科技期刊编辑如何利用统计知识筛查学术不端行为

How can editors of scientific journals use statistical method to screen academic misconduct

论文的统计学质量是学术期刊质量控制的重中之重。本文介绍几种常用统计学检验方法的规则,结合案例分析,分享《新医学》在文章统计学质量把关方面的经验:1) t检验和秩和检验中|t|或者|Z|若<1.96,对应的P值必然>0.05,且在相同样本量的情况下,|t|或者|Z|越大,P值越小。2)四格表的卡方检验,χ2值若<3.84,则对应的P值>0.05;在样本量相同时,χ2值越大,P值越小。3) logistic回归分析或Cox生存分析中,自然对数的底(e)的回归系数(B)次方,即exp(B),必须等于RO或RH值;RO或RH值的95%CI的下限=exp(B-1.96Es),95%CI的上限=exp(B+1.96Es);且RO或RH值的95%CI如果包含1,对应的P值必然>0.05;RO或RH值的95%CI必然包含RO或RH值exp(B)。编辑人员应掌握基本统计知识,并增强与作者的沟通,刊发文章应有多次统计审稿,并要求作者提供具体的统计值。通过上述规则和经验,希望帮助科技期刊编辑初步筛查可能存在统计结果错误的文章,以避免刊出因统计问题所致的结果造假类学术不端论文。

Statistical quality is the top priority of academic journal quality control.This article introduces the rules of several commonly used statistical tests combined with case analysis to share our experience in the statistical quality control of articles.1) If|t|or|Z|in the t test and rank sum test was<1.96,the corresponding P value must be>0.05.And in the case of the same sample size,the larger|t|or|Z|,the smaller the P value.2) If the χ2 value is<3.84 in the Pearson chi-square test of the 2×2 table,the corresponding P value must be>0.05.When the sample size is the same,the larger the χ2 value,the smaller the P value.3) In logistic regression analysis or Cox survival analysis,the exp(B)(B is regression coefficient) must be equal to the RO(odds ratio) or RH(hazard ratio) value.The Lower limit of 95%CI of the RO or RH value=exp (B-1.96Es)(Esis standard error),upper limit=exp(B+1.96Es).And if the 95%CI of RO or RH value contains 1,then the corresponding P value must be>0.05.The 95%CI of the RO or RH value must include the RO or RH value(exp(B)).Journal editors should master basic statistical knowledge,and enhance communication with authors.Published articles should have multiple statistical review,and the authors should required to provide specific statistical values.We hope the above-mentioned rules and experience can help editors initially screen for articles that may have incorrect statistical results,and avoid publishing academic misconduct papers with falsified data results.