|
以P<0.05作为判定“显著性”的标准,无疑给判定研究结论带来了便利,但同时也引起了大量的反对声音。比如P=0.049和P=0.052之间有什么本质差异么?以P<0.05作为效应是否存在的二项判定依据,数十年来一直充满争议,甚至有人提出要废除P值,但也有不少铁粉誓死捍卫P值。2022年,生态学第一刊Trends in Ecology & Evolution上发表了“Rewriting results sections in the language of evidence”一文,对P值的使用提出了新的建议。
首先,作者简要回顾了P值的历史和确切定义:
The P-value is the probability of observing a specific data summary (e.g., an average) that is at least as extreme as the one observed, given that the null hypothesis (H0) is correct.
然后,作者分析了8本生态进化领域的主流期刊(Ecology letters, Ecology, Journal of ecology, Journal of applied Ecology, ProceedingsB, Evolutionary_Applications, Biology_Letters, Evolution_Letters)2021年第一期上发表的所有论文中P值的使用情况。一通研究之后,作者发现,当前,P<0.05标准依旧十分风靡:97.3%的论文依然用到了“significance”一词。
We used the January 2021 issues (December 2020 if January 2021 was a special issue) of eight major journals in ecology and evolution and checked all research papers containing at least one statistical analysis (n = 137, see the supplemental information online). Of those, 113 (82.5%) reported results based on the NHST philosophy: 104/113 (92%) of the dichotomous decisions were based on the P-value, while seven used the 95% CIs, and two used an information criterion. A total of 110/113 (97.3%) reported their findings using the ‘significance’ terminology. It appears as if the decades with waving warning flags had relatively little impact on the routines in our field when it comes to writing the results sections of scientific papers.
可见,对于P值应用是否得当的大讨论,对P值在实际应用中的影响微乎其微。同时,作者指出,有的方法,虽然不直接用P值,比如AIC,BIC值,95% CI,bayes factors这些方法,其实都可以与P值转化或者等价。所以在应用统计中想完全废弃P值,并不现实。
于是作者提出了一种针对P值的一种新的表述方式,即P值依然可用,但表述有所变化。这种新的表述方法,主要是依据P值的范围,将其描述为对应的证据性语言(evidence language),而不是显著性语言。P值和Evidence强弱的对应关系为:
Figure 1. Suggested ranges to approximately translate the P-value into the language of evidence.
同时作者也给出了一些更改表达的演示(第一列为显著性语言表达,第二列为证据性语言表达):
针对前面提到的八大期刊中的一些论文中的表达,作者也给出了对应的evidence language下的表达方式(第一列为论文中的表达,第二列为作者建议的证据性表达):
据Web of science查询可以,自2022年3月发表以来,目前该论文已获得304次引用,可见这一建议已经获得了较高的认可度,值得推荐!
其中就包括上周Nature的封面Meta分析论文:
文献来源:
Muff, S., E. B. Nilsen, R. B. O’Hara, and C. R. Nater. 2022. Rewriting results sections in the language of evidence. Trends in Ecology & Evolution 37:203-210.
欢迎大家关注鄙人的个人公众号:二傻统计
Archiver|手机版|科学网 ( 京ICP备07017567号-12 )
GMT+8, 2024-12-22 10:22
Powered by ScienceNet.cn
Copyright © 2007- 中国科学报社