Examining multidimensional structure in view of unidimensional and multidimensional item response theory Tek boyutlu ve çok boyutlu madde tepki kuramına göre çok boyutlu yapıların incelenmesi

Creative Commons License

GÜL E., Koç N.

Hacettepe Egitim Dergisi, vol.32, no.2, pp.312-326, 2017 (Scopus) identifier

  • Publication Type: Article / Article
  • Volume: 32 Issue: 2
  • Publication Date: 2017
  • Doi Number: 10.16986/huje.2016019932
  • Journal Name: Hacettepe Egitim Dergisi
  • Journal Indexes: Scopus, TR DİZİN (ULAKBİM)
  • Page Numbers: pp.312-326
  • Keywords: Dimensionality, Multidimensional data, Multidimensional item response theory, Unidimensional item response theory
  • Hakkari University Affiliated: Yes


The purpose of the study is to compare item-person parameters that are estimated based on unidimensional and multidimensional Item Response Theory (IRT) of dichotomously scored multidimensional constructs, under the following conditions: different sampling sizes, inter-dimensional correlation and the number of dimensions. The study which is carried out with simulative data is an example of basic research. The standard errors of the item-person parameters estimated on the basis of the both models are evaluated with root mean square error. According to the research findings, the two-dimensional data constructs and the errors in the item parameter values obtained from unidimensional and multidimensional Item Response Theory, where the inter-dimensional correlation value is found high, do not significantly vary. Additionally, the number of errors in the item parameter values obtained from unidimensional IRT, in which there is three-dimensional and five-dimensional data construct increases. The standard errors of the item parameters have lower values as the sampling size increases. When the standard errors obtained from the person parameter estimations are considered, it is seen that multidimensional IRT under any conditions estimates with lower errors. As a result of the study, it is concluded that multidimensional IRT is found to give better results in the analysis of multidimensional constructs and to give more accurate results than unidimensional IRT, particularly in person parameter estimations and decision making. It could be advisable to employ multidimensional models since large scale assessments on national and international scales have various sub-dimensions for personal skills estimation.