Bootstrapping with R to Make Generalized Inference for Regression Model

Bootstrap is a resampling procedure drawn from an original sample data with replacement allocation method to build a sampling distribution of a statistic for statistical inference. This paper focuses to validate the generalized linear regression model by using the bootstrap method in order to make generalization of statistical inference to the different settings outside the original. The first application involved the bootstrap regression coefficients of predictors in the classical regression model while the others emphasized the bootstrap responses for binary outcomes in the logistic regression and for count data in the Poisson regression. The results on the bootstrap regression coefficients perform well even if the original data were restricted with small sample sizes and/or non-normal errors. The confidence intervals based upon the normal theory is quite narrower than the percentile interval and the bootstrap t interval. For the results of the bootstrap responses along a single predictor, both percentile confidence intervals of logistic and Poisson regression models perform well with a nice bandwidth of bootstrap responses for generalization.