医学论文中常见统计学错误例析

An analysis of common statistical errors in medical papers

列举医学论文中常见的统计学问题,包括t检验的常见误用和χ~2检验的常见误用。t检验的常见误用有误用t检验代替配对t检验、误用t检验代替单因素方差分析、误用t检验代替重复测量数据的方差分析,χ~2检验的常见误用有误用χ~2检验代替Fisher确切概率法、误用χ~2检验代替秩和检验、误用χ~2检验代替配对χ~2检验。通过对医学论文中常见的统计学问题进行案例分析发现,统计学方法误用的主要原因有忽略统计学方法的应用前提,未能根据实际情况和统计分析的目的来正解选用统计学方法,未能充分理解研究资料是否满足参数检验的条件,不能正确判断计量资料所对应的实验设计类型。为了提高统计学应用的准确性,首先,应聘请统计学审稿专家对稿件进行质量把关;其次,编辑人员应加强自身统计学专业素养,提高统计学专业知识水平,在编辑稿件过程中加强对稿件中统计学处理的思考和把握,遇到有疑问的地方要及时翻阅统计学资料或请教相关统计学专家以求正确解决;最后,作者在进行科研设计时,要多与统计学专家交流合作,从源头上避免设计上的漏洞和偏差。从作者、编辑、审稿专家等三方提高统计学的应用能力,不仅能保证科研的科学性、可靠性和有效性,同时能减少对读者的错误信息的引导。

The common statistical mistakes seen in medical researches, including common misuse of t test and χ2 test, are enumerated in this paper. The common misuse of t test includes t test instead of paired t test, t test instead of one-way ANOVA, and t test instead of a repeated measures analysis of variance. χ2 test is incorrectly used instead of Fisher exact probability method, rank test, or paired χ2 test. The case analysis on the common statistical problems in medical papers shows that the main reasons for misuse of statistical methods are the application premise error of a special statistical meth- od, incorrectly choice of statistical methods divorced from the actual situation and the purpose of statistical analysis, not fully understanding whether the research data to meet the conditions of parameter testing, and incorrectly determine the measurement data corresponding to the experimental design type. To ensure the accuracy of statistical applications in a sci- entific journal, the editorial department should invite statistical review experts to control the quality of the manuscript; the editors should strengthen their own statistical professional quality, improve the level of statistical professional knowledge, deeply think and grasp the statistical analysis of the manuscripts in the process of editing manuscripts, timely look over the statistical data and consult the relevant statistical expert where you have a question. Finally, the authors should have more exchanges with the statistical experts in the research design, from the source to avoid the design of the loopholes and deviations. It is suggested to improve the application ability of statistics from the authors, editors and peer review experts to en- sure the scientificalness,reliability and validity of scientific research, and reduce the induction of the error messages to readers.