REML Models for unreplicated treatments

[Mark.Hearnden@nt.gov.au]

Dear S-listers

I have a client with data from a large unreplicated monitoring trial. I haven't had much experience with REML's but this has been suggested as the best approach.

Treatments are applied to individual paddocks, i.e. one treatment per paddock,  and the data (yield etc) collected from quadrats along transects (by multiple observers) nested within each paddock.

I'm using lme with paddock and transect as random factors similar to the example in Crawley (2002) pp 673-676  (however my design is unbalanced).

If the treatments are A, and the transects are B is this the correct model:

model <- lme(y ~ 1, random=~1|A/B, data = "">

or should I be treating A as fixed and using a model like:

model <- lme(y ~ A, random=~1|B, data = "">


What's the best way to add a continuous covariate (y ~ c + A, random= ~1 | B)?  

Any prod in the right direction would be appreciated.

(Using S+ 6.2 with XP)

Regards
MNH

=================================================================
Dr MARK HEARNDEN
Department of Business Industry and Resource Development
GPO Box 3000
DARWIN NT 0801

Tel: (08) 8999 2131 
Fax: (08) 8999 2200
Email: Mark.Hearnden@nt.gov.au
=================================================================