##
Retrospective Power Calculations

This screen performs "after the fact" power analyses. After you have performed
a statistical analysis and have a p-value, this screen tells you: (1) the
power of your original analysis to detect this difference with this sample
size, (2) the minimum difference detectable with a sample this size, and
(3) the minimum sample size to detect this difference. This sort of analysis
is most often done after a non-significant (p>0.05) result has been obtained,
although it can be used with any value for p.

It should be noted that the whole idea of retrospective power calculations
is a controversial one. Purists maintain that the very concept is meaningless,
and that power calculations must be performed prior to the execution of the
experiment. (Download the
PDF document
by Russel Lenth to get this side of the story. And anyone feeling that
they need to do one of these calculations might want to first check out
Richard Stevens' interesting
web page, which can be thought of as a "pre-retrospective-power-calculation
screening test", to see if you really need to do one.) Yet the major statistical
software products (SAS, SPSS, etc.) routinely provide this kind of post-hoc
analysis. It is perhaps best thought of as using the results of your
just-completed experiment as if it were a pilot run to help you design your
next experiment. Anyway, for good or for bad, here's a web page that carries
out a typical set of basic retrospective power calculations.

This method used here is based on formulas for comparing two large unpaired
samples, but it appears to be applicable to a wide variety of two-variable
tests.

Note: Before using this page for the first time, make sure you read the
JavaStat user interface guidelines for important
information about interacting with JavaStat pages.