ABSTRACT
Chapter 8 highlights the ‘EZ principle’, which states that large-sample power for a wide variety of tests is dictated solely by the expected z-score. Consequently, formulas for power, sample size, or effect size for these tests all flow from one simple formula; for 80%, 85\%, or 90% power in a 2-tailed, level 0.05 test, equate the expected z-score to 2.80, 3.00, or 3.24, respectively. We apply this simple rule to t-tests, tests of proportions, the logrank test, and other settings with asymptotically normal test statistics. We include adjustments to account of correlation in a cluster-randomized trial. We also provide R programs for computing exact power for the t-test and Fisher's exact test. Practical considerations in sample size calculations are also discussed, including estimating nuisance parameters and specifying treatment effects.
