Abstract

Bayesian spatial models often forms the basis of fairness evaluations in public health or economics. In this talk I show that approximate Bayesian inference through the INLA methodology offers a computationally efficient way to fit models for fairness audits while accounting for uncertainty through the fully Bayesian framework. Various stakeholder metrics can be calculated and produced from the obtained posteriors, enabling an efficient auditing framework that enables policymakers to evaluate the fairness of the decision framework in question.