Type I errors and power of the parametric bootstrap goodness‐of‐fit test: Full and limited information

In sparse tables for categorical data well‐known goodness‐of‐fit statistics are not chi‐square distributed. A consequence is that model selection becomes a problem. It has been suggested that a way out of this problem is the use of the parametric bootstrap. In this paper, the parametric bootstrap goodness‐of‐fit test is studied by means of an extensive simulation study; the Type I error rates and power of this test are studied under several conditions of sparseness. In the presence of sparseness, models were used that were likely to violate the regularity conditions. Besides bootstrapping the goodness‐of‐fit usually used (full information statistics), corrected versions of these statistics and a limited information statistic are bootstrapped. These bootstrap tests were also compared to an asymptotic test using limited information. Results indicate that bootstrapping the usual statistics fails because these tests are too liberal, and that bootstrapping or asymptotically testing the limited information statistic works better with respect to Type I error and outperforms the other statistics by far in terms of statistical power. The properties of all tests are illustrated using categorical Markov models.     

Limited‐information goodness‐of‐fit testing of item response theory models for sparse 2P tables

Bartholomew and Leung proposed a limited‐information goodness‐of‐fit test statistic (Y) for models fitted to sparse 2^P contingency tables. The null distribution of Y was approximated using a chi‐squared distribution by matching moments. The moments were derived under the assumption that the model parameters were known in advance and it was conjectured that the approximation would also be appropriate when the parameters were to be estimated. Using maximum likelihood estimation of the two‐parameter logistic item response theory model, we show that the effect of parameter estimation on the distribution of Y is too large to be ignored. Consequently, we derive the asymptotic moments of Y for maximum likelihood estimation. We show using a simulation study that when the null distribution of Y is approximated using moments that take into account the effect of estimation, Y becomes a very useful statistic to assess the overall goodness of fit of models fitted to sparse 2^P tables.     

Testing the Rasch model by means of the mixture fit index

Rudas, Clogg, and Lindsay (RCL) proposed a new index of fit for contingency table analysis. Using the overparametrized two‐component mixture, where the first component with weight 1?w represents the model to be tested and the second component with weight w is unstructured, the mixture index of fit was defined to be the smallest w compatible with the saturated two‐component mixture. This index of fit, which is insensitive to sample size, is applied to the problem of assessing the fit of the Rasch model. In this application, use is made of the equivalence of the semi‐parametric version of the Rasch model to specifically restricted latent class models. Therefore, the Rasch model can be represented by the structured component of the RCL mixture, with this component itself consisting of two or more subcomponents corresponding to the classes, and the unstructured component capturing the discrepancies between the data and the model. An empirical example demonstrates the application of this approach. Based on four‐item data, the one‐ and two‐class unrestricted latent class models and the one‐ to three‐class models restricted according to the Rasch model are considered, with respect to both their chi‐squared statistics and their mixture fit indices.     

Influence analysis of structural equation models with polytomous variables

The estimation of model parameters in structural equation models with polytomous variables can be handled by several computationally efficient procedures. However, sensitivity or influence analysis of the model is not well studied. We demonstrate that the existing influence analysis methods for contingency tables or for normal theory structural equation models cannot be applied directly to structural equation models with polytomous variables; and we develop appropriate procedures based on the local influence approach of Cook (1986). The proposed procedures are computationally efficient, the necessary bits of the proposed diagnostic measures are readily available following an usual fit of the model. We consider the influence of an individual cell frequency with respect to three cases: when all parameters in an unstructured model are of interest, when the unstructured polychoric correlations are of interest, and when the structural parameters are of interest. We also consider the sensitivity of the parameters estimates. Two examples based on real data are presented for illustration.The work described in this paper was partially supported by a Chinese University of Hong Kong Postdoctoral Fellows Scheme and a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (RGC Ref. No. CUHK4186/98P). We are indebted to P.M. Bentler and M.D. Newcomb for providing the data set, Michael Leung for his assistance, and the Editor and the referees for some very valuable comments.     

A general approach to categorical data analysis with missing data,using generalized linear models with composite links

A general approach for analyzing categorical data when there are missing data is described and illustrated. The method is based on generalized linear models with composite links. The approach can be used (among other applications) to fill in contingency tables with supplementary margins, fit loglinear models when data are missing, fit latent class models (without or with missing data on observed variables), fit models with fused cells (including many models from genetics), and to fill in tables or fit models to data when variables are more finely categorized for some cases than others. Both Newton-like and EM methods are easy to implement for parameter estimation.The author thanks the editor, the reviewers, Laurie Hopp Rindskopf, and Clifford Clogg for comments and suggestions that substantially improved the paper.     

Bias-corrected estimation of the Rudas–Clogg–Lindsay mixture index of fit

Rudas, Clogg, and Lindsay (1994, J. R Stat Soc. Ser. B, 56, 623) introduced the so-called mixture index of fit, also known as pi-star (π*), for quantifying the goodness of fit of a model. It is the lowest proportion of ‘contamination’ which, if removed from the population or from the sample, makes the fit of the model perfect. The mixture index of fit has been widely used in psychometric studies. We show that the asymptotic confidence limits proposed by Rudas et al. (1994, J. R Stat Soc. Ser. B, 56, 623) as well as the jackknife confidence interval by Dayton ( 2003 , Br. J. Math. Stat. Psychol., 56, 1) perform poorly, and propose a new bias-corrected point estimate, a bootstrap test and confidence limits for pi-star. The proposed confidence limits have coverage probability much closer to the nominal level than the other methods do. We illustrate the usefulness of the proposed method in practice by presenting some practical applications to log-linear models for contingency tables.     

Sample size and bentler and Bonett's nonnormed fit index

Bentler and Bonett's nonnormed fit index is a widely used measure of goodness of fit for the analysis of covariance structures. This note shows that contrary to what has been claimed the nonnormed fit index is dependent on sample size. Specifically for a constant value of a fitting function, the nonnormed index is inversely related to sample size. A simple alternative fit measure is proposed that removes this dependency. In addition, it is shown that this new measure as well as the old nonnormed fit index can be applied to any fitting function that measures the deviation of the observed covariance matrix from the covariance matrix implied by the parameter estimates for a model.     

Bayesian analysis of structural equation models with mixed exponential family and ordered categorical data

Structural equation models are very popular for studying relationships among observed and latent variables. However, the existing theory and computer packages are developed mainly under the assumption of normality, and hence cannot be satisfactorily applied to non‐normal and ordered categorical data that are common in behavioural, social and psychological research. In this paper, we develop a Bayesian approach to the analysis of structural equation models in which the manifest variables are ordered categorical and/or from an exponential family. In this framework, models with a mixture of binomial, ordered categorical and normal variables can be analysed. Bayesian estimates of the unknown parameters are obtained by a computational procedure that combines the Gibbs sampler and the Metropolis–Hastings algorithm. Some goodness‐of‐fit statistics are proposed to evaluate the fit of the posited model. The methodology is illustrated by results obtained from a simulation study and analysis of a real data set about non‐adherence of hypertension patients in a medical treatment scheme.     

结构方程模型常用拟合指数检验的实质

拟合指数检验是评价结构方程模型(SEM)的重要环节。从协方差结构分析的角度将SEM与传统的回归模型比较,容易理解为什么SEM需要拟合指数。揭示了目前几种流行的拟合指数检验的实质:基于卡方的绝对拟合指数(如RMSEA)检验的实质是重新设定卡方检验的显著性水平(不同于通常的.05),相对拟合指数(如NNFI和CFI)检验的实质是基于虚模型设定均方(卡方与自由度之比)降低到的比例;在NNFI大于临界值后,报告和检验CFI是不必要的。根据研究结果提出了一些方便实用的拟合检验建议。     

Is there an optimal factor structure of the Positive and Negative Syndrome Scale in patients with first‐episode psychosis?

The Positive and Negative Syndrome Scale (PANSS) is the most widely used scale to assess a variety of symptoms in patients with schizophrenia and other psychoses. The factor structure of the PANSS has been examined with confirmatory factor analyses in several studies, but not in a well‐defined first‐episode psychosis sample. The aim of this paper is to examine the statistical fit of five different PANSS models in a first‐episode, non‐affective psychosis sample. Confirmatory factor analyses were performed on PANSS data (n = 588). A main criterion for best fit was defined as the Expected Cross Validation Index (ECVI). No tested model revealed an optimally satisfactory model fit index. The Wallwork/Fortgang five‐factor model demonstrated the most optimal psychometric properties. The corresponding subscales of all evaluated five‐factor models were strongly intercorrelated. The Wallwork/Fortgang five‐factor model was found to be statistically and clinically ideal among patients with first‐episode psychosis. Therefore, we recommend this model in forthcoming studies among patients with first‐episode psychosis. However, to prevent the loss of clinically valuable information on an item level, we do not recommend removing any items from the original form. Our study also implies that the specific choice of model will not have a substantial effect on outcome results in studies on the course and outcome in first‐episode psychosis.     

SEM of another flavour: Two new applications of the supplemented EM algorithm

The supplemented EM (SEM) algorithm is applied to address two goodness‐of‐fit testing problems in psychometrics. The first problem involves computing the information matrix for item parameters in item response theory models. This matrix is important for limited‐information goodness‐of‐fit testing and it is also used to compute standard errors for the item parameter estimates. For the second problem, it is shown that the SEM algorithm provides a convenient computational procedure that leads to an asymptotically chi‐squared goodness‐of‐fit statistic for the ‘two‐stage EM’ procedure of fitting covariance structure models in the presence of missing data. Both simulated and real data are used to illustrate the proposed procedures.     

On specifying the null model for incremental fit indices in structural equation modeling

In structural equation modeling, incremental fit indices are based on the comparison of the fit of a substantive model to that of a null model. The standard null model yields unconstrained estimates of the variance (and mean, if included) of each manifest variable. For many models, however, the standard null model is an improper comparison model. In these cases, incremental fit index values reported automatically by structural modeling software have no interpretation and should be disregarded. The authors explain how to formulate an acceptable, modified null model, predict changes in fit index values accompanying its use, provide examples illustrating effects on fit index values when using such a model, and discuss implications for theory and practice of structural equation modeling.     

Evaluating models of remember-know judgments: Complexity, mimicry, and discriminability

Remember-know judgments provide additional information in recognition memory tests, but the nature of this information and the attendant decision process are in dispute. Competing models have proposed that remember judgments reflect a sum of familiarity and recollective information (the one-dimensional model), are based on a difference between these strengths (STREAK), or are purely recollective (the dual-process model). A choice among these accounts is sometimes made by comparing the precision of their fits to data, but this strategy may be muddied by differences in model complexity: Some models that appear to provide good fits may simply be better able to mimic the data produced by other models. To evaluate this possibility, we simulated data with each of the models in each of three popular remember-know paradigms, then fit those data to each of the models. We found that the one-dimensional model is generally less complex than the others, but despite this handicap, it dominates the others as the best-fitting model. For both reasons, the one-dimensional model should be preferred. In addition, we found that some empirical paradigms are ill-suited for distinguishing among models. For example, data collected by soliciting remember/know/new judgments--that is, the trinary task--provide a particularly weak ground for distinguishing models. Additional tables and figures may be downloaded from the Psychonomic Society's Archive of Norms, Stimuli, and Data, at www.psychonomic.org/archive.     

Applying Occam’s razor in modeling cognition: A Bayesian approach

In mathematical modeling of cognition, it is important to have well-justified criteria for choosing among differing explanations (i.e., models) of observed data. This paper introduces a Bayesian model selection approach that formalizes Occam’s razor, choosing the simplest model that describes the data well. The choice of a model is carried out by taking into account not only the traditional model selection criteria (i.e., a model’s fit to the data and the number of parameters) but also the extension of the parameter space, and, most importantly, the functional form of the model (i.e., the way in which the parameters are combined in the model’s equation). An advantage of the approach is that it can be applied to the comparison of non-nested models as well as nested ones. Application examples are presented and implications of the results for evaluating models of cognition are discussed.     

The Problem with Having Two Watches: Assessment of Fit When RMSEA and CFI Disagree

The root mean square error of approximation (RMSEA) and the comparative fit index (CFI) are two widely applied indices to assess fit of structural equation models. Because these two indices are viewed positively by researchers, one might presume that their values would yield comparable qualitative assessments of model fit for any data set. When RMSEA and CFI offer different evaluations of model fit, we argue that researchers are likely to be confused and potentially make incorrect research conclusions. We derive the necessary as well as the sufficient conditions for inconsistent interpretations of these indices. We also study inconsistency in results for RMSEA and CFI at the sample level. Rather than indicating that the model is misspecified in a particular manner or that there are any flaws in the data, the two indices can disagree because (a) they evaluate, by design, the magnitude of the model's fit function value from different perspectives; (b) the cutoff values for these indices are arbitrary; and (c) the meaning of “good” fit and its relationship with fit indices are not well understood. In the context of inconsistent judgments of fit using RMSEA and CFI, we discuss the implications of using cutoff values to evaluate model fit in practice and to design SEM studies.     

Estimation of nonlinear mixed-effects continuous-time models using the continuous-discrete extended Kalman filter

Many intensive longitudinal measurements are collected at irregularly spaced time intervals, and involve complex, possibly nonlinear and heterogeneous patterns of change. Effective modelling of such change processes requires continuous-time differential equation models that may be nonlinear and include mixed effects in the parameters. One approach of fitting such models is to define random effect variables as additional latent variables in a stochastic differential equation (SDE) model of choice, and use estimation algorithms designed for fitting SDE models, such as the continuous-discrete extended Kalman filter (CDEKF) approach implemented in the dynr R package, to estimate the random effect variables as latent variables. However, this approach's efficacy and identification constraints in handling mixed-effects SDE models have not been investigated. In the current study, we analytically inspect the identification constraints of using the CDEKF approach to fit nonlinear mixed-effects SDE models; extend a published model of emotions to a nonlinear mixed-effects SDE model as an example, and fit it to a set of irregularly spaced ecological momentary assessment data; and evaluate the feasibility of the proposed approach to fit the model through a Monte Carlo simulation study. Results show that the proposed approach produces reasonable parameter and standard error estimates when some identification constraint is met. We address the effects of sample size, process noise variance, and data spacing conditions on estimation results.     

Sensitivity to the Sampling Process Emerges From the Principle of Efficiency

Humans can seamlessly infer other people's preferences, based on what they do. Broadly, two types of accounts have been proposed to explain different aspects of this ability. The first account focuses on spatial information: Agents' efficient navigation in space reveals what they like. The second account focuses on statistical information: Uncommon choices reveal stronger preferences. Together, these two lines of research suggest that we have two distinct capacities for inferring preferences. Here we propose that this is not the case, and that spatial‐based and statistical‐based preference inferences can be explained by the assumption that agents are efficient alone. We show that people's sensitivity to spatial and statistical information when they infer preferences is best predicted by a computational model of the principle of efficiency, and that this model outperforms dual‐system models, even when the latter are fit to participant judgments. Our results suggest that, as adults, a unified understanding of agency under the principle of efficiency underlies our ability to infer preferences.     

Limited Information Goodness-of-fit Testing in Multidimensional Contingency Tables

We introduce a family of goodness-of-fit statistics for testing composite null hypotheses in multidimensional contingency tables. These statistics are quadratic forms in marginal residuals up to order r. They are asymptotically chi-square under the null hypothesis when parameters are estimated using any asymptotically normal consistent estimator. For a widely used item response model, when r is small and multidimensional tables are sparse, the proposed statistics have accurate empirical Type I errors, unlike Pearson's X ². For this model in nonsparse situations, the proposed statistics are also more powerful than X ². In addition, the proposed statistics are asymptotically chi-square when applied to subtables, and can be used for a piecewise goodness-of-fit assessment to determine the source of misfit in poorly fitting models. This research has been supported by the Department of Universities, Research, and Information Society (DURSI) of the Catalan Government, by grant BSO2003-08507 of the Spanish Ministry of Science and Technology, and an NSERC Canada grant. We are grateful to the referees for comments leading to improvements.