Significant baseline, Oddball paradigm

Question

I recently conducted an EEG experiment using a passive Oddball paradigm with two experimental conditions : standard stimuli (occurring with a 85% probability) and deviant stimuli (occurring with a 15% probability). This experiment was conducted on 10 participants.

ERP were recorded with 64 electrodes.

An ANOVA test on the prestimulus baseline period yielded a significant effect of the condition. I have observed a difference between standard stimuli's wave and deviant stimuli's wave. There shouldn't be such a difference between these two but I understood that this difference is caused by a smaller number of trials in the deviant condition than in the standard one.

However, I'm still wondering whether this significant difference can be caused by the small number of participants (n=10) or by too small inter-stimulus interval (ISI=1500ms) ? How do you suggest I should deal with this significant baseline regarding the understanding of MMN/P3 effects ?

I looked on cogsci.SE and I didn't find anything relevant to my question.

Ana · Accepted Answer · 2013-12-01T20:32:45.823

How many trials do you have per condition? With a small number of trials in the deviant condition, and a small number of participants, these things can happen.

The ISI would not cause this per se, however, have you considered looking at effects of the previous trial? You can analyze the baseline intervals as a function of what type the previous trial was, and see if there is a difference in the period following deviants or standards.

You could also try to randomly sample as many standard trials as you have deviants per participant, and then re-do your analysis and see what happens. You can make a loop and do this many times, and then look at the median p-value.

It could also be that some small number of subjects is introducing some skew to your results. You could try to re-do your analysis on a single subject level, comparing standard and deviant trials within subjects. Then you could see whether all your subjects have this effect or only some of them. If it's only some, you could look at their data more carefully for potential outliers. If it's a general thing, I'd look into the effects of the previous trial.

And if it's the previous trial effect, you can just remove all standard trials preceded by a deviant and try again.

All that being said, even if you have a baseline difference, it might not really matter: this is what baseline correction is for. But I agree that it's disconcerting.

I'm also a bit confused why you would use an ANOVA to compare two conditions, and not a t-test. Maybe I'm not understanding you design completely.

I have more than two conditions : Condition (standard versus deviant), Emotion (happiness, fear, sadness or neutral). Per condition I have 246 standard stimuli and 42 deviant stimuli. I will try to do a T-test (however, I thought ANOVA could be used to test if two means are equal or not) and see if the difference between the means is still significant and re-do my within-subjects analysis. Can I get back to you after doing some more analysis ? — Mathilde, Dec 03 '13 at 20:23
Hi @Mathilde, of course you can :) If you google me and the institute where I work (you can find my web page in my profile if you click on my name), you can also find my e-mail address and contact me directly if you prefer. But I do read this board regularly, so it's safe to ask further questions here as well. — Ana, Dec 03 '13 at 20:27
@Mathilde - oh and by the way, of all the things I suggested, I would first try removing standards preceded by deviants. It's a much cleaner comparison that way in any case. — Ana, Dec 03 '13 at 20:29
That's very nice of you. For a given stimulus, I don't know if it is possible to retrieve whether the stimulus is followed by a deviant or by a standard. I'll try ! — Mathilde, Dec 03 '13 at 20:54

Significant baseline, Oddball paradigm

1 Answers1