I am trying to run a replication of a study which used a factorial design (2x2) with three DVs (dependent variables) and subsequently ran three ANOVAs to test their hypotheses. Given that the DVs in their study are highly correlated (they reflect different proportions and therefore add up to 1), I want to do a MANOVA instead.

To calculate the sample size necessary for my replication I can use the effect sizes from their ANOVAs (around Eta squared = 0.03/0.05, so quite small) and compute an average N. Instead, I want to estimate a necessary sample size for a MANOVA design which has more power. Given that I don't have Pillai's trace to begin with this leads to guessing. Anybody know a better idea?


You can use sample size determination software that has MANOVA capability, such as PASS. Be advised, however, that you will need to estimate the components of the variance-covariance matrix, as this is part of the input.

  • $\begingroup$ Yes, I am using GPower, which has this function too. The problem is rather in choosing the correct solution. Using multiple ANOVAs to somehow estimate a sample size for a MANOVA or ignoring the MANOVA option and redo the original with three ANOVAs. $\endgroup$
    – timfaber
    May 22 '15 at 14:50
  • $\begingroup$ I don't know why you feel that you need to run three ANOVAs in order to estimate a sample size for MANOVA. Indeed, running multiple ANOVAs won't provide any info that would help you estimate the covariances that must be input into a MANOVA sample size routine. $\endgroup$
    – user44867
    May 23 '15 at 7:40
  • $\begingroup$ It's not that I feel like it, I rather want to understand what the right choice is. Is there anyway of estimating a sample size for MANOVA based on the results on the paper or do I just have to skip doing a MANOVA and replicate the three ANOVAs (meaning a significantly higher sample size). $\endgroup$
    – timfaber
    May 24 '15 at 9:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.