科技期刊编辑如何利用统计知识筛查学术不端行为

郑巧兰; 林燕薇; 洪悦民; 杨江瑜

文章摘要

郑巧兰,林燕薇,洪悦民,杨江瑜.科技期刊编辑如何利用统计知识筛查学术不端行为.编辑学报,2021,(1):74-77

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

DOI：10.16811/j.cnki.1001-4314.2021.01.017

中文关键词: 学术不端统计学科技期刊规则

英文关键词: academic misconduct statistics scientific journals rules

基金项目:

作者	单位
郑巧兰	中山大学附属第三医院期刊中心《新医学》编辑部,510630,广州
林燕薇	中山大学附属第三医院期刊中心《新医学》编辑部,510630,广州
洪悦民	中山大学附属第三医院期刊中心《新医学》编辑部,510630,广州
杨江瑜	中山大学附属第三医院期刊中心《新医学》编辑部,510630,广州

摘要点击次数: 906

全文下载次数: 712

中文摘要:

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RO或RH值；RO或RH值的95%CI的下限＝exp(B－1.96E_s),95%CI的上限＝exp(B＋1.96E_s)；且RO或RH值的95%CI如果包含1,对应的P值必然＞0.05；RO或RH值的95%CI必然包含RO或RH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。

英文摘要:

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 χ^2value, 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 RO(odds ratio) or RH(hazard ratio) value. The Lower limit of 95% CI of the RO or RH value ＝exp(B－1.96E_s)(E_s is standard error), upper limit ＝exp(B＋1.96E_s). And if the 95%CI of RO or RH value contains 1, then the corresponding P value must be ＞ 0.05. The 95% CI of the RO or RH value must include the RO or RH 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.

查看全文查看/发表评论下载PDF阅读器

关闭