This paper is from Session 8B: Research on developing students’ reasoning using simulation methods for introductory statistical inference: Session 2

which comes under Topic 8: Research in statistics education

Bootstrapping for learning statistics

Presenter

• Tim Hesterberg (Google, United States)

Abstract

Collecting data, producing plots such as histograms and scatter plots, and calculating numerical statistics such as means, medians, and regression coefficients are relatively concrete operations. In contrast, ideas related to the random variability of those statistics—sampling distributions, standard error, confidence intervals, central limit theorems, hypothesis tests, P-values, and statistical significance—are relatively abstract, and more difficult for students to understand. Bootstrap methods and permutation tests take those concrete tools, that students are used to using with data, and apply them to sampling distributions. This promotes understanding. We demonstrate using two examples—one involving linear regression, the other comparing two sample means. We finish by discussing why the bootstrap works, and what to watch out for.