One of the main problems in cluster analysis is that of determining the number of groups in the data. In general, the approach taken depends on the cluster method used. For K-means, some of the most widely employed criteria are formulated in terms of the decomposition of the total point scatter, regarding a two-mode data set of N points in p dimensions, which are optimally arranged into K classes. This paper addresses the formulation of criteria to determine the number of clusters, in the general situation in which the available information for clustering is a one-mode $N\times N$\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$N\times N$$\end{document} dissimilarity matrix describing the objects. In this framework, p and the coordinates of points are usually unknown, and the application of criteria originally formulated for two-mode data sets is dependent on their possible reformulation in the one-mode situation. The decomposition of the variability of the clustered objects is proposed in terms of the corresponding block-shaped partition of the dissimilarity matrix. Within-block and between-block dispersion values for the partitioned dissimilarity matrix are derived, and variance-based criteria are subsequently formulated in order to determine the number of groups in the data. A Monte Carlo experiment was carried out to study the performance of the proposed criteria. For simulated clustered points in p dimensions, greater efficiency in recovering the number of clusters is obtained when the criteria are calculated from the related Euclidean distances instead of the known two-mode data set, in general, for unequal-sized clusters and for low dimensionality situations. For simulated dissimilarity data sets, the proposed criteria always outperform the results obtained when these criteria are calculated from their original formulation, using dissimilarities instead of distances.

This paper considers the reflection unidentifiability problem in confirmatory factor analysis (CFA) and the associated implications for Bayesian estimation. We note a direct analogy between the multimodality in CFA models that is due to all possible column sign changes in the matrix of loadings and the multimodality in finite mixture models that is due to all possible relabelings of the mixture components. Drawing on this analogy, we derive and present a simple approach for dealing with reflection in variance in Bayesian factor analysis. We recommend fitting Bayesian factor analysis models without rotational constraints on the loadings—allowing Markov chain Monte Carlo algorithms to explore the full posterior distribution—and then using a relabeling algorithm to pick a factor solution that corresponds to one mode. We demonstrate our approach on the case of a bifactor model; however, the relabeling algorithm is straightforward to generalize for handling multimodalities due to sign invariance in the likelihood in other factor analysis models.

The ideal point classification (IPC) model was originally proposed for analysing multinomial data in the presence of predictors. In this paper, we studied properties of the IPC model for analysing bivariate binary data with a specific focus on three quantities: (1) the marginal probabilities; (2) the association structure between the two binary responses; and (3) the joint probabilities. We found that the IPC model with a specific class point configuration represents either the marginal probabilities or the association structure. However, the IPC model is not able to represent both quantities at the same time. We then derived a new parametrization of the model, the bivariate IPC (BIPC) model, which is able to represent both the marginal probabilities and the association structure. Like the standard IPC model, the results of the BIPC model can be displayed in a biplot, from which the effects of predictors on the binary responses and on their association can be read. We will illustrate our findings with a psychological example relating personality traits to depression and anxiety disorders.

A penalized likelihood (PL) method for structural equation modeling (SEM) was proposed as a methodology for exploring the underlying relations among both observed and latent variables. Compared to the usual likelihood method, PL includes a penalty term to control the complexity of the hypothesized model. When the penalty level is appropriately chosen, the PL can yield an SEM model that balances the model goodness-of-fit and model complexity. In addition, the PL results in a sparse estimate that enhances the interpretability of the final model. The proposed method is especially useful when limited substantive knowledge is available for model specifications. The PL method can be also understood as a methodology that links the traditional SEM to the exploratory SEM (Asparouhov & Muthén in Struct Equ Model Multidiscipl J 16:397–438, 2009). An expectation-conditional maximization algorithm was developed to maximize the PL criterion. The asymptotic properties of the proposed PL were also derived. The performance of PL was evaluated through a numerical experiment, and two real data illustrations were presented to demonstrate its utility in psychological research.

A Generalized INDCLUS model, termed GINDCLUS, is presented for clustering three-way two-mode proximity data. In order to account for the heterogeneity of the data, both a partition of the subjects into homogeneous classes and a covering of the objects into groups are simultaneously determined. Furthermore, the availability of information which is external to the three-way data is exploited to better account for such heterogeneity: the weights of both classifications are linearly linked to external variables allowing for the identification of meaningful classes of subjects and groups of objects. The model is fitted in a least-squares framework, and an efficient Alternating Least-Squares algorithm is provided. An extensive simulation study and an application on benchmark data are also presented.

The multiple-group categorical factor analysis (FA) model and the graded response model (GRM) are commonly used to examine polytomous items for differential item functioning to detect possible measurement bias in educational testing. In this study, the multiple-group categorical factor analysis model (MC-FA) and multiple-group normal-ogive GRM models are unified under the common framework of discretization of a normal variant. We rigorously justify a set of identified parameters and determine possible identifiability constraints necessary to make the parameters just-identified and estimable in the common framework of MC-FA. By doing so, the difference between categorical FA model and normal-ogive GRM is simply the use of two different sets of identifiability constraints, rather than the seeming distinction between categorical FA and GRM. Thus, we compare the performance on DIF assessment between the categorical FA and GRM approaches through simulation studies on the MC-FA models with their corresponding particular sets of identifiability constraints. Our results show that, under the scenarios with varying degrees of DIF for examinees of different ability levels, models with the GRM type of identifiability constraints generally perform better on DIF detection with a higher testing power. General guidelines regarding the choice of just-identified parameterization are also provided for practical use.

Model selection is a popular strategy in structural equation modeling (SEM). To select an “optimal” model, many selection criteria have been proposed. In this study, we derive the asymptotics of several popular selection procedures in SEM, including AIC, BIC, the RMSEA, and a two-stage rule for the RMSEA (RMSEA-2S). All of the results are derived under weak distributional assumptions and can be applied to a wide class of discrepancy functions. The results show that both AIC and BIC asymptotically select a model with the smallest population minimum discrepancy function (MDF) value regardless of nested or non-nested selection, but only BIC could consistently choose the most parsimonious one under nested model selection. When there are many non-nested models attaining the smallest MDF value, the consistency of BIC for the most parsimonious one fails. On the other hand, the RMSEA asymptotically selects a model that attains the smallest population RMSEA value, and the RESEA-2S chooses the most parsimonious model from all models with the population RMSEA smaller than the pre-specified cutoff. The empirical behavior of the considered criteria is also illustrated via four numerical examples.

Functional principal component analysis (FPCA) and functional multiple-set canonical correlation analysis (FMCCA) are data reduction techniques for functional data that are collected in the form of smooth curves or functions over a continuum such as time or space. In FPCA, low-dimensional components are extracted from a single functional dataset such that they explain the most variance of the dataset, whereas in FMCCA, low-dimensional components are obtained from each of multiple functional datasets in such a way that the associations among the components are maximized across the different sets. In this paper, we propose a unified approach to FPCA and FMCCA. The proposed approach subsumes both techniques as special cases. Furthermore, it permits a compromise between the techniques, such that components are obtained from each set of functional data to maximize their associations across different datasets, while accounting for the variance of the data well. We propose a single optimization criterion for the proposed approach, and develop an alternating regularized least squares algorithm to minimize the criterion in combination with basis function approximations to functions. We conduct a simulation study to investigate the performance of the proposed approach based on synthetic data. We also apply the approach for the analysis of multiple-subject functional magnetic resonance imaging data to obtain low-dimensional components of blood-oxygen level-dependent signal changes of the brain over time, which are highly correlated across the subjects as well as representative of the data. The extracted components are used to identify networks of neural activity that are commonly activated across the subjects while carrying out a working memory task.

The elicitation of an ordinal judgment on multiple alternatives is often required in many psychological and behavioral experiments to investigate preference/choice orientation of a specific population. The Plackett–Luce model is one of the most popular and frequently applied parametric distributions to analyze rankings of a finite set of items. The present work introduces a Bayesian finite mixture of Plackett–Luce models to account for unobserved sample heterogeneity of partially ranked data. We describe an efficient way to incorporate the latent group structure in the data augmentation approach and the derivation of existing maximum likelihood procedures as special instances of the proposed Bayesian method. Inference can be conducted with the combination of the Expectation-Maximization algorithm for maximum a posteriori estimation and the Gibbs sampling iterative procedure. We additionally investigate several Bayesian criteria for selecting the optimal mixture configuration and describe diagnostic tools for assessing the fitness of ranking distributions conditionally and unconditionally on the number of ranked items. The utility of the novel Bayesian parametric Plackett–Luce mixture for characterizing sample heterogeneity is illustrated with several applications to simulated and real preference ranked data. We compare our method with the frequentist approach and a Bayesian nonparametric mixture model both assuming the Plackett–Luce model as a mixture component. Our analysis on real datasets reveals the importance of an accurate diagnostic check for an appropriate in-depth understanding of the heterogenous nature of the partial ranking data.

We construct a framework for meta-analysis and other multi-level data structures that codifies the sources of heterogeneity between studies or settings in treatment effects and examines their implications for analyses. The key idea is to consider, for each of the treatments under investigation, the subject’s potential outcome in each study or setting were he to receive that treatment. We consider four sources of heterogeneity: (1) response inconsistency, whereby a subject’s response to a given treatment would vary across different studies or settings, (2) the grouping of nonequivalent treatments, where two or more treatments are grouped and treated as a single treatment under the incorrect assumption that a subject’s responses to the different treatments would be identical, (3) nonignorable treatment assignment, and (4) response-related variability in the composition of subjects in different studies or settings. We then examine how these sources affect heterogeneity/homogeneity of conditional and unconditional treatment effects. To illustrate the utility of our approach, we re-analyze individual participant data from 29 randomized placebo-controlled studies on the cardiovascular risk of Vioxx, a Cox-2 selective nonsteroidal anti-inflammatory drug approved by the FDA in 1999 for the management of pain and withdrawn from the market in 2004.

Whether, when, and why perceivers are able to accurately infer the personality traits of other individuals is a key topic in psychological science. Studies examining this question typically ask a number of perceivers to judge a number of targets with regard to a specific trait. The resulting data are then analyzed by averaging the judgments across perceivers or by computing the respective statistic for each single perceiver. Here, we discuss the limitations of the average-perceiver and single-perceiver approaches. Furthermore, we argue that and illustrate how cross-classified structural equation models can be used for the flexible analysis of accuracy data.

Parameter recovery and item utilization were investigated for different designs for online test item calibration. The design was adaptive in a double sense: it assumed both adaptive testing of examinees from an operational pool of previously calibrated items and adaptive assignment of field-test items to the examinees. Four criteria of optimality for the assignment of the field-test items were used, each of them based on the information in the posterior distributions of the examinee’s ability parameter during adaptive testing as well as the sequentially updated posterior distributions of the field-test item parameters. In addition, different stopping rules based on target values for the posterior standard deviations of the field-test parameters and the size of the calibration sample were used. The impact of each of the criteria and stopping rules on the statistical efficiency of the estimates of the field-test parameters and on the time spent by the items in the calibration procedure was investigated. Recommendations as to the practical use of the designs are given.

This commentary concerns the theoretical properties of the estimation procedure in “A General Method of Empirical Q-matrix Validation” by Jimmy de la Torre and Chia-Yi Chiu. It raises the consistency issue of the estimator, proposes some modifications to it, and also makes some conjectures.

This rejoinder responds to the commentary by Liu (Psychometrika, 2015) entitled “On the consistency of Q-matrix estimation: A commentary” on the paper “A general method of empirical Q-matrix validation” by de la Torre and Chiu (Psychometrika, 2015). It discusses and addresses three concerns raised in the commentary, namely the estimation accuracy when a provisional Q-matrix is used, the consistency of the Q-matrix estimator, and the computational efficiency of the proposed method.