Some brief notes on papers I digested from the recent crop of astro ph submissions:
(1) Cibrika et al.’s CODEX Weak Lensing: Concentration of Galaxy Clusters at z∼0.5: A study of the halo mass-concentration relation with a stacked weak lensing analysis. As input to their likelihood function the authors estimate the precision matrix for a multivariate error term across radial bins designed to account for various noise sources in the stacking such as intrinsic profile differences at fixed mass. In particular, a bootstrap procedure is used to produce an estimate of the covariance matrix to which is applied the de-biasing correction noted by Hartlap et al. (2007), because this estimator for the covariance matrix is seen to be noisy and known to be biased (though consistent). My concern would be along much the same lines as Selentin & Heavens (2015) in the sense that if the covariance matrix estimator is noisy, then so too is the precision matrix estimator and its uncertainty should be carried through into the Bayesian analysis stage. I would also note that the de-biasing formula used is only for the case of independent observations, an assumption violated by the bootstrapping approach (as pointed out in Hartlap et al.’s example).
(2) White et al.’s Prospects for Distinguishing Dark Matter Models Using Annual Modulation: A prospective analysis of the potential for direct detection experiments to distinguish particular dark matter models. This is a good example of sanity checking Bayesian model selection approaches through calibration of frequentist style Type I/II error rates with mock datasets a la Jenkins & Peacock (2011).
(3) El Badry et al.’s The statistical challenge of constraining the low-mass IMF in Local Group dwarf galaxies: Similar to the above but looking at parameter estimation and model selection via the BIC for distinguishing log-Normal IMFs from power-law IMFs; cautions against drawing conclusions where the data just aren’t sufficient to support a strong conclusion either way: more-or-less where the presence or absence of a turn over just isn’t identifiable at the depth of the available imaging.
Some non-astro statistics:
– A recent episode of BBC Radio 4’s More-or-Less podcast featured work by Peter Diggle’s group at Lancaster in much the same vein as what I do at the Malaria Atlas Project, but in this case for childhood malnutrition. Hopefully I will hear more about this paper by Justice Aheto at the (yet to be advertised) Lancaster Workshop on Disease Mapping in September. (Not to be confused with the Spatial Statistics 2017: One World, One Health conference also held at Lancaster this year, but in July!).