The MCPOWER1 macro will perform Monte Carlo power calculations across a range of sample sizes, for linear, logistic, and Poisson regression, and will interpolate to nice round power values and make plots as well. MCPOWER1 macro.
The macro is described in this SAS Global Forum paper.
The MCPOWER_WINDOW macro contains a menu-based access to the first macro. MCPOWER_WINDOW macro.
Here's an example of tabled output. It shows that for a power of 0.9 to detect a slope of 1.5 you need 422 subjects, while you can detect a slop of 1.75 with 307 subjects. Since these are derived via Monte Carlo methods, they are estimates rather than solutions, and CI on the number of cases are also relevant.
nsubs lowclnsubs upclnsubs
Obs altslope desiredpower Req Req Req
1 1.50 0.5 127 113 141
2 1.50 0.6 172 159 184
3 1.50 0.7 235 212 257
4 1.50 0.8 329 312 346
5 1.50 0.9 422 411 436
6 1.75 0.8 162 143 180
7 1.75 0.9 307 276 338
Here's an example of the (unlovely but useful) graphical output. The title details that these values come from a logistic regression with t-distributed predictor (with 10 df), and that 25 data sets were simulated under the alternative for each point.
:
