A method for comparing the cross validated classification accuracies of linear and quadratic classification rules is presented under varying data conditions for the k-group classification problem. With this method, separate-group as well as total-group proportions of correct classifications can be compared for the two rules. McNemar's test for…

The Pearson and likelihood ratio statistics are frequently used for assessing the absolute fit of probability models. Researchers are often interested in comparing fits provided by different models which may have a subsuming or non-subsuming relation. A subsuming relation exists when the parameters of the reduced model form a subset of those…

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…

The testing of a priori contrasts, post hoc contrasts, and experimental error rates are discussed. Methods for controlling the experimental error rate for a set of a priori contrasts tested simultaneously have been developed by Dunnett, Dunn, Sidak, and Krishnaiah. Each of these methods is discussed and contrasted as to applicability, power, and…

The effectiveness of two statistical tests of the area between item response functions (exact signed area and exact unsigned area) estimated in different samples, a measure of differential item functioning (DIF), was compared with Lord's chi square. Lord's chi square was found the most effective in determining DIF. (SLD)

A general model is described which can be used to represent the four common types of meta-analysis: (1) estimation of effect size by combining study outcomes; (2) estimation of effect size by contrasting study outcomes; (3) estimation of statistical significance by combining study outcomes; and (4) estimation of statistical significance by…

Comparisons were made of three Angoff Design V linear equating methods (two forms equated to a common test, two forms predicted by a common test, or two forms used to predict a common test) and Tucker's and R. Levine's linear methods, under common item linear equating with non-equivalent populations. Forms of a professional certification test…

In this paper a Bayesian procedure is developed for the simultaneous estimation of the reading ability and difficulty parameters which are assumed to be factors in reading errors by the multiplicative Poisson Model. According to several criteria, the Bayesian estimates are better than comparable maximum likelihood estimates. (Author/JAZ)

Standard maximum likeliheed estimation was compared using two forms of robust estimation, BIWEIGHT (based on Tukey's Biweight) and AMTJACK (AMT-Robustified Jackknife), and Rasch model person analysis. The two procedures recovered the generating parameters, but Rasch person analysis also helped to identify the nature of a response disturbance. (GDC)

Four methods for determining the dimensionality of a set of test items were compared: (1) linear factor analysis; (2) residual analysis; (3) nonlinear factor analysis; and (4) Bejar's method. Five artificial test data sets (for 40 items and 1500 examinees) were generated, consistent with the three-parameter logistic model and the assumption of…

Despite advances in common factor analysis, a review of 89 studies published in four selected journals between 1963 and 1976 indicated that behavioral scientists preferred principal components analysis, followed by varimax or orthogonal rotation. Resultant row sums of squares of factor matrices from principal component analyses of real data sets…

To examine and clarify background conditions for understanding variables which affect salary, the salary and compensation programs at two industrial and three educational organizations were subjected to a statistical audit. Data were available on 272 employees. Ten compensation variables were studied as having direct or indirect effects on salary:…

A procedure for assessing content-area and total-test dimensionality which uses response function discrepancies (RFD) was studied. Three different versions of the RFD procedure were compared to Bejar's principal axis content-area procedure and Indow and Samejima's exploratory factor analytic technique. The procedures were compared in terms of the…

The TESTFACT computer program and full-information factor analysis of test items were used in a computer simulation conducted to correct for the guessing effect. Full-information factor analysis also corrects for omitted items. The present version of TESTFACT handles up to five factors and 150 items. A preliminary smoothing of the tetrachoric…