Sample size and predictive factors (proc power logistic)

[Elisa97]

Hi everyone,

I want to calculate a sample size to study predictive factors.

I don't understand why i don't get the same NTOTAL in these two procedures :

1) Procedure 1

proc power;
logistic
vardist("fp1") = binomial(0.1,1)
testpredictor = "fp1"
responseprob = 0.3
testoddsratio = 2.25
ntotal = .
power = 0.8;
run;

2) Procedure 2

proc power;
logistic
vardist("fp1") = binomial(0.1,1)
testpredictor = "fp1"
responseprob = 0.7
testoddsratio = 2.25
ntotal = .
power = 0.8;
run;

Could someone explain to me, please.

Have a nice day.

[jiltao · Accepted solution]

Your responseprob = values are different in the two programs, and that makes the resulting sample sizes to be different.

If you "reverse" the definition of "success", then you also need to "reverse" the odds ratio, that is, the new value would be testoddsratio=0.44444 (which is 1/2.25), then you would get consistent results.

Hope this helps,

Jill

[Elisa97]

Thank you very much!! Have a nice day.

Elisa