Statistical power in biological psychiatry.

The use of statistical power and power analysis in both the design and evaluation of experiments in biological psychiatry is described. The possible consequences of low power investigations are discussed, and guidelines are provided to facilitate the application of power analysis. Power curves are provided for sample sizes ranging from 10 to 100 for the Student's t test, and for testing the significance of an obtained Pearson correlation coefficient at both the 0.01 and 0.05 alpha levels. Additionally, difference scales are provided for plasma cortisol and for several neurotransmitter metabolites that are frequently measured in cerebrospinal fluid (CSF). Power evaluation of selected CSF studies measuring neurotransmitter metabolites in depressives and controls suggests the majority had less than a 50% chance of detecting a medium size difference before the experiment was actually performed.