Within-subject vs. between-subject with trials doubled

Hi Mike and friends,

I’m looking for others’ experience to help me decide between a within-subject and a between-subject design. Any input is welcome:

I’m working on an EEG experiment using the oddball paradigm. The experiment already has 2 within-subject factors. I want to introduce a third one but I don’t want to make the experiment too long.

If I make the third-factor within-subject, I will have to halve the trials; if I keep the number of trials, I will have to make the factor between-subject.

I was made aware of a few studies on the inter-subject and inter-session variance of the brain data. But I don’t know if the loss of power by using a between-subject design can be offset by doubling the trials.

I’m wondering, from your experience of the EEG data, which gives more power, a within-subject factor or a between-subject factor with trials doubled?

Thank you!

Hi Chao. Interesting predicament. In general, it’s better to have more trials and fewer subjects, than fewer trials and more subjects (unless you’re doing individual differences analyses).

Is it possible to have your subjects come in twice? Then you can have 2x as many trials spread over two sessions. Or perhaps ask them to come in for a really long session (e.g., 4 hours) but give them a 1-hour break in the middle? (Or two 30-minutes breaks.)

Also: How do you know you need so many trials? Have you done piloting or sample-size calculations? Perhaps an idea is to start with 1-2 really dedicated subjects to do two full sessions. Then you can analyze the data using all trials and 50% of trials, and see whether you really need such a large N.

Those are my initial thoughts; hope it’s helpful!

Thanks for your input, Mike. You are right, it’s better to empirically find out how many subjects/trials we need. :smiley:

Hi there,
I know this is an old thread, but you might find this paper useful: How Many Trials Does It Take to Get a Significant ERP Effect? It Depends It includes some simulations to show how number of trials and subjects interact and how it affects statistical power. However, the paper is referring to ERPs only, although it might be a good reference.

Thank you for the paper, Maciek.