The authors propose a new Item Response Model which can be regarded as a leaner model. Unique features of this model are as follows: 1. The model explains learners'solving process to test items. 2. The parameters of this model can be divided into two kinds, namely, item parameters and incidental parameters. Learners'characteristics are described by the incidental parameters, i. e., latent multidimensional variables. 3. Learners'solving process is expressed in terms of the network, and the transition probability is modeled by Markov Process. We apply this model to real data, and show that the parameters are estimated reliably and they reflect well the characteristics of the learners and items.
OSMOD(Saito&Otsu, 1988), which is a principal component analysis for mixed mesurement data, and its extensions are exarnined with applications to artificial data and analytical considerations. Two important conclusions are obtained.(1)In many cases, Minimizing Generalized Variance criterion(MGV)works better than Maximizing Total Variance criterion(MTV).(2)OSMOD gives more stable category scores than Multiple Correspondence Analysis(MCA).
Most of the statistical methods have been devised for the purpose of applying to the simple random samples. However, it is often difficult to carry out simple random sampling strictly in the actual surveys. So some analysts apply the techniques, which are generally utilized, to the obtained sample without examining the sampling methods. In this paper, we discuss the effects of disagreement between the sampling methods and the estimators on the estimation of a population mean. As a result some pieces of advice on the way to check it are pointed out.
When quantitative data are obtained by sample surveys, we usually estimate a population variance as well as a population mean. It is well known that an unbiased estimator for a population variance is usually used, if a sample is taken under simple random sampling with replacement. However, it is rare that the suitable estimators for other sampling methods are utilized in the actual surveys. These problems arise from misunderstanding about the sampling methods or lack of knowledge of the estimators. In this paper, we investigate the effects of the improper use of estimators on the estimation of a population variance.
Parameter estimation of the Canonical Correlation Analysis(CCA)can be often difficult in the presence of multicolinearity. The purpose of this paper is to make the parameter estimates more stable by structuring relevant covariance matrices using the prior knowledge. More concretely, we assumed factor analytic model and covariance structure model, then in terms of the estimated parameters in these models, we reproduced the covariance matrices. We substituted these covariance matrices in the CCA intead of the maximum likelihood estimates, i. e. sample covariance matrices, then obtained the estimates for the CCA vectors. In this paper, we consider only the first principal CCA vectors, because we assume here our interest is in constructing the index connecting the two sets of variables. So, the stability of the estimated parameters was judged by the cross validation techniques. The cross validation study showed that the parameter estimation using the structured covariance matrices yielded more effective prediction than the traditional method.
Semantic Differential technique was applied to synesthetic tendencies or common affective effects between various stimulus categories of different sensory modalities; classical music, sounds, colours, forms, symbolic words, and movies. Factor analyses indicated three stable factors; evaluation, activity, and lightness(negative potency)across these six different stimulus categories. Classification of patterns of factor scores of 58 stimuli in different categories and cluster analysis based on these factor scores had been shown useful to classify these stimuli into affective groups independently of stimulus categories. Key words: Synesthesia, Semantic Differential, Affective meaning