Post hoc multiple comparison procedures useful in assessing differences in population variability are formulated for three nonparametric alternatives to the parametric Bartlett test. The three nonparametric tests are the generalized Puri K-sample extensions of the Siegel-Tukey, mood, and Klotz tests. (Author/CTM)

The purpose of this paper is to calculate the upper limit of the correlation between normal and dichotomous variables. An empirically obtained correlation should be evaluated in view of this limit, instead of the usual limit of Pearson correlation. (Author)

Statistical issues concerning the analysis of multilevel data common in educational studies are discussed. Computer simulation results are presented to argue for a modification of two currently popular approaches to multilevel or contextual analytical procedures. (JKS)

A Bayesian solution is presented for the Johnson-Neyman problem (whether or not the distance between two regression lines is statistically significant over a finite interval of the independent variable). (Author/CTM)

A study was conducted using generated data sets that contained specified amounts of error to determine empirically which of two large sample approximations for the coefficient of concordance or weighted average tau was more appropriate for various numbers of judges and numbers of objects. (CTM)

The robustness of the statistic for complex contrasts in analysis of variance is compared to the statistic developed by Welch. The Welch statistic is recommended as the benchmark test for complex contrasts. (Author/JKS)

Illustrates a simple nonparametric alternative that can be used to test a hypothesis that two proximity matrices on the same set of variables or objects reflect a similar pattern of high and low entries. (RC)

Disagreements have arisen about the robustness of the t test in normal populations with unequal variances. Employing liberal but objective standards for assessing robustness, it is shown that the t test is not always robust to the assumption of equal population variances even when sample sizes are equal. (Author/JKS)

This paper presents the generalized symmetric means approach to the estimation of population covariances, complete with derivations and examples. Particular attention is paid to the problem of missing data, which is handled very naturally in the incidence sampling framework. (CTM)

The negative bias of biserial correlations in the presence of guessing is demonstrated by Monte Carlo studies, and another estimator is described that is free of such bias. The usual biserial-correlation assumptions and assumptions about guessing are discussed. Brogden's coefficient of selective efficiency and the triserial correlation are…

An Edgeworth approximation for accurate significance probabilities for the Wilcoxon two-sample test is substantially simplified. A method is developed that allows quick calculations of very accurate probabilities. Exact formulas are given for most of the remaining cases, and tables are presented comparing the new simplification to likely…

Two stage multiple-comparison procedures give an exact solution to problems of power and Type I errors, but require equal sample sizes in the first stage. This paper suggests a method of evaluating the experimentwise Type I error probability when the first stage has unequal sample sizes. (Author/BW)

Discusses difficulties encountered in use of the Scheffe procedure to locate the most significant parametric function within a linear statistical model that has been tested and rejected by, for example, analysis of variance. A solution to the problems is presented. (TJH)

Two alternative procedures are described for testing the significance of differences of group means. The first consists of a reduction in the critical value when comparing the largest and smallest means. The other alternative uses the unmodified range test without a preliminary F test. An example is provided. (Author/CTM)

If the quantitative result of a study is observed only when the mean difference is statistically significant, the observed mean difference, variance, and effect size are biased estimators of corresponding population parameters. The exact distribution of sample effect size and the maximum likelihood estimator of effect size are derived. (Author/BW)