EEG waves measuring statistical significance of an event I have recently gotten into a problem with EEG activity measurement.
I need to prove (or disprove) that certain event triggered a statistically significant reaction in EEG activity. I know exactly at time the event happened and the brain activity (at certain parts) at that times.
What I did so far was that I measured absolute brain wave values and made linear regression and I discovered a statistically significant impact on absolute values of the waves. That means some waves increased in size because of the event.
My question is, is this enough for a proof? Surely, I must do some validation, etc., but would finding statistically significant effect (possible on few different participants) prove that this event does trigger something in the brain.
For those unaware, EEG waves come in cycles, have mean 0 and have quite a lot of noise in them.
Also how strong is this conclusion? Let's say it says that when event is triggered, the brain activity increases for, say 2% (since the data is very noisy, I won't have large increases/decreases). What interpretation can I have out of that that would be excepted in scientific community?
In case this does not prove anything, what would be other methods to try? I know ANOVA is the most standard procedure here, but what to do when it fails? I should still at least prove there is no effect then...
 A: Seems you only have a single type of event with (hopefully) multiple instances. You need to show that the amplitude of the EEG signal after the event is significantly above the level of activity prior to the event.
You will have to take a relatively arbitrary length of time prior to the event (a baseline period), estimate its mean (should be 0 but this must be demonstrated), with a certain deviation. You can then measure how many standard deviations above the baseline activity your evoked activity is. This will give you an indication of the reliability of the resulting pattern of EEG activity induced by the event.
Another possibility is to use a mass-univariate analysis testing each time point at each measured EEG channel against zero (the presumed baseline activity mean). The p-values typically associated with these T-values cannot be used to determine significance since these multiple independent tests will suffer from the multiple comparisons problem and you'll have an inflated false positive rate. However there are multiple ways to deal with this problem, chief among them permutation analysis. See this paper for a review of methods. I happen to have a matlab based toolbox which has several analysis options.
Good luck!
