The use and abuse of hypothesis tests: how to present P values

Abstract

This overview highlights some of the key issues involved in performing and interpreting hypothesis tests. We describe the general approach taken in performing a hypothesis test with a focus on how to state the null and alternative hypothesis, and why two-sided tests are usually more appropriate than one-sided tests. We describe best practice techniques in performing and presenting the results of hypothesis tests. We recommend that, alongside any p-values, authors should also present estimates of the size of any treatment effects and their confidence intervals. Furthermore, they should specify the exact p-value rather than using terms such as 'NS' or the commonly used asterix notation. We discuss other pitfalls that are encountered at the analysis stage such as the use of repeated observations on individuals, the use of multiple tests on the data and the erroneous use of parametric tests when data are not normally distributed and vice versa. We highlight these points using two different examples: one looking at the use of compression stockings for preventing the occurrence of DVT on long-haul flights and a second hypothetical study comparing laser versus surgery techniques for the removal of varicose veins.