Dear Thang,
Usually it should be possible to create "dummy" trials slightly before the onset of the paradigm (negative onsets) or onsets very close to the end, maybe it's related to the duration of the conditions.
In case of "empty" conditions I would thus just discard the regressor for that session. This might require some manual editing of the contrast vectors, but as long as it's just a single subject this shouldn't take much time. Note that even if you create a regressor based on some artificial onsets, resulting in some beta estimates, these estimates would be meaningless.
However, maybe it's best to discard the whole session. If one of the conditions is "empty" you cannot obtain beta estimates for the condition. And without beta estimates you (probably) can't average across sessions in the way you would do for the other conditions. As you can't rule out session effects you would introduce a bias if e.g. estimates for A and B are based on six sessions while those for C are based on only five.
Best,
Helmut
|