Statistical power: planning for it, estimating it post hoc, and reporting it

Abstract

To test the theories that drive technical communication research, investigators may statistically analyze data gathered for descriptive or experimental studies. In such analyses, investigators often set a very small statistical risk of rejecting a true null hypothesis of no relationship between variables to avoid subsequently incorrectly accepting an alternative hypothesis that there is a relationship. By this normal procedure, investigators may unwittingly lower the statistical power to reject a false null hypothesis of no relationship, and, thereafter, they may incorrectly fail to accept the statistically alternative hypothesis that there is a relationship. Our purpose is to demonstrate how to use a statistical table for planning ahead to gain acceptable power and how to report the power fully in the results. Even after an experiment has been completed, investigators can still estimate and report the power. Careful attention to power contributes to more meaningful tests of theories, and good reporting gives readers a clearer picture of the meaning of the tests.