Skip to content
Prev 5383 / 7420 Next

Resource selection- correlation between variables

Teresa:  There probably are no simple short cuts here - you need to
investigate the correlations structure for each of your possible
comparisons.  You can use the variance inflation factor function vif() in
the car package for glms, which includes an extension for categorical
predictors.  I recommend the vif over pairwise correlations as it is the
linear correlation among multiple predictors that creates issues.  Note
that really large VIFs (e.g., >10 or so) are likely to indicate instability
with standard errors for regression coefficient estimates.  Smaller VIFs
1-5 largely indicate an issue with how to interpret regression coefficients
as partial effects.  VIFs close to 1 indicate no linear correlation.  You
don't necessarily need to eliminate predictor variables from a model just
because there is some multicollinearity, e.g., VIFs in the range 1-5.  You
just need to understand that the interpretation of the regression
coefficient as a partial effect for a unit change in the predictor variable
really needs to be interpreted as a unit change in the part of the
predictor that is not linearly related to the other predictors (see Cade
2015.  Ecology 96:2370-2382).  This, of course, is why it is so wonderful
to have perfectly uncorrelated predictors - the interpretation of partial
effects is simpler. But that is not realistic for most resource selection
analyses.

Brian

Brian S. Cade, PhD

U. S. Geological Survey
Fort Collins Science Center
2150 Centre Ave., Bldg. C
Fort Collins, CO  80526-8818

email:  cadeb at usgs.gov <brian_cade at usgs.gov>
tel:  970 226-9326


On Sat, Jun 4, 2016 at 9:54 AM, Teresa Oliveira <mteresaoliveira92 at gmail.com