Drasgow, Fritz.
Application of Unidimensional Item Response Theory Models to Multidimensional Data [microform] / Fritz Drasgow and Charles K. Parsons. - [Washington, D.C.] : Distributed by ERIC Clearinghouse, 1982. - 16 p.
ERIC Note: In: Item Response Theory and Computerized Adaptive Testing Conference Proceedings (Wayzata, MN, July 27-30, 1982) (TM 850 744).
The effects of a multidimensional latent trait space on estimation of item and person parameters by the computer program LOGIST are examined. Several item pools were simulated that ranged from truly unidimensional to an inconsequential general latent trait. Item pools with intermediate levels of prepotency of the general latent trait were also constructed. These item pools were used to determine the degree of prepotency that is required by LOGIST in order to recover the general latent trait and not be drawn to a latent trait underlying a cluster of items. The types of multidimensionality studied have several effects on the estimation techniques programmed in LOGIST. Perhaps most important is that as the prepotency of the general factor decreases, LOGIST is gradually drawn to the strongest group factor. Estimates of item difficulty occasionally become excessively large in magnitude when actual data sets are analyzed by LOGIST, although the most recent version has options that may reduce this problem. The results obtained here indicate that this phenomenon may partially be due to multidimensional item pools. However, unidimensional models do provide a good description of multidimensional data sets when the dominant latent trait is sufficiently prepotent. (PN)
Researchers. ericd
Microfiche.
[Washington D.C.]:
ERIC Clearinghouse
microfiches : positive.
ED264268 ERIC
Computer Simulation.
Computer Software.
Difficulty Level.
Item Analysis.
Item Banks.
Latent Trait Theory.
Mathematical Models.
Maximum Likelihood Statistics.
Psychometrics.
Test Items.
Testing Problems.
Parametric Analysis Unidimensional Scaling LOGIST Computer Program
Reports, Research.
Speeches/Meeting Papers.
Application of Unidimensional Item Response Theory Models to Multidimensional Data [microform] / Fritz Drasgow and Charles K. Parsons. - [Washington, D.C.] : Distributed by ERIC Clearinghouse, 1982. - 16 p.
ERIC Note: In: Item Response Theory and Computerized Adaptive Testing Conference Proceedings (Wayzata, MN, July 27-30, 1982) (TM 850 744).
The effects of a multidimensional latent trait space on estimation of item and person parameters by the computer program LOGIST are examined. Several item pools were simulated that ranged from truly unidimensional to an inconsequential general latent trait. Item pools with intermediate levels of prepotency of the general latent trait were also constructed. These item pools were used to determine the degree of prepotency that is required by LOGIST in order to recover the general latent trait and not be drawn to a latent trait underlying a cluster of items. The types of multidimensionality studied have several effects on the estimation techniques programmed in LOGIST. Perhaps most important is that as the prepotency of the general factor decreases, LOGIST is gradually drawn to the strongest group factor. Estimates of item difficulty occasionally become excessively large in magnitude when actual data sets are analyzed by LOGIST, although the most recent version has options that may reduce this problem. The results obtained here indicate that this phenomenon may partially be due to multidimensional item pools. However, unidimensional models do provide a good description of multidimensional data sets when the dominant latent trait is sufficiently prepotent. (PN)
Researchers. ericd
Microfiche.
[Washington D.C.]:
ERIC Clearinghouse
microfiches : positive.
ED264268 ERIC
Computer Simulation.
Computer Software.
Difficulty Level.
Item Analysis.
Item Banks.
Latent Trait Theory.
Mathematical Models.
Maximum Likelihood Statistics.
Psychometrics.
Test Items.
Testing Problems.
Parametric Analysis Unidimensional Scaling LOGIST Computer Program
Reports, Research.
Speeches/Meeting Papers.