Effect of zero imputation methods for log-transformation of independent variables in logistic regression
Communications for Statistical Applications and Methods · 31.07.2024 · 31 (4) · 409-425
DOI: 10.29220/csam.2024.31.4.409 ↗Kurz gefasst
Fachartikel in der Zeitschrift Communications for Statistical Applications and Methods, erschienen am 31.07.2024. Verfasst von Seo Young Park. Die Arbeit wurde bisher 9-mal zitiert (Crossref) und nennt 9 Literaturangaben. Gefördert von: Korea National Open University.
Abstract (Originalsprache)
Logistic regression models are commonly used to explain binary health outcome variable using independent variables such as patient characteristics in medical science and public health research. Although there is no distributional assumption required for independent variables in logistic regression, variables with severely right-skewed distribution such as lab values are often log-transformed to achieve symmetry or approximate normality. However, lab values often have zeros due to limit of detection which makes it impossible to apply log-transformation. Therefore, preprocessing to handle zeros in the observation before log-transformation is necessary. In this study, five methods that remove zeros (shift by 1, shift by half of the smallest nonzero, shift by square root of the smallest nonzero, replace zeros with half of the smallest nonzero, replace zeros with the square root of the smallest nonzero) are investigated in logistic regression setting. To evaluate performances of these methods, we performed a simulation study based on randomly generated data from log-normal distribution and logistic regression model. Shift by 1 method has the worst performance, and overall shift by half of the smallest nonzero method, replace zeros with half of the smallest nonzero method, and replace zeros with the square root of the smallest nonzero method showed comparable and stable performances.
Förderung
- Korea National Open University
Zitieren
APA
Park, S. Y. (2024). Effect of zero imputation methods for log-transformation of independent variables in logistic regression. Communications for Statistical Applications and Methods, 31(4), 409-425. https://doi.org/10.29220/csam.2024.31.4.409
Vancouver
Park SY. Effect of zero imputation methods for log-transformation of independent variables in logistic regression. Communications for Statistical Applications and Methods. 2024;31(4):409-425. doi:10.29220/csam.2024.31.4.409
BibTeX
@article{Park2024,
author = {Park, Seo Young},
title = {Effect of zero imputation methods for log-transformation of independent variables in logistic regression},
journal = {Communications for Statistical Applications and Methods},
year = {2024},
volume = {31},
number = {4},
pages = {409--425},
doi = {10.29220/csam.2024.31.4.409}
}
EUVIMED