[FieldTrip] posthoc stats after ft_regressconfound

[Zita Eva Patai]

Dear FTers

I have a stats/double dipping question. Here is my problem & reasoning:

1. I found a significant effect between condition A & B at times x-y in the
theta band.

2. I also found that overall trials and subjects there is no significant
difference between conditions (A&B) on variance in saccade behaviour (as
measured by the EOG signals and extracting the values from the ICA for that
component). But there is a difference between condition A & C, which is why
i do not look at this comparison, as i can't be certain that eyemovements
aren't driving all my neural differences (i have other reasons to be
cautious in comparing these latter two conditions)

3. I could also look at trial-by-trial differences, on data where the ICA
component for saccades has been regressed out (ft_regressconfound). Would
this be a more statistically sound approach? In this case, I could compare
condition A and B at the subject level (indepsamplesT) and take the
surviving t-values to the group level and compare against zero.

My problem is how to restrict this analysis...i can restrict myself to the
theta band, but i have a very long epoch (4 seconds) and I have a lot of
sensors, none of which I can come up with a good apriori reason to select.

>>>*Is it double dipping to report the effect from point1, and then run a
posthoc test at the subject level, only for the relevant times, and see if
i still get a significant difference between A & B, when correcting for
eyemovements at the single trial level?*

Thanks so much for your help in advance,
z




-- 
Eva Zita Patai, DPhil
Postdoctoral Researcher
Institute of Behavioural Neuroscience
UCL
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20180812/b1ab84ba/attachment-0001.html>