Variable importance: There can be many related variables to a system. We want to know which ones are most important to the observation data.
Model averaging
A gam model fitted to a data provides p-values of the coefficients used. Model averaging computes averaged significance of each term by averaging coefficients by information criterion (e.g. Akaike weights) across candidate models.
Example with GAM
> library(mgcv)
> library(MuMIn)
> gam1 <- bam(response~l1+l2+... +s(x1)+s(x2)+...,data=data, method="REML")
> dd <- dredge(gam1)
> summary(model.avg(dd))
This summary() shows significance of each coefficient.
References
