CCR-Logistic Based Variable Importance Visualization: Differentiating Prime and Suppressor Variables in Logit Models

Ana Perišić, Ana Perišić, Ivan Sever

2025

Abstract

Logistic regression typically involves assessing variable importance. This task becomes considerably more challenging in the presence of correlated variables (predictors) and suppression. We present a procedure for determining variable importance in multiple logistic regression models that can distinguish between suppressor variables and prime predictors. We propose a simple visualization tool for representing variable importance that can help practitioners to determine important prime and suppressor variables when building the multiple logistic regression model. The methodology relies on the extension of the Correlated Component Regression approach to logistic regression (CCR-Logit), which utilizes linear combinations of predictors instead of original predictors and can easily be generalized to various regression models. CCR-logistic methodology can handle a large number of predictors and is especially useful when dealing with correlated predictors. The variable importance is quantified by observing standardized regression coefficients from univariate models and higher-order component models, where univariate models capture the direct effect on the outcome, while the higher-order component models capture the suppressor effects. The proposed methodology is presented on a real-world dataset within the field of tourism.

Paper Citation

in Harvard Style

Perišić A. and Sever I. (2025). CCR-Logistic Based Variable Importance Visualization: Differentiating Prime and Suppressor Variables in Logit Models. In Proceedings of the 14th International Conference on Data Science, Technology and Applications - Volume 1: DATA; ISBN 978-989-758-758-0, SciTePress, pages 43-52. DOI: 10.5220/0013461700003967

in Bibtex Style

@conference{data25,
author={Ana Perišić and Ivan Sever},
title={CCR-Logistic Based Variable Importance Visualization: Differentiating Prime and Suppressor Variables in Logit Models},
booktitle={Proceedings of the 14th International Conference on Data Science, Technology and Applications - Volume 1: DATA},
year={2025},
pages={43-52},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013461700003967},
isbn={978-989-758-758-0},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 14th International Conference on Data Science, Technology and Applications - Volume 1: DATA
TI - CCR-Logistic Based Variable Importance Visualization: Differentiating Prime and Suppressor Variables in Logit Models
SN - 978-989-758-758-0
AU - Perišić A.
AU - Sever I.
PY - 2025
SP - 43
EP - 52
DO - 10.5220/0013461700003967
PB - SciTePress