Reply to: Overconfidence in Bayesian analyses of galaxy rotation curves

Abstract

Cameron {\it et al.}~2019 (hereafter C19) recommends a more cautious and rigorous approach to statistical analysis in astronomy. We welcome this particular side of their communication as it helps stimulating the effort towards the adoption of better statistical methods in galaxy rotation curves, an effort to which we contributed with Rodrigues {\it et al.}~2018 (hereafter R18). Indeed, R18 was the first work that, in order to conclude on the universality of the acceleration scale $a_0$, studied the posterior distributions on $a_0$ of a large set of galaxies. As C19 agrees, the credible intervals were found within the Bayesian framework, that is, the marginalized posteriors on $a_0$ were found using Bayes' theorem to update the priors in light of the observational data; this process was done without introducing any approximation. Considering R18, C19 also remarks that: i) better methods to select the nuisance parameters and the corresponding priors could be used; ii) a quality cut based on $\chi^2$ values should not be used, and iii) the compatibility of the posteriors should be assessed in a more robust way. In the following, after first clarifying the context of our work, we address these criticisms.