[R-sig-ME] What is the difference between Matlab Anovan and R mixed Model?

[Ben Bolker]

Gordon, Joanne <jcgordon at ...> writes:

> 
> Dear all,

> I would really like to perform a linear mixed effects model in R but
> have limited maths and computational skill.  I have a matlab Anovan
> function that does the job but I like doing basic things in R and
> would like to try the analysis there.
 
> I believe my data is unbalanced and that I would like to perform a
> crossed analysis (if my self teaching is correct).  My dependent
> variable ( A) is continuous.  I then have three independent
> categorical variables (B, C- fixed) (D= random) My random variable
> pertains to 'Individual'.

  so D is 'individual'?
 
> Several people have advised my that the main important thing is that
> I have included the effects of the random factor interactions with
> the other fixed factors- for example the advice was "try adding the
> first level interaction terms?  Of particular importance is the
> interaction between the fixed effect and random effect (BirdNo),
> because the correct calculation of the F-statistic for a mixed model
> should be the MSE of the fixed effect (e.g.  A) over the MSE for the
> interaction term (e.g.  A*D).  It looks like R doesn't automatically
> do this if the model doesn't include the interaction terms."

  see e.g. Schielzeth and Nakagawa _Methods in Ecology and Evolution_
on this topic.
 
> And someone who uses R suggested:
> lmer(A ~ B + C + (1|D), data=emgDat)
> lmer(A ~ B + C + (1|D) + (1|D*B*C),data=emgDat)

  I think you want

 lmer(A ~ B*C + (B*C|D), data=emgDat)

 **BUT** this only makes sense if the treatments B and C vary within
individuals, or at least within some individuals.  Otherwise you
can't estimate the interaction between B and C (and their interaction)
and D , which is what that last term does ...

   Hope that helps.