Local calibration of safety performance functions for multi-lane rural highways in Sonepat

This study develops a locally calibrated safety performance function (SPF) using a negative binomial (NB) model to estimate crash frequency on NH-44 and NH-334B corridors in Sonepat, Haryana. Crash data from 2018–2022 comprising 1230 cases were integrated with traffic, roadway, and pavement condition variables. Mixed NB models were not adopted due to the absence of significant unobserved heterogeneity after inclusion of key explanatory factors. Results indicate that annual average daily traffic is the most influential variable, with an elasticity of 1.387, implying that a 1% increase in traffic leads to ∼1.39% rise in crashes. Heavy-vehicle proportion also showed a significant positive effect, increasing crashes by 2.3% for every 1% increase. Pavement roughness (International Roughness Index) increased crash frequency by 9.3% per 1 m/km rise, while each additional metre of shoulder width reduced crashes by 13.2%. Earthen medians and absence of medians increased crash risk by 60% and 170%, respectively, compared to concrete barriers. Model validation demonstrated improved predictive accuracy over the national SPF, with reductions of 20.3% in mean absolute error, 26.2% in root mean squared error, and 25% in mean absolute percentage. The findings highlight the importance of incorporating local roadway and traffic characteristics for effective highway safety planning.