Dear Rita,
I am not entirely sure that I understand the questions of interest, but I will describe some things that might match.
To model your experiment I would have 4 basic EVs (A_remember, A_forgot, B_remember, B_forgot) as well as another 4 EVs that model the modulation of the BOLD response by the dprime values. These latter four EVs would have the same timing information as the first four, but in the third column (of the 3-column format specification) you would include demeaned values of the dprime measure. However, you don't have to split it into the four groups if you want to model the modulation as being the same in different groups (e.g. A_remember and A_forgot). This really depends on you and what you think is an appropriate model. If you include all four EVs then you allow for a different modulation (or slope or correlation) in each of the separate subgroups. Also, if you want to "control for" any systematic differences in the dprime values between groups then you should demean the values as a single set of values and then split them into the different EVs. If you do not want to control for differences in this way, then you can demean each subgroup after you've split them.
As for contrasts, if we go with the 8 EV model, then a contrast on one of the last four EVs would test for a linear relationship (i.e. a non-zero correlation) between BOLD and dprime within the subgroup (e.g. [0 0 0 0 0 1 0 0] would be the test for this within the A_forgot subgroup). If you want to look at the difference in the relationship (or equivalently, the slope or correlation) between subgroups then you would have a contrast that took the difference between two, such as [0 0 0 0 1 0 -1 0] which would look for where the relationship was greater in A_remember than in B_remember). You should be aware that such differences can arise from two positive, two negative, or one positive and one negative value, so it is also useful to distinguish between these cases using contrast masking.
If you want to restrict your results to areas where the responses on average are also significant (not just the correlation with BOLD) then you should include contrasts on the first four EVs (e.g. [1 0 0 0 0 0 0 0] for A_remember on its own, or [1 0 -1 0 0 0 0 0] for A_remember>B_remember, using the average activations). To combine these with the results of testing with dprime, then you want to use contrast masking to see when both things were true. You don't need to set up contrast masking at the first level (although you can), as the non-masked values get passed up to the higher level. It is normally at the higher level that it is necessary to use contrast masking to ask your specific question of interest.
I hope this helps.
All the best,
Mark
On 8 Apr 2014, at 21:47, Rita Elena Loiotile <[log in to unmask]> wrote:
> Hi,
>
> I am interested in correlating differences in memory performance under 2 conditions to fMRI activation (amongst subjects).
>
> So, for example, I have Context A and Context B and have set up the following regressors for my GLM: A_remember, A_forgot, B_remember, B_forgot. I also have 2 behavioral measures: dprime (memory sensitivity) for context A, and dprime for context B.
> I'm not entirely sure how to implement my correlation question of interest.
>
> For example, I have entered the following contrast from each individual subject into FEAT: context_A > context_B x remember > forget (A_remember + B_forgot - A_forgot + B_remember).
> For EV's I am using all 1s (to capture mean) and a demeaned difference divided by their sum of each subject's behavioral measures (i.e. (dprimeA - dprimeB)/(dprimeA + dprimeB)). My thoughts are that this latter EV will implement correlating the behavioral difference to the contrast difference.
>
> Is there a more appropriate way for me to go about this. Either
> (1) using Beta values and doing some sort of weighting based on behavioral measure
> (2) using a different neuroimaging contrast (e.g. context A > context B at remember = A_remember - B_remember)
> (3) performing a different transformation on the two behavioral measures for each subject (e.g. ratio of the 2 measures, or a difference divided by the average of the two measures)
>
> Thank you in advance,
> Rita
>
>
>
>
>
|