论文部分内容阅读
对数转换的方法在生物医学和社会心理研究中处理非正态数据时被广泛应用。本文重点介绍该传统方法在处理非正态数据时存在的严重问题。尽管通常认为对数转换可以减少数据的变异性,使数据更符合正态分布,但是通常并非如此。此外,对数转换后的数据得出的标准统计测试结果往往和未转化的原始数据不相关。我们通过使用模拟数据示例来说明这些问题。我们认为如果采用数据转换,必须非常谨慎应用。我们建议研究者在大多数情况下摒弃这些处理非正态数据的传统方法,选择采用较新的不依赖于数据分布的方法:如广义估计方程(GEE)。
Logarithmic transformation methods are widely used to deal with non-normal data in biomedical and psychosocial research. This article highlights some of the serious problems with this traditional method of dealing with non-normal data. Although logarithmic transformations are generally considered to reduce data variability and make the data more in line with normal distributions, this is usually not the case. In addition, the standard statistical test results obtained from logarithmically transformed data are often not correlated with the unconverted original data. We illustrate these problems by using simulation data examples. We think we must be very cautious if we use data conversion. We suggest that in most cases, researchers abandon these traditional methods of dealing with non-normal data and choose newer methods that do not rely on data distribution: for example, Generalized Estimation Equations (GEEs).