Kazuo Shigemasu

Decline or improvement? Age-related differences in facial expression recognition

We examined age-related differences in facial expression recognition in association with potentially interfering variables such as general cognitive ability (verbal and visuospatial abilities), face recognition ability, and the experiences of positive and negative emotions. Participants comprised 34 older (aged 62-81 years) and 34 younger (aged 18-25 years) healthy Japanese adults. The results showed not only age-related decline in sadness recognition but also age-related improvement in disgust recognition. Among other variables, visuospatial ability was moderately related to facial expression recognition in general, and the experience of negative emotions was related to sadness recognition. Consequently, age-related decline in sadness recognition was statistically explained by age-related decrease in the experience of negative emotions. On the other hand, age-related improvement in disgust recognition was not explained by the interfering variables, and it reflected a higher tendency in the younger participants to mistake disgust for anger. Possible mechanisms are discussed in terms of neurobiological and socio-environmental factors.

Application of the somatic marker hypothesis to individual differences in decision making.

The somatic marker hypothesis (Damasio, Tranel, & Damasio, 1991) is a controversial theory asserting that somatic activities implicitly bias human behavior. In this study, we examined the relationship between choice behaviors in the Iowa Gambling Task and patterns of skin conductance responses (SCRs) within a healthy population. Results showed that low SCRs for appraising the monetary outcome of risky decisions were related to persistence in risky choices. Such adherence to risky decisions was not related to poor explicit knowledge about the task. On the other hand, anticipatory SCRs and the effect of them on performance were not confirmed. Our findings suggest that a variation in covert physiological appraisal underlies individual differences in decision making.

A new discrimination method for the Concealed Information Test using pretest data and within-individual comparisons.

A latent class discrimination method is proposed for analyzing autonomic responses on the concealed information test. Because there are significant individual differences in autonomic responses, individual response patterns are estimated on the pretest. Then an appropriate discriminant formula for the response pattern of each individual is applied to the CIT test results. The probability that the individual concealed information is calculated by comparing the discriminant formula value of the crime-related item to that of non-crime-related items. The discrimination performance of the latent class discrimination method was higher than those of the logistic regression method and the discriminant analysis method in an experimental demonstration applying the three methods to the same data set.

Standard Errors of Estimated Latent Variable Scores With Estimated Structural Parameters

The authors propose a concise formula to evaluate the standard error of the estimated latent variable score when the true values of the structural parameters are not known and must be estimated. The formula can be applied to factor scores in factor analysis or ability parameters in item response theory, without bootstrap or Markov chain Monte Carlo methods. The proposed method is valid when both the number of items and the number of subjects are large. However, simulation results suggest that the proposed method provides valid estimates for standard errors of latent abilities even when the numbers of items and subjects are not large.

Within-individual discrimination on the Concealed Information Test using dynamic mixture modeling.

Whether an examinee has information about a crime is determined by the Concealed Information Test based on autonomic differences between the crime-related item and other control items. Multivariate quantitative statistical methods have been proposed for this determination. However, these require specific databases of responses, which are problematic for field application. Alternative methods, using only an individuals data, are preferable, but traditionally such within-individual approaches have limitations because of small data sample size. The present study proposes a new within-individual judgment method, the hidden Markov discrimination method, in which time series-data are modeled with dynamic mixture distributions. This method was applied to experimental data and showed sufficient potential in discriminating guilty from innocent examinees in a mock theft experiment compared with performance of previous methods.

Bayesian multidimensional scaling for the estimation of a Minkowski exponent

The Minkowski property of psychological space has long been of interest to researchers. A common strategy has been calculating the stress in multidimensional scaling for many Minkowski exponent values and choosing the one that results in the lowest stress. However, this strategy has an arbitrariness problem—that is, a loss function. Although a recently proposed Bayesian approach could solve this problem, the method was intended for individual subject data. It is unknown whether this method is directly applicable to averaged or single data, which are common in psychology and behavioral science. Therefore, we first conducted a simulation study to evaluate the applicability of the method to the averaged data problem and found that it failed to recover the true Minkowski exponent. Therefore, a new method is proposed that is a simple extension of the existing Euclidean Bayesian multidimensional scaling to the Minkowski metric. Another simulation study revealed that the proposed method could successfully recover the true Minkowski exponent. BUGS codes used in this study are given in the Appendix.

Bayesian Factor Analysis Model and Choosing the Number of Factors Using a New Informational Complexity Criterion

This paper introduces two forms of informational complexity ICOMP criteria of Bozdogan (1988, 1990,1994) as a decision rule for model selection and evaluation in Bayesian Confirmatory Factor Analysis (BAYCFA) model due to Press and Shigemasu (1989) in contemporaneously choosing the number of factors and determining the “best” approximating factor pattern structure. A Monte Carlo simulation example with a known factor pattern structure and known actual number of factors is shown to demonstrate the utility and versatility of the new approach in recovering the true structure.

BMDS: A Collection of R Functions for Bayesian Multidimensional Scaling

Bayesian MDS has recently attracted a great deal of researchers’ attention because (1) it provides a better fit than classical MDS and ALSCAL, (2) it provides estimation errors of the distances, and (3) the Bayesian dimension selection criterion, MDSIC, provides a direct indication of optimal dimensionality; see the original paper by Oh & Raftery (2001). However, Bayesian MDS is not yet widely applied in practice. One of the reasons can be attributed to the apparent lack of software: there is none except for the original Oh & Raftery’s code, which requires good experience in Fortran programming and the IMSL library, which is a commercial library for numerical calculation. It may be difficult to require such environment for many researchers. Considering this situation, we propose a set of R functions, BMDS, to perform Bayesian MDS and to evaluate the results. Using BMDS, researchers can (1) perform Bayesian estimation in MDS, (2) check the convergence of Markov chain Monte Carlo (MCMC) estimation, (3) evaluate the optimal number of dimensions, (4) evaluate the estimation errors and (5) plot the resultant configurations. Also, using BMDS users can comparatively evaluate the result of Bayesian and classical MDS in terms of the value of stress and the plot of observed and estimated distances. In our functions, we made use of WinBUGS (Spiegelhalter, Thomas, Best, & Lunn, 2007) via R2WinBUGS package (Sturtz, Ligges, & Gelman, 2005) for MCMC estimation. Because the Bayesian MDS model is rather complex and it is impossible to use single WinBUGS script for any model, our bmds() function automatically produces a BUGS script that is adequate for the current data every time we run the R function. By using WinBUGS in this way we can speed-up the MCMC estimation while maintaining the readability of the code, which tends to be complex in Bayesian estimation.

A Batesian Semiparametric Generalized Linear Model with Random Effects Using Dirichlet Process Priors

The purpose of this paper is to propose a parameter estimation method that doesn’t need to set the number of heterogeneous populations in generalized linear models. We use a finite dimensional Dirichlet process mixed model (Ishwaran and James 2001). Due to the use of Dirichlet process, we make no assumption about the number of subgroups that are mixed. The proposed model can be considered as the direct extension of the model of Lenk and DeSarbo (2000) in the sense that the proposed method needs no assumption for the number of mixed latent classes in their model.

Bayesian Analysis of Structural Equation Modeling

A Bayesian procedure to make exact distributional inferences about all structural parameters and latent variables was proposed. This procedure handles the problem associated with the fixed parameters by means of conditinalization, and uses the Gibbs sampler to derive the posterior distribution for each unknown quantitiy. A simulation study was conducted to evaluate the performance of the proposed procedure.