CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: ABC in a compartmental model for smoking habit dynamics Authors:  Alessio Lachi - University of Florence (Italy) [presenting]
Cecilia Viscardi - Univeristy of Florence (Italy)
Michela Baccini - University of Florence (Italy)
Abstract: Smoking is the main risk factor for many common chronic diseases. In order to describe the evolution of the population's smoking habits in Tuscany (Italy), we have developed a compartmental model. Compartmental models assume that at any given time, the population is divided into groups called ``compartments'' and that individuals can move from one to the other following simple probabilistic rules described by a system of differential equations. The population is divided into Never, Current, and Former smokers, with ex-smokers allowed to relapse smoking. The likelihood function of the model is complex to evaluate analytically, and the model requires specific estimation methods. We investigated the use of approximate Bayesian computation (ABC), a class of likelihood-free algorithms, as a tool to perform inference in both a frequentist and a fully Bayesian context. From a frequentist perspective, we used ABC as a method for the "stochastic search" for optimal parameters, using the deterministic version of the model and compared the results with those obtained from standard optimization algorithms. From a fully Bayesian perspective, we used ABC to sample from the joint posterior distribution of model parameters. Our results suggest that ABC is a powerful method to provide solutions in complex compartmental models.