What if Proc Power Can’t Help?  Calculating Power via Simulation, Example with Count Regression

Reviewers for an article comparing opioid use after surgery questioned the amount of power for a Zero-Inflated Negative Binomial (ZINB) model.  Power was estimated by simulation because PROC POWER does not have a ZINB option. In this presentation, I will demonstrate how to use PROCs GENMOD and COUNTREG to estimate the distribution parameters for Poisson, Negative Binomial, ZIP, and ZINB models.  Then, the STREAMINIT() and RAND() functions will be used to simulate distributions in a DATA step.  Finally, the outcomes model on the simulated data will be run with 1000 reps and the power will be computed.

About the Presenter

Brandy Sinco is a statistician and database programmer at the University of Michigan Medical School. She has worked on a wide variety of research projects.  She holds a bachelors degree in engineering, and masters degrees in mathematics and statistics.  The American Public Health Association awarded Brandy a student research award for her statistics masters thesis on path analysis.  She has also received “Advanced Analytics” and “Data for Good” awards from the Midwest SAS Users Group conference.

Outside of work, she enjoys playing the piano, composing music, singing in choir, martial arts aerobics, yoga, and tai chi.