Wileman et al (Reference Wileman, Eagles and Andrew2001) evaluated bright white v. dim red light therapy for seasonal affective disorder in primary care and reported no significant difference in the proportions of responders in either group. Working on the basis of small trials having a large type II error, the group size for each group can be estimated. For most statistical tests, tables are available that show the power of the test to detect specified differences for a given α and sample size, as well as tables that show the required sample size to achieve selected power for specified difference and given α (Reference Machin and CampbellMachin & Campbell, 1987; Reference CohenCohen, 1988). In designing a clinical trial, we should select the power 1 — β to be at least 0.80, so that there is a chance of one in five or less of missing an important difference between treatments.

The differences in response rates between the two groups even using the broad remission criterion was 16.9%. To detect this difference at a significance level α=0.05, power 80%, 160 patients are required in each group (Reference Freeman and TyrerFreeman & Tyrer, 1992). However, only 57 patients in total were enrolled by Wileman et al. This reduced the power of the tests to about 20% and strongly suggests a type II error. It would therefore be misleading to conclude that bright white light is not associated with greater improvement.