Sample Size calculation for repeated measures ANOVA

[Ben Hando]

Hi All,

I need to perform a sample size calculation for a single group study and I'm not clear what values to put for some of the options in Stata.

My scenario - We need to determine a sample size for a cohort that will be measured at baseline, 8 weeks and 12 weeks (they will get 6 weeks of treatment). The primary outcome measure will be continuous/ratio level data (pain pressure threshold) and measured with means.

My understanding is that I would use the following syntax in Stata for this:

power repeated, varerror(#) corr(#) repeated(3) ngroups(1) varwithin(#)

My question: I'm not clear what values to put in the varerror, corr and varwithin options. In the documentation examples, they offer sample numbers, but these appear arbitrary. I don't have pilot data or even data in the literature to inform our decision on what #'s to put here. I.e. how correlated are the repeated measures? .2? .3? .5? and the varerror and varwithin - the Stata documentation example uses 42 for varerror and 4.22 for varwithin.

Thanks for any insight here,

Ben

[Rich Goldstein]

when I am in this situation, I try a range of numbers to see how sensitive the needed N is to various values; a substantive expert (not me, in other words) needs to be involved in setting these ranges; if the values of N differ "widely" across the range, then the substantive expert again needs to be involved; my experience is that there is little variation in N for many of these values unless the range used is so wide as to be unbelievable

[Ben Hando]

Thanks, Rich. Yeah, I'm going to go back to the lit and find some real world examples. Thanks for your help/insight here.

Ben

[Carlo Lazzaro]

Ben:
as an aside to Rich's helpful advice, you may want to skim though Joseph Coveney ' s posts, who provided interesting guidance on similar (as well as other) topics