Robust Approximate Bayesian Inference with Synthetic Likelihood

Knowingly or unknowingly, a web of complex models underpins many features of modern life. Models are used to predict traffic patterns (via mechanistic network models or interconnected dynamic queues), help environmental agencies control pest populations (such as invasive toad species), and underpin policy decisions regarding epidemiological restrictions for disease prevention (such as restrictions aimed at reducing the spread of COVID-19). However, in general the models that are used to carry out such tasks are at best an approximation to the observed data, and an unwavering belief in the validity of these approximations can lead to severe and unintended consequences.

In this manuscript, ACEMS researchers develop a novel statistical approach to the parameter estimation of complex models. This approach allows researchers to ascertain the reliability/accuracy of complex models, deliver robust parameter estimates, and understand what features of the underlying model may be inaccurate in light of observed data. Notably, this method can be used in many situations where there currently exists no other mechanism to diagnose model reliability.

The proposed method is employed to analyze the empirical accuracy of a specific model for the behavior of invasive toad species. The empirical evidence suggests that a commonly employed model for predicting the behavior of certain invasive toads may be unreliable.