Evaluating the Reliability and Consistency of Statistical Models for Speed Distribution Analysis at Hazardous and Non-Hazardous Roadway Locations: Multifraction Data Set Approach

Understanding the statistical dynamics of traffic speeds at hazardous and non-hazardous locations is essential for effective roadway safety interventions. This study investigates the distinct characteristics of spot speed distributions across six Indian highway segments, including National and State Highways. It uses continuous probability distributions and hypothesis testing to assess the statistical significance of speed differences between hazardous and non-hazardous locations. The analysis is based on observed spot speed measurements, stratified into four data fractions (25%, 50%, 75%, and 100%), obtained using a simple random sampling with replacement approach. Seven continuous probability distributions, including normal, lognormal, gamma, logistic, Weibull, Burr, and generalized extreme value (GEV), have been fitted independently for each location type and data fraction to capture their distributional characteristics. The location, scale, and shape parameters of the models have been estimated using maximum likelihood estimation. However, model adequacy has been confirmed using Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values. Furthermore, a two-sample Kolmogorov–Smirnov test has been conducted to assess the statistical difference in speed profiles between hazardous and non-hazardous locations. The results reveal that the GEV distribution consistently outperforms other models across all locations and data fractions, demonstrating strong parameter stability and model adequacy. Larger data fractions improved model performance and hypothesis testing power, indicating greater distributional robustness. To address the potential effect of vehicle interactions and transient congestion during the observation periods, a modified Kaplan–Meier (KM) framework is used to estimate congestion-adjusted desired speed distributions. The KM-based findings show that hazardous roadway locations exhibit higher desired speed potential and greater upper-tail speed characteristics compared with non-hazardous locations. Interestingly, statistically significant speed differences have been found in nearly all settings, confirming the notion that crash-prone zones exhibit distinct speed dynamics. These findings have significant implications for road safety policy and infrastructure design, as well as the need for location-specific speed management strategies.