Methods to evaluate the effectiveness of advanced stop lines for motorcycles in improving road safety

In Barcelona, Advanced Stop Lines (ASL) for motorcycles, were implemented since 2009. This paper aims to describe the process followed in determining the best statistical model to analyse the effectiveness of ASL in preventing road traffic injury collisions. A quasi-experimental design of an evaluation study of an intervention with comparison group was performed. • The starting model is the Poisson regression model, including type of area (ASL, comparison zone), period (pre and post-intervention), linear trend of the number of collisions, period and type of area interaction, as explanatory variables. • Various models are tested to correct existing overdispersion in the starting model: Poisson -Lognormal, Poisson- Gamma, Zero-Inflated Poisson, Zero-Inflated Negative Binomial, Hurdle model with Poisson distribution and with Negative Binomial distribution. To select the best model the Akaike Information Criterion is used. The final model is the Poisson-lognormal, adding the area as random factor (for each area repeated measures for different years are available). • The coefficients of the model parameters are interpreted in terms of relatives risks (RR), and the percentage change in the number of collisions is estimated in the post regards the pre-intervention period, from the RRs (- (1-RR)), to quantify the impact of ASL. The interaction allows assessment of whether the effect of the intervention differed between ASL and comparison zones.