Dear Statalist,
I'm wishing to calculate a sample size based on the comparison of means between two groups, adjusting for a baseline measure, i.e. using an ANCOVA
I used to do this with sampsi, but power is now the recommended alternative. I do not seem to be getting the same answer if I use "sampsi" with the pre() argument or if I use "power repeated". I show some example code below. Any assistance for where I'm going wrong or have misunderstood the power repeated command would be much appreciated.
I'm wishing to calculate a sample size based on the comparison of means between two groups, adjusting for a baseline measure, i.e. using an ANCOVA
I used to do this with sampsi, but power is now the recommended alternative. I do not seem to be getting the same answer if I use "sampsi" with the pre() argument or if I use "power repeated". I show some example code below. Any assistance for where I'm going wrong or have misunderstood the power repeated command would be much appreciated.
Code:
** Using Stata 15.1 ** Say I had two groups, with means 0.74 at baseline (in both groups) and SD=0.18 (again in both groups) ** At follow-up, we wish to calculate the sample size based on a change in the means to 0.76 and 0.74 for the two groups ** The correlation between baseline and follow-up measures is 0.5 ** For 80% power, alpha=0.05, using the pre(1) argument to specify 1 baseline measure sampsi 0.76 0.74, sd(0.18) power(0.8) pre(1) r01(0.5) ** This gives the following answers: POST 1,272 per group, CHANGE 1,272 per group, ANCOVA 954 per group ** If I try and use the power repeated command I get a very different answer that does not correspond with any of the three analyses above. matrix M = (0.74, 0.76 \ 0.74, 0.74) power repeated M, power(0.8) corr(0.5) varerror(0.0324) factor(between) ** 3,816 per group ** Furthermore, if I set the correlation between the baseline and follow-up measures to zero I thought I might get the same sample size using a power twomeans calculation as using ANCOVA ** using sampsi sampsi 0.76 0.74, sd(0.18) power(0.8) pre(1) r01(0) ** 1272 per group using ANCOVA ** using power twomeans power twomeans 0.76 0.74, sd(0.18) power(0.8) ** 1273 per group -- this agrees with sampsi, pre() above ** using power repeated matrix M = (0.74, 0.76 \ 0.74, 0.74) power repeated M, corr(0) varerror(0.0324) factor(between) ** 2544 per group
