Do not model average the regression coefficients. This makes no sense (see
Cade 2015. Model averaging and muddled multimodel inferences). And, no
there is no reasonable way to model average regression coefficients ala
Burnham and Anderson approach when some of your models include interactions
among predictors and some don't. Stick to model averaging the predicted
responses and you can do something that is sensible and that can be applied
to any combination of predictor variables including those with interactions.
Brian Cade
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, Feb 20, 2018 at 2:19 PM, Hannah van Noort <hannahvnoort at
gmail.com>
wrote:
> Dear moderator,
>
> If possible I would like to send in the following question for R-help:
>
> I am analyzing a small data set using PGLS with phylogenetic uncertainty
> taken into account and thereby including 100 potential phylogenetic tree
> scenarios. I've managed to run models on all of the different trees and
> performed model averaging to get parameter estimates for the intercept and
> most of the predictor variable slopes using the *AICcmodavg* package and
> *APE*, but I seem to get stuck with some of the variables that are also
> included in a two-way interaction in the model. I can obtain values for the
> two-way interaction but not for the variables separately.
>
> The error I receive is:
>
> ?Error in modavg.AICgls(parm = "agefirstbreed", cand.set =
Cand.models4, :
>
> Some models include more than one instance of the parameter of interest.
> This may be due to the presence of interaction/polynomial terms, or
> variables with similar names: see "?modavg" for details on
variable
> specification and "exclude" argument?
>
>
>
> I'm familiar with the *exclude* function of the *AICcmodavg* package,
but
> this only works when different models have been specified including ones
> without these specific two-way interactions and my models are all similar
> with regards to the interactions, only the phylogenetic tree scenarios
> differ.
>
>
>
> I was wondering if there is another method to obtain these averaged
> parameter estimates for the concerning predictor variable slopes? Not
> excluding the (significant) two-way interaction in which they are also
> involved?
> Thank you in advance for your reply.
>
> Kind regards,
>
> Hannah van Noort
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/
> posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
>
[[alternative HTML version deleted]]