Factors influencing the Mantel-Haenszel procedure in the detection of differential item functioning.
Item
-
Title
-
Factors influencing the Mantel-Haenszel procedure in the detection of differential item functioning.
-
Identifier
-
AAI9224863
-
identifier
-
9224863
-
Creator
-
Uttaro, Thomas Edward.
-
Contributor
-
Adviser: Roger Millsap
-
Date
-
1992
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Education, Educational Psychology | Education, Tests and Measurements
-
Abstract
-
A continuing inquiry has focused on the development of statistical procedures for DIF analysis which require smaller sample sizes and are more practical and less costly than IRT based procedures, (Lord, 1980; Rudner, Getson & Knight, 1980; Linn & Harnish, 1981). The application of the Mantel Haenszel procedure to DIF study (Mantel & Haenszel, 1959; Holland & Thayer, 1988) is the most recent of continuing efforts to cultivate chi-square methodology. The Mantel-Haenszel procedure is intended to approximate the theoretically preferred IRT definition of DIF (Lord, 1980). Correspondence has been reported between the MH procedure and IRT methods in several DIF comparison studies (Thissen, Steinberg & Wainer, 1988; Hambleton & Rogers, 1989).;Simulated data sets were generated from IRT DIF models and the MH procedure was applied to recover the IRT DIF. The MH procedure was evaluated in terms of the statistical accuracy of chi-square type I and type II error rates, {dollar}\alpha\sb{lcub}\rm mh{rcub}{dollar} and {dollar}\Delta\sb{lcub}\rm mh{rcub}{dollar} estimators, variances and ETS DIF classifications. The results of the study suggest that there are serious obstacles to the simple, direct application of the Mantel-Haenszel procedure for DIF analyses.;The procedure may lead to inaccuracies given certain item ICC parameterizations of studied items and items in the matching criteria and given discrepancies between focal and reference group {dollar}\Theta{dollar} levels. In the Rasch case the expected behavior of {dollar}\alpha{dollar} is verified but instability of MH{dollar}\sb{lcub}\rm CHISQ{rcub}{dollar} is demonstrated under discrepant ability and matching criterion item difficulty conditions. Factors influencing the empirical and estimated (Phillips & Holland, 1987) variances of {dollar}\alpha{dollar} are examined. It is shown that the estimated variance is sensitive to ICC parameterizations and ability discrepancies. The reliability of total score given known ability ({dollar}\Theta{dollar}) estimates is evaluated and shown to interact with ability discrepancies, influencing the MH procedure.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
