A Study of Multicollinearity Effects on a Binary Response
No Thumbnail Available
Date
2000-06-12
Journal Title
Journal ISSN
Volume Title
Publisher
Faculty of Science, University of Ilorin
Abstract
Regression analysis involves the study of existence of relationships between the dependent variable (the response) and sets of independent variables. If the correlation that exists between the independent variables is however high, then the estimates of the regression coefficients of these predictors are unstable and therefore unreliable. This is referred to as multicollinearity effect in this paper.
In this paper we consider the multicollinearity effect on a binary response using more than one independent variable. Without loss of generality we used the logistic function, which represents the family of monotonic functions, to model the response.
We observed that the expected response could differ sometimes remarkedly but similarly monotonic when multicollinearity exists between the predictors. This fact is useful in obtaining a value of the response of the logistic function which could be greater than the values of any of the individual function when the levels of the independent variables are carefully chosen.
Description
Keywords
Multicollinearity, Binary Response, Regression
Citation
Adejumo, A. O. and Jolayemi, E.T (2000). A Study of Multicollinearity Effects on a Binary Response; Nigerian Journal of Pure and Applied Sciences. 15:1088 – 1093. (Published by the Faculty of Science, University of Ilorin). https://www.unilorin.edu.ng/ejournals/index.php/njpas