CODEWITHSUNDEEP
r
hn.phuong
hn.phuong

Reputation: 835

Could we do backward elimination with mixed model using lmer

I used the following syntax for the mixed model and then step but it did not work.

Does it normally work like this or I actually can not use backward elimination with lmer? Thanks!

fullmodel<-lmer(Eeff~NDF+ADF+CP+NEL+DMI+FCM + (1|Study),data=na.omit(phuong))
step(fullmodel, direction = "backward", trace=FALSE )

Upvotes: 6

Views: 16018

Answers (3)

Mikko
Mikko

Reputation: 7755

You can do this with lmerTest package: 

library(lmerTest)
step(fullmodel)

After testing this function with my rather complex data, it does seem to produce feasible model alternatives.

Upvotes: 7

bokov
bokov

Reputation: 3534

The function you want is stepAIC from the MASS package.

stepAIC (and step) use AIC by default, which is asymptotically equivalent to leave-one-out cross validation.

As for the trenchant criticisms, expert knowledge is a great starting point for model selection, but I too often see this used as an excuse to pass the responsibility for making complex statistical decisions off to an applied researcher who doesn't understand statistics.

Edit: sorry, my bad, misread your question, I thought you said 'lme' instead of 'lmer'. I have no idea whether stepAIC supports lmer.

Upvotes: -1

John
John

Reputation: 23758

You could do it, just not with the step function. Since your model is just additive it shouldn't take that long to do by hand.

Upvotes: 2

Related Questions