Describes available variance component tests and presents three new score tests. One test uses the asymptotic normal distribution of the test statistic as a reference distribution; the others use a Satterthwaite approximation for the null distribution of the test statistic. Evaluates the performance of these tests through Monte Carlo simulation.…

Compared the analysis of variance (ANOVA) "F" and several nonparametric competitors for two-way designs for empirical alpha and power through simulation. Results suggest the ANOVA "F" suffers from conservative alpha and power for the mixed normal distribution, but is generally recommended. (Author/SLD)

Methods for calculating confidence intervals for the mean are reviewed for the case where the data come from a log-normal distribution. In a simulation study it is found that a variation of the method suggested by Cox works well in practice. An approach based on Generalized confidence intervals also works well. A comparison of our results with…

The normality of number distributions generated by various random-number generators were studied, focusing on when the random-number generator reached a normal distribution and at what sample size. Findings suggest the steps that should be followed when using a random-number generator in a Monte Carlo simulation. (SLD)

Using simulated data, the MULTILOG and PARSCALE software packages were compared for their recovery of item and trait parameters under the graded response and generalized partial credit item response theory models. The shape of the latent population distribution (normal, skewed, or uniform) and the sample size (250 or 500) were varied. Parameter…

Previous studies that investigated the effect of unequal ability distributions on the Type I error (TIE) of the Mantel-Haenszel chi-square test for detecting differential item functioning (DIF) simulated ability distributions that differed only in means. This simulation study suggests that the magnitude of TIE inflation is increased, and the type…

Studied whether the theoretical sampling distribution of the U3 person-fit statistic is in agreement with the simulated sampling distribution under different item response theory models and varying item and test characteristics. Simulation results suggest that the use of standard normal deviates for the standardized version of the U3 statistic may…

Presents a hierarchical Bayes approach to modeling parameter heterogeneity in generalized linear models. The approach combines the flexibility of semiparametric latent class models that assume common parameters for each subpopulation and the parsimony of random effects models that assume normal distributions for the regression parameters.…

Assessed the relative merits of the Student "t" test and the Wilcoxon rank sum test under four population distributions and six sample-size pairings through Monte Carlo methods. The Wilcoxon rank sum test demonstrated an advantage in statistical power for nonnormal distributions (but not normal distributions), with fewer Type III errors…

This paper evaluated multidimensional linking procedures with which multidimensional test data from two separate calibrations were put on a common scale. Data were simulated with known ability distributions varying on two factors which made linking necessary: mean vector differences and variance-covariance (v-c) matrix differences. After the…

To satisfy the security requirements of computerized adaptive tests (CATs), efforts have been made to control the exposure rates of optimal items directly by incorporating statistical methods into the item selection procedure. Since differences are likely to occur between the exposure control parameter derivation stage and the operational CAT…

A generalization of the usual random-effects model based on trimmed means is proposed. The resulting test of no differences among J randomly sampled groups has advantages in terms of Type I errors and can yield gains in power when distributions have heavy tails and outliers. (SLD)

A person-fit method that allows researchers to identify nonfitting response vectors is the l(z) statistic. Simulation results show that l(z) may not perform as expected when estimated person parameters are used rather than true person parameters. Other considerations in using true and estimated person parameters are discussed. (SLD)

Studied the effects of sample size on the cumulative distribution of selected fit indices using Monte Carlo simulation. Generally, the comparative fit index exhibited very stable patterns and was less influenced by sample size or data types than were other fit indices. (SLD)

The assumption of normality underlies much of the standard statistical methodology. Knowing how to determine whether a sample of measurements is from a normally distributed population is crucial both in the development of statistical theory and in practice. W. Ware and J. Ferron have developed a new test statistic, modeled after the K-squared test…