 1000), number of items (10, 30), number of latent classes (2, 3) missing data type (complete, missing at random (MAR) and missing not at random (MNAR)) and percentage of missing data (10%, 20%) variables are manipulated variables. R program was used to generate the data. In the analysis of the data, the "Mplus Automation" package, which allows the automation of the R program and the Mplus program, was used. As a result of the research, mean RMSE and bias values were obtained for item difficulty, item discrimination and pseudo-chance parameter estimation. Multi-way ANOVA was performed to examine the interaction effects of selected factors in the estimation of item parameters. When the interaction effects of the selected factors are examined, it can be said that the factors of sample size, number of items, latent classes, missing data type, and percentage of missing data are important factors in estimating item parameters for Mixture IRT models. It was observed that the mean RMSE values obtained for the Mixture Rasch model were lower than the Mixture 2PL and Mixture 3PL models. In addition, classification accuracy percentages were calculated for the data obtained from the combination of the considered factors. It was observed that the highest classification accuracy percentage was reached in the Mixture Rasch model with 30 item, 2 class, 1000 sample size and complete data conditions. Finally, simulation data was supported using TIMSS 2015 8th science test data for Turkey. It has been determined that the results obtained from the real data and the results obtained from the simulation data are consistent with each other.