Revisiting the empirical fundamental relationship of traffic flow in highways using a causal econometric approach

Abstract

The fundamental diagram of traffic flow, or in other words, the vehicular flow-speed or flow-density relationship, for a highway section is determined by both physical characteristics of the section and attributes of the population of drivers and vehicles travelling through it. The empirical estimation of this relationship by fitting a regression curve to a cloud of observations of traffic variables may not be robust as the relationship may suffer from confounding from observed and unobserved attributes of driving behaviour and vehicular characteristics. This paper adopts a causal econometric approach to obtain a more reproducible characterisation of the fundamental relationship. We relax the strict stationary state assumption and instead focus on the cause-effect interactions between flow and speed to model the spatio-temporal variations in traffic at a macroscopic level. We use traffic data from a highway section in California. We adopt a fully-flexible nonparametric specification for the relationship and apply instrumental variables estimation to control for the aforementioned confounding bias. We deliver some new empirical insights into the macroscopic level dynamics of traffic flow in a highway section alongside verifying some existing ones.