[R] Analyses of covariation with lme() or lm()

From: CG Pettersson <cg.pettersson_at_evp.slu.se>
Date: Wed 05 Oct 2005 - 23:10:18 EST

Hello all!

I have a problem that calls for a better understanding, than mine, of how lme() uses the random part of the call.

The dataset consists of eleven field trials (Trial) with three replicates (Block) and four fertiliser treatments (Treat). Analysing for   example yield with lme() is easy:

m1 <- lme(Yield ~ Treat, data=data,

           random =~1| Trial/Block)

giving estimates of Treat effects with good significances. If I compare m1 with the model without any random structure:

m2 <- lm(Yield ~ Treat, data=data),
m1 is, naturally, much better than m2. So far so good.

Now I have one (1) measure from each Trial, of soil factors weather and such, that I want to evaluate. Remember: only one value of the covariate for each Trial. The suggestion I have got from my local guru is to base this in m1 like:

m3 <- lme(Yield ~ Treat + Cov1 + Treat:Cov1, data=data,

           random =~1| Trial/Block)

thus giving a model where the major random factor (Trial) is represented   both as a (1) measure of Cov1 in the fixed part and by itself in the random part. Trying the simpler call:

m4 <- lm(Yield ~ Treat + Cov1 + Treat:Cov1, data=data)

gives back basically the same fixed effects as m3, but with better significances for Cov1. Tested with anova(m3,m4) naturally gives the answer that m3 is better than m4. Ok, what about dealing with Trial in the fixed call? :

m5 <- lm(Yield ~ Trial + Treat + Cov1 + Treat:Cov1, data=data)

lm() swallows this, but silently moves out Cov1 from the analysis, an action that feels very logical to me.

My guru says that using the random call secures you from overestimating the p-values of the covariate. I fear that the risk is as big that you underestimate them with the same action. Working on a paper, I naturally want to be able to do some sort of discussion on the impact of covariates... ;-)

What is the wise solution? Or, if this is trying to make other people do my homework, could anyone tell me where the homework is? (Ive got both Pinhiero & Bates and MASS as well as some others in the bookshelf.)


CG Pettersson MSci. PhD.Stud.
Swedish University of Agricultural Sciences (SLU)
Dep. of Crop Production Ecology (VPE).

R-help@stat.math.ethz.ch mailing list
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Received on Wed Oct 05 22:49:16 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:40:36 EST