Hi Alberto,
>ok. Since i have three sessions (that i "merge" through a 2nd level analysis) and my reaction time changes across sessions (they reduce, subjects learn).
>Should I use as RT_m the single session mean?, rather than the 3 sessions mean. it should be so.
Probably, yes.
>>Your contrast of interest for this design matrix is [0 1]. That is, the (cA-iA)-(cB-iB) contrast is implicit in the second regressor, so you
>>won't need to do a 1 -1 -1 1 or 1 -1 contrast and also no need to add RT to the second level analysis.
>
>Here I lost you. I have 6 different task conditions each represented by two EV (the target and the answer), I add a third EV, the b), with a dynamic 3rd column. my contrast (cA-iA)-(cB-iB) account for differences between the two different congruent/incongruent comparison. how could [0 1] do this?
If you are performing a typical conflict monitoring task, then you will be looking at RT differences between congruent and incongruent (and maybe neutral) trials. The difference between RT means, i.e. incongRT - congRT, is thought to represent the amount of "conflict" in the decision. Thus, in standard conflict monitoring models, incongRT-congRT is proportional to conflict. If you are interested in detecting conflict voxels, then you should create a regressor in which the amplitude of each BOLD response is proportional to the amount of conflict.
So, if congruent trials are the "baseline" condition, then assign them a value of zero. Incongruent trials have a conflict level of incongRT-congRT and neutral trials will presumable have a conflict level of neutRT-congRT. If you think that neutral trials are the baseline, then assign them a value of zero and make the others incongRT-neutRT and congRT-neutRT. In all cases, the conflict activity should last until the conflict is resolved, which presumably occurs when the response is made. Thus, the regressor will have a duration of RT_trial and a height of xRT-baselineRT, where xRT is one of the three conditions.
It is still important to control for non-specific effects of task, so you will need a regressor that models trial-to-trial RTs. This is modeling variance unrelated to conflict and thus doesn't change from trial to trial but is present for the duration of the trial. It will have a height of 1 but a duration of RT_trial.
The design matrix will only have two regressors: one for non-specific, unmodulated activity and one for conflict-modulated activity. You are presumably interested in the [0 1] contrast.
Note: the construction of any regressor requires a number of assumptions. In this case, we assume that (1) conflict is present until the response is made, (2) conflict is proportional to RT differences between conditions, and (3) learning does not affect the estimate of conflict. Make sure that these assumptions are consistent with the cognitive model you are using to interpret your data.
>I setup such an analysis, the feat_model warned me over the similarity among EV. of course EV1 and EV3 of each condition have same onset and duration. can i simply ignore this warning?
No. If you ever get this warning, your model is rank deficient and will not provide meaningful results.
cheers,
jack
|