Power Analysis Calculator

This calculator sizes a study before you run it. Give it the difference you want to detect, the standard deviation, the significance level and the power you are aiming for, and it returns the smallest number of participants per group that reaches it, along with the power one participant short so you can see the answer really is the minimum. A second action goes the other way and reports the power of a sample size you already have, plus what you would need for 80, 90 and 95 percent. A third does the same job for two proportions. The means calculations use the noncentral t distribution rather than the normal approximation, which is why they reproduce the published tables exactly: 394, 64 and 26 per group for small, medium and large effects at the usual settings.

Calculator Type
In the units you measure in. For a paired design this is the mean of the differences. To work from Cohen's d directly, put d here and 1 below.
The within-group standard deviation, assumed the same in both groups. For a paired design it is the standard deviation of the differences, not of the raw scores.
0.05 is the usual choice. It is the chance of calling a real difference when there is none.
80 percent is the convention and 90 percent is common for a trial that has to be right. Power is the chance of finding the effect if it is really there.
Leave at 1 for equal groups, which is the most efficient split. Ignored for a paired design.
Widget
Ignored for a paired or one-sample design.
0.05 is the usual choice. It is the chance of calling a real difference when there is none.
Widget
As a decimal, so 50 percent is 0.5.
0.05 is the usual choice. It is the chance of calling a real difference when there is none.
80 percent is the convention and 90 percent is common for a trial that has to be right. Power is the chance of finding the effect if it is really there.
Widget
Related Tools
Confidence Interval Calculator T Score Calculator ANOVA Calculator Critical Value Calculator