Bootstrap-Based Confidence Intervals: Methods, Applications, and Challenges

Abstract

Bootstrap methods offer a flexible and generally applicable framework for statistical inference, especially when classical parametric assumptions such as normality and large-sample approximations are violated. This paper offers a systematic review of confidence intervals constructed via bootstrap methods, including their conceptual foundations, methodological variants, and applications. After presenting an overview of the motivation for bootstrap confidence intervals and their advantages over classical analytical methods, three major methods are reviewed and discussed: the percentile method, the bias-corrected and accelerated method, and the bootstrap-t method. These methods are compared and evaluated according to their bias correction, complexity, and accuracy. Some applications of these methods, such as group comparisons, correlation coefficient estimation, and mediation effects, are reviewed and discussed to illustrate their applicability and effectiveness under non-normal and complex data conditions. Finally, some methodological challenges and limitations, such as small sample size, method selection, and computational issues, are reviewed and discussed. By synthesizing recent research and applied studies on confidence intervals via bootstrap methods, this review article attempts to highlight the applicability and relevance of confidence intervals via bootstrap methods as a robust and flexible statistical tool for modern statistical practices.