SENSITIVITAS MENDETEKSI BIAS BUTIR METODE UJI BEDA TARAF SUKAR, KHI-KUADRAT LORD DAN DISTRIBUSI SAMPLING EMPIRIS

Penelitian ini bertujuan untuk mengetahui perbedaan sensitivitas metode DIF: (1) Uji Beda Taraf Sukar, (2) Khi-Kuadrat Lord, dan (3) Distribusi Sampling Empiris. Penelitian ini merupakan penelitian deskriptif eksploratif. Data diperoleh dari Puspendik, berupa data respons siswa SMA Program IPA Provinsi DKI Jakarta yang berjumlah 4.869 orang. Butir soal yang dianalisis adalah butir soal UN Matematika tahun pelajaran 2014/2015 kode paket 1101. Untuk mendeteksi bias butir dilakukan replikasi secara acak sebanyak 30 kali, pada kelompok fokal dan referensi masing-masing berjumlah 1000 orang. Data dianalisis menggunakan Program BILOG-MG, SPSS 22.0, dan Excel 2010. Pengujian hipotesis menggunakan statistik One_Way ANOVA dan uji Post Hoct Tukey. Hasil penelitian menunjukkan bahwa: (1) metode Distribusi Sampling Empiris lebih sensitif mendeteksi bias butir daripada metode Khi-Kuadrat Lord dengan perbedaan rerata 3.200 dan nilai sig. 0.000<0.05; (2) metode Khi-Kuadrat Lord lebih sensitif mendeteksi bias butir daripada metode Uji Beda Taraf Sukar dengan perbedaan rerata 2.367 dan nilai sig. 0.001<0.05.The objective of this research was to understand the sensitivity difference of DIF methods: (1) test of difficulty level difference, (2) Lord’s chi-square, (3) empirical sampling distributions to detect item bias. This research is of descriptive exploratory nature. Data were obtained from Puspendik in a form of response data of a high school student majoring in science in DKI Jakarta Province of 4,869 people. The item bias test being tested was the mathematics national exam of the 2014/2015 academic year with package code 1101. To detect item bias, replication is done 30 times, the samples are randomly selected from both the reference group and focal group, each consisting of 1000 people. Data were analyzed by using the Bilog-MG Program, SPSS 22.0, and Microsoft Excel 2010. The hypothesis is tested by using One_Way ANOVA statistical and Tukey Post Hoc test. The results showed that: (1) the empirical sampling distributions method is more sensitive than the Lord’s chi-square method to detect item bias with mean difference of 3.200 and value-sig.0.000<0.05; (2) the Lord’s chi-square method was more sensitive than the test of b difference method for detecting item bias, with mean difference of 2.367 and value-sig.0.001<0.05