<![CDATA[Purpose – This study examined the effects of test length, sample size, and Item Response Theory (IRT) models on the reliability of Generalizability Theory (GT) coefficients.Methodology – Simulated dichotomous and polytomous data were generated using WinGen and analyzed using EduG. The study employed several combinations of sample sizes (N = 200, 300, 500, 1000, and 5000), test lengths (n = 5, 10, 30, 50, and 100), dichotomous models (1PL, 2PL, and 3PL), and polytomous models (GRM and GPCM) with different response categories.Findings – The findings indicate that test length had a stronger influence on G-coefficient reliability than sample size. In dichotomous data, the 1PL and 2PL models generally produced higher G-coefficients than the 3PL model, especially with a test length of 50 items. In polytomous data, the GPCM model produced higher and more stable G-coefficients than the GRM model, particularly when the number of response categories and test length increased. The results indicate that optimal combinations of test length, sample size, parametric model, and response categories are important for achieving high measurement reliability. This study also demonstrates the effectiveness of Generalizability Theory and EduG software in evaluating reliability for both dichotomous and polytomous data.Contribution – This research helps researchers determine appropriate model parameters, sample size, and test length. Furthermore, it can strengthen theory in empirical research]]>
Copyrights © 2026
