Jung S H, Kang S H
Division of Biostatistics, Department of Medicine, Indiana University School of Medicine, Indianapolis 46202-5289, USA.
Stat Med. 2001 Mar 15;20(5):785-94. doi: 10.1002/sim.705.
Ordered categorical data summarized in a 2 x K table usually consist of two-sample multinomial or K-sample binomial observations. In analysing these data, we usually assign scores to the K columns and perform a testing for the equality of two multinomial distributions in the former case and no trend among K binomial proportions in the latter case. Among the most popular score tests are the Wilcoxon rank sum test and the Armitage's linear trend test. In this paper we extend the score tests to be used for clustered data under diverse study designs. Our methods do not require correct specification of the dependence structure within clusters. The proposed tests are based on the asymptotic normality for large number of clusters and are a generalization of the standard tests used for independent data. Simulation studies are conducted to investigate the finite-sample performance of the new methods. The proposed methods are applied to real-life data.
汇总在2×K表中的有序分类数据通常由两样本多项分布或K样本二项分布观测值组成。在分析这些数据时，我们通常会给K列赋值，并在前一种情况下对两个多项分布的相等性进行检验，在后一种情况下对K个二项比例之间是否存在趋势进行检验。最常用的得分检验包括Wilcoxon秩和检验和Armitage线性趋势检验。在本文中，我们将得分检验扩展到用于不同研究设计下的聚类数据。我们的方法不需要正确指定聚类内的依赖结构。所提出的检验基于大量聚类的渐近正态性，是用于独立数据的标准检验的推广。进行了模拟研究以考察新方法的有限样本性能。所提出的方法被应用于实际数据。
