A Comparison of the Fit of Empirical Data to Two Latent Trait Models. Report No. 92.

Autor/in	Hutten, Leah R.
Institution	Massachusetts Univ., Amherst. School of Education.
Titel	A Comparison of the Fit of Empirical Data to Two Latent Trait Models. Report No. 92.
Quelle	(1979), (32 Seiten)Verfügbarkeit
Sprache	englisch
Dokumenttyp	gedruckt; Monographie
Schlagwörter	Ability Identification; Achievement Tests; College Entrance Examinations; Comparative Analysis; Elementary Secondary Education; Equated Scores; Factor Analysis; Goodness of Fit; Item Analysis; Mathematical Models; Raw Scores; Research Reports; Sample Size; Statistical Analysis; Statistical Data; Test Interpretation; Test Length + Suchen Sie Ihr Suchwort? Achievement test; Achievement; Testing; Test; Tests; Leistungsbeurteilung; Leistungsüberprüfung; Leistung; Testdurchführung; Testen; Aufnahmeprüfung; Faktorenanalyse; Itemanalyse; Mathematical model; Mathematisches Modell; Research report; Forschungsbericht; Statistische Analyse; Test analysis; Testauswertung
Abstract	Goodness of fit of raw test score data were compared, using two latent trait models: the Rasch model and the Birnbaum three-parameter logistic model. Data were taken from various achievement tests and the Scholastic Aptitude Test (Verbal). A minimum sample size of 1,000 was required, and the minimum test length was 40 items. Results indicated that for data having items equal in discrimination, the Rasch model provided better fit to empirical data. A method for determining equality of item discrimination, using classical point-biserials, was suggested. Results suggested that the maximum likelihood estimate of discrimination may be inadequate at this time. Generally, guessing could not be estimated for this data. Since guessing and discrimination were thus confounded, it was impossible to determine whether use of the guessing parameter might have improved fit using Birnbaum's model. Results also indicated: (1) in both models, fit improved when data showed extremely strong first factor variance; (2) ability estimates from short tests were reasonably good, but item estimates from small person samples were not; and (3) when the LOGIST computer program was used with known item parameters, the cost of estimation was equivalent. In estimating item parameters simultaneously with ability, the savings using the Rasch model was considerable. (Author/GDC)
Erfasst von	ERIC (Education Resources Information Center), Washington, DC