Dear Olivia,
I would include both EVs (weighted with all ones, and another weighted by the scores). Make sure that you demean the scores prior to using them as the weights (and I'm assuming you mean the third column of the three-column EV format when you are referring to weights - that is what I mean).
You can then form the appropriate contrast at the first level and feed this up.
If you demean the scores before using them as weights (i.e. the third column values) then that will avoid rank deficiency issues.
All the best,
Mark
On 18 Feb 2013, at 16:52, Olivia Maynard <[log in to unmask]> wrote:
> Hi all,
>
> I'm currently analysing data from an fMRI study, with subjects looking at visual stimuli with different valences.
> In my first level analysis, I have EVs which represent individual trials and are currently weighted as 1.
> However, I also have a behavioural measure that describes the subjects' performance per trial and would like to include this information and so have created EVs weighted by subjects' scores on this measure.
>
> When designing the model for the first-level analysis, should I be including the performance naïve regressors (i.e. weighted 1) as well as the behaviourally scaled regressors and then compare them at this level, or should I run them as two separate analyses and then make a higher-level comparison? - to see whether the behaviourally scaled EVs are better at describing my data?
>
> Also, is there anything I need to do to ensure my design is not rank deficient?
>
> Any general hints or insights into interpretation would be much appreciated
>
> Olivia
|