Options
Resampling methods for testing hypotheses
Author
Ng, Cheng Yee
Supervisor
Yap, Sook Fwe
Abstract
In this academic exercise, we examine the use of two resampling methods: permutation tests and bootstrap method for testing hypotheses on one and two sample problems concerning means and variances.
We first look at one-sample test on mean. The theory of the traditional i-test is reviewed and the bootstrap method introduced Two guidelines for bootstrap hypothesis testing are also highlighted. On two-sample test on means, we reviewed the traditional two-sample 1-test. Permutation test is introduced and the bootstrap method for testing two-sample means discussed. A new test statistic for bootstrap hypothesis testing is also proposed. On two-sample test on variances, we reviewed the F-test and discussed various permutation and bootstrap tests for comparing variances. At the end of each chapter on one and two sample problems, a simulation study was conducted to examine the performance of the various bootstrap plans and to compare the traditional tests with the two resampling methods.
Our simulation results show that bootstrap method can be applied more widely over the range of distribution types considered and their performance is comparable to permutation tests when testing equality of distributions.
We first look at one-sample test on mean. The theory of the traditional i-test is reviewed and the bootstrap method introduced Two guidelines for bootstrap hypothesis testing are also highlighted. On two-sample test on means, we reviewed the traditional two-sample 1-test. Permutation test is introduced and the bootstrap method for testing two-sample means discussed. A new test statistic for bootstrap hypothesis testing is also proposed. On two-sample test on variances, we reviewed the F-test and discussed various permutation and bootstrap tests for comparing variances. At the end of each chapter on one and two sample problems, a simulation study was conducted to examine the performance of the various bootstrap plans and to compare the traditional tests with the two resampling methods.
Our simulation results show that bootstrap method can be applied more widely over the range of distribution types considered and their performance is comparable to permutation tests when testing equality of distributions.
Date Issued
1998
Call Number
QA277 Ng
Date Submitted
1998