On the choice of the prior distribution in the bayesian analysis for the evaluation of mini roundabouts as a road safety measure

Abstract

An integral art of the Bayesian approach which is not present in the classical approach is the prior distribution. Different researchers may have different level of prior knowledge regarding the parameter of interest before seeing the data. Sometimes different prior distributions can result in different decisions, as such investigations have to be careful in making the choice of the prior distribution. In this paper, we compare results from the Bayesian analyses based on three possible choices of the prior distributions, which are uniform prior, lognormal prior and an improper prior, in the evaluation of the effectiveness of mini-roundabouts. Data from five before and after studies into the effect of mini-roundabouts when replacing priority junctions are used. The effects of the different prior distributions are distinguishable from the analysis of an anamolous 'desk-drawer' study. The uniform and improper prior pull the estimated treatment effect away from one more than the lognormal prior. The results based on lognormal prior depict a less worst scenario of the ineffectiveness of mini-roundabouts and this may correspond to the deficiency in engineering design at only a few sites. Consequently, it is more appropriate to use the lognormal prior in the analysis of mini-roundabouts as a road safety measure.