Background We investigate the specification of roadway lighting for safety to understand the elements needed in statistical analysis of road collisions during night time. Several goals were targeted. First, which type of response is best, or whether both responses should be used. Second, which indicator of lighting should we favour? Third, which other factors should be included in the analysis and fourth, how effective is lighting in reducing nigh-time collision.

Methods The case study comprised illuminance and luminance measurements collected for the Arthabaska region in Quebec, along with available operational and geometric variables expected to explain roadway collisions. A zero-inflated negative-binomial model was used to analyse the impact of predictors on collision frequency and severity using classical maximum likelihood validated by a Full Bayesian regression

Results It was found that collision severity is best, resulting in more factors being significant in the expected sense of contribution. Luminance was the best indicator for road lighting. A correlation matrix aided in the identification of linearly dependencies between factors and the response or other factors. The last goal was investigated by comparing daytime with night-time collision analysis. The night time analysis included luminance and glare. The results were very close between day and night, with luminance proving to be an effective countermeasure for night collisions. A three-times difference on the coefficient for traffic volume was found. The use of a dummy variable related to standard levels of illumination is presented and will be key in future research for the estimation of effective levels of lighting.

Conclusions A connexion between roadway lighting and crash history can be used to support the warrant of lighting and identification of levels by using the statistical methods herein proposed adapted to test effectiveness of lighting levels and explanatory power of variables surviving co-linearity and significance tests.