All parametric analysis focuses on the "synthetic" variables created by applying weights to "observed" variables, but these synthetic variables are called by different names across methods. This paper explains four ways of computing the synthetic scores in factor analysis: (1) regression scores; (2) M. S. Bartlett's algorithm…

Among the computer-based methods used for the construction of trees such as AID, THAID, CART, and FACT, the only one that uses an algorithm that first grows a tree and then prunes the tree is CART. The pruning component of CART is analogous in spirit to the backward elimination approach in regression analysis. This idea provides a tool in…

A globally optimal solution is presented for a class of functions composed of a linear regression function and a penalty function for the sums of squared regression weights. A completing-the-squares approach is used, rather than calculus, because it yields global minimality easily in two of three cases examined. (SLD)

The Gibbs sampling algorithms presented by M. H. Seltzer (1993) are fully generalized to a broad range of settings in which vectors of random regression parameters in the hierarchical model are assumed multivariate normally or multivariate "t" distributed across groups. The use of a fully Bayesian approach is discussed. (SLD)

R. A. Bailey and J. C. Gower explored approximating a symmetric matrix "B" by another, "C," in the least squares sense when the squared discrepancies for diagonal elements receive specific nonunit weights. A solution is proposed where "C" is constrained to be positive semidefinite and of a fixed rank. (SLD)

A method is presented that simultaneously estimates cluster membership and corresponding regression functions for a sample of observations or subjects. This methodology is presented with the simulated annealing-based algorithm. A set of Monte Carlo analyses is included to demonstrate the performance of the algorithm. (SLD)

A detailed comparison of four computer programs for analyzing hierarchical linear models is presented. The programs are: VARCL; HLM; ML2; and GENMOD. All are compiled, stand-alone, and specialized. All use maximum likelihood (ML) estimation for decomposition of the variance into different parts; and in all cases, computing the ML estimates…

Practical problems with multilevel techniques are discussed. These problems relate to terminology, computer programs employing different algorithms, and interpretations of the coefficients in either one or two steps. The usefulness of hierarchical linear models (HLMs) in common situations in educational research is explored. While elegant, HLMs…

It is unrealistic to suppose that standard item response theory (IRT) models will be appropriate for all new and currently considered computer-based tests. In addition to developing new models, researchers will need to give some attention to the possibility of constructing and analyzing new tests without the aid of strong models. Computerized…

Research on automated scoring is extended by comparing alternative automated systems for scoring a computer simulation of physicians' patient management skills. A regression-based system is more highly correlated with experts' evaluations than a system that uses complex rules to map performances into score levels, but both approaches are feasible.…

An approach is described that reveals the hierarchical test structure (HTS) based on the cognitive demands of the test items, and conducts a linear trait modeling by using the HST elements as item difficulty components. This approach, referred to as the Hierarchical Latent Trait Approach (HLTA), employs an algorithm that allows all test items to…

The use of stepwise regression has been criticized for both interpretive misunderstandings and statistical aberrations. A major statistical problem with stepwise regression and other procedures that involve multiple significance tests is the inflation of the Type I error rate. General approaches to control the family-wise error rate such as the…

An alternating least squares method for iteratively fitting the longitudinal reduced-rank regression model is proposed. For social and behavioral science applications, the developed algorithm does not rely on the assumption of multinomial errors and allows for scaling of input and output variables. (SLD)

The statistical power of analysis of covariance (ANCOVA) and its advantages over simple analysis of variance are examined in some experimental situations, and an algorithm is proposed for its proper application. In nonrandomized experiments, an ANCOVA is generally not a good approach. (SLD)