การเปรียบเทียบระหว่างวิธี Hglm กับวิธี Sibtest =The effect of samplesize on the detection of differential item functioning using grade 6 o-net results and two methods : hgml and sibtest
Abstract:
The objectives of this research were to analyze the item quality of the O-NET examination across eight groups of curriculum implementation, to examine the differential functions of O-NET items by using HGLM and SIBTEST methods, and to compare the performance of the functions of the O-NET items under the condition of three different sample sizes, that is, small (n = 300), medium (n = 1,000), and large (n = 2,000) and also between HGLM and SIBTEST methods. Secondary data, O-NET examination results, were obtained from the National Institute of Educational Testing Service, involving Grade 6 students in the academic year 2556, and were examined for all eight groups of curriculum implementation: 1) Thai, 2) social studies, religion and culture, 3) foreign languages, 4) mathematics, 5) science, 6) health and physical education, 7) art, and 8) work, career, and technology. The three-parameter logistic model of item response theory was used to examine item quality. Statistical analysis was performed using Xcalibre Version 4.2.2. The results showed that: 1. The O-NET items of grade six across eight groups of curriculum implementation had good discrimination levels, and quite a high difficulty level. The science and art curriculum implementation had the highest levels of item difficulty, with item guessing parameters below 0.30. 2. Examination of differential item function of O-NET for grade six across eight curriculum implementation with HGLM and SIBTEST methods under the three different sample sizes (small, medium, and large) indicated that sample size affected differential item functioning (DIF). The larger sample size had a better examination of DIF than the medium and small sample sizes. The HGLM method outperformed SIBTEST method in terms of DIF detection. 3. The effectiveness of the performance results of differential item function of O-NET items between HGLM and SIBTEST methods, when considering the rate of type I error, demonstrated that small sample size had a lower rate. Furthermore, both HGLM and SIBTEST were comparable in the effectiveness of the examination results differential item functioning.