Landis J R, Miller M E, Davis C S, Koch G G
Department of Biostatistics, School of Public Health, University of Michigan, Ann Arbor 48109.
Stat Med. 1988 Jan-Feb;7(1-2):109-37. doi: 10.1002/sim.4780070114.
This paper is concerned with the analysis of multivariate categorical data from epidemiologic and clinical studies with longitudinal designs. An expository discussion of pertinent hypotheses for such situations is provided within the context of two relevant data sets. Appropriate large-sample tests of these hypotheses are developed through the application of weighted least squares to generate Wald statistics. These procedures are illustrated with extensive analyses of one of these data sets. In some situations, the resulting cross-classification of the response variables leads to extremely sparse frequency data, especially when the number of subjects is not large. For such repeated measurement designs in which a single variable is measured repeatedly over time, this paper considers the use of a generalized Mantel-Haenszel strategy for tests of marginal homogeneity (symmetry). These randomization model methods are illustrated for data in which the repeated measurement variable is reported on an ordinal scale. This paper also focuses on the available computing software to implement these methods within the version 5 release of the SAS system. The randomization model approach can be implemented within the FREQ procedure and a broad range of models and hypotheses can be investigated within the CATMOD procedure.
本文关注对具有纵向设计的流行病学和临床研究中的多变量分类数据的分析。在两个相关数据集的背景下，对这类情况的相关假设进行了阐述性讨论。通过应用加权最小二乘法生成 Wald 统计量，开发了针对这些假设的适当大样本检验。通过对其中一个数据集的广泛分析来说明这些程序。在某些情况下，响应变量的交叉分类会导致频率数据极其稀疏，尤其是在样本数量不大时。对于在一段时间内对单个变量进行重复测量的此类重复测量设计，本文考虑使用广义 Mantel-Haenszel 策略进行边际同质性（对称性）检验。针对重复测量变量按有序尺度报告的数据，举例说明了这些随机化模型方法。本文还重点介绍了在 SAS 系统版本 5 中用于实现这些方法的可用计算软件。随机化模型方法可在 FREQ 过程中实现，并且可以在 CATMOD 过程中研究广泛的模型和假设。
