Population inference

When we want to say something about a population — all the stars of a certain kind, the distribution of planet sizes, the mass function of black holes — we almost never observe that population fairly. Surveys see bright things more easily than faint ones, nearby things more easily than distant ones, and simple things more easily than complicated ones. The map between what's out there and what we observe is the selection function, and getting it wrong quietly biases everything downstream.

A recurring thread in my research is doing population-level inference with imperfect and often poorly-known selection functions, rather than pretending they don't exist. The goal is to recover the underlying distribution of objects — and honest uncertainties on it — given that our view of them is incomplete and distorted.

What this involves:

• Building hierarchical models that separate the true population from the observational process.
• Characterising and propagating selection effects instead of ignoring them.
• Making this tractable for the very large samples that modern surveys provide.

These methods are deliberately general, which is why they show up across my work — from stellar astrophysics and astrometry to gravitational waves.