Less is More: Traffic-Improving Interventions with Braess's Paradox.

Abstract

This project investigates the presence of Braess’ paradox in a real-world setting using empirical traffic data from the city of Calgary, Alberta. Given a road network, Braess's paradox is the observation that removing a road can sometimes speed up total travel time across drivers. This effect occurs due to the selfish nature of drivers when they attempt to minimize their own travel time. To analyze this effect, the road network is modeled as a directed graph with travel times derived from the Greenshields' Traffic model. The travel times are calculated using observed traffic volumes, speed limits, and estimated jam densities. Traffic assignment is performed under Wardrop’s First Principle, which is approximated using both a sequential assignment and the Method of Successive Averages (MSA). Finally, individual roads and pairs of roads are blocked. Metrics after the roads are blocked are compared with metrics of the original baseline network. Results show consistent evidence that a subset of roads exhibits behavior aligned with Braess’ paradox, with total travel time reductions ranging from 0.02% to 0.23%. These findings are robust across both sequential assignment methods and MSA assignment methods. They also stay relatively consistent with varying levels of additional origin–destination demand. Furthermore, removing pairwise combinations of the subset of roads demonstrate approximately additive improvements, suggesting limited interaction between these affected routes. While the percentage that the total travel time decreases by may not be large, the results highlight the potential for network interventions to improve traffic efficiency without increasing infrastructure. This project contributes a computational framework for detecting Braess's paradox in real-world traffic networks.