PERBANDINGAN SENSITIVITAS ANTARA METODE MANTEL-HAENSZEL, REGRESI LOGISTIK DAN IRT-LIKELIHOOD RATIO PADA PENDETEKSIAN DIFFERENTIAL ITEM FUNCTIONING

KERDID SIMBOLON, . (2019) PERBANDINGAN SENSITIVITAS ANTARA METODE MANTEL-HAENSZEL, REGRESI LOGISTIK DAN IRT-LIKELIHOOD RATIO PADA PENDETEKSIAN DIFFERENTIAL ITEM FUNCTIONING. Doktor thesis, UNIVERSITAS NEGERI JAKARTA.

	Text COVER.pdf.pdf Download (1MB)
	Text BAB 1.pdf.pdf Restricted to Registered users only Download (514kB) | Request a copy
	Text BAB 2.pdf.pdf Restricted to Registered users only Download (1MB) | Request a copy
	Text BAB 3.pdf.pdf Restricted to Registered users only Download (535kB) | Request a copy
	Text BAB 4.pdf.pdf Restricted to Registered users only Download (3MB) | Request a copy
	Text BAB 5.pdf.pdf Restricted to Registered users only Download (424kB) | Request a copy
	Text DAFTAR PUSTAKA.pdf.pdf Restricted to Registered users only Download (285kB) | Request a copy
	Text LAMPIRAN.pdf.pdf Restricted to Registered users only Download (13MB) | Request a copy

Abstract

ABSTRAK Penelitian ini bertujuan menguji perbedaan sensitivitas 3 metode deteksi DIF (Differential Item Functioning) yakni: IRT-Likelihood Ratio (IRT-LR), Mantel- Haenszel (MH) dan Regresi Logistik (RL), pada pendeteksian DIF Ujian Nasional (UN) Matematika SMA Program IPA Tahun 2015 di wilayah Sumatera Utara, Indonesia. Tepatnya bertujuan menentukan metode paling sensitif dalam mendeteksi DIF atau “mengendus” keberadaan DIF bahkan ketika derajat DIF kecil. Butir DIF adalah butir yang menguntungkan kelompok tertentu dan merugikan kelompok lain dengan kemampuan yang sama. DIF ditinjau dari dua jenis pengelompokan, pertama jenis kelamin (JK), dimana laki-laki sebagai kelompok referensi dan perempuan sebagai kelompok fokus; kedua, jenis sekolah (JS), dimana SMA Negeri sebagai kelompok referensi dan SMA Swasta sebagai kelompok fokus. Penelitian ini mengunakan metode eksperimen dengan disain 3x1, 1 faktor (metode) dengan 3 perlakuan (metode deteksi DIF yang berbeda). Populasi mencakup keseluruhan 5 jenis paket soal UN Matematika 2015 (kode 1107, 2207, 3307, 4407 dan 5507). Paket 2207 terambil sebagai sampel terdiri atas 5000 peserta (3067 perempuan,1933 laki-laki; 2427 siswa SMA Swasta, 2573 siswa SMA Negeri), dengan 40 butir soal UN. Seleksi butir teori tes klasik (TTK) diterapkan terhadap 40 butir UN dan meloloskan 32 butir. Seleksi teori respon butir (TRB) meloloskan 18 butir yang memenuhi. Dengan program R 3.333 dan IRTLRDIF 2.0 pada kelompok JK didapat 5 butir DIF oleh IRT-LR, 4 butir oleh RL, dan 3 butir oleh MH. Pada kelompok JS didapat 18 butir DIF oleh IRT-LR, 32 butir oleh RL, dan 25 butir oleh MH. Dalam uji sensitivitas metode DIF, dibentuk 6 kelompok analisis: (4400,40), (4400,32), (4400,18), (3000,40), (3000,32), (3000,18), serta dibangkitkan 40 set data acak dan dilakukan 40 kali deteksi DIF pada setiap kelompok. Dengan data tidak fit model, model logistik 3 parameter (3PL) dipilih sebagai model paling sesuai. Hasil uji Tukey menunjukkan metode IRT-LR lebih sensitif dari MH dan RL dalam kelompok (4400,40) dan (3000,40). IRT-LR teruji lebih sensitif dari MH pada kelompok (4400,32) dan (3000,32) tetapi tidak lebih sensitif dari RL. Kedua metode MH dan RL berbalik menjadi lebih sensitif dari IRT-LR dalam kelompok (4400,18) dan (3000,18). Metode RL konsisten teruji lebih sensitif dari MH pada seluruh kelompok analisis. Sementara pada kelompok JS, seluruh hipotesis penelitian teruji signifikan dimana RL dan MH berbalik menjadi lebih sensitif dari IRT- LR, dan metode RL konsisten lebih sensitif dari MH. ******************************** ABSTRACT The purpose of this study was to examine the differences in sensitivity of 3 DIF (Differential Item Functioning) detection methods, namely: IRT-Likelihood Ratio (IRT-LR), Mantel-Haenszel (MH) and Logistic Regression (RL), in detecting DIF Mathematics National Examination (UN) academic year of 2015 in North Sumatra, Indonesia. Precisely to determine the most sensitive method of detecting DIF or "sniffing" the presence of DIF even when the DIF degree is small. DIF item is item that benefit certain group participants and disadvantage other groups with similar abilities. The presence of DIF is reviewed on two types of grouping of participants. First by gender (JK) where men as a reference group (R) and women as focus group (F); second by type of school (JS), where public high school as a reference group (R) and private high school as focus group (F). This study uses an experimental method with a 3x1 design, with 1 factor (method) with 3 treatments (different DIF detection methods). The population includes all 5 types of 2015 Mathematics UN question packages (codes: 1107, 2207, 3307, 4407 and 5507). The 2207 package was taken as a sample consisting of 5000 participants (3067 women, 1933 men; 2427 Private High School students, 2573 Public High School students), with 40 UN items. Classic test theory (TTK) item selection was applied to 40 UN items and passed 32 items. Item response theory selection (TRB) passed 18 items that fulfilled. With the R 3.333 program and IRTLRDIF 2.0, in the JK group, 5 items were detected as DIF by IRT-LR, 4 items by RL, and 3 items by MH. In the JS group, there were 18 items were detected as DIF by IRT-LR, 32 items by RL, and 25 items by MH. In the sensitivity test of the DIF methods, 6 analysis groups were formed: (4400,40), (4400,32), (4400,18), (3000,40), (3000,32), (3000,18), 40 data sets were generated and 40 times DIF detection was performed in each group. With data not not fit models, a three parameter logistic model (3PL) is chosen as the most suitable model. In the JK group, Tukey's test results showed the IRT-LR method was more sensitive than MH and RL in the groups (4400,40) and (3000,40). IRT-LR was tested to be more sensitive than MH in the groups (4400,32) and (3000,32) but not more sensitive than RL. Both MH and RL methods turn out to be more sensitive than IRT-LR in groups (4400,18) and (3000,18). The RL method was consistently tested to be more sensitive than MH in the entire analysis group. While in the JS group, all research hypotheses were tested as significant where RL and MH turned to be more sensitive than IRT-LR, and the RL method was consistently more sensitive than MH.

Item Type:	Thesis (Doktor)
Additional Information:	1). Prof. Dr. Yetti Supriyati, M.Pd. ; 2). Prof. Dr. Ir. Dali Santun Naga, MMSI.
Subjects:	Pendidikan > Teori, Penelitian Pendidikan
Divisions:	PASCASARJANA > S3 Penelitian Dan Evaluasi Pendidikan
Depositing User:	Users 2444 not found.
Date Deposited:	24 Apr 2020 14:37
Last Modified:	24 Apr 2020 14:37
URI:	http://repository.unj.ac.id/id/eprint/7149

Actions (login required)

View Item