R help - Apr 2004 - modelling nested random effects with interactions in R

Hi there

Please excuse this elementary question, but I have been fumbling with this for 
hours and can't seem to get it right.

I have a nested anova, with random factor "lakefac" nested within 
factor "fishfac" (fixed), with an additional fixed factor
"Habfac". If I
consider everything as fixed effects, it's addmittedly not the correct
model,
but I can at least get this to work:
> nested <- anova(lm(ltotinv ~ habfac + fishfac/lakefac + habfac:fishfac +
habfac:(lakefac %in% fishfac)))
> nested
Analysis of Variance Table

Response: ltotinv
                       Df  Sum Sq Mean Sq F value    Pr(>F)    
habfac                  2 17.3140  8.6570 25.0568 3.534e-08 ***
fishfac                 1  0.9131  0.9131  2.6428   0.11057    
fishfac:lakefac         2  2.3802  1.1901  3.4447   0.04000 *  
habfac:fishfac          2 13.0101  6.5050 18.8281 9.196e-07 ***
habfac:fishfac:lakefac  4  3.0103  0.7526  2.1783   0.08557 .  
Residuals              48 16.5838  0.3455                      
---

So now I try to run it using the lme4 package, treating lakefac as random;
> 
> lakernd <- lme(ltotinv ~ habfac + fishfac/lakefac + habfac:fishfac +
habfac:
(lakefac %in% fishfac), random = ~ lakefac)
Error in .class1(object) : Argument "data" is missing, with no
default
> lakernd
Error: Object "lakernd" not found

The lme help file suggests that if "data" is not specified, that it
defaults to
whatever object is currently in use in the environment (as was the case in the 
fixed effects model- I am using a matrix called "use" in this
example). If I
(naively) simply try to add data= use to the above formula, this happens:
> lakernd <- lme(ltotinv ~ habfac + fishfac/lakefac + habfac:fishfac +
habfac:
(lakefac %in% fishfac), data = use, random = ~ lakefac)
Error in as(data, "data.frame") : No method or default for coercing
"matrix"
to "data.frame"

So how do I get lme to use my data matrix "use" for the model? My
guess is that
my syntax is off, but does anyone have any suggestions on how to fix it? 
Unfortunately, the only resources I have available to me at the moment are 
an "S-plus v. 4. guide to statistics", the archived help files on the
R
website, an the examples and help files in R.  

Your help with this problem is greatly appreciated- I do hope to hear from 
someone as to how I might get this to work.

Sincerely,   

-- 
Michael Rennie
Ph.D. Candidate
University of Toronto at Mississauga
3359 Mississauga Rd. N.
Mississauga ON  L5L 1C6
Ph: 905-828-5452  Fax: 905-828-3792

Hi Michael,
> I have a nested anova, with random factor "lakefac" nested within
> factor "fishfac" (fixed), with an additional fixed factor 
> "Habfac". If I consider everything as fixed effects, it's
addmittedly
> not the correct model, but I can at least get this to work:
> ...
> So now I try to run it using the lme4 package, treating 
> lakefac as random;
I am not quite sure about lme4 but in limrary ('nlme') you would need to
do
something like:
> lakernd <- lme(ltotinv ~ habfac * fishfac,
                 random = ~ 1 | fishfac/lakefac,
                 data= use)

with this model you have the fixed effect of habfac and fishfac and their
interaction, lme should get the df's of the model ok if you specify it that
way. In this way, the random term is only for the intercept (which to my
understanding is the same as saying there is no interaction between random
and fixed effects). If you want to include the interactions you need to do
something like this:
> lakernd.int  <- lme(ltotinv ~ habfac * fishfac,
                      random = ~ habfac * fishfac | fishfac/lakefac,
                      data= use)

Thus, you specify random effects for the habfac, fishfac and their
interactions, which is the same as saying that there are interactions
between random and fixed effects.

Possibly you might need to do something like the following (because fishfac
is in the fixed and the random term, thus you specify interactions
seperately for both levels:
> lakernd.int  <- lme(ltotinv ~ habfac * fishfac,
                      random = list (~ 1 | fishfac,
                                     ~ habfac * fishfac | lakefac)
                      data= use)

To test whether interactions are significant you need to compare models.
Thus the following would test whether all these included interactions
togethter lead to a statistically better model:

anova (lakernd, lakernd.int)

I hope this helps to get you on the right track.

Regards, Lorenz
- 
Lorenz Gygax
Tel: +41 (0)52 368 33 84 / lorenz.gygax at fat.admin.ch      

Center for proper housing of ruminants and pigs
Swiss Veterinary Office
agroscope FAT T??nikon, CH-8356 Ettenhausen / Switzerland
Fax : +41 (0)52 365 11 90 / Tel: +41 (0)52 368 31 31

Where Can I Have More information about Formulae in R (Nesti8ng,
Crossing Factors, Etc). I??ve tried to obtain this information in R Help
and that source were a kind of inconclusive for my Doubts.

[]s
Leonard Assis
Estat??stico
CONFE 7439