Improving BAHAMAS: Reducing Selection Effects Bias in Bayesian Hierarchical Supernova Cosmological Inference

We present improved inference results from BAyesian HierArchical Modelling for

the Analysis of Supernova cosmology (BAHAMAS) via the incorporation of selection effects bias in our probabilistic model. This approach improves upon previous renditions of supernova hierarchical modelling by calibrating aspects to simulations of the newly-released Dark Energy Survey (DES) data. Type Ia Supernovae (SNIa) observations are subject to selection bias, which affects inference of cosmological parameters extracted from SNIa observations. Using DES-type SNANA simulations of SNIa, we train binary classifiers to assign probabilities of selection to observed SNIa spectral data such that we propagate a bias for selection in a forward manner, instead of back-correcting data using the standard modulus correction approach. Given simulated SNIa light curve parameters, we test our hierarchical model with both probit and logit-trained selection effects. We estimate posterior distributions of cosmological and latent-layer parameters simultaneously. With selection effects, we report ≈20% decrease in Ω_m-bias, but an increase in w-bias by ≈25%. Future work will reduce our model’s w sensitivity to simulated intrinsic scatter, σ^int, and the shape of population-level color distribution.