A modified multinomial baseline logit model with logit functions having different covariates

Abstract

The multinomial logistic regression is a useful tool in the health and life sciences. In this paper, we propose a modified multinomial baseline logit model for nominal polychotomous data. The modified model is suitable for use in the situation where separate logistic models may be functions of different covariates. An estimation procedure is presented. The modified model is an alternative to the multinomial baseline logit model and the multivariate sparse group lasso. Simulation shows that this modified model outperforms the multinomial baseline logit model and the multinomial sparse group lasso. A real data set about an adolescent placement study is analyzed to demonstrate flexibility and efficiency of the modified model.

Keywords:

• Logit function
• Multinomial baseline logit model
• Multinomial logistic regression
• Multinomial sparse group lasso
• Nominal polychotomous data

Mathematical Subject Classification:

• Primary 62H12
• Secondary 62J07; 62C15

Acknowledgments

The authors would like to sincerely thank the Editor-in-Chief, Professor N. Balakrishnan and an anonymous referee for their constructive comments that led to the current substantially improved version.

Additional information

Funding

Ding’s research was supported by National Natural Science Foundation of China (NSFC) Grant 11571219. Liu’s research was supported by NSFC Grant 71601123 and MOE (Ministry of Education in China) Project of Humanities and Social Sciences (Project No. 15YJC910004).