Say, i have a fitted multi-variable linear model, multicollinearity may leads to small effects of some variables, how can i say one variable is just noise(not related to the target variable) rather than effected by multicollinearity?
2
-
1$\begingroup$ Please make the question clear, what do you mean by " multicollinearity may leads to small effects of some variables" ? $\endgroup$– Vishaal SudarsanCommented Mar 13, 2018 at 6:43
-
$\begingroup$ @VishaalSudarsan the fitted parameter value could be small for variable a if a is correlated with variable b though both a&&b are highly related with the target variable. $\endgroup$– yeren1989Commented Mar 14, 2018 at 3:50
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
There are tests for multicolinearity. My personal choice is condition indexes and proportion of variance explained. This has been discussed here before, but you can also see the work of David Belesley (two books and ) or Peter Flom's (!) dissertation.
Noise will not have high condition indexes.
answered Mar 13, 2018 at 13:25