Hi David,
This is quite a tricky question, as having two measures of performance means there is more than one question you could ask about the modulation of response and performance. Personally, I would favour reducing the dataset to a single difference measure for each subject (by taking the difference of the appropriate COPEs) and then you can more straightforwardly look at the average difference (A-B or B-A) as well as the relationship/correlation with performance. With this approach you also do not have to worry about statistical issues with the repeated measures and problems of multiple variance components.
For the performance relationship/correlation you have to decide on how to summarise the two performance measures from each subject into a single value (e.g. take the mean of the two values, or maybe the difference if performance _improvement_ is more interesting). This will not only be easier to test but be easier to interpret. Other people might have other suggestions (as I'm sure there will be several options that could be taken here) but this is what I would do.
All the best,
Mark
On 15 Apr 2013, at 22:05, David Soto <[log in to unmask]> wrote:
> hi - some advise would be much welcome
> this seems slighty different from previous posts.
>
> I've got a sample of subjects and copes for task condition A and B.
> now I want to assess whether A>B is modulated by task performance
> so there is a different performance score for each subject on A and B
> my design looks like -
> Note that behaviourally overall memory performance is superior in A than in B
>
> I thought of something like this,
> rows: 1-4 are for the cope in condition A
> rows: 5-8 are for the cope in condition B
>
> Group EV1 EV2 EV3 EV4 EV5 EV6 Performance EV
> subj1 1 1 1 0 0 0 0.7
> subj2 1 1 0 1 0 0 0.97
> subj3 1 1 0 0 1 0 0.85
> subj4 1 1 0 0 0 1 0.75
> subj1 1 -1 1 0 0 0 0.55
> subj2 1 -1 0 1 0 0 0..65
> subj3 1 -1 0 0 1 0 0.53
> subj4 1 -1 0 0 0 1 0.56
>
>
> From prior posts it appears that I only need to multiply each performance score
> by 1 or -1 (i.e adding +-signs accordingly for the performance EV)
>
> then orthogonalise the Performance EV wrt EV2 (assume this is the same as overall demeaning
> the performance EV, if only looking at covariation with this)
>
> then specify contrast
>
> 0 0 0 0 0 0 +1 (to test for regions where A>B differences is modulated by performance EV
> 0 0 0 0 0 0 -1
>
> my only concern is the fact that the behavioral mean for condition A is superior to the mean for conditions B,
> so may reflect different things in quantity and quality
> but guess this does not matter for the model?
>
> cheers
>
> david
|