Tests of Random Assignment to Peers in the Face of Mechanical Negative Correlation: An Evaluation of Four Techniques

Abstract

I present a novel method of testing for random assignment to peer groups-a 'split sample test'-and evaluate it using simulations alongside three other commonly used tests: a permutation test, a test of joint significance of peer group dummies, and a test which controls for the mean type in the urn of potential peers. All four tests are designed to address finitesample mechanical negative correlation between peers. While the first three tests have correct size and similar power against an alternative hypothesis of positive correlation, the split sample test is the only one which provides an estimate of the magnitude of correlation between peers. The test which controls for the mean type in the urn of potential peers has low power in all simulations and incorrect size in some.