topic Re: Sample size for binary outcome and continuous (%) predictor in Statistical Procedures

Sample size for binary outcome and continuous (%) predictor

sas_question123 — Mon, 07 Dec 2020 15:37:02 GMT

I have a project in which I have 60-70 samples.

Outcome = binary (positive or negative case)
Exposure = % of the time participants followed new guidelines (0-100%)
Question - would it be correct to use logistic regression to analyze this?
I am guessing a sample size of 60 or 70 is too small.

What would be the best way to calculate the sample size? Difficult to find examples where the predictor is continuous and outcome is binary.
Also generally, how would the proportion of positive cases affect the analysis? I don't expect there'll be much, if any, but would like to know how to calculate the sample size in both scenarios (regular number of positive cases and low number of positive cases). Thank you.

Re: Sample size for binary outcome and continuous (%) predictor

StatDave — Mon, 07 Dec 2020 15:40:00 GMT

Use the LOGISTIC statement in PROC POWER.

Re: Sample size for binary outcome and continuous (%) predictor

sas_question123 — Mon, 07 Dec 2020 16:07:51 GMT

Thank you! I'm very new to power/sample size calculations. I pasted my code below. If I expect the predictor to be really skewed to the left, would exponential be a good vardist to choose?

proc power ;
logistic
vardist ("predictor") = exponential ()
testpredictor ="predictor"
power = 0.8
ntotal=.;
run ;

Re: Sample size for binary outcome and continuous (%) predictor

StatDave — Mon, 07 Dec 2020 16:19:45 GMT

Exponential and gamma are both reasonable distributions for positively valued and right skewed data