[R] F values from a Repeated Measures aov

[Peter Dalgaard]

Alex Baugh wrote:
> Hi Folks,
>
> I have repeated measures for data on association time (under 2
> acoustic condtions) in male and female frogs as they grow to adulthood
> (6 timepoints). Thus, two within-subject variables (Acoustic
> Condition: 2 levels, Timepoint: 6 levels) and one between-subject
> variable (Sex:male or female).
>
> I am pretty sure my distributions depart from normality but I would
> first like to simply run a RM anova on the data. My problem is that
> when I do this I generate different values of F for my main effects
> and interaction when I do the analysis in [R] and SPSS - so I don't
> know which one to believe.
>
> Here is my code in R:
>
>
>   
>> mydata.tab=read.delim("mydata.txt", header=T)   #read in my data
>>     
>
>   
>> mydata.tab$Timepoint=as.factor(mydata.tab$Timepoint)    #col headings
>>     
> are factors so df are correct
>
>   
>> mydata.tab$Acx.Cond=as.factor(mydata.tab$Acx.Cond)
>>     
>
>   
>> mydata.tab$Sex=as.factor(mydata.tab$Sex)
>>     
>
>   
>> aov.F=aov(Targ.Assoc.Time~(Timepoint*Acx.Cond*Sex) + Error(Subject/(Timepoint*Acx.Cond))+(Sex), data=mydata.tab)
>>     
>
> #run aov where i look at the main effects of Timepoint, Acoustic
> Condition and Sex as well as all the interactions therein on the
> amount of time a frog spends associating with the target sound.
> Include anything to do with Subject in the error term.
>
>
>
>
> Does this look right for a Repeated Measures ANOVA, or am I missing
> something to make it RM and that explains the large discrepancies in
> my F-values between [R] and SPSS?
>
>   
Nothing obviously wrong to my eyes. There's a stray (Sex) term, but I 
don't think that actually does anything. I assume that your data are 
balanced and complete. Apart from that it coincides with my expectation 
of what SPSS would do.

You'd give us a better chance of helping if you actually included some 
output on the two systems.

If you approach this as a multivariate linear model (with 12-dimensional 
response), then you can also use the features of anova.mlm (the example 
on the help page is quite similar to your setup). This takes a bit more 
work, but it give the "epsilon" corrections that people like to 
calculate for these models.
> As soon as I get this canonical aov code figured out I want to derive
> my p-values by bootstrapping my F distributions, but first I need
> those canonical F's.
>
>
> Thanks
> -Alex
>
>   


-- 
   O__  ---- Peter Dalgaard             Øster Farimagsgade 5, Entr.B
  c/ /'_ --- Dept. of Biostatistics     PO Box 2099, 1014 Cph. K
 (*) \(*) -- University of Copenhagen   Denmark      Ph:  (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)              FAX: (+45) 35327907