[Eeglablist] Comparing average ERPs based on different number of observations

[Cigir Kalfaoglu]

Dear EEGLAB forum,
I am comparing EEG activity during naturally occurring response errors with
correct responses where the average accuracy across participants is 76%. So
the number of events in the error category/condition is very different from
that in the correct response condition.
I am especially interested in comparing different categories of errors to
correct responses, which makes the difference between the number of
observations from different conditions even larger. For example, I have an
average of around 1000 vs. 34 observations in correct responses vs. a
specific kind of error response. This makes the variability around the
means of different categories very different, which is not ideal especially
when using parametric tests to compare those means.
How does the STUDY environment handle ERP comparisons between conditions
with such large sample size differences between conditions? Since it is not
ideal to compare ERPs which are the averages of different numbers of
observations, I was wondering if there are any matching or correcting
procedures within the STUDY environment. And if yes, whether they can be
turned on/off so that their effect on the results can be seen.
I am asking this because I am not getting identical results when I conduct
the analyses using EEGLAB functions within vs. outside the STUDY
environment (for a small subset of ERP comparisons). More specifically,
some significant differences observed within the STUDY environment
disappear when I test the data outside the STUDY. I believe this contrast
is due to STUDY analysis having more statistical power than my manual
comparison of ERPs outside the STUDY environment, but I would like to
understand why that is the case.
Thanks in advance,
Cigir Kalfaoglu

-- 
*Çığır Kalfaoğlu*