Comparison of Bootstrap Methods and T-Methods: Capture Rates of Confidence Intervals and Probability of Type I Errors in Hypothesis Tests

• Activity Number: 366
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Education
• Author(s): Jeff Kollath*
• Companies: Oregon State University
• Keywords: bootstrap ; randomization tests ; technology in classroom

Abstract

Statistical educators have emphasized using technology in introductory statistics classes. Given the rapid growth of technology in our society, particularly with the current generation of students, incorporating a range of various technologies into the classroom engages students to get involved and more readily participate in the classroom discussion. In particular, some educators have proposed using bootstrap samples and randomization tests in the introduction to inference in introductory statistics courses. However, it is important to illustrate to the students in these classes how well the bootstrap methods perform compared to t-methods. Based on simulations from normal and skewed populations, this paper will compare the capture rates of population parameters of 95% confidence intervals for means using t-methods with 95% confidence intervals for means and medians using bootstrap methods. In addition, the performance of hypothesis tests will be compared. In particular, the percent of times a correct null hypothesis is rejected at a 5% significance level will be compared between using t-test with bootstrap methods.