[R] Mixed-effects model for nested design data

[Federico Calboli]

quote:

> I am using nlme for data from nested design.  That is, "tows" are
nested
> within "trip",  "trips" nested within "vessel", and "vessels" nested
> within "season".  I also have several covariates, say "tow_time",
> "latitude" and "depth"
> My model is
>    y = season + tow_time + latitude + depth + vessel(season) +
> trip(season, vessel) + e
> In SAS, the program would be
> proc mixed NOCLPRINT NOITPRINT data=obtwl.x;
>   class vessel trip tow season depth;
>   model y = season depth latitude /solution;  <----------fixed effects
>   random vessel(season) trip(season vessel);
> run;
> My question is:  How this nested mixed-effects model can be 
> fitted in R-> "nlme"?


> I do not know about SAS but I would guess that your model should be
> fitted
> as something like:
> 
> lme (fixed= y ~ season + tow_time + latitude + depth,
>      random= ~ 1 | season/vessel/trip)
> 
> Maybe you should do some reading in the book by Pinheiro & Bates?
> They explain well how to set up models.



I would create a grouped data variable, to avoid having season a both a
random and fixed effect:

your.data$SV<-getGroups(your.data, form=~1|season/vessel, level=2)

the effect is to create a variable that groups vessels %in% season. BTW,
according to your coding of the data, this stem is not always necessary.

HTH

Federico Calboli
-- 



=================================

Federico C. F. Calboli

Dipartimento di Biologia
Via Selmi 3
40126 Bologna
Italy

tel (+39) 051 209 4187
fax (+39) 051 209 4286

f.calboli at ucl.ac.uk
fcalboli at alma.unibo.it