Sleep Deprivation and Performance of Residents: The Power of the Study

Abstract

To the Editor.— I read with interest the comments 1 elicited by the article by Deaconson et al 2 on sleep deprivation among residents. Frankly, I was surprised that no one challenged the power statement of Deaconson et al. It read: The study was designed to detect a difference as small as 10% or less for each of the five psychometric tests. That is, when comparing sleep-deprived and non—sleep-deprived subjects, there is a 90% chance (power equals 0.90) of detecting a difference for each of the five variables when there is only a 10% difference between the two sleep states. Note the sample sizes of 10 subjects in August, 9 in September, and only 7 in October! When the outcome is a continuous variable (eg, scores on a psychometric test), power is determined by the effect size (in this study the authors chose 10%), alpha error (.05), and characteristics of the