Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Correcting Multicollinearity

From   Nick Cox <>
Subject   Re: st: Correcting Multicollinearity
Date   Thu, 29 Mar 2012 14:18:28 +0100

Another answer is that multicollinearity has computational,
statistical and scientific sides.

It's (mostly) Stata's job to do the best with the computation it can
given (tendencies to) multicollinearity among the predictors.

It is your job to think about choice of predictors.

But you can think about the statistical side by looking at the
structure of interdependence among predictors statistically. One part
of the statistical world is seemingly obsessed by the idea that this
must mean some yes-or-no test yielding a P-value or some omnibus,
factotum or portmanteau statistic quantifying how bad the problem is;
their punishment is to miss what may be learned from looking at
correlations and e.g. -graph matrix-.

Scientifically, what you know about the problem should guide revised
idea about whether the predictor choice is overkill.


On Thu, Mar 29, 2012 at 12:33 PM, Maarten Buis <> wrote:
> On Thu, Mar 29, 2012 at 1:22 PM, D. Demetriou wrote:
>>> Is there any Stata tool to (partly) account for this
>>> defect(multicollinearity) after this has been indicated by, for example, vif
>>> or collin commands?
> No, you can only do something about that before collecting your data
> by choosing a specific design. After you have collected the data, the
> correlations in the data are just what they are and you will just have
> to learn to live with them.
*   For searches and help try:

© Copyright 1996–2016 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index