The Examination of Item Difficulty Distribution, Test Length and Sample Size in Different Ability Distribution


Creative Commons License

ŞAHİN M. G., Yildirim Y.

JOURNAL OF MEASUREMENT AND EVALUATION IN EDUCATION AND PSYCHOLOGY-EPOD, cilt.9, sa.3, ss.277-294, 2018 (ESCI) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 9 Sayı: 3
  • Basım Tarihi: 2018
  • Doi Numarası: 10.21031/epod.385000
  • Dergi Adı: JOURNAL OF MEASUREMENT AND EVALUATION IN EDUCATION AND PSYCHOLOGY-EPOD
  • Derginin Tarandığı İndeksler: Emerging Sources Citation Index (ESCI), TR DİZİN (ULAKBİM)
  • Sayfa Sayıları: ss.277-294
  • Anahtar Kelimeler: Item response theory, examinee distribution, item difficulty distribution, sample size, test length, NOMINAL RESPONSE MODEL, MARGINAL MAXIMUM-LIKELIHOOD, MONTE-CARLO ESTIMATION, PARAMETER RECOVERY, RASCH MODELS, 2-PARAMETER, CATEGORIES, MULTILOG
  • Gazi Üniversitesi Adresli: Evet

Özet

This is a post-hoc simulation study which investigates the effect of different item difficulty distributions, sample sizes, and test lengths on measurement precision while estimating the examinee parameters in right and left-skewed distributions. First of all, the examinee parameters were obtained from 20-item real test results for the right-skewed and left-skewed sample groups of 500, 1000, 2500, 5000, and 10000. In the second phase of the study, four different tests were formed according to the b parameter values: normal, uniform, left skewed and right skewed distributions. A total of 80 conditions were formed within the scope of this research by selecting 20-item and 30-item condition as the test length variable. In determining the measurement precision, the RMSE and AAD values were calculated. The results were evaluated in terms of the item difficulty distributions, sample sizes, and test lengths. As a result, in right-skewed examinee distribution, the highest measurement precision was obtained at the normal b distribution and the lowest measurement precision was obtained at the right skewed b distribution. A higher measurement precision was obtained in the 30-item test, however, it was observed that the change in the sample size didn't affect the measurement precision significantly in right-skewed examinee distribution. In the left skewed distribution, the highest measurement precision was obtained at the normal b distribution and the lowest measurement precision was obtained at the left-skewed b distribution. Also it was observed that the change in the sample size and test length didn't affect the measurement precision significantly in the left-skewed distribution.