David R. Smith Govinda Weerakkody Patrick D. Gerard 1998 <p>Power analysis after study completion has been suggested to interpret study results. We present 3 methods of estimating power and discuss their limitations. We use simulation studies to show that estimated power can be biased, extremely variable, and severely bounded. We endorse the practice of computing power to detect a biologically meaningful difference as a tool for study planning but suggest that calculation of confidence intervals on the parameter of interest is the appropriate way to gauge the strength and biological meaning of study results.</p> application/pdf 10.2307/3802357 en Wiley Limits of retrospective power analysis article