The statistical significance revolution

Abstract Statistical significance has long relied on the criterion of P less than or equal to .05. Although this threshold has generally functioned well, it has engendered some negative practices to circumvent it and been criticized as too inflexible. We concur with the statisticians and methodologists who are currently arguing for more flexibility to the P value and more reliance on the 95% confidence interval, a shift that is likely to change future practice in data analysis and interpretation for oncology.