[R-sig-eco] How to decrease inflation factor of environmental variables in CCA.

[Jari Oksanen]

On 6/05/11 04:54 AM, "刘朔孺" <lsrzggod at gmail.com> wrote:

> Dear list,
> As I know(according to other related papers), the inflation factor of
> environmental variables in CCA should be less than 20. But, when I conducted
> it(in CANOCO version 4.5),  the final result revealed the inflation factor of
> two key environmental variables are 25 and 26(>20). The two variables are
> percent of forest landuse and impervious surface, which I want to keep in the
> analysis.
> My question is whether I could transform the two factors to decrease the
> inflation factor or not? Or there is any better choice for me?
> Maybe this is not an "R" issue. But I really couldn't find a better place to
> propose such a question. Thanks in advance for your time and suggestion.

Dear Shuoru,

Variance inflation factor (VIF) is based on linear relationship. So the
answer is: yes, nonlinear transformations can change VIF which can either be
reduced or increased. Most people would argue that it makes no sense to
transform data because of VIF: you need better reasons for transformation.

You didn't say how many environmental variables you had in you analysis in
addition to the two key variables you want to keep. VIF is a measure of
*multi*collinearity. Removing some other variables than those two key ones
may reduce the VIF of the remaining. Construct your model carefully.

If the VIF still disturbs you, don't calculate the VIF.

Cheers, Jari Oksanen