Korean J Anesthesiol.  2020 Dec;73(6):503-508. 10.4097/kja.20137.

Data transformation: a focus on the interpretation

• Lee DK

• KMID: 2510863
• DOI: http://doi.org/10.4097/kja.20137

Abstract

Several assumptions such as normality, linear relationship, and homoscedasticity are frequently required in parametric statistical analysis methods. Data collected from the clinical situation or experiments often violate these assumptions. Variable transformation provides an opportunity to make data available for parametric statistical analysis without statistical errors. The purpose of variable transformation to enable parametric statistical analysis and its final goal is a perfect interpretation of the result with transformed variables. Variable transformation usually changes the original characteristics and nature of units of variables. Back-transformation is crucial for the interpretation of the estimated results. This article introduces general concepts about variable transformation, mainly focused on logarithmic transformation. Back-transformation and other important considerations are also described herein.

Keyword

Normality; Homoscedasticity; Transformation; Logarithmic; Power; Box-Cox transformation; Skewed distribution; Back-transformation; Retransformation