What Do We Learn From an Equivalence Study Without Statistical Power? Kyle Peyton, MA, MPhil Annals of Emergency Medicine Volume 72, Issue 2, Pages 223-224 (August 2018) DOI: 10.1016/j.annemergmed.2018.03.001 Copyright © 2018 American College of Emergency Physicians Terms and Conditions
Figure Two simulated versions of the study by Linden et al, with idealized versus implemented sample sizes. Each plot shows a random sample of 100 results from 10,000 simulations of each design. Estimated differences in composite morbidity are displayed as 90% confidence intervals. Vertical dotted lines are the author’s selected equivalence range of 8%. Because equivalence is true in these simulations, gray intervals are correct rejections and black intervals are incorrect rejections. The idealized study has 0.84 power to detect equivalence, but the implemented study has 0 power. An explanation of the simulations presented here with accompanying R code is available at https://kylepeyton.github.io/assets/aem_sims.html. Annals of Emergency Medicine 2018 72, 223-224DOI: (10.1016/j.annemergmed.2018.03.001) Copyright © 2018 American College of Emergency Physicians Terms and Conditions