Titolo della tesi: Advancing knowledge on speed management and developing a speed prediction model for a mixed traffic
Road traffic fatalities and injuries pose a significant global challenge, particularly in low- and middle-income countries (LMICs), where a major cause is consistently linked to speeding (i.e., excessive or inappropriate speeds). Meanwhile, LMICs continue to face challenges in managing speeding risks. Part of the problem is the absence of knowledge on speed management that can guide effective policies, with LMICs frequently resorting to adopting solutions from high-income countries (HICs), which may not always be suitable for their unique contexts. Critical gaps in the literature include a lack of robust speed prediction models suitable for LMICs, a lack of synthesis and understanding of the role of speeding in contributing to crashes and fatalities, and unclear linkages between speed management and various transport externalities, including emissions, travel time, and road user behaviour.
This thesis aimed to fill these gaps through two principal objectives: (1) to synthesize and enhance knowledge on speed, speed management, and speed outcomes under different thematic areas of speed with an emphasis on LMICs; (2) to calibrate and validate a generalized and transferable desired speed prediction model for mixed traffic to predict mean and operating speeds on both urban and rural LMIC roads.
The methodology of this thesis combined various literature reviews, quantitative synthesis, and empirical data analysis to achieve its objectives. Literature reviews, including exploratory, scoping, and systematic reviews, were conducted to synthesize existing research on speed-related issues. This was complemented by quantitative synthesis to synthesize estimates. For the development of speed models, data on the spot speed of 39,413 vehicles and road attributes were collected at 64 sites across five LMICs, including Argentina, Colombia, Ecuador, Kenya, and Uganda. This data was used to calibrate and validate the models using robust multilevel modelling techniques (MLM) and multiple linear regression (MLR). Multilevel models were particularly used to address the hierarchical nature of the data and to account for within-group dependencies and between-group variability.
The results of the quantitative synthesis showed that speeding contributes to approximately 54% of fatalities worldwide, 57% in LMICs, and 28% in HICs and that reducing speed limits had positive impacts on emissions in both high-speed and low-speed environments. The results of the empirical analysis revealed distinct speed hierarchies and variabilities of vehicle types between rural and urban areas. The rural MLR models identified posted speed limit, access point density, lane width, and land use type as significant predictors, with high explanatory power. Urban models had fewer significant predictors but still achieved good performance. A combined rural-urban model performed well overall but had larger validation errors than the rural and urban models. In both rural and urban settings, all-vehicle models outperformed car-only models. The MLM highlighted the importance of considering the hierarchical nature of the data, as they revealed significant country-level effects on individual driver speeds. This country effect was not significant for aggregated speed data models, however, underscoring the issue of ecological fallacy. All models were validated on an independent data set from the five countries.
These findings advance the understanding and knowledge of speed management issues in LMICs and provide evidence-based, LMIC-focused solutions, and reliable and adaptable speed prediction models that transportation practitioners and policymakers can use to improve road safety in LMICs, and globally