Dr Ewan Cameron, PhD (Astronomy & Astrophysics, RSAA/ANU/MSO).
(List of my publications on google scholar).
Former astronomer, currently a statistician/epidemiologist with the Malaria Atlas Project at the University of Oxford. Moving to Perth (Australia) to join the Telethon Kids Institute & Curtin University (along with the rest of the MAP team, relocating en masse) in February 2020.
My experience and interests lie mostly within the realm of Bayesian inference, specifically:
- Approximate Bayesian Computation (a method for approximating the Bayesian posterior when the likelihood function is intractable but simulating mock datasets is cheap);
- Bayesian semi/non-parametrics (i.e., hierarchical Bayesian models built with stochastic process priors); particularly methods that exploit results from asymptotic theory;
- Methods for improving Bayesian inference under model specification (i.e., how to estimate model parameters when we know that the model is only a crude approximation to the true data generating process);
- Bayesian optimisation (a method for approximating the log likelihood surface when evaluations are computationally expensive and for adaptive scheduling of new evaluation points);
- Bayesian approaches to kernel methods (e.g. Bayesian distribution regression);
- computational methods for marginal likelihood estimation (computing the normalisation constant for a Bayesian posterior, as used in Bayesian model selection)
This is the second incarnation of this blog; the first ran from approx 2015-2018. My ambition with this blog has always been to highlight the huge volume of bad statistical analyses published in astronomical journals with a hope that calling out bad practice will in some way lead to reform. More importantly, given that a large proportion of these bad analyses are touted as “Bayesian”, it is my concern that once it becomes clear how many mistaken conclusions are being reached in this way that opinion will turn unfairly against the statistical framework rather than the people who’ve been mis-using it. If so, we might see journals starting to ‘ban’ Bayesian studies just like they now ‘ban’ p-values. Sadly, we may find that one day this blog can become a historical record of how the credibility of Bayesian methods in astronomy was destroyed.
The nature of the academic system is that the burden of work regarding the planning and implementation of large astronomical data analyses falls on the shoulders of PhD students and postdocs, since professors either lack the skills and knowledge to help with this task or simply don’t care to take part in it. Unfortunately, for this reason the authors of many of the studies I criticise here are junior researchers, which risks that this blog becomes a platform for ‘punching down’. Being aware of this, I will try my best to be respectful to junior researchers at all times and I ask that commentators remember this policy as well. On the other hand, professors responsible for egregiously bad statistical analyses are fair game for ridicule!