Analysing Complex Choice Experiments Data from a Data-Driven Perspective : Machine Learning Methods for Participatory Value Evaluation Experiments

This paper investigates the potential of association rules learning and random forests to analyse data from complex choice experiments, and to obtain relevant insights for choice modellers and practitioners. We apply these data-driven methods on data from a so-called Participatory Value Evaluation (PVE) choice experiment. In a PVE choice experiment, respondents choose a combination of alternatives, subject to a resource constraint. More specifically, we show that both methods can successfully assist choice modellers to specify their (theory-driven) choice models. We clarify how association rules learning can be used to identify interactions that would otherwise potentially go unnoticed; and how random forests can be used to determine the importance of each attribute, and as a mechanism to test the validity of the behavioural assumptions of a theory-driven choice model. By applying these methods, we are able to obtain considerable model fit improvements, as well as to reduce the time invested on the model specification phase. As such, our results support the notion that data-driven and theory-driven methods are complementary tools for choice modellers