Logistic regression is one of the most widely used statistical models in epidemiological studies and many other fields. It can be used in cross-sectional or longitudinal data analysis to model risk factors and provide a reasonable estimate of odd ratios. In our study, we assessed the stability of logistic regression model on modeling low birth weight by bootstrapping and Monte Carlo simulation. Results from bootstrapping and Monte Carlo simulations were compared against asymptotic statistical inference. Our study confirmed that a small sample size could cause a large variance and bias in logistic regression models. We also observed that the biases of coefficients of covariates by Monte Carlo simulations were less than those by bootstrapping methods. Results from our study confirmed that the variances and biases decreased as sample size increased. Our study provided users of logistic regression insights in model building.

logistic regression, bootstrapping, Monte Carlo simulation, low birth weight.