Re: The ongoing tyranny of statistical significance testing in biomedical research

Excerpt

I would like to thank Stang et al. [1] for an excellent commentary on the limitations and problems with significance testing as currently practiced in biomedical research. I would, however, like to also put in a good word for the P-value by way of defending its being reported, along with confidence intervals, in statistical analyses. It is of course the case that the confidence interval generally renders the P-value redundant. From the point estimate, along with a confidence interval and knowledge of the model, one could perform the necessary transformations to calculate the standard error estimate and thence the P-value. I do nevertheless believe that, in spite of its redundancy, the P-value is generally worth reporting. As noted by Stang et al., the P-value, at least under the Fisherian interpretation, gives a measure of evidence against the null. The difference between a P-value of 0.02 and 0.0002 is of consequence and I do not believe the average reader of the biomedical literature is in general capable of distinguishing between these two given the confidence interval alone. …