A comparison of item-fit statistics for the three-parameter logistic model

Cornelis A.W. Glas, Juan Carlos Suarez Falcon

Research output: Contribution to journalArticleAcademic

80 Citations (Scopus)

Abstract

In this article, the Type I error rate and the power of a number of existing and new tests of fit to the 3-parameter logistic model (3PLM) are investigated. The first test is a generalization of a test for the evaluation of the fit to the 2-parameter logistic model (2PLM) based on the Lagrange multiplier (LM) test or the equivalent efficient score test. This technique is applied to two model violations: deviation from the 3PLM item characteristic curve and violation of local stochastic independence. The LM test for the first violation is compared with the Q1 – G²j and S – G²j tests, respectively. The LM test for the second violation is compared with the Q3 test and a new test, the S3 test, which can be viewed as a generalization of the approach of the S – G²j test to the evaluation of violation of local independence. The results of simulation studies indicate that all tests, except the Q1 – G²j test, have a Type I error rate that is acceptably close to the nominal significance level, and good power to detect the model violations they are targeted at. When, however, misfitting items are present in a test, the proportion of items that are flagged incorrectly as misfitting can become undesirably high, especially for short tests.
Original languageUndefined
Pages (from-to)81-100
Number of pages20
JournalApplied psychological measurement
Volume27
Issue number2
DOIs
Publication statusPublished - 2003

Keywords

  • IR-60161
  • METIS-215710
  • modification indices
  • item response theory (item fit)
  • item fit
  • efficient score test
  • power studies
  • three-parameter logistic model
  • Lagrange multiplier test
  • Type I error
  • Goodness of Fit
  • Chi-square distribution

Cite this