Thank you for your reply - yes, conditional model should be used for matched data. I liked your simulation part, too! ... View more

Thank you, Rick, for all your help! Just so you know, I ran the simulations using your program just to make sure mine is correct. Results are very close with yours is consistently a little bit more conservative than mine (1000 simulations, if mine has 81.9% power, yours would be 80.5). I ended up using results from your program to be safe. I really appreciate your expertise and help!!! ... View more

Thank you, Rich for your help! I went with our original thought, generated larger data from the model I'm testing, used Proc Surveyselect to select a 1:1 ratio case/control, and achieved what I wanted. The solution your sent is definitely more elegant. I'm saving it for the future. Thank you for your prompt replies, suggestions, and help! ... View more

Thank you for your reply!!!   This is actually very helpful. Yes I'm doing power for a case-control data (1:1 ratio) that aimed to test an interaction between a continuous variable and race. Generating logistic data using an interaction term is no problem for me, but I got stuck at making half the sample control, half case. Seems you are suggesting generating larger data and sample from it to get the 1:1 case-control data that I want? That is a great idea! Question though: Will sampling from the larger data set change the effect size I used to generate the larger data set?   Thank you! I'm so glad I posted the question after scratching my head in front of my computer for a whole day!! ... View more