A Bayesian reanalaysis of the quasar dataset

We investigate recent claims of spatial variation in the fine structure constant on cosmic distance scales based on estimates of its extra-galactic–to–on-Earth ratio recovered from “many multiplet” fitting of quasar absorption spectra. To overcome the limitations of previous analyses requiring the assumption of a strictly unbiased and Normal distribution for the “unexplained errors” of this quasar dataset we employ a Bayesian model selection strategy with prior-sensitivity analysis. A particular strength of the hypothesis testing methodology advocated herein is that it can handle both parametric and semi-parametric models self-consistently through a combination of recursive marginal likelihood estimation and importance sample reweighting. We conclude from the presently-available data that the observed trends are more likely to arise from biases of opposing sign in the two telescopes used to undertake these measurements than from a genuine large-scale trend in this fundamental “constant”.