Procedures are developed for testing the hypothesis that Cronbach's alpha reliability coefficient is equal for two tests given to the same subjects. (Author/JKS)

Some results on how the Alexander-Govern heteroscedastic analysis of variance (ANOVA) procedure (R. Alexander and D. Govern, 1994) performs under nonnormality are presented. This method can provide poor control of Type I errors in some cases, and in some situations power decreases as differences among the means get large. (SLD)

Discusses how taking a multilevel perspective on item response modeling addresses certain analytic problems in using latent variables as outcomes in regression analyses. Application of the models to measurement error in the dependent variable of a between-student regression illustrates the adequacy of some National Assessment of Educational…

Proposes an analytic approach to standard error and confidence interval estimation of scale reliability with fixed congeneric measures. The method is based on a generally applicable estimator stability evaluation procedure, the delta method. The approach, which combines wide-spread point estimation of composite reliability in behavioral scale…

Investigated the impact of categorization on confirmatory factor analysis parameter estimates, standard errors, and five ad hoc fit indexes through simulation studies. Results replicate some previous studies but also suggest that tests of parameter estimates will be underestimated and the amount of underestimation will increase as saturation…

The focus on the individual that is possible in analyzing behavioral data provides the possibility of investigating sequencing effects. Autocorrelation--as illustrated with classroom data from a previous study--can cause standard procedures to underestimate the magnitude of measurement error. Recommendations are made to reduce the effects of…

Concludes that constant errors of several degrees of visual angle may result if group-mean parameter values are used instead of individually determined parameter values. Because infants are relatively inaccurate in directing fixations toward a selected target, repeated calibration trials are required for collection of sufficient data on individual…

The paradox that a Student t-test based on pretest-posttest differences can attain its greatest power when the difference score reliability is zero was explained by demonstrating that power is not a mathematical function of reliability unless either true score variance or error score variance is constant. (SLD)

Describes a measurement lab that introduces measurement and presents a simple example of how to use error analysis with an obvious illustration of its value. The experiment measures the diameters of pinhole images of the sun and uses them to calculate the heights of the leafy canopy that created the images. (JRH)

A maximum test in which the test statistic is the more extreme of the Brown-Forsythe and in which O'Brien's test statistics are developed, with estimated Type I error rates and power for all three tests. For study conditions, Type I error rates for the maximum test are near the nominal level. (SLD)

Three problems with stepwise research methods are explored. Computer packages may use incorrect degrees of freedom in stepwise computations. In addition, stepwise methods do not identify correctly the best variable set of a given size. A third problem is that stepwise methods tend to capitalize on sampling error. (SLD)

A Statistical Analysis System (SAS) macro procedure for performing a jackknife analysis on structure coefficients in discriminant analysis is described together with issues and caveats about its use in multivariate methods. An example of use of the SAS macro is provided. (SLD)

Proposes a new method, reliability generalization, for meta-analysis. Reliability generalization characterizes the typical reliability of scores for a test across studies, the amount of variability in reliability coefficients, and the sources of this variability. Analysis of 87 reliability coefficients for two scales of the Bem Sex Role Inventory…

A new measure of work time correlates well with the standard self-reported workweek method. However, a closer look reveals that reference periods (last week versus last year) have an impact on some workers in the distribution. (Author)

Yen and Burket suggested that shrinkage in vertical equating cannot be understood apart from multidimensionality. Reviews research on reliability, multidimensionality, and scale shrinkage, and explores issues of practical importance to educators. (SLD)