Erratum

“Demystify Statistical Significance – Time To Move On From The P Value To Bayesian Analysis” by J. Jack Lee [J Natl Cancer Inst 2011;103(1): 2–3 ]. On page 3, the sentence on lines 13–14 should read as follows: “Based on the binomial one-sided test, Trial 1 has a P value of .121.” As a result, the next sentence should read as follows: “Does this mean that the chance of the null hypothesis being true is 1 in 9 and the chance of the alternative hypothesis being true is 8 in 9?” Also, lines 24–26 would read as follows: “Therefore, the posterior odds of the null hypothesis against the alternative hypothesis are only 1:3, not 1:9 as when inferring from the frequentist P value.” On page 3, the second sentence in the paragraph that begins with “Now, assume…” should read as follows: “Trial 2 observes 27 responses in 100 patients ( $\widehat{\theta }=0.27$ ) and has a P value of .056.” Furthermore, to attain the P values reported in the original editorial, Trial 1 should have 5 responses in 10 patients ( P = .033) and Trial 2 should have 28 responses in 100 patients ( P = .034). Even though both trials have comparable P values, based on the Bayes factor, the odds are 1 in 7.6 in favor of the alternative hypothesis for Trial 1 and 1 in 3.7 in favor of the null hypothesis for Trial 2.

The author regrets these errors.