Significance testing - are we ready yet to abandon its use?
Understanding of the damaging effects of significance testing has steadily grown. Reporting p values without dichotomizing the result to be significant or not, is not the solution. Confidence intervals are better, but are troubled by a non-intuitive interpretation, and are often misused just to see whether the null value lies within the interval. Bayesian statistics provide an alternative which solves most of these problems. Although criticized for relying on subjective models, the interpretation of a Bayesian posterior probability is more intuitive than the interpretation of a p value, and seems to be closest to intuitive patterns of human decision making. Another alternative could be using confidence interval functions (or p value functions) to display a continuum of intervals at different levels of confidence around a point estimate. Thus, better alternatives to significance testing exist. The reluctance to abandon this practice might be both preference of clinging to old habits as well as the unfamiliarity with better methods. Authors might question if using less commonly exercised, though superior, techniques will be well received by the editors, reviewers and the readership. A joint effort will be needed to abandon significance testing in clinical research in the future.