Krummenauer Frank
Institut für Medizinische Biometrie, Epidemiologie und Informatik der Universität Mainz.
Klin Monbl Augenheilkd. 2002 Nov;219(11):817-20. doi: 10.1055/s-2002-36323.
Standard statistics software packages offer a variety of significance tests. The major problem, however, is the correct choice of an appropriate significance test for the underlying data and design setting. In general, the choice of significance tests should decide between two sample versus one sample (i.e. interindividual versus intraindividual) analyses; a further determinant is the clinical endpoint's scale level (mainly continuous or categorical). Two sample comparisons can be performed using the Wilcoxon test for continuous endpoints and the exact Fisher test for binary endpoints. Intraindividual comparisons become feasible using the sign test for continuous and the McNemar test for binary endpoints. Time to event data can be analysed using the log rank test. If one measurement series has to be compared to a single reference value, the sign test can be applied for continuous data, the binomial test for binary data.
标准统计软件包提供了多种显著性检验。然而，主要问题在于为基础数据和设计设置正确选择合适的显著性检验。一般来说，显著性检验的选择应在两样本分析与单样本分析（即个体间分析与个体内分析）之间做出决定；另一个决定因素是临床终点的量表水平（主要是连续型或分类型）。对于连续终点，可以使用 Wilcoxon 检验进行两样本比较；对于二元终点，可以使用精确 Fisher 检验进行两样本比较。对于个体内比较，对于连续终点可以使用符号检验，对于二元终点可以使用 McNemar 检验。生存时间数据可以使用对数秩检验进行分析。如果要将一个测量系列与单个参考值进行比较，对于连续数据可以应用符号检验，对于二元数据可以应用二项式检验。
