Covariates and sample size effects on parameter estimation for binary logistic regression model

• Published in: Malaysian Journal of Science
• Main Author: Hamid H.A.; Wah Y.B.; Xie X.-J.
• Format: Article
• Language: English
• Published: Malaysian Abstracting and Indexing System 2016
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85031934401&doi=10.22452%2fmjs.vol35no1.7&partnerID=40&md5=358dad442cca3f4998c587dcf13efaa6

The types of covariate and sample size may influence many statistical methods. This study involves a rigorous Monte Carlo simulation to illustrate the effect of different types of covariate and sample size on parameter estimation for binary logistic regression model. The simulation study covers different sample sizes and types of covariate (continuous, count, categorical). This study shows how the MLE parameter estimates are affected by different types of covariate. The simulation results confirm that the parameter estimates improves as sample size increases. Results for single normal, two normal, categorical and count covariate show that sample size below 50 produced highly biased estimates. For model with skewed covariate, sample size of 150 and below produced biased estimates. The variability of parameter estimate increases when of the Poisson distribution increases. An application to a real data set confirms the results of the simulation study.