Shizuhiko Nishisato
Biographic Data
| ID | 4344627 |
|---|---|
| NAME | Shizuhiko Nishisato |
| GIVEN NAMES | Shizuhiko |
| FAMILY NAME | Nishisato |
| SIGNATURE | NISHISATO S |
| AFFILIATIONS | The Ontario Institute for Studies in Education and the University of Toronto |
| VERIFIED | No |
| TOTAL WORKS | 8 |
| TOTAL CITATIONS | 0 |
| AUTHOR COUNT | 8 |
| EDITOR COUNT | 0 |
| FIRST PUBLICATION YEAR | 1970 |
| LATEST PUBLICATION YEAR | 2003 |
| H-INDEX | 0 |
Total Information in Multivariate Data from Dual Scaling Perspectives
It is an established matter that the total information in multivariate data is defined as the sum of eigenvalues of the variance-covariance matrix. In this article I challenge this time-honored tradition and look at another definition of the total information in data from a dual scaling perspective. This proposal is a step toward unifying the concept of information for both discrete and continuous variables
Dual Scaling of Successive Categories Data
A method was developed to determine values of stimuli and category boundaries through differential weighting of subjects. The method transforms the data into the subjects-by-parameters (scale values of stimuli and category boundaries) matrix of incidences, and derives both weights for the subjects and estimates of the parameters which are the most discriminative in the least squares sense. A numerical example was presented to illustrate the proce…
Principal Components of Deviation Scores and Standardized Scores
The study compared principal components of the variance-covariance matrix V and those of the corresponding correlation matrix R, and identified 2 contributors to the discrepancies between the 2 sets of principal components; the dispersion of the variances of the variables σ(v) and the dispersion of the differences of the corresponding eigenvalues of V and R, σ(λ-λ*). Although both contribute independently to the discrepancies, the effect of σ(v) …
Partially Optimal Scaling of Items With Ordered Categories
A method of partially optimal scaling is developed by incorporating a simple subjective scoring method into optimal scaling. The method mitigates practical problems of the data size and the a priori order of response categories. On the basis of Monte Carlo computations, some guidelines for differential use of partially optimal scaling, optimal scaling or the combination of the two are suggested
Effects of Categorizing Continuous Normal Variables on Product-Moment Correlation
Transform Factor Analysis
A generalized model of factor analysis, called transform factor analysis, was proposed so as to provide useful guidelines for further developments in factor analytic methodology. Topics discussed were: a generalized model, existing models as its special cases, appropriate transformations, analysis of all forms of association measures and variance-covariance structures of original variables. A numerical example was given to illustrate applicabilit…
Structure and Probability Distribution of Dichotomous Response Patterns
A view that dichotomous response patterns with negligible probabilities should be excluded from data analysis was pursued mainly to mitigate the difficulty in dealing with an enormous number of possible response patterns. Based on an assumption of latent structure, a geometric model of dichotomous response patterns was proposed. Monte Carlo computations revealed, among others, that the structural consideration of binary information can lead to an…
Probability Estimation of Dichotomous Response Patterns by Logistic Fractional-Factorial Representation
A method for estimating probabilities of 2n dichotomous response patterns is proposed. Incorporated in the method are two techniques: the fractional replicate principle, which would mitigate the size problem (the number of possible patterns and the number of parameters): a logistic representation of probabilities, which would assure the efficient use of the fractional principle, and which eliminates the possibility that an estimate may exceed the…
No prominent works on this page.
Structure and Probability Distribution of Dichotomous Response Patterns
A view that dichotomous response patterns with negligible probabilities should be excluded from data analysis was pursued mainly to mitigate the difficulty in dealing with an enormous number of possible response patterns. Based on an assumption of latent structure, a geometric model of dichotomous response patterns was proposed. Monte Carlo computations revealed, among others, that the structural consideration of binary information can lead to an…
Probability Estimation of Dichotomous Response Patterns by Logistic Fractional-Factorial Representation
A method for estimating probabilities of 2n dichotomous response patterns is proposed. Incorporated in the method are two techniques: the fractional replicate principle, which would mitigate the size problem (the number of possible patterns and the number of parameters): a logistic representation of probabilities, which would assure the efficient use of the fractional principle, and which eliminates the possibility that an estimate may exceed the…
Effects of Categorizing Continuous Normal Variables on Product-Moment Correlation
Transform Factor Analysis
A generalized model of factor analysis, called transform factor analysis, was proposed so as to provide useful guidelines for further developments in factor analytic methodology. Topics discussed were: a generalized model, existing models as its special cases, appropriate transformations, analysis of all forms of association measures and variance-covariance structures of original variables. A numerical example was given to illustrate applicabilit…
Partially Optimal Scaling of Items With Ordered Categories
A method of partially optimal scaling is developed by incorporating a simple subjective scoring method into optimal scaling. The method mitigates practical problems of the data size and the a priori order of response categories. On the basis of Monte Carlo computations, some guidelines for differential use of partially optimal scaling, optimal scaling or the combination of the two are suggested
Principal Components of Deviation Scores and Standardized Scores
The study compared principal components of the variance-covariance matrix V and those of the corresponding correlation matrix R, and identified 2 contributors to the discrepancies between the 2 sets of principal components; the dispersion of the variances of the variables σ(v) and the dispersion of the differences of the corresponding eigenvalues of V and R, σ(λ-λ*). Although both contribute independently to the discrepancies, the effect of σ(v) …
Dual Scaling of Successive Categories Data
A method was developed to determine values of stimuli and category boundaries through differential weighting of subjects. The method transforms the data into the subjects-by-parameters (scale values of stimuli and category boundaries) matrix of incidences, and derives both weights for the subjects and estimates of the parameters which are the most discriminative in the least squares sense. A numerical example was presented to illustrate the proce…
Total Information in Multivariate Data from Dual Scaling Perspectives
It is an established matter that the total information in multivariate data is defined as the sum of eigenvalues of the variance-covariance matrix. In this article I challenge this time-honored tradition and look at another definition of the total information in data from a dual scaling perspective. This proposal is a step toward unifying the concept of information for both discrete and continuous variables
Mathematics (8 works) · Statistics (8 works) · Computer Science (6 works) · Artificial Intelligence (4 works) · Artificial Intelligence (3 works) · Covariance (3 works) · Econometrics (3 works) · Multidimensional scaling (3 works) · Physics (3 works) · Scaling (3 works)