Box's method of approximation for the null distributions of likelihood criteria is described. It simplifies the formulas, describes a method of obtaining f, phi, and rho directly from given values, and provides two illustrations of the method. (DJ)

Most articles on missing values assume the missing data are "missing at random" and ignore the process that "caused" the missing values. The condition under which this procedure is justified is explored here: the concept of missing at random is precisely defined, several examples are discussed, and two simple conditions are given which are…

In multivariate analysis of variance the canonical variates of one effect may be correlated with the canonical variates of another effect. When the two effects are an interaction and a main effect this correlation interferes with the interpretation of the main effect. When the two effects are both main effects the interpretation of the common…

The area investigated in the present study is the comparison of the permutation t-test with Student's t-test and the Mann-Whitney U-test. The comparison was made for small samples for three distributions, including a normal distribution, a uniform distribution, and a skewed distribution. The properties of each test compared were the probability of…

The stepped-up reliability coefficient does not have the same standard error as an ordinary correlation coefficient. Fisher's Z -transformation should not be applied to it. Appropriate procedures are suggested. (Author)

Bounds for the tails of Dirichlet integrals are established by showing that each integral as a function of the limits is a Schur function. In particular, it is shown how these bounds apply to the simultaneous analysis of variance test and to the multinomial distribution. (Author)

When examining a repeated measures design with independent groups for a significant group by trial interaction, classical analysis of variance or multivariate procedures can be used if the assumptions underlying the tests are met. Neither procedure may be justified for designs with small sample sizes and dichotomous dependent variables. An omnibus…

A summary of the technical problems encountered in performing Bayesian m group regression is given. Grade-point averages for students entering a vocational-technical program are predicted using ability assessments from the Career Planning Profile (CPP), a development of The American College Testing Program (ACT). The theory derived by Lindley (see…

Factorial modeling is justified as a method for analyzing correlations in support of causal inference. The method is illustrated and compared to path analysis, LISREL-type analysis, canonical correlation, and commonality analysis. Predictions of impacts of policy manipulations are demonstrated. (Author/CP)

In large-scale surveys, it is no longer uncommon for repeated measurements to be obtained from respondents and analyses performed to gauge the magnitiude of nonsampling errors. This is particularly true for periodic surveys and longitudinal surveys where a very large investment in data collection is made. This technical note, aimed at the analysis…

A partial least squares method is described for estimating parameters of linear structural relation models. This method is an extension of Herman Wold's proposal for estimation parameters without distributional assumptions, using some algorithms worked out by Paul Horst. The method (LISPLS) determines a different number of latent variables from…

The robustness of the Johnson-Neyman technique and analysis of covariance (ANCOVA) to violations of assumptions of homoscedasticity and homogeneity of variance was tested through the use of Monte Carlo computer procedures. The study simulated a one-way, fixed-effects analysis with two treatment groups, one criterion, Y, and one covariate, X. Five…

A procedure was developed within hypothesis-testing logic that allows researchers to support a hypothesis that has traditionally been the statistical or null hypothesis. Four activities involved in attainment of this goal were discussed: (1) development of statistical logic needed to define the sampling distribution associated with the hypothesis…

A general class of regression models for ordered categorised data is developed and discussed. Thier properties are contrasted with log linear models and other methods of analysis. The general problem of fitting by maximum likelihood is discussed and solved. The models are shown to be multivariate extensions of generalised linear models and the…

Human performance reliability for tasks in the time-space continuous domain is defined and a general mathematical model presented. The human performance measurement terms time-to-error and time-to-error-correction are defined. The model and measurement terms are tested using laboratory vigilance and manual control tasks. Error and error-correction…