[ Home ]                                                        Research Methods in Psychology,  S. J. Gilbert, SUNY-Oneonta

Statistical POWER is the probability that a study will produce a statistically significant result when its hypothesis is, in fact, true.  A POWER rating of
.8 (80%) or better is considered desirable.


Suppose that I do a simple experiment to test the hypothesis that women smile more than men.  And let us further suppose that nature actually does predispose women to smile more than men (i.e., my hypothesis is TRUE).

I obtain a sample of women and a sample of men, and I measure the mean number of times per minute that they smile.  The results show a higher mean for women than for men.  Fine.  But will the difference between the means that I obtained in my experiment be statistically significant?  That depends on the statistical power of my experiment and the test I am using.

If it has HIGH power, say .90, then the chances are very high (9 in 10) that the results of my experiment will be statistically significant, thus enabling me to conclude that the results actually DO SUPPORT the hypothesis.  But suppose the statistical power is very LOW, say .50.  Then, even though my hypothesis is true, the likelihood that my results will produce statistical significance is only 50%.  Even though I am right, I can only say so half the time!

So, it is important that our experiments have high (.80 or better) statistical power.  But what contributes to statistical power?  That's what the figure below shows.  Look it over carefully.  True to figure it out.  Then we will talk about it in class!


  [ Home ]