Beta
314448

A Suggested Biased Estimator for Correcting Multicollinearity in Multinomial Logistic Regression

Article

Last updated: 28 Dec 2024

Subjects

-

Tags

-

Abstract

Multinomial logistic model suffers from multicollinearity that causes wider confidence intervals and incorrect decisions for testing hypotheses for the regression parameters. Biased estimators were used for correcting multicollinearity in linear regression, binomial logistic and recently multinomial logistic regression [2,8,11,12]. This paper introduces a new biased estimator for the multinomial logistic regression model. This estimator is an extension of Liu biased estimator [18], which originally was developed for correcting multicollinearity in linear regression model and then extended by Mansson et al., [19] for the binomial logistic regression model. The suggested biased estimator combines the advantages of Stien estimators and ridge regression estimators, which are: 1) reducing the mean square errors (MSES) for the estimates of the parameters, and 2) improving the conditioning of the estimated weighted information matrix in multinomial logistic regression. So it is expected that the performance of the suggested biased estimator would be better than other biased estimators. A procedure of the successive steps for determining the suggested biased estimator is introduced. Using a data set comparison between the results of applying the suggested biased estimator with the results of applying the existing estimators (multinomial logistic Stien estimators and multinomial logistic ridge regression estimators) show that the suggested biased estimator is superior in terms of a reduction in the variances of the multinomial logistic regression estimates and also a reduction in the Mean Squared Errors of Responses (MSER).  

DOI

10.21608/esju.2014.314448

Keywords

Biased Estimators, Multicollinearity - Multinomial Logistic Regression - Multinomial Stien Estimates - Multinomial Ridge Regression

Volume

58

Article Issue

2

Related Issue

43135

Issue Date

2014-12-01

Publish Date

2014-12-01

Page Start

183

Page End

197

Print ISSN

0542-1748

Online ISSN

2786-0086

Link

https://esju.journals.ekb.eg/article_314448.html

Detail API

https://esju.journals.ekb.eg/service?article_code=314448

Order

5

Type

Original Article

Type Code

1,914

Publication Type

Journal

Publication Title

The Egyptian Statistical Journal

Publication Link

https://esju.journals.ekb.eg/

MainTitle

A Suggested Biased Estimator for Correcting Multicollinearity in Multinomial Logistic Regression

Details

Type

Article

Created At

28 Dec 2024