Comment on `Tainted evidence: cosmological model selection versus fitting', by Eric V. Linder and Ramon Miquel (astro-ph/0702542v2)

Abstract

In astro-ph/0702542v2, Linder and Miquel seek to criticize the use of Bayesian model selection for data analysis and for survey forecasting and design. Their discussion is based on three serious misunderstandings of the conceptual underpinnings and application of model-level Bayesian inference, which invalidate all their main conclusions. Their paper includes numerous further inaccuracies, including an erroneous calculation of the Bayesian Information Criterion. Here we seek to set the record straight.