This paper describes the graded response model. The graded response model represents a family of mathematical models that deal with ordered polytomous categories, such as: (1) letter grading; (2) an attitude survey with "strongly disagree, disagree, agree, and strongly agree" choices; (3) partial credit given in accord with an…

It is shown that it is not easy to apply the EM algorithm to latent class models in the general case with equality constraints because a nonlinear equation has to be solved. A simpler condition is given in which the EM algorithm can be easily applied. (SLD)

The perturbation theory of the generalized eigenproblem is used to derive influence functions of each squared canonical correlation coefficient and the corresponding canonical vector pair. Three sample versions of these functions are described, and some properties are noted. Two obvious applications, multiple correlation and correspondence…

Multiple imputations for latent variables are constructed so that analyses treating them as true variables have the correct expectations for population characteristics. Analyzing multiple imputations in accordance with their construction yields correct estimates of population characteristics, whereas analyzing them as multiple indicators generally…

A new version of the standardized estimated parameter change that is invariant to the original metrics of the measured and latent variables is suggested for use in model modification. A multivariate estimated parameter change for a set of fixed parameters to be freed simultaneously is also introduced. (SLD)

Two ideal-point probabilistic choice models from the external analysis of preferences are presented that allow for more flexible distributions of ideal-points. The first extends the probit model of W. Kamakura and R. Srivastava. The second is based on simplifying assumptions that lead to a multidimensional histogram of ideal-points. (SLD)

How longitudinal factor score estimation--the estimation of the evolution of factor scores for individual examinees over time--can profit from the Kalman filter technique is described. The Kalman estimates change more cautiously over time, have lower estimation error variances, and reproduce the LISREL program latent state correlations more…

Type I error rates were estimated for three tests that compare means by using data from two independent samples: the independent samples t test, Welch's approximate degrees of freedom test, and James's second order test. Type I error rates were estimated for skewed distributions, equal and unequal variances, equal and unequal sample sizes, and a…

In multiple regression analysis, where resulting predictive equation effectiveness is subject to shrinkage, it is especially important to evaluate result replicability. Double cross-validation is an empirical method by which an estimate of invariance or stability can be obtained from research data. A procedure for double cross-validation is…

Point-estimate and confidence-interval procedures are proposed for the (1) combined Mann-Whitney U test; (2) combined two-sample normal-scores test; (3) combined matched-pair Wilcoxon signed-ranks test; and (4) combined matched-pair normal-scores test. The four techniques are illustrated with hypothetical examples and are considered either…

The use of a covariate for randomized response (RRT) research has been shown to reduce standard errors of sensitive trait proportion estimates. At the same time, the model has been shown to be subject to serious misspecification when the relationship between the covariate and the sensitive trait is non-monotonic. The RRT covariate model is adapted…

The problem of combining information to estimate standardized partial regression coefficients in a linear model is considered. A combined estimate obtained from the pooled correlation matrix is proposed, and its large sample distribution is obtained. This estimate can be generalized to address situations in which not every study measures every…

Supporting the use of separate analyses of measurement and structural portions of latent or mixed manifest and latent variable models, limited information (single equation) procedures are presented for estimating structural parameters. These procedures are recommended for testing specific causal hypotheses and locating specific structural model…

The problems posed for latent variable analysis by populations with different sets of parameter values are discussed. An overview of methodology to address heterogeneity is presented, including a review of structural modeling by the multiple indicator-multiple causes model and the use of the LISCOMP computer program. (SLD)

Two methods of estimating parameters in the Rasch model are compared. The equivalence of likelihood estimations from the model of G. J. Mellenbergh and P. Vijn (1981) and from usual unconditional maximum likelihood (UML) estimation is demonstrated. Mellenbergh and Vijn's model is a convenient method of calculating UML estimates. (SLD)