Psychometric Quality Evaluation of the U-GRT Figural Subtest Using the Rasch Model among Job Selection Participants

Abstract

This study aimed to evaluate the psychometric quality of the Figural subtest of the USU General Reasoning Test (U-GRT) using the Rasch Model to determine its suitability for employee selection. A descriptive quantitative approach was employed using secondary data obtained from 728 job applicants who completed a Computer-Based Test (CBT). Data were analyzed using the Rasch Model to examine item difficulty, item fit statistics (infit and outfit mean square), unidimensionality through Principal Component Analysis of Residuals (PCAR), person reliability, and score categorization based on person ability estimates transformed from logit measures into T-scores. The findings indicated that item difficulty ranged from −1.645 to 4.001 logits, with most items clustered at moderate difficulty levels. Nineteen of the twenty items met the Rasch fit criteria, while one item displayed an outfit statistic exceeding the recommended threshold and therefore requires further revision. The PCAR results supported the assumption of unidimensionality, whereas person reliability was classified as acceptable. Ability estimates were subsequently converted into five interpretive categories: very low, low, moderate, high, and very high. Overall, the Figural subtest of the U-GRT demonstrated adequate psychometric properties, although additional items representing extreme difficulty levels are recommended to improve measurement precision across the full range of ability.

Penelitian ini bertujuan mengevaluasi kualitas psikometrik subtes Figural pada USU General Reasoning Test (U-GRT) menggunakan Model Rasch sebagai dasar penilaian kelayakan instrumen dalam seleksi kerja. Penelitian menggunakan pendekatan kuantitatif deskriptif dengan data sekunder yang berasal dari 728 peserta seleksi kerja berbasis Computer-Based Test (CBT). Analisis dilakukan melalui Model Rasch untuk mengevaluasi tingkat kesukaran aitem, kesesuaian aitem terhadap model (infit dan outfit mean square), unidimensionalitas menggunakan Principal Component Analysis of Residuals (PCAR), reliabilitas person, serta penyusunan kategorisasi skor berdasarkan estimasi kemampuan dalam satuan logit yang ditransformasikan menjadi T-score. Hasil penelitian menunjukkan bahwa tingkat kesukaran aitem berada pada rentang −1,645 hingga 4,001 logit dengan sebagian besar aitem berada pada kategori kesukaran sedang. Sebanyak 19 dari 20 aitem memenuhi kriteria kesesuaian model Rasch, sedangkan satu aitem menunjukkan nilai outfit di atas batas yang direkomendasikan sehingga memerlukan evaluasi lebih lanjut. Analisis PCAR mendukung asumsi unidimensionalitas, sementara nilai person reliability berada pada kategori cukup. Kategorisasi skor menghasilkan lima tingkat kemampuan, yaitu sangat rendah, rendah, sedang, tinggi, dan sangat tinggi. Temuan ini menunjukkan bahwa subtes Figural U-GRT memiliki kualitas psikometrik yang memadai, meskipun pengembangan aitem pada tingkat kesukaran ekstrem masih diperlukan untuk meningkatkan presisi pengukuran.