Bagaimana hasil penyetaraan paket tes USBN pada mata pelajaran matematika dengan teori respon butir?

Abstract

Dalam melakukan standarisasi pendidikan di Indonesia maka dilakukan Ujian Sekolah Berstandar Nasional (USBN).Karena perangkat tes yang digunakan dalam USBN tersebut sebagian besar butir tesnya dikembangkan oleh guru, perangkat tes yang diujikan kepada siswa di daerah yang satu akan berbeda dengan daerah yang lainnya meskipun sama-sama mengacu pada kisi-kisi dari pemerintah. Oleh karena itu perlu dilakukan penyetaraan perangkat tersebut. Penelitian deskriptif eksploratif dengan pendekatan kuantitif ini bertujuan untuk mendeskripsikan kesetaraan perangkat USBN tahun 2018/2019 pada mata pelajaran matematika wajib. Pengumpulan data dilakukan melalui dokumentasi respon siswa pada USBN 2018/2019 untuk mata pelajaran matematika wajib. Respon siswa tersebut berasal dari lima paket soal dari empat sekolah menengah atas di Provinsi D.I. Yogyakarta dan Kalimantan Selatan. Data yang terkumpul dianalisis menggunakan teknik equating berdasarkan teori respon butir dengan metode mean-mean, mean-sigma, Haebara, dan Stocking Lord. Estimasi parameter butir dan equating dilakukan dengan bantuan program R. Hasil penyetaraan menggunakan empat metode menunjukkan bahwa lima paket tes USBN 2018/2019 untuk mata pelajaran matematika wajib cenderung setara satu sama lain dan penyetaraan menggunakan metode Haebara menghasilkan kesetaraan yang lebih baik dibandingkan dengan tiga metode equating lainnya. Dalam artikel ini juga disajikan contoh butir sulit beserta peluangnya untuk digunakan sebagai referensi dalam meningkatkan kualitas pembelajaran matematika. How are the results of the equating of test packages of mathematics USBN with item response theory?AbstractIn standardizing education in Indonesia, then Ujian Sekolah Berstandar Nasional (USBN) or National-Standardized School Examination was conducted. Because most of the test items contained in the test packages which were used in USBN were developed by the teachers, the test packages which were administered to students in a region would be different from other regions. Therefore, there was a need to do equating towards those test packages. This exploratory descriptive research with a quantitative approach was focused on describing the equality of test packages of USBN year 2018/2019 for compulsory mathematics subject. Data collection was done through documentation of students’ responses to the USBN 2018/2019 for compulsory mathematics subject. These students’ responses were collected from five test packages from four senior high schools in Province of Special Region of Yogyakarta and South Kalimantan, Indonesia. The collected data were analyzed by using an equating technique based on the item response theory with the methods of mean-mean, mean-sigma, Haebara, and Stocking Lord. The item parameter estimation and equating were conducted with the aid of the R program. The results of equating showed that the five test packages of the USBN 2018/2019 for compulsory mathematics subject tend to be equal to each other and the equating through Haebara method yields better equality than the other three equating methods. This article also presents the example of a difficult item as well as its opportunity to be used as a reference for enhancing the quality of mathematics learning.

Equating of standardized science subjects tests using various methods: which is the most profitable?

“…Of the four equalization methods used, it was found that equalization using the Stocking-Lord method produced the most equivalent scores between packages. These findings are the same as those of previous studies (Uysal & Kilmen, 2016) but different from the findings of other studies (Yusron et al, 2020a). This indicates that the literature has not been able to provide strong evidence that one method is better than the other equalization methods.…”

“…The difference in findings between studies is very likely due to differences in several things, such as the characteristics of the test takers and the types of test items. For example, the study by (Yusron et al, 2020a) used a test kit for high school level students. This study is clearly different from our study, in which we used a test for junior high school students.…”

“…However, this study did not explore the equivalence of scores among test takers on the test packages. In a study by Yusron et al (2020b) the equalization of the national standardized school exam test package in mathematics at the senior high school level was investigated. This study not only described the equivalence of the test packages but also compared the results of four equalization methods based on the 3PL item response theory.…”

The Use of Stocking-Lord and Haebara Methods in Horizontal Equating: A Case of Indonesian Madrasah Competence Assessment

“…Both methods have more moderate error estimates, and the results are more accurate when using more anchor items (Born et al, 2019). Similarly, Yusron, Retnawati, and Rafi (2020) show that the Haebara method can produce the smallest average RMSE compared to the mean-mean, mean-sigma, and Stocking Lord methods.…”

Which one is more accurate, BILOG or R program? (a comparison for score test equating)

“…Based on IRT, there are three testing equivalence designs that were employed: i) Anchor testing design; ii) Equivalent group design; and iii) Single group design [36]. In an equivalent group design, two tests were done to two equivalent groups drawn randomly from the same population (presumed that they had the same level of ability) [37], [38]. The equivalent group design may not only lessen the impact of activity and boredom, but it may also cause bias since the groups' levels of ability were probably not equivalent [39].…”