Files in this item



application/pdf8803147.pdf (3MB)Restricted to U of Illinois
(no description provided)PDF


Title:Refinement of Stout's Procedure for Assessing Latent Trait Unidimensionality
Author(s):Nandakumar, Ratna
Doctoral Committee Chair(s):Stout, William F.
Department / Program:Education
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Education, Tests and Measurements
Psychology, Psychometrics
Abstract:Currently, the most popular method of measuring an individual's ability is based on Item Response Theory (IRT). One of the most critical assumptions on which this IRT methodology is based is "unidimensionality". In general this assumption means that the items measure one and only one dimension or ability. In practice, however, this assumption cannot be strictly met because a multitude of factors influence test performance. What is actually required is that only one dominant trait influence test performance as a whole.
Stout (1982) has developed a statistical significance test procedure for assessing the unidimensionality of a set of items consistent with the notion of counting the number of dominant traits being measured. The purpose of this thesis has been to complete a detailed investigation and modification of Stout's procedure based on theoretical and empirical reasoning.
It was shown that unacceptably large bias occurs in the value of the statistic when most of the items in a test have high discrimination power. Three methods were proposed to correct for this bias, out of which one method, called Difficulty Check, worked. An algorithm was developed to determine the size of the subtests needed to compute Stout's statistic and a nonparametric index called AHAT has been developed as a crude estimate of item discrimination parameter and compared with other indices in the literature.
Based on Monte Carlo simulations it was observed that the bias was completely eliminated. The procedure, in the case of unidimensionality, adheres to the desired level of significance, and the power of the statistical test, in the case of multidimensionality, is very good even when the correlation between the abilities is as high as 0.5.
Based on this study, it is evident that Stout's procedure can be used for assessing unidimensionality of tests in many practical situations. Further research is recommended regarding the application of Stout's procedure for other purposes of educational measurement like detecting test bias and test equating.
Issue Date:1987
Description:105 p.
Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 1987.
Other Identifier(s):(UMI)AAI8803147
Date Available in IDEALS:2014-12-15
Date Deposited:1987

This item appears in the following Collection(s)

Item Statistics