Cox C, Reeder J E, Robinson R D, Suppes S B, Wheeless L L
Division of Biostatistics, University of Rochester Medical Center, New York 14642.
Cytometry. 1988 Jul;9(4):291-8. doi: 10.1002/cyto.990090404.
A number of methods have previously been considered for the statistical comparison of flow cytometric frequency distributions. For two distributions, the foremost of these is the Kolmogorov-Smirnov (K-S) test, which has been criticized as "too sensitive." We discuss some alternative methods based on the Poisson distribution. The assumption of Poisson variation within channels allows the use of channel-by-channel confidence intervals and chi-square tests. These are simple and more appropriate for discrete data than the K-S test. Graphical displays of these and other techniques are presented. We also attempt to set the problem in an appropriate context. We argue that any statistical procedure must rest on a reasonable understanding of the nature of the variability in the system. This understanding takes the form of an appropriate probability model, which may be approximate but must provide a reasonably accurate description of the data. Incomplete understanding of the data can lead to inappropriate analysis. We discuss the assumptions that underlie our techniques and consider extensions to more complex situations.
此前，人们已经考虑了许多用于流式细胞术频率分布统计比较的方法。对于两种分布而言，其中最重要的是Kolmogorov-Smirnov（K-S）检验，但该检验被批评为“过于敏感”。我们讨论了一些基于泊松分布的替代方法。通道内泊松变化的假设允许使用逐个通道的置信区间和卡方检验。这些方法简单，并且比K-S检验更适合离散数据。本文展示了这些方法以及其他技术的图形化展示。我们还尝试将该问题置于适当的背景中。我们认为，任何统计程序都必须基于对系统变异性本质的合理理解。这种理解以适当的概率模型的形式呈现，该模型可能是近似的，但必须对数据提供合理准确的描述。对数据的不完全理解可能导致不恰当的分析。我们讨论了我们技术所基于的假设，并考虑了对更复杂情况的扩展。
