In some areas of measurement item parameters should not be modeled as fixed but as random. Examples of such areas are: item sampling, computerized item generation, measurement with substantial estimation error in the item parameter estimates, and grouping of items under a common stimulus or in a common context. A hierarchical version of the…

This study investigated the applicability to computerized adaptive testing (CAT) data of a differential item functioning (DIF) analysis that involves an empirical Bayes (EB) enhancement of the popular Mantel Haenszel (MH) DIF analysis method. The computerized Law School Admission Test (LSAT) assumed for this study was similar to that currently…

This examination of data selection preceding multivariate analysis compares results grained with "gentle" and "draconian" variable elimination. To acquire comparable results, two stages of statistical exploration into an integrated model of motivation, learning strategies, and quality of teaching were used. The goal of the…

The accuracy of Gibbs sampling, a Markov chain Monte Carlo procedure, was considered for estimation of item and ability parameters under the two-parameter logistic model. Memory test data were analyzed to illustrate the Gibbs sampling procedure. Simulated data sets were analyzed using Gibbs sampling and the marginal Bayesian method. The marginal…

(See also TM 501 087, TM 501 088, TM 501 090.)

(See also TM 501 087, TM 501 088, and TM 501 089.)

To test the use of Bayes's theorem to adjust for nonresponse bias, 600 hospitals were used in a simulated sample survey. On the basis of known information on five variables, Bayes's formula correctly predicted the status of 92 of the 100 "nonrespondents" relative to a sixth variable. (Author/BW)

The Multiattribute Utility Model combines subjective goal definition with objective data analysis. Goals are defined, ranked, and weighted. Subjective opinions about their attainment are assigned against decision alternatives considered by the school. Bayesian analysis of data enables revision of prior opinions about the realization of goals…

A class of two-stage models is developed to accommodate three common characteristics of behavioral data: (1) its multivariate nature; (2) the typical small sample size; and (3) the possibility of missing observations. The model, as illustrated, permits estimation of the full spectrum of plausible measurement error structures. (SLD)

Introduces a new technique for estimating the parameters of models with continuous latent data. To streamline presentation of this Markov Chain Monte Carlo (MCMC) method, the Rasch model is used. Also introduces a new sampling-based Bayesian technique, the DA-T-Gibbs sampler. (SLD)

Proposes a modified version of commonly employed item response models in a fully Bayesian framework and obtains inferences under the model using Markov chain Monte Carlo techniques. Demonstrates use of the model in a series of simulations and with operational data from the North Carolina Test of Computer Skills and the Test of Spoken English…

The equating of results from the PC-BILOG computer program to an underlying metric was studied through simulation when a two-parameter item response theory model was used. Results are discussed in terms of the identification problem and implications for test equating. (SLD)

The concept of consensus ordering is defined, and formulas for exact and approximate posterior probabilities for consensus ordering are developed under the assumption of a generalized Mallows' model with a diffuse conjugate prior. These methods are applied to a data set concerning 98 college students. (SLD)

A Bayes estimation procedure for Rasch-type model estimation that has statistical and computational advantages over existing methods is described. It involves constructing posterior distributions based on sample data and artificial data reflecting prior information. Its use for some Rasch-type cases, and how it can improve parameter estimation are…

A unified treatment is presented for conditions that should allow detection of measurement bias using statistical procedures involving only observed or manifest variables. Computational results demonstrate that methods for studying bias that rely exclusively on manifest variables are not generally diagnostic of the presence or absence of…