[R] MuMIn package, problem using model selection table from manually created list of models

[Kamil Bartoń]

Dnieper 2012-01-17 10:51, Dunbar, Michael J. piste:
>
> The subject says it all really.
>
> Question 1.
> Here is some code created to illustrate my problem, can anyone spot where I'm going wrong?
>
> Question 2.
> The reason I'm following a manual specification of models relates to the fact that in reality I am using mgcv::gam, and I'm not aware that dredge is able to separate individual smooth terms out of say s(a,b). Hence an additional request, if anyone has example code for using gam in a multimodel inference framework, especially with bivariate smooths, I'd be most grateful.

You can model average the coefficients, but not the terms.
>
> Cheers and Thanks in Advance
> Mike
>
> require(MuMIn)
> data(Cement)
> # option 1, create model.selection object using dredge
> fm0<- lm(y ~ ., data = Cement)
> print(dd<- dredge(fm0))
> fm1<- lm(formula = y ~ X1 + X2, data = Cement)
> fm2<- lm(formula = y ~ X1 + X2 + X4, data = Cement)
> fm3<- lm(formula = y ~ X1 + X2 + X3, data = Cement)
> fm4<- lm(formula = y ~ X1 + X4, data = Cement)
> fm5<- lm(formula = y ~ X1 + X3 + X4, data = Cement)
> # ranked with AICc by default
> # obviously this works
> model.avg(get.models(dd, delta<  4))
>
> #  option 2: the aim is to produce a model selection object comparable to that from get.models(dd, delta<  4)
> # but from a manually-specified list of models
> my.manual.selection<- mod.sel(list(fm1, fm2, fm3, fm4, fm5))
> # works
> model.avg(list(fm1, fm2, fm3, fm4, fm5)) # or jut model.avg(fm1, fm2, fm3, fm4, fm5)
> # doesn't work
> model.avg(my.manual.selection)

> # hence this doesn't work
> get.models(my.manual.selection, delta<  4)

There is no need to recreate the models (which is what get.models does) once you have them already 
as a list.

models <- list(fm1, fm2, fm3, fm4, fm5)
my.manual.selection <- mod.sel(models)
model.avg(models[ my.manual.selection$delta < 4 ])