Statistics, biomedical scientists, and circulation research.
Statistics, biomedical scientists, and circulation research.
复制标题
统计学、生物医学科学家和循环研究。
DOI：
10.1161/01.res.42.6.739
复制
发表时间：
1978
影响因子：
20.1
通讯作者：
B. Hoffman
中科院分区：
文献类型：
--
作者：
M. Rosen;B. Hoffman
在我们担任《循环研究》编辑期间，我们收到了许多改进该杂志的建议。一些记者对《华尔街日报》撰稿人使用和可能滥用统计方法表示关切。斯坦顿·格兰茨博士写了一篇广泛的评论，其中部分写道：“未配对和配对的学生/测验可能是检验生物医学研究中假设的最流行的统计学方法。这一统计数据检验了两组人来自同一人群的零假设。然而，除了检验这一假设之外，它还被广泛误用于通过测试每一对组来检验两个以上组之间的差异。例如，使用重复的/-测试将同一对照组与多个不同的治疗组或反应在不同的时间进行比较是很常见的。这一过程完全是不正确的，违反了t分布背后的假设，产生了高得令人误解的显著水平(P值)。格兰茨博士还对《华尔街日报》最近的一卷进行了详细的分析，并指出了他认为哪些论文使用了适当的方法，哪些他认为选择了不当的方法。在79篇论文中，他说：“20篇文章(25%)没有使用统计学方法来比较他们的试验组；16篇(20%)报告的工作数据分为两组，因此适合使用配对或非配对/-检验进行分析；7篇(9%)使用方差或协方差分析正确地分析了多组数据；36篇(46%)用/-检验不正确地分析了多组数据。”这些交流使我们探讨了两个领域：第一，《华尔街日报》发表的论文在多大程度上错误地使用了统计方法，以及过去在多大程度上错误地使用了统计方法；第二，可以做些什么来确保今后尽量减少不适当地使用统计方法。
During our tenure as editors of Circulation Research, we have received a number of suggestions to improve the Journal. Several correspondents have been concerned with the use and possible misuse of statistical methods by contributors to the Journal. Dr. Stanton Glantz wrote an extensive critique which stated, in part," The unpaired and paired Student's/-tests are probably the most popular statistical methods to test hypotheses in biomedical research. This statistic tests the null hypothesis that two groups were drawn from the same population. In addition to testing this hypothesis, however, it is widely misused to test for differences between more than two groups by testing each pair of groups. For example, it is commonplace to compare the same control group with a number of different treatment groups or responses at different times using repeated/-tests. This procedure is simply incorrect and, by violating the assumptions underlying the t distribution, produces significance levels (P values) which are misleadingly high." Dr. Glantz also provided a detailed analysis of a recent volume of the Journal and indicated which papers he believed had used appropriate methods and those which he believed had chosen inappropriate methods. Of 79 papers, he states," twenty articles (25%) did not use statistical methods to compare their experimental groups; 16 (20%) reported work in which the data fell into two groups and, hence, were appropriate for analysis using the paired or unpaired/-tests; 7 (9%) analyzed multigroup data correctly using an analysis of variance or covariance; and 36 (46%) analyzed multigroup data incorrectly with/-tests." These communications have led us to explore two areas: first, what is the extent to which statisstical methods are and have in the past been used incorrectly in papers published in the Journal and second, what can be done to ensure that inappropriate use of statistical methods will be minimized in the future.