Network architecture and traffic flows: Experiments on the Pigou-Knight-Downs and Braess Paradoxes

This paper presents theory and experiments to investigate how network architecture influences route-choice behavior. We consider changes to networks that, theoretically, exhibit the Pigou-Knight-Downs and Braess Paradoxes. We show that these paradoxes are specific examples of more general classes of network change properties that we term the "least congestible route" and "size" principles, respectively. We find that technical improvements to networks induce adjustments in traffic flows. In the case of network changes based on the Pigou-Knight-Downs Paradox, these adjustments undermine short-term payoff improvements. In the case of network changes based on the Braess Paradox, these adjustments reinforce the counter-intuitive, but theoretically predicted, effect of reducing payoffs to network users. Although aggregate traffic flows are close to equilibrium levels, we see some systematic deviations from equilibrium. We show that the qualitative features of these discrepancies can be accounted for by a simple reinforcement learning model.