Hi all,

I'm hoping someone may be able to help me with this. If not, doesn't matter really, but I can't figure out the answer and it's doing my nut in!

So, a typical experiment design in my PhD is a 2 by 2 repeated measures design where BOTH variables are repeated measures. So, I've got 2 IVs, both with 2 levels, both repeated measures. Sometimes I have a mixed design - 2 by 2 by 2 with 2 repeated measures IVs and 1 between participants factor.

Does anyone know how I can calculate the power in this design? I get statistically significant interactions (with pretty strong effect sizes according to partial eta squared) but the simple main effects are non-significant. So, that indicates a lack of power. I would like to know how much power it lacks, and how many participants would be needed to prevent the type 2 error that I appear to be experiencing. (Not that I'm going to go out and run more participants, but I think the argument I'm presenting would be stronger if I had the numbers).

I've been using G*Power (hello high learning curve!) and I've figured out how to do a power analysis for a one-way repeated measures ANOVA, as well as a power analysis for a Mixed ANOVA (as long as it's 1 within and 1 between factor). But, I can't figure out how to enter the data so I get the correct degrees of freedom and accommodate the correlations between my multiple repeated measures variables.

Does anyone know how to do this? Either using G*Power or with some other (Mac) software or online resource?

THANKS in advance,
Anna

Retrospective power calculations are bad!!!

How did you decide on your original sample size?

Sometimes. Not always.

You not what is pretty much always bad though? Making overstated generalisations based on limited knowledge. (And yes, I see the irony that I am making a generalisation here, but my knowledge in this area is certainly not limited).



My sample size is appropriate for my broader discipline. Hell, I've seen plenty of peer reviewed published research with far smaller samples than mine. There is no other previous quantitative work which would give me the data needed to have done an a priori sample size calculation. Personally, I don't see the point in picking an arbitrary effect size, they're about as useful as a chocolate teapot. So I took a punt, which as I'm in an emerging area is a completely valid approach. I mean, someone has to collect the first data on a subject so others can make sensible sample size estimations in the future, right?

The punt didn't quite work out. So, in my thesis discussion I want to consider how many participants would have been needed to prevent a type 2 error. And then I can include this in the discussions of future work together with approaches for learning more about this thing that everyone is fascinated by, and everyone has opinions on but no one understands. But, I can't quite figure out how, and my deadline is approaching fast.

So, if anyone out there does have any ideas of resources that could help, please share!