A computer simulation compared significance tests of correlation coefficients calculated from initial scores, from ranks assigned by the Spearman method, and from three kinds of modified ranks. Implications of findings for the idea that rank correlation is a nonparametric correlation method are discussed. (SLD)

This study examined the sampling distribution of the analysis of variance F ratio in the two sample cases when it followed a preliminary test for variance equality. When the population variances were equal, the sampling distribution approximated the theoretical F distribution quite well, but not when population variances differed. (JAZ)

Effects of different weighting schemes on selecting the optimal number of observations in multivariate-multifacet generalizability designs are studied when cost constraints are imposed. Comparison of four schemes through simulation indicates that all four produce similar optimal values and that reliability should be similar. (SLD)

Rasch model total-fit statistics and between-item fit statistics were compared for their ability to detect measurement disturbances through the use of simulated data. Results indicate that the between-fit statistic appears more sensitive to systematic measurement disturbances and the total-fit statistic is more sensitive to random measurement…

Comparisons of the Wilcoxon test, Friedman test, and repeated-measures analysis of variance (ANOVA) on ranks in a computer simulation show that the Friedman test performs like the sign test whereas the ANOVA performs like the Wilcoxon test. Classification of these tests in introductory statistics textbooks should be revised. (SLD)

The stepwise regression method of selecting predictors for computer assisted multiple regression analysis was compared with forward, backward, and best subsets regression, using 16 data sets. The results indicated the stepwise method was preferred because of its practical nature, when the models chosen by different selection methods were similar…

Three measures of differential item functioning for the dichotomous response model are extended to include Samejima's graded response model. Two are based on area differences between item true score functions, and one is a chi-square statistic for comparing differences in item parameters. (SLD)

The functioning of two population-based Mantel-Haenszel (MH) common-odds ratios was compared. One ratio is conditioned on the observed test score, while the other is conditioned on a latent trait or true ability score. When the comparison group distributions are incongruent or nonoverlapping to some degree, the observed score represents different…

Using simulated and real data, this study compares the performance of three methodologies for assessing unidimensionality: (1) DIMTEST; (2) the approach of Holland and Rosenbaum; and (3) nonlinear factor analysis. All three models correctly confirm unidimensionality, but they differ in their ability to detect the lack of unidimensionality.…

C. L. Olson (1976, 1979) suggests the Pillai-Bartlett trace (V) as an omnibus multivariate analysis of variance (MANOVA) test statistic for its superior robustness to heterogeneous variances. J. Stevens (1979, 1980) contends that the robustness of V, Wilk's lambda (W) and the Hotelling-Lawley trace (T) are similar, and that their power functions…

An attempt was made to determine which item response theory (IRT) equating method results in the least amount of equating error or "scale drift" when equating scores across one or more test forms. An internal anchor test design was employed with five different test forms, each consisting of 30 items, 10 in common with the base test and 5…

A modification of a test of the equality of nonindependent alpha reliability coefficients is proposed. It avoids the limitation that the product of the number of test parts times the number of subjects be quite large. Monte Carlo studies indicate that this test can be used in comparing interrater reliabilities. (SLD)

Rasch between and total weighted and unweighted fit statistics were compared using varying test lengths and sample sizes. Two test lengths (20 and 50 items) and three sample sizes (150, 500, and 1,000 were crossed. Each of the six combinations were replicated 100 times. In addition, power comparisons were made. Results indicated that there were no…

A population data set was randomly generated from which a random sample was drawn. This sample was randomly divided into two data sets, one of which was used to generate parameter estimates, which were then used in the second data set for cross-validation purposes. The best variable subset models were compared between the two data sets on the…

In this paper a quadratic curve equating method for different test forms under a random-group data-collection design is proposed. Procedures for implementing this method and related issues are described and discussed. The quadratic-curve method was evaluated with real test data (from two 30-item subtests for a professional licensure examination…