VIF <10 : Paper

[Jorge Vela]

Hello everybody

I have several GMM estimates and the average value of the VIF are around 1.43 ( the highest value of one variable is 2.30).

I have found an old paper that sets the top value of 10.

Marquaridt, D. W. Generalized inverses, ridge regression, biased linear estimation, and nonlinear estimation. Technometrics, 1970, vol. 12, no 3, p. 591-612

In page 610 : "A rule of thumb for choosing the amount of bias to allow with ill conditioned data, whether by ridge or generalized inverse, is that the maximum variance inflation factor usually should be larger than 1.0 but certainly not as large as 10"

Could this article be an adequate reference to refer to the "possible" VIF limit in GMM?

Thank you very much!

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output and sample data using dataex.

Colinearity means that you have trouble accurately estimating a parameter for a variable because that variable is similar to others (or combinations of others) in the model. There is no absolute cutoff - with larger samples, you can get reasonably accurate estimates even with colinearity. The rule of thumb of 10 is just that - a rule of thumb. Whether a rule of thumb for the models Marquaridt discussed apply to GMM is not clear. Whether your discipline will see that article as an adequate reference is not something folks outside your discipline can judge.

[Carlo Lazzaro]

Jorge:
as an aside to Phil's helpful advice, you may also want to take a look at Examples 11-13, -regress postestimation- entry, Stata .pdf manual and related references.

[Steve Samuels]

See Paul Allison's blog and this article entitled "A Caution Regarding Rules of Thumb for Variance Inflation Factors". The abstract concludes:

Values of the VIF of 10, 20, 40, or even higher do not, by themselves, discount the results of regression analyses, call for the elimination of one or more independent variables from the analysis, suggest the use of ridge regression, or require combining of independent variable into a single index

[Steve Samuels]

Unintentional duplicate.