This study compared the performance of the LOGIST and BILOG computer programs on item response theory (IRT) based scaling and equating for the Test of English as a Foreign Language (TOEFL) using real and simulated data and two calibration structures. Applications of IRT for the TOEFL program are based on the three-parameter logistic (3PL) model.…

Developed a new regression equation to estimate the mean value of eigenvalues in parallel analysis and studied the performance of the equation in comparison with previously published regression equations through simulation. Performance of the new equation was comparable to that of the LCHF equation of G. Lautenschlager and others (1989). (SLD)

Properties of the pseudolikelihood method of estimating Rasch model item parameters, which is based on comparing responses to pairs of items without regard to other items, are studied. Simulation found pseudolikelihood estimates comparable to conditional and marginal maximum likelihood estimates. (SLD)

The benefits of item response theory (IRT) will only accrue to a testing program to the extent that model assumptions are met. Obtaining accurate item parameter estimates is a critical first step. However, the sample sizes required for stable parameter estimation are often difficult to obtain in practice, particularly for the more complex models.…

The theoretical approaches taken by the LOGIST and BILOG computer programs for estimating the parameters of the three-parameter logistic item response model, and their practical consequences, are compared. Guidelines for choosing between the programs for particular applications are offered. (SLD)

Using computer simulated data, the Type I error rate and statistical power were empirically estimated for several pairwise multiple comparison strategies for situations where population variances differ. Focus was on comparing modified Bonferroni procedures with Dunnett's solutions, and determining whether or not J. P. Shaffer's suggestion of…

The predicted scores produced by regression with (1) single best predictor, (2) two best predictors, (3) all predictors with observed values, and (4) all predictors with or without observed values were compared with variable means as estimates of missing values. The study was conducted in a simulation mode on nx8 data matrices using various levels…

Monte Carlo simulation compared chi-square statistics, parameter estimates, and root mean square error of approximation values using normal and elliptical estimation methods. Three research conditions were imposed on the simulated data: sample size, population contamination percent, and kurtosis. A Bentler-Weeks structural model established the…

A simulation study was conducted to investigate the application of expected a posteriori (EAP) trait estimation in computerized adaptive tests (CAT) based on the partial credit model and compare it with maximum likelihood estimation (MLE). Results show the conditions under which EAP and MLE provide relatively accurate estimation in CAT. (SLD)

Procedures for implementing parallel analysis (PA) criteria in practice were compared, examining regression equation methods that can be used to estimate random data eigenvalues from known values of the sample size and number of variables. More internally accurate methods for determining PA criteria are presented. (SLD)

A method of comparing one-step M-estimates of location for heavy tailed distributions is proposed and investigated. Simulations indicate that the new procedure provides good control over Type I errors and has more power than do some other methods for dealing with heavy tailed distributions. (SLD)

The influence of sample size (N) and model parsimony on a set of 22 goodness of fit indices was investigated, including those typically used in confirmatory factor analysis and some recently developed indices. For sample data simulated from 2 known population data structures, values for 6 of 22 fit indices were reasonably independent of N and were…

An investigation of the effects of randomly missing data in two-predictor regression analyses is described. The differences in the effectiveness of five common treatments of missing data on estimates of R-squared values and each of the two standardized regression weights is also investigated. Bootstrap sample sizes of 50, 100, and 200 were drawn…

When latent class parameters are estimated, maximum likelihood and Pearson chi-square statistics can be derived for assessing the fit of the model to the data. This study used simulated data to compare these two statistics, and is based on mixtures of latent binomial distributions, using data generated from five dichotomous manifest variables.…

Robustness of the school-level item response theoretic (IRT) model to violations of distributional assumptions was studied in a computer simulation. In situations where school-level precision might be acceptable for real school comparisons, expected a posteriori estimates of school ability were robust over a range of violations and conditions.…