Institute for Advanced Study/Princeton University Early Universe/Cosmology Lunch Discussion

While cosmological parameter inference is predominantly Bayesian, in this talk, I will discuss how frequentist statistics can be used as a complementary method for parameter inference. I will introduce the frequentist profile likelihood and discuss in which cases it can give different results than the ones inferred from Bayesian posteriors. As an example, I will focus on the constraints of the early dark energy (EDE) model, a proposed solution to the Hubble tension. For this model, frequentist and Bayesian inference under Planck cosmic microwave background and BOSS galaxy clustering data give different results: while the Bayesian analysis clearly disfavours the EDE model, the frequentist analysis allows for sufficient amounts of EDE to resolve the tension. As a way forward, more data can help to restore the concordance between the two statistical approaches.