Modelling monotonic effects of ordinal predictors in Bayesian regression models

Ordinal predictors are commonly used in regression models. They are often incorrectly treated as either nominal or metric, thus under- or overestimating the information contained. Such practices may lead to worse inference and predictions compared to methods which are specifically designed for this purpose. We propose a new method for modelling ordinal predictors that applies in situations in which it is reasonable to assume their effects to be monotonic. The parameterization of such monotonic effects is realized in terms of a scale parameter b representing the direction and size of the effect and a simplex parameter modelling the normalized differences between categories. This ensures that predictions increase or decrease monotonically, while changes between adjacent categories may vary across categories. This formulation generalizes to interaction terms as well as multilevel structures. Monotonic effects may be applied not only to ordinal predictors, but also to other discrete variables for which a monotonic relationship is plausible. In simulation studies we show that the model is well calibrated and, if there is monotonicity present, exhibits predictive performance similar to or even better than other approaches designed to handle ordinal predictors. Using Stan, we developed a Bayesian estimation method for monotonic effects which allows us to incorporate prior information and to check the assumption of monotonicity. We have implemented this method in the R package brms, so that fitting monotonic effects in a fully Bayesian framework is now straightforward.     

An elicitation method for multiple linear regression models

This paper describes a method of quantifying subjective opinion about a normal linear regression model. Opinion about the regression coefficients and experimental error is elicited and modeled by a multivariate probability distribution (a Bayesian conjugate prior distribution). The distribution model is richly parameterized and various assessment tasks are used to estimate its parameters. These tasks include the revision of opinion in the light of hypothetical data, the assessment of credible intervals, and a task commonly performed in cue-weighting experiments. A new assessment task is also introduced. In addition, implementation of the method in an interactive computer program is described and the method is illustrated with a practical example.     

Methods for describing ordinal data with cardinal models

Several methods, both new and old, for describing ordinal data with a cardinal model are discussed in a newly developed and general context which broadens their applicability beyond their traditional use in multidimensional scaling. The relationships between the methods is investigated. It is shown that the two most commonly used methods (Guttman's rank-image principle and Kruskal's least-square monotonic transformation) are the boundary conditions of a newly proposed single parameter family of methods. An additional method is proposed which is shown to yield an equal-density model space. The commonly made distinction between “transformation” methods and “transformation-free” methods is shown to be a pseudodistinction. It is observed that all the methods can be stated as matrix operations on the model space with the conclusion that they try to optimize a linear combination of the model space.     

On the use of heterogeneous thresholds ordinal regression models to account for individual differences in response style

This paper proposes a general approach to accounting for individual differences in the extreme response style in statistical models for ordered response categories. This approach uses a hierarchical ordinal regression modeling framework with heterogeneous thresholds structures to account for individual differences in the response style. Markov chain Monte Carlo algorithms for Bayesian inference for models with heterogeneous thresholds structures are discussed in detail. A simulation and two examples based on ordinal probit models are given to illustrate the proposed methodology. The simulation and examples also demonstrate that failing to account for individual differences in the extreme response style can have adverse consequences for statistical inferences.The author is grateful to Ulf Böckenholt, an associate editor, and three anonymous reviewers for helpful comments, and Kristine Kuhn and Kshiti Joshi for providing the data.     

Evaluation of dominance‐based ordinal multiple regression for variables with few categories

Dominance‐based ordinal multiple regression (DOR) is designed to answer ordinal questions about relationships among ordinal variables. Only one parameter per predictor is estimated, and the number of parameters is constant for any number of outcome levels. The majority of existing simulation evaluations of DOR use predictors that are continuous or ordinal with many categories, so the performance of the method is not well understood for ordinal variables with few categories. This research evaluates DOR in simulations using three‐category ordinal variables for the outcome and predictors, with a comparison to the cumulative logits proportional odds model (POC). Although ordinary least squares (OLS) regression is inapplicable for theoretical reasons, it was also included in the simulations because of its popularity in the social sciences. Most simulation outcomes indicated that DOR performs well for variables with few categories, and is preferable to the POC for smaller samples and when the proportional odds assumption is violated. Nevertheless, confidence interval coverage for DOR was not flawless and possibilities for improvement are suggested.     

Polytomous IRT models and monotone likelihood ratio of the total score

In a broad class of item response theory (IRT) models for dichotomous items the unweighted total score has monotone likelihood ratio (MLR) in the latent trait. In this study, it is shown that for polytomous items MLR holds for the partial credit model and a trivial generalization of this model. MLR does not necessarily hold if the slopes of the item step response functions vary over items, item steps, or both. MLR holds neither for Samejima's graded response model, nor for nonparametric versions of these three polytomous models. These results are surprising in the context of Grayson's and Huynh's results on MLR for nonparametric dichotomous IRT models, and suggest that establishing stochastic ordering properties for nonparametric polytomous IRT models will be much harder.Hemker's research was supported by the Netherlands Research Council, Grant 575-67-034. Junker's research was supported in part by the National Institutes of Health, Grant CA54852, and by the National Science Foundation, Grant DMS-94.04438.     

Graphical regression models for polytomous variables

When modeling the relationship between two nominal categorical variables, it is often desirable to include covariates to understand how individuals differ in their response behavior. Typically, however, not all the relevant covariates are available, with the result that the measured variables cannot fully account for the associations between the nominal variables. Under the assumption that the observed and unobserved variables follow a homogeneous conditional Gaussian distribution, this paper proposesRC(M) regression models to decompose the residual associations between the polytomous variables. Based on Goodman's (1979, 1985)RC(M) association model, a distinctive feature ofRC(M) regression models is that they facilitate the joint estimation of effects due to manifest and omitted (continuous) variables without requiring numerical integration. TheRC(M) regression models are illustrated using data from the High School and Beyond study (Tatsuoka & Lohnes, 1988). This article was accepted for publication, when Willem J. Heiser was the Editor ofPsychometrika. This research was supported by grants from the National Science Foundation (#SBR96-17510 and #SBR94-09531) and the Bureau of Educational Research at the University of Illinois. We thank Jee-Seon Kim for comments and computational assistance.     

IRTree models with ordinal and multidimensional decision nodes for response styles and trait-based rating responses

IRTree models decompose observed rating responses into sequences of theory-based decision nodes, and they provide a flexible framework for analysing trait-related judgements and response styles. However, most previous applications of IRTree models have been limited to binary decision nodes that reflect qualitatively distinct and unidimensional judgement processes. The present research extends the family of IRTree models for the analysis of response styles to ordinal judgement processes for polytomous decisions and to multidimensional parametrizations of decision nodes. The integration of ordinal judgement processes overcomes the limitation to binary nodes, and it allows researchers to test whether decisions reflect qualitatively distinct response processes or gradual steps on a joint latent continuum. The extension to multidimensional node models enables researchers to specify multiple judgement processes that simultaneously affect the decision between competing response options. Empirical applications highlight the roles of extreme and midpoint response style in rating judgements and show that judgement processes are moderated by different response formats. Model applications with multidimensional decision nodes reveal that decisions among rating categories are jointly informed by trait-related processes and response styles.     

A maximum likelihood method for latent class regression involving a censored dependent variable

The standard tobit or censored regression model is typically utilized for regression analysis when the dependent variable is censored. This model is generalized by developing a conditional mixture, maximum likelihood method for latent class censored regression. The proposed method simultaneously estimates separate regression functions and subject membership in K latent classes or groups given a censored dependent variable for a cross-section of subjects. Maximum likelihood estimates are obtained using an EM algorithm. The proposed method is illustrated via a consumer psychology application.     

Two-group classification in latent trait theory: Scores with monotone likelihood ratio

This paper deals with two-group classification when a unidimensional latent trait,, is appropriate for explaining the data,X. It is shown that ifX has monotone likelihood ratio then optimal allocation rules can be based on its magnitude when allocation must be made to one of two groups related to. These groups may relate to probabilistically via a non-decreasing functionp(), or may be defined by all subjects above or below a selected value on.In the case where the data arise from dichotomous items, then only the assumption that the items have nondecreasing item characteristic functions is enough to ensure that the unweighted sum of responses (the number-right score or raw score) possesses this fundamental monotone likelihood ratio property.     

Digit ratios,but not facial width-to-height ratios,are associated with the priority placed on attending to faces versus bodies

Testosterone has been linked to a sexual preference for bodies rather than faces. Low digit ratio (2D:4D) and high facial width-to-height ratio (WHR) are associated with high prenatal testosterone. We tested whether low 2D:4D and high facial WHR were correlated with a preference for bodies over faces. Our sample consisted of 109 college students (64 males). A two-way analysis of variance demonstrated a significant main effect of digit ratio on the priority placed on paying attention to faces or bodies such that low 2D:4D was linked to a preference for bodies, but the effect of sex was not significant. There were no significant interaction effects. Another two-way analysis of variance revealed no significant effects for facial WHR and attentional priority but significant associations between sex and attentional priority. There were no significant interaction effects. Our findings indicate that individuals with low digit ratios tended to pay more attention to bodies than to faces compared with individuals with higher digit ratios, independently of sex. We also found that males tended to pay more attention to bodies than to faces compared with females, independently of facial WHR.     

Bayesian Gaussian distributional regression models for more efficient norm estimation

A test score on a psychological test is usually expressed as a normed score, representing its position relative to test scores in a reference population. These typically depend on predictor(s) such as age. The test score distribution conditional on predictors is estimated using regression, which may need large normative samples to estimate the relationships between the predictor(s) and the distribution characteristics properly. In this study, we examine to what extent this burden can be alleviated by using prior information in the estimation of new norms with Bayesian Gaussian distributional regression. In a simulation study, we investigate to what extent this norm estimation is more efficient and how robust it is to prior model deviations. We varied the prior type, prior misspecification and sample size. In our simulated conditions, using a fixed effects prior resulted in more efficient norm estimation than a weakly informative prior as long as the prior misspecification was not age dependent. With the proposed method and reasonable prior information, the same norm precision can be achieved with a smaller normative sample, at least in empirical problems similar to our simulated conditions. This may help test developers to achieve cost-efficient high-quality norms. The method is illustrated using empirical normative data from the IDS-2 intelligence test.     

A general class of latent variable models for ordinal manifest variables with covariate effects on the manifest and latent variables

Previous work on a general class of multidimensional latent variable models for analysing ordinal manifest variables is extended here to allow for direct covariate effects on the manifest ordinal variables and covariate effects on the latent variables. A full maximum likelihood estimation method is used to estimate all the model parameters simultaneously. Goodness‐of‐fit statistics and standard errors are discussed. Two examples from the 1996 British Social Attitudes Survey are used to illustrate the methodology.     

A two-stage logistic regression model for analyzing inter-rater agreement

Studies of agreement commonly occur in psychiatric research. For example, researchers are often interested in the agreement among radiologists in their review of brain scans of elderly patients with dementia or in the agreement among multiple informant reports of psychopathology in children. In this paper, we consider the agreement between two raters when rating a dichotomous outcome (e.g., presence or absence of psychopathology). In particular, we consider logistic regression models that allow agreement to depend on both rater- and subject-level covariates. Logistic regression has been proposed as a simple method for identifying covariates that are predictive of agreement (Coughlin et al., 1992). However, this approach is problematic since it does not take account of agreement due to chance alone. As a result, a spurious association between the probability (or odds) of agreement and a covariate could arise due entirely to chance agreement. That is, if the prevalence of the dichotomous outcome varies among subgroups of the population, then covariates that identify the subgroups may appear to be predictive of agreement. In this paper we propose a modification to the standard logistic regression model in order to take proper account of chance agreement. An attractive feature of the proposed method is that it can be easily implemented using existing statistical software for logistic regression. The proposed method is motivated by data from the Connecticut Child Study (Zahner et al., 1992) on the agreement among parent and teacher reports of psychopathology in children. In this study, parents and teachers provide dichotomous assessments of a child's psychopathology and it is of interest to examine whether agreement among the parent and teacher reports is related to the age and gender of the child and to the time elapsed between parent and teacher assessments of the child.The authors thank the Associate Editor and the referees for their helpful comments and suggestions. We also thank Gwen Zahner for use of data from the Connecticut Child Study, which was conducted under contract to the Connecticut Department of Children and Youth Services. This research was supported by grants HL 69800, AHRQ 10871, HL52329, HL61769, GM 29745, MH 54693 and MH 17119 from the National Institutes of Health.     

Pairwise likelihood estimation for confirmatory factor analysis models with categorical variables and data that are missing at random

Methods for the treatment of item non-response in attitudinal scales and in large-scale assessments under the pairwise likelihood (PL) estimation framework and under a missing at random (MAR) mechanism are proposed. Under a full information likelihood estimation framework and MAR, ignorability of the missing data mechanism does not lead to biased estimates. However, this is not the case for pseudo-likelihood approaches such as the PL. We develop and study the performance of three strategies for incorporating missing values into confirmatory factor analysis under the PL framework, the complete-pairs (CP), the available-cases (AC) and the doubly robust (DR) approaches. The CP and AC require only a model for the observed data and standard errors are easy to compute. Doubly-robust versions of the PL estimation require a predictive model for the missing responses given the observed ones and are computationally more demanding than the AC and CP. A simulation study is used to compare the proposed methods. The proposed methods are employed to analyze the UK data on numeracy and literacy collected as part of the OECD Survey of Adult Skills.     

A new proof for monotone likelihood ratio for the sum of independent bernoulli random variables

By use of an inequality of Marcus and Lopes for elementary symmetric functions, a new proof is presented for the following result by Ghurye and Wallace: Given that the independent random variablesX _j are Bernoulli with success probabilityp _j () strictly between 0 and 1 and nondecreasing in, the sum X _j has monotone likelihood ratio.     

A novel method for assessing rival models of recognition memory

A general comparison is made between the multinomial processing tree (MPT) approach and a strength-based approach for modeling recognition memory measurement. Strength models include the signal-detection model and the dual-process model. Existing MPT models for recognition memory and a new generic MPT model, called the Multistate (MS) model, are contrasted with the strength models. Although the ROC curves for the MS model and strength model are similar, there is a critical difference between existing strength models and MPT models that goes beyond the assessment of the ROC. This difference concerns the question of stochastic mixtures for foil test trials. The hazard function and the reverse hazard function are powerful methods for detecting the presence of a probabilistic mixture. Several new theorems establish a novel method for obtaining information about the hazard function and reverse hazard function for the latent continuous distributions that are assumed in the strength approach to recognition memory. Evidence is provided that foil test trials involve a stochastic mixture. This finding occurred for both short-term memory procedures, such as the Brown–Peterson task, and long-term list-learning procedures, such as the paired-associate task. The effect of mixtures on foil trials is problematic for existing strength models but can be readily handled by MPT models such as the MS model. Other phenomena, such as the mirror effect and the effect of target-foil similarity, are also predicted accurately by the MPT modeling framework.     

Pseudo maximum likelihood estimation and a test for misspecification in mean and covariance structure models

Using the theory of pseudo maximum likelihood estimation the asymptotic covariance matrix of maximum likelihood estimates for mean and covariance structure models is given for the case where the variables are not multivariate normal. This asymptotic covariance matrix is consistently estimated without the computation of the empirical fourth order moment matrix. Using quasi-maximum likelihood theory a Hausman misspecification test is developed. This test is sensitive to misspecification caused by errors that are correlated with the independent variables. This misspecification cannot be detected by the test statistics currently used in covariance structure analysis.For helpful comments on a previous draft of the paper we are indebted to Kenneth A. Bollen, Ulrich L. Küsters, Michael E. Sobel and the anonymous reviewers of Psychometrika. For partial research support, the first author wishes to thank the Department of Sociology at the University of Arizona, where he was a visiting professor during the fall semester 1987.     

Digit ratios and preferences for aggressive content in entertainment

Digit ratios (2D:4D) contain information concerning an individual’s propensity towards aggression. Our study adds the first clue to better understanding the relationship between 2D:4D and exposure to aggressive contents in entertainment products. Our findings suggest that individuals with low 2D:4D prefer aggressive contents such as action films, sports telecast, killing and achieving games, hip-hop music, and erotic video clips rather than do individuals with high 2D:4D. Also individuals with low 2D:4D tend to demonstrate less preference for romance films than individuals with high 2D:4D. In addition, we found that low 2D:4D was associated with a preference for sports instead of other genres of entertainment products. Therefore, 2D:4D (a putative correlate of prenatal sex steroids) helps us to better understand the rationale of individuals’ preferences for media violence.