Hierarchical cluster analysis is shown to be an effective method for forming scales from sets of items. Comparisons with factor analytic techniques suggest that hierarchical analysis is superior in some respects for scale construction. (Author/JKS)

This article compares two methods for analyzing small sets of repeated measures data under normal and non-normal heteroscedastic conditions: a mixed model approach with the Kenward-Roger correction and a multivariate extension of the modified Brown-Forsythe (BF) test. These procedures differ in their assumptions about the covariance structure of…

Latent variable models exist with continuous, categorical, or both types of latent variables. The role of latent variables is to account for systematic patterns in the observed responses. This article has two goals: (a) to establish whether, based on observed responses, it can be decided that an underlying latent variable is continuous or…

The impact of misspecification on the estimation, testing, and improvement of structural equation models was assessed via a population study in which a prototypical latent variable model was misspecified. Results provide insights into the maximum likelihood estimator versus a limited two-stage least squares estimator in LISREL. (TJH)

A method is proposed for partially analyzing additive ipsative data (PAID). Transforming the PAID according to a developed equation preserves the density of the transformed data, and maximum likelihood estimation can be carried out as usual. Simulation results show that the original structural parameters can be accurately estimated from PAID. (SLD)

Examined the relative performance of two commonly used methods of parameter estimation in factor analysis, maximum likelihood (ML) and ordinary least squares (OLS) through simulation. In situations with a moderate amount of error, ML often failed to recover the weak factor while OLS succeeded. Also presented an example using empirical data. (SLD)

A maximum likelihood approach to correlational outlier identification is introduced and compared to the Mahalanobis D squared and Comrey D statistics through Monte Carlo simulation. Identification performance depends on the nature of correlational outliers and the measure used, but the maximum likelihood approach is the most robust performance…

A method is presented that eliminates some interpretational limitations arising from assumptions implicit in the use of arbitrary rules of thumb to interpret exploratory factor analytic results. The bootstrap method is presented as a way of approximating sampling distributions of estimated factor loadings. Simulated datasets illustrate the…

A mathematical formula is introduced for the effect of integrating data. A method is then derived to eliminate the effect from correlations of variables, including mean composites, thus allowing for a clustering algorithm that requires allocation of variables according to the magnitude of their correlations. Examples illustrate the procedure. (SLD)

Used simulation to compare the ability of maximum likelihood (ML) and generalized least-squares (GLS) estimation to provide theoretic fit in models that are parsimonious representations of a true model. The better empirical fit obtained for GLS, compared with ML, was obtained at the cost of lower theoretic fit. (Author/SLD)

The person-response function (PRF) relates the probability of an individual's correct answer to the difficulty of items measuring the same latent trait. Local deviations of the observed PRF from the expected PRF indicate person misfit. We discuss two new approaches to investigate person fit. The first approach uses kernel smoothing to estimate…

Ordered latent class analysis (OLCA) can be used to approximate unidimensional latent distributions. The main objective of this study is to evaluate the method of OLCA in detecting non-normality of an unobserved continuous variable (i.e., a common factor) used to explain the covariation between dichotomous item-level responses. Using simulation,…

Seven data sets (namely, clinical data on children) were subjected to clustering by seven algorithms--the B-coefficient, Linear Typal Analysis; elementary linkage analysis, Numerical Taxonomy System, Statistical Analysis System hierarchical clustering method, Taxonomy, and Bolz's Type Analysis. The little-known B-coefficient method compared…

In a simulation study, factor analysis and multidimensional item response theory (IRT) models are compared with respect to estimates of item parameters. For multidimensional data, a common factor analysis on the matrix of tetrachoric correlations performs at least as well as the multidimensional IRT model. (SLD)

Use of parallel analysis (PA), a selection rule for the number-of-factors problem, is investigated from the viewpoint of permutation assessment through a Monte Carlo simulation. Results reveal advantages and limitations of PA. Tables of sample eigenvalues are included. (SLD)