04-30-2018 07:50 AM
Guys, I really need help with this:
I am trying to do a sample size calculation, for comparison of means of three groups as follows:
%let ngroup=34; /*n per group*/ %let sd=15; /*standard deviation*/ %let power=0.80; /*Required power*/ %let m1=20; /*Mean group 1*/ %let m2=30; /*Mean group 2*/ %let m3=30; /*Mean group 3*/ %let alpha=0.05; /*Mean group 3*/ proc power; onewayanova test = overall groupmeans = &m1 | &m2 | &m3 stddev = &sd npergroup = &ngroup power = . alpha=α plot x=n min=1 max=100 npoints=20; run;
OK so this is telling me that with 34 per group I will get a power of 0.80
Correct me if I am wrong, but this will only tell you how strong your power is to detect overall difference, and not between group differences.
I am also interested in taking into account pairwise comparisons when doing the sample size calculation, I can think of two strategies and I dont know if they are adequate:
1. use Bonferroni correction dropping the alpha to 0.05/ngroup(3)=0.017
or use contrast but i am not sure if contrasts are right and what /how to use them
Could anyone please please try o help me out with these questions?