A New Measure of Multicollinearity in Linear Regression Models
Kovács, Péter ; Petres, Tibor ; Tóth, László
Internat. Statist. Rev., Tome 73 (2005) no. 1, p. 405-412 / Harvested from Project Euclid
Text on Project Euclid
PDF
Alt PDF
Databases with a lot of data very often mean little information. It is because of the collinearity of variables which consist of the data of the database. This collinearity is in fact a kind of redundancy of the database. In the study a new indicator is given. With this indicator, which contains the eigenvalues of the variables' correlation matrix, it is possible to quantify the percentage of collinearity: from 0% (all the eigenvalues are equal to 1) to 100% (all the eigenvalues, except the first, are equal to 0).
Publié le : 2005-12-14
Classification:  Redundancy of databases,  Multicollinearity,  Spectral decomposition of the correlation matrix 
@article{1133819161,
     author = {Kov\'acs, P\'eter and Petres, Tibor and T\'oth, L\'aszl\'o},
     title = {A New Measure of Multicollinearity in Linear Regression Models},
     journal = {Internat. Statist. Rev.},
     volume = {73},
     number = {1},
     year = {2005},
     pages = { 405-412},
     language = {en},
     url = {http://dml.mathdoc.fr/item/1133819161}
}

Kovács, Péter; Petres, Tibor; Tóth, László. A New Measure of Multicollinearity in Linear Regression Models. Internat. Statist. Rev., Tome 73 (2005) no. 1, pp.  405-412. http://gdmltest.u-ga.fr/item/1133819161/