Princeton University Gravity Initiative Seminar

Abstract: Gravitational-wave astronomy is entering a regime in which detections are sufficiently numerous to support population-level statistical inference. The standard tool is hierarchical Bayesian inference, which is powerful but does not natively provide a notion of goodness of fit. This is often acceptable when population models are well understood, but becomes problematic when they are not. In that regime, broad priors on nuisance parameters can combine into an effectively informative prior structure, and inference may converge to parameter regions that are weakly informed by the data. Similarly, model misspecification can drive convergence to biased or otherwise poor solutions while yielding apparently stable fits in model space. In this talk, I present a goodness-of-fit framework designed to operate in observed space rather than only in model space, providing a direct way to diagnose misspecification in hierarchical population analyses. I will show preliminary results for black-hole spin populations, and discuss how the framework may extend to more general population settings.