Dear all, if I use only continuous explanatory variables in the adonis() function it is like a multiple linear regression with dissimilarities. In multiple regression with highly correlated exp. variables I have the problem with multicollinearity and an increasing standard error of the coefficents. Is it the same for the adonis() function? I would think so since adonis() is "analogous" to a MANOVA, but I am not sure. And is there any possibility to estimate the influence of multicollinearity in adonis(); like the variance inflation factor for lm? Thanks & Best, Jonas
adonis() and collinearity
2 messages · Jonas Kuppler, Jari Oksanen
Dear Jonas Kuppler,
On 17/09/2014, at 16:06 PM, Jonas Kuppler wrote:
if I use only continuous explanatory variables in the adonis() function it is like a multiple linear regression with dissimilarities. In multiple regression with highly correlated exp. variables I have the problem with multicollinearity and an increasing standard error of the coefficents. Is it the same for the adonis() function?
Yes, it is.
I would think so since adonis() is "analogous" to a MANOVA, but I am not sure. And is there any possibility to estimate the influence of multicollinearity in adonis(); like the variance inflation factor for lm?
Not directly: adonis does not return all intermediate results that are useful to find the vif. It could be modified to return those items and then it would be very simple to calculate vif. Not yet, though (neither in plans). There are some tricks that may work even with the current code: check Legendre & Legendre latest edition. Cheers, Jari Oksanen