[R] help on nested mixed effects ANOVA

[Ben Bolker]

Mukta Chakraborty <chakraborty.mukta <at> gmail.com> writes:

> 
> Hello,
> I am trying to run a mixed effects nested ANOVA but none of my codes
> are giving me any meaningful results and I am not sure what I am doing
> wrong. I am a new user on R and would appreciate some help.
> The experimental design is that I have some frogs that have been
> exposed to three acoustic Treatments and I am measuring neural
> activity (egr), in 12 brain regions. Some frogs also called different
> calls than others so I am testing for effect of Call as a fixed
> effect.
> I have Treatment (Trt), Brain Region (Region), and Call (Call) as
> fixed effects and I am modeling Subject (Subj) as a random effect with
> brain region nested within subject. I am using lme4 package and I am
> not sure if that is the correct one to use.
> I also have some missing values as NA as part of the dataset. I have
> also set Subj, Region, and Call as factors using as.factor.
> I have run the model which I am sure is wrong:
> model1<-anova(lme(egr ~ Trt* Region, random=~1|Subj/Region,na.action=na.omit))
> 
> The output I am getting is:
>    numDF           denDF        F-value           p-value
>  Min.   : 1.00   Min.   : 32   Min.   :  2.891   Min.   :0.000e+00
>  1st Qu.: 1.75   1st Qu.:197   1st Qu.: 12.120   1st Qu.:0.000e+00
>  Median : 6.50   Median :252   Median : 20.275   Median :1.146e-05
>  Mean   : 9.00   Mean   :197   Mean   : 87.353   Mean   :1.383e-05
>  3rd Qu.:13.75   3rd Qu.:252   3rd Qu.: 95.508   3rd Qu.:2.530e-05
>  Max.   :22.00   Max.   :252   Max.   :305.972   Max.   :3.240e-05
> 
> Any help on what I am doing wrong on the coding would be very helpful.
> Thanks a lot.
> Mukta


 1.  it would probably be a good idea to ask further questions on this
topic on r-sig-mixed-models instead.

 2. you probably want to look just at the lme() result, and not immediately
wrap it an anova() frame.  I'm a little surprised by your results -- but
since you haven't given a reproducible example I don't know exactly
what's wrong. When I run the lme example in the nlme package and
then do

> anova(fm1)

I get

            numDF denDF   F-value p-value
(Intercept)     1    80 3096.4889  <.0001
age             1    80   85.8464  <.0001

which doesn't look like your answer.

3. I don't think you want to use Region as both a random factor and
a fixed factor.  I would have guessed either

lme(egr ~ Trt*Call, random=~1|Subj/Region,na.action=na.omit)) 

OR

lme(egr ~ Trt*Call*Region, random=~1|Subj,na.action=na.omit)) 

I can also imagine that you might be interested in a crossed
design in this case, in which case you should probably use lme4:

library(lme4)
lmer(egr~Trt*Call+(1|Subj)+(1|Region) ...)

[I'm not sure how na.action works for lmer]

 good luck
   Ben Bolker