DEVELOPMENT OF A MULTILEVEL MIXTURE IRT MODEL USING ALTERNATIVE IRT MODEL WITH GUESSING การพัฒนาโมเดลทฤษฎีการตอบสนองข้อสอบแบบผสมพหุระดับโดยใช้โมเดลทฤษฎีการตอบสนองข้อสอบทางเลือกที่คำนึงถึงการเดา

Abstract

This study had three purposes: 1) to develop a multilevel mixture IRT model applied from the 2PLE model (MMix2PLE) and an appropriate estimation method; 2) to compare the accuracy of detection of latent classes between AIC and BIC for the MMix3PL and MMix2PLE models; and 3) to compare the precision of estimation: the latent classes, mean ability, and item parameters between the MMix3PL and MMix2PLE models. The data were simulated from the Monte Carlo method under 32 conditions (2 simulation models x 2 sample sizes x 8 candidate models), each condition was replicated 10 times, yielding a total of 320 data sets. The research results were as follows:
1. A Bayesian estimation algorithm for the MMix2PLE model developed in this research was found to be quite suitable. It was able to estimate unique parameter and recover parameter quite well under simulated conditions.
2. Comparison of the accuracy of detection of latent classes, AIC tends to be more accurate than BIC for both the MMix3PL and MMix2PLE models.
3. Comparison of the precision of estimation: the latent classes, mean ability, and item parameters, the MMix2PLE model tended to be more precise than the MMix3PL model, especially when the sample size was large (N = 2,500).