Transformation techniques for cross-sectional and longitudinal endocrine data: Application to salivary cortisol concentrations
Transformation techniques for cross-sectional and longitudinal endocrine data: Application to salivary cortisol concentrations
复制标题
DOI：
10.1016/j.psyneuen.2012.09.013
复制
发表时间：
2013-06-01
影响因子：
3.7
通讯作者：
Plessow, Franziska
中科院分区：
文献类型：
--
作者：
Miller, Robert;Plessow, Franziska
内分泌时间序列往往缺乏正态性和均方差，这很可能是由于其自然决定因素的非线性动力学和生化分析工具的内在特征。因此，数据转换（例如，对数转换）经常被应用于支持基于模型的一般线性分析。然而，到目前为止，数据转换技术在不同的研究中有很大的不同，哪种是最佳的功率转换的问题仍有待解决。本报告旨在通过系统地比较不同的功率转换对数据正态性和均方差的影响，为内分泌时间序列的分析提供一个共同的解决方案。为此，将Box-Cox家族的各种权力转换应用于309名健康参与者的唾液皮质醇数据，这些参与者在时间上接近社会心理压力源（特里尔社会压力测试）。尽管我们的分析表明，在满足正态性和均方差性方面，非和对数转换数据都较差，但它们也为两者提供了最佳转换，反映分布浓度平衡的横断面皮质醇样本和纵向皮质醇时间序列，包括系统改变的激素分布，这些分布是由同时引发的脉动变化和连续消除过程引起的。考虑到这些内分泌振荡的动态，在测试glm之前进行数据转换似乎是必要的，以尽量减少偏差结果。(C) 2012 Elsevier Ltd.版权所有。
Endocrine time series often lack normality and homoscedasticity most likely due to the non-linear dynamics of their natural determinants and the immanent characteristics of the biochemical analysis tools, respectively. As a consequence, data transformation (e.g., log-transformation) is frequently applied to enable general linear model-based analyses. However, to date, data transformation techniques substantially vary across studies and the question of which is the optimum power transformation remains to be addressed. The present report aims to provide a common solution for the analysis of endocrine time series by systematically comparing different power transformations with regard to their impact on data normality and homoscedasticity. For this, a variety of power transformations of the Box-Cox family were applied to salivary cortisol data of 309 healthy participants sampled in temporal proximity to a psychosocial stressor (the Trier Social Stress Test). Whereas our analyses show that un- as well as log-transformed data are inferior in terms of meeting normality and homoscedasticity, they also provide optimum transformations for both, cross-sectional cortisol samples reflecting the distributional concentration equilibrium and longitudinal cortisol time series comprising systematically altered hormone distributions that result from simultaneously elicited pulsatile change and continuous elimination processes. Considering these dynamics of endocrine oscillations, data transformation prior to testing GLMs seems mandatory to minimize biased results. (C) 2012 Elsevier Ltd. All rights reserved.