I was having a browse through the latest paper to apply a hierarchical Bayesian model to population level inference of dark matter halo mass & stellar IMFs by Sonnenfeld et al. and I noticed a citation to a familiar name: Patil et al. (2010). Anand Patil was effectively my predecessor as computational statistician in the Malaria Atlas Project (if we ignore Samir Bhatt, sorry Sam!) and, aside from his work on (e.g.) recovering the prevalence-incidence relationship from horribly noisy data, Anand was a keen developer of Python tools: including PyMC. Looking back at the >130 citations to his PyMC paper one can see >30 (cited) uses of this code in astronomical studies! One implication of this surprising connection is that thanks to his work on PyMC Anand has managed to make a significant impact on my field before I have even been able to scratch the surface of his. But another more important point is that of the value of cross-disciplinary exchange, which (especially where computational statistics is concerned) will (I believe) soon be recognised as being essential for making real progress in modern astronomy.*
As for the Sonnenfeld et al. paper my quick browse suggested the hierarchical modelling has been done carefully and successfully; the only comment I would make is that by using a Monte Carlo integration approximation of the likelihood in their code (for the integrals in Section 4 with the measurement likelihood function of each galaxy as proposal) it sounds like the authors have drifted into the territory of pseudo-marginal MCMC and might therefore find some useful tips on tuning the number of proposals and acceptance rate from the corresponding theory (e.g. Doucet et al. 201[2/4]).
* To that end I should note that the BASP Frontiers conference is one example of a step in the right direction: http://www.baspfrontiers.org .