Abstract:
This study had four purposes: 1) to study the form of the MMix2PLE model and its parameter estimation method; 2) to compare the accuracy of model selection between AIC, BIC, CAIC, and DBIC for the MMix3PL and MMix2PLE models under simulated conditions; 3) to study the precision of parameter estimation: classes, items, and mean ability between the MMix3PL and MMix2PLE models under simulated conditions; and 4) to determine the number of classes, latent class parameters, item parameters, and mean ability in classes from empirical data using the MMix3PL and MMix2PLE models. The data consisted of 32 simulated data from the Monte Carlo method (2 models x 2 sample sizes x 8 candidate models), each data was replicated 10 times, yielding a total of 320 data sets, and the empirical data from O-NET in Mathematics Grade 12 (2019), of 2,500 students, 50 schools, 4 affiliations: OBEC, OPEC, MHESI, and DLA. The research results were as follows: 1. The form of the MMix2PLE model, which described the chance of a correct guess with the characteristics of respondents and items can avoid problems may arise from estimation of c parameters of the MMix3PL model. As for the Bayesian estimation tended to provide fairly accurate estimates, correlate to real values, be specifiable, and be robust in parameter estimation when the sample size in the class was small. 2. Comparison of the accuracy of model selection, AIC tended to be most accurate, followed by DBIC, and the lowest being BIC and CAIC under 2 school-level and 2 student-level latent classes conditions. 3. The precision of parameter estimation, the MMix2PLE and MMix3PL model tended to be highly precise, except c parameters for the MMix3PL model which did not correlate with the real values when the analyzed model was the same as the population model. 4. Empirical data analysis using the MMix3PL and MMix2PLE models yielded consistent results regarding latent classes and ability. The data consisted of 2 school-level and 2 student-level latent classes, total of 4 ability groups: 1) high, 2) low, 3) moderate, and 4) very low. Most of the first two groups were students whose parents earn more than 300,000 baht, studied in MHESI schools, Bangkok, while most of the latter two groups were students whose parents earn no more than 300,000 baht, studied in OBEC, OPEC, and DLA schools, not Bangkok. The item parameter estimation found that both models gave inconsistent estimates as a result of different arrangement of members in the latent classes.