2004 Fiscal Year Final Research Report Summary
Some New Findings on Selection of Variables in Multivariate Analysis
| Project/Area Number |
15500182
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Statistical science
|
|---|
| Research Institution | Tokyo Institute of Technology |
|---|
Principal Investigator |
MIYAKAWA Masami Tokyo Institute of Technology, Graduate School of Decision Science and Technology, Industrial Engineering and Management, Professor, 大学院・社会理工学研究科, 教授 (90157595)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
YAJIMA Yasutoshi Tokyo Institute of Technology, Graduate School of Decision Science and Technology, Industrial Engineering and Management, Assistant Professor, 大学院・社会理工学研究科, 助教授 (80231645)
|
|---|
| Project Period (FY) |
2003 – 2004
|
| Keywords | Discrimination / Mahalanobis distance / Partial least squares / Multicollinearity / Causal analysis / Total effect |
|---|
| Research Abstract |
There are several fundamentally different situations in which it may be desired to select a subset from a large number of variables. In this research, firstly, the problem of selection of variables is investigated in asymmetric discrimination using Mahalanobis distance. As a countermeasure against multicollinearity, two kinds of distance are proposed which enable us to distinguish the redundant measurements from the inherent linear constraints of a normal group. In the asymmetric discrimination problem, not only the unconditional error rate but also the conditional error rates of a future observation from a normal group are also investigated. Secondly, several basic properties of the predicted values obtained by the partial least squares method are discussed, Furthermore, in a causal diagram, graphical criteria for selecting both covariates and variables caused by a response variable are proposed in order to identify total effects in studies with the unobserved response variable.
|
