Appropriate Null Hypothesis for Cosmological Birefringence

A recent paper (Nodland and Ralston, PRL 78, 3043, astro-ph/9704196) claims to have detected evidence for birefringence in the propagation of radio waves across cosmological distances. In order to assess the statistical significance of their results, the authors analyze simulated data sets, finding a stronger correlation in the real data than in the simulations. Unfortunately, the procedure used for generating the simulated data sets is based on an incorrect null hypothesis. Furthermore, the correct null hypothesis would lead to a stronger correlation in the simulated data sets, weakening the case for birefringence. We conclude that the paper's analysis does not provide statistically significant evidence of birefringence.