Wang Hongyue, Chen Guanqing, Lu Xiang, Zhang Hui, Feng Changyong
Departments of Biostatistics and Computational Biology, University of Rochester, Rochester, NY.
Department of Biostatistics, St. Jude Children's Research Hospital, Memphis, TN.
Shanghai Arch Psychiatry. 2015 Oct;27(5):319-25. doi: 10.11919/j.issn.1002-0829.215121.
The sample geometric mean has been widely used in biomedical and psychosocial research to estimate and compare population geometric means. However, due to the detection limit of measurement instruments, the actual value of the measurement is not always observable. A common practice to deal with this problem is to replace missing values by small positive constants and make inferences based on the imputed data. However, no work has been carried out to study the effect of this naïve imputation method on inference. In this report, we show that this simple imputation method may dramatically change the reported outcomes of a study and, thus, make the results uninterpretable, even if the detection limit is very small.
样本几何平均数已广泛应用于生物医学和心理社会研究中，以估计和比较总体几何平均数。然而，由于测量仪器的检测限，测量的实际值并非总是可观测到的。处理这个问题的一个常见做法是用小的正常数替换缺失值，并基于插补后的数据进行推断。然而，尚未开展任何工作来研究这种简单插补方法对推断的影响。在本报告中，我们表明，即使检测限非常小，这种简单的插补方法也可能显著改变研究报告的结果，从而使结果无法解释。
