A Study of The Performance for Variable Selection of Ordinal Regression Models under Multicollinearity Using Reversible Jump Algorithm

Authors

  • Chonlanan Sukthuayat Department of Mathematics and Statistics Faculty of Science and Technology Thammasat University,Rangsit Center
  • Saengla Chaimongkol Department of Mathematics and Statistics Faculty of Science and Technology Thammasat University,Rangsit Center

Keywords:

Proportional odds model (PO), Non-Proportional odds model (NPO), Partial proportional odds model (PPO), Reversible Jump Markov Chain Monte Carlo (RJ-MCMC)

Abstract

The purpose of this study was to investigate the performance of variable selection in the ordinal regression models when explanatory variables were correlated using a reversible jump method. Three models were considered: Proportional odds model, Non-Proportional odds model and Partial proportional odds model. The criteria of this study is percentage of average misclassification. Each model includes a response variable with three levels and five explanatory variables   with four difference types of correlation matrices: independent, constant correlation structure, Toeplitz, and Hub Toeplitz. The results show that variable selection performance using RJ-MCMC does not depend on the relationship structure and the level of relationship between the explanatory variables. The performance of variable selection of PO models was higher than that of NPO models.

References

Agresti, A. (2010). Categorical Data Analysis. 3rd ed. Canada: John Wiley & Sons.

Ari, E., and Yildiz, Z. (2014). Parallel Linear Assumption in Ordinal Logistic Regression and Analisis Approaches. International Interdisciplinary Journal of Scientific Research, 1(3), 1-16.

Gelman, A., Carlin, J. B., Stern, H. S., David B. Dunson, Vehtari, A., and Rubin, D. B. (2014). Bayesian Data Analysis. 3rd ed. United States: CRC Press.

Hardin, J., Garcia, S. R., and Golan, D. (2013). A Method for Generating Realistic Correlation Matrices. The Annals of Applied Statistics, 7(3), 1733-1762. doi: 10.1214/13-AOAS638.

Hastie, D., and Green, P. J. (2011). Model choice using reversible jump Markov Chain Monte Carl. Statistica Neerlandica, 66(3), 309–338. doi: 10.1111/j.1467-9574.2012.00516.x.

McKinley, T. J., Morters, M., and Wood, J. L. N. (2015). Bayesian Model Choice in Cumulative Link Ordinal Regression Models. International Society for Bayesian Analysis, 10,1-30. doi: 10.1214/14-BA884.

Peterson, B., Harrell, F. E., and Jr. (1990). Partial Proportional Odds Models for Ordinal Response Variables. Journal of the Royal Statistical Society Series C (Applied Statistics), 39(2), 205-217.

Tutz, G., and Scholz, T. (2003). Ordinal regression modelling between proportional odds and non-proportional odds. Technical report, Institute of Statistics, University of Munich.

Downloads

Published

2018-12-31

How to Cite

Sukthuayat, C., & Chaimongkol, S. (2018). A Study of The Performance for Variable Selection of Ordinal Regression Models under Multicollinearity Using Reversible Jump Algorithm. Journal of Applied Statistics and Information Technology, 3(2), 19–30. Retrieved from https://ph02.tci-thaijo.org/index.php/asit-journal/article/view/195410