Tests for the Severity of Multicolinearity in Regression Analysis: A Comment
The Review of Economics and Statistics
1975
det xoo'xoo 1 ) (N n + 1)/(n-2) has the det xo'xo F-distribution with n 2 and N (n 1) degrees of freedom, assuming normality of the underlying distributions. If the statistic is (almost) zero, det xoo'xoo (almost) equals det xo'xo, and the multicollinearity among the columns of xo is (almost) wholly attributable to the columns of xoo. Each step in the proposed third stage is virtually a replica of the second stage, the only difference being that fewer variables are involved. Thus, to show why the proposed third stage is effective in detecting multicollinearity patterns, it suffices to refer to the justification of the second stage, in footnote 3. 4 In practice, calculation of a few, strategically chosen, principil minors should suffice.
- DOI
- 10.2307/1923927
- Volume
- 57 (3)
- Pages
- 368
- Export
- BibTeX
- Sources
- openalex crossref