Skip to content
Prev 5258 / 7420 Next

AIC in R: back-transforming standardized model parameters (slopes)

Just to echo Bob O'Hara's comment and elaborate a bit more - Don't model
average the regression coefficients, especially if you are considering
models with and without interactions among the predictors. Follow the link
provided by Bob to Cade (2015.  Model averaging and muddled multimodel
inferences. Ecology 96:2370-2382) to see why model-averaged regression
coefficients as conventionally done following Burnham and Anderson provides
meaningless estimates of partial effects in the presence of
multicollinearity and addresses a concept that doesn't exist (model
uncertainty in regression coefficients) when there is no multicollinearity.
  What you perhaps really need to think about is why you want standardized
predictors. Some times they are useful and some times not.  Here you seem
to be going to a lot of trouble to standardize and then to get back to
unstandardized estimates (Drew and Phil have provided good advice about
recovering estimates in unstandardized scale) but without any indication of
how standardization is aiding your interpretations.  Note that in general,
when standardizing regression coefficients it would be more consistent with
the algebra of the regression coefficients to actually standardize them by
their partial standard deviations that adjusts for the linear relationship
(multicollinearity) among predictors (see the Cade 2015 paper to see why
this suggestion originally by Bring 1994 works).  This may be less relevant
to your simple one continuous predictor (time) model with an interaction
with two levels of a factor (air or ice), but it wasn't clear what other
models you might be considering.

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 Tue, Jan 12, 2016 at 8:16 AM, Bob O'Hara <bohara at senckenberg.de> wrote: