I have run some single-trial analyses in preparation for trying to explore MVPA. This was run for each subject and run with one regressor per trial and a set of nuisance regressors. I’ve attached a picture of the design for one run.
However, afterward, the runs are highly discriminable. I have attached a plot of the data. To visualize this a ran an ICA and plotted each trial along the first few independent components. The run associated with each trial is a different color.
This seems too specific to the run to be a feature of autocorrelation. Is there a problem with running single-trial models separately for each run? Should I not be including so many nuisance regressors? Let me know if you’ve seen this before or have an idea about what might be the problem/ how to fix it.