Befitting bootstrap analysis: A case study

Abstract

Bootstrapping is a popular method for inference and uncertainty quantifications. The key element of bootstrapping analysis is computing distributions of statistical estimators by resampling, with replacement, of a given data set. However, in practice, data often have some inherent data structure reflecting the data-generation process. The point in this article is that the corresponding bootstrap analysis needs to incorporate such information to enhance the quality of inference and uncertainty quantification. We propose applying a befitting bootstrap analysis (BBA) method reflecting the data generation structure. The proposed befitting bootstrap analysis method generalizes findings to a population frame with similar data generation processes. It is a follow up to the befitting cross validation (BCV) method proposed earlier by the same authors. A case study is used to elaborate the merits of the befitting bootstrap analysis method, in comparison with several conventional bootstrapping methods. The Python code used in the analysis is available in an openly available Github repository.