R help - May 2011 - lmer with 2 random effects with only two levels

Dear all,

I am analysing a data set based on 6 groups of individuals.   Each group is
observed for 10 days. 5 days with one manipulation 5 days with another
manipulation. I therefore have 6 replicate groups (n=6) each with one mean
measurement for manipulation A and manipulation B.  Each group consists of a
set of males and females. An independent group of males for each group
replicate, however there are only 2 sets of females each replicated 3 times
within the six groups.  

data takes the from:

group      response     female set   treatment
 
1               3.0                 A           high
2               3.0                 B          high
3               2.8                 A           high
4               2.6                B             high
5               2.6                 A            high
6               2.9                B              high
1               1.5                A              low
2               1.4               B                low
3               1.7              A               low
4               1.9              B               low
5               2.0              A               low
6                2.1              B              low

The order of treatment is counterbalanced and I would assume I would choose
to fit the model:
> model1<-lme(response~treat, random=~1|femaleset/group)
or
> model2<-lmer(response~treat+(1|femaleset/group))
However I am concerned with two aspcts:  my small sample size of course but
also the use of a random effect of female set only has two levels (A and B).

Is there a more appropriate way to handle this analysis?  A glm with female
group as an explanatory for instance such as:

model3<-glm(response~treatment+femaleset+treatment*femaleset).  Although
yhis will not properly account for the pseudoreplication.  Ant assistance or
help would be greatly appreciated.

Best 
Colin


--
View this message in context:
http://r.789695.n4.nabble.com/lmer-with-2-random-effects-with-only-two-levels-tp3536791p3536791.html
Sent from the R help mailing list archive at Nabble.com.

What you should and can do depends on the expectations you have regarding the
correlation structure of the data and is limited by your degrees of freedom.
For example, what is wrong about:

reg<-lmer(response~femaleset+treatment+(1|group))
 
?

This assumes that there are constant group effects on a group's responses
(i.e., it allows for differences in mean responses between groups) and that
this constant group effect is independent of the effect of femaleset and
treatment on a group.

Best,
Daniel


--
View this message in context:
http://r.789695.n4.nabble.com/lmer-with-2-random-effects-with-only-two-levels-tp3536791p3537184.html
Sent from the R help mailing list archive at Nabble.com.