2

I'm trying to run a power analysis in Stata, using chi2 (even though I am using logistic regression, I just can't find an option for it and I'm new to power analysis) and it seems I have to choose how the groups are allocated.

My study is a cohort study so there is only one group.

Is power analysis not computable without certain group allocation? I know how many have the outcome and how many don't.

Paze
  • 1,751
  • 7
  • 21

1 Answers1

2

No, you can't compute power without taking into account the allocation between the two groups. As an extreme example, imagine all the patients turned out to be men and you have 0 women. You wouldn't have any ability to tell how the rates differed by sex, even if the effect is really obvious. In general, you would have the best power if the allocation were 50-50. It may help you to read my answer to How should one interpret the comparison of means from different sample sizes? Although the example uses t-tests instead of chi-squared tests, the principle is the same.

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
  • I'm a little confused now because I have to choose both group allocation and effect size. I can't tell the difference between the two. Picture of UI: https://prnt.sc/qvtxvl My study is looking at surgery outcome (binary good or bad) and I know how many are estimated to have a good outcome. So I can't compute the power of my study (or an estimated sample size?) because it's not a randomized control trial? – Paze Jan 31 '20 at 22:24
  • @Paze, I don't follow your setup. Can you post the data you expect to find? – gung - Reinstate Monica Feb 01 '20 at 01:17
  • 1
    Let's pretend this is the exact data you expect to find: https://gofile.io/?c=bh7U2x You want to see if aggregateScore can predict the outcome. Can you retrospectively do a power analysis of this data and walk me through how you did so? – Paze Feb 03 '20 at 09:17