Centre for Population Health Studies, Usher Institute of Population Health Sciences and Informatics, University of Edinburgh, Edinburgh, Scotland, UK.
Stat Med. 2019 Sep 30;38(22):4264-4269. doi: 10.1002/sim.8294. Epub 2019 Jul 1.
Two-tailed significance testing for 2 × 2 contingency tables has remained controversial. Within the medical literature, different tests are used in different papers and that choice may decide whether findings are adjudged to be significant or nonsignificant; a state of affairs that is clearly undesirable. In this paper, it is argued that a part of the controversy is due to a failure to recognise that there are two possible alternative hypotheses to the Null. It is further argued that, while one alternative hypothesis can lead to tests with greater power, the other choice is more applicable in medical research. That leads to the recommendation that, within medical research, 2 × 2 tables should be tested using double the one-tailed exact probability from Fisher's exact test or, as an approximation, the chi-squared test with Yates' correction for continuity.
2×2 列联表的双侧显著性检验一直存在争议。在医学文献中，不同的论文使用不同的检验方法，而这种选择可能决定研究结果是否被认为具有统计学意义；这种情况显然是不理想的。本文认为，争议的部分原因是未能认识到存在两种可能的备择假设。进一步认为，虽然一种备择假设可以导致更有效的检验，但另一种选择在医学研究中更适用。这导致了这样的建议，即在医学研究中，2×2 表应该使用 Fisher 精确检验的双侧单尾概率的两倍进行检验，或者作为近似值，使用带有 Yates 连续性校正的卡方检验。
