Network Econometrics and Traffic Flow Analysis

Congratulations to Alireza Ermagun for successfully defending his dissertation: “Network Econometrics and Traffic Flow Analysis.” He will soon be off to do a post-doc. img_2498


Short-term traffic forecasting aims to predict the number of vehicles on a link during a given time slice, typically less than an hour. For decades, transportation analysts tackled the forecasting of traffic conditions, while focusing on the temporal dependency of traffic conditions in a road segment. Following the emergence of spatial analysis in traffic studies, a growing interest has aimed to embed spatial information in forecasting methods. These approaches generally take advantage of the information that many of the cars that will be on one link soon are already on the network upstream of the relevant location, and of typical patterns of flow.

While embedding the spatial component in forecasting methods acts as a catalyst, its functioning is hindered by the constraints of spatial weight matrices. The positivity of components in spatial weight matrices postulates that traffic links have a positive spatial dependency. In essence, this hypothesis is necessary to represent complementary (upstream and downstream) traffic links. For simple single facility corridors, this may be sufficient. On the flip side of the coin is the competitive nature of traffic links. This nature demonstrates that competitive links bear a significant proportion of diverted vehicles, when one of them is saturated or closed. Short-term traffic forecasting is initially confined to scrutinizing complementary links. In consequence, the competitive nature of traffic links has been overlooked in the spatial weight matrix configuration and short-term traffic forecasting.

This dissertation overcomes this challenge by introducing concepts, theories, and methods dealing with network econometrics to gain a deeper understanding of how the components are interact in a complex network. More precisely, it introduces distinctive network weight matrices, yet alike in concepts and theories, to extract the existing spatial dependency between traffic links. The network weight matrices stem from the concepts of betweenness centrality and vulnerability in network science. Their elements are a function not simply of proximity, but of network topology, network structure, and demand configuration. The network weight matrices are tested in congested and uncongested traffic conditions in both simulation-based and real-world environments.


From the simulation-based viewpoint, a 3 × 3 grid network and Nguyen-Dupuis network are designed and adopted as main test networks along with several toy networks for pedagogical purposes. To simulate traffic flow, a macroscopic traffic flow model is selected due to the purpose of this dissertation, which deals with traffic flow in a link in a specific time slice and does not include the single vehicle-driver units. From the real-word viewpoint, a grid-like sub-network is selected from the Minneapolis – St. Paul highway system, which comprises 687 detectors and 295 traffic links. The traffic flow of each link is extracted in 30 seconds increments for 2015 as the most recent year.

The results of the analysis lead to a clear and unshakable conclusion that traditional spatial weight matrices are unable to capture the realistic spatial dependency between traffic links in a network. Not only do they overlook the competitive nature of traffic links, but they also ignore the role of network topology and demand configuration in measuring the spatial dependence between traffic links. Neglecting these elements is not simply information loss. It has nontrivial impacts on the outcomes of research and policy decisions. However, using the proposed network weight matrices as a substitute for traditional spatial weight matrices exhibit the capability to overcome these deficiencies. The network weight matrices are theoretically defensible in account of acknowledging traffic theory. As the elements of the network weight matrix more closely reflect the dependence structure of the traffic links, the weight matrix becomes more accurate and defensible. They capture the competitive and complementary nature of links and embed additional network dynamics such as cost of links and demand configuration.

Building on real-world data analysis, the results contribute inexorably to the conclusion that in a network comprising links in parallel and series, both negative and positive correlation showe up between links. The strength of the correlation varies by time-of- day and day-of-week. Strong negative correlations are observed in rush hours, when congestion affects travel behavior. This correlation occurs mostly in parallel links, and in far upstream links where travelers receive information about congestion (for instance from media, variable message signs, or personal observations of propagating shockwaves) and are able to switch to substitute paths. Irrespective of time-of-day and day-of-week, a strong positive correlation is observed between upstream and downstream sections. This correlation is stronger in uncongested regimes, as traffic flow passes through the consecutive links in a shorter time and there is no congestion effect to shift or stall traffic.


Although this dissertation tests and validates the network weight matrices in the road traffic network problem to derive the realistic spatial dependency between traffic links, they have potential for implementation in other disciplines such as geography, regional, and social network sciences. The network weight matrices have further applications not only in models of physical flow, but also in social networks for which links or nodes may be either competitive or complementary with each other.

You have seen other work we have done together, including research related to this dissertation, earlier on the blog. He has a quite a few publications with me, and he was only at Minnesota two years, (and is not counting papers with others since he’s been here) including: