How to handle wrong responses in first-level GLM

I have an fMRI experiment where participants perform a task unrelated to the cognition process of interest. In this task, the color of the target changes once per trial, and participants must detect the color change. The purpose of this task is to ensure subjects remain attentive to the targets and to test whether they attend targets well.

I’m considering how to handle trials with incorrect responses during this task. One approach is to simply exclude these incorrect trials. However, this method results in varying total lengths of time series across different subjects, which could pose a problem for DCM analysis.

Another option is to include an additional regressor in the first-level GLM to encode the incorrect trials. This regressor would consist of binary values: 0 for correct trials and 1 for wrong trials.

Do you have any recommendations or alternative methods for handling wrong responses in this context?

Thank you!