Johnny O’Meara, University of Notre Dame
This paper makes two contributions. First is proposing a permutation test for testing equality of variances in two populations. The test is easy to implement, has exact size in finite samples under the sharp null, and has correct size in large samples. Monte Carlo simulations allow for a comparison between the permutation test and classical tests for equality of variances that point to many settings where the permutation test outperforms classical tests. Second is to demonstrate that testing for clustering in regression analysis can be thought of as testing for equality of variances. Finally, the permutation test is empirically illustrated using data from Tennessee’s Project STAR. The permutation test indicates that clustering should be done at the classroom level, which is consistent with previous findings.
Read the full paper here.