Improved approximate Bayesian computation methods via empirical likelihood

Approximate Bayesian Computation (ABC) is a method of statistical inference that is used for complex models where the likelihood function is intractable or computationally difficult, but can be simulated by a computer model. As proposed by Mengersen et al. (Proc Natl Acad Sci 110(4):1321-1326, 2013), when additional information about the parameter of interest is available, empirical likelihood techniques can be used in place of model simulation. In this paper we propose an improvement to Mengersen et al. (2013) ABC via empirical likelihood algorithm through the addition of a testing procedure. We demonstrate the effectiveness of our proposed method through a nanotechnology application where we assess the reliability of nanowires. The efficiency and improved accuracy is shown through simulation analysis.