Consequences of Model Misspecification in Approximate Bayesian Computation

点击翻译按钮获取中文摘要

In almost any empirical application, the model the analyst is working with constitutes a misspecified description of the true process that has generated the data. While the method of Approximate Bayesian computation (ABC) is now a staple in the toolkit of the applied modeller, the impact of misspecification in ABC is unknown. This project aims to undertake a rigorous study into the behaviour of ABC under model misspecification. Expected outcomes include new theoretical results for ABC under misspecification and new methods capable of detecting/mitigating model misspecification. This project will provide significant benefits in all spheres where reliable, robust statistical inference methods are required in order to make reliable decisions.