Tales Mistold: Does congestion pricing increase throughput

In the FHWA publication (and widely repeated elsewhere): Congestion Pricing: A Primer the claim is made (p.3) that the throughput on priced lanes exceeds those on unpriced lanes, using California SR91 as an example. Quoting:

“Effects of Pricing on Vehicle Throughput Vehicle “throughput” on a freeway is the number of vehicles that get through over a short period such as an hour. Once freeway traffic exceeds a certain threshold level, both vehicle speed and vehicle throughput drop precipitously. Data show that maximum vehicle throughput occurs at free flow speeds ranging from 45 mph to 65 mph. The number of vehicles that get through per hour can drop by as much as 50 percent when severe con- gestion sets in. At high traffic levels, the freeway is kept in this condition of “collapse” for several hours after the rush of commuters has stopped. This causes further unnecessary delay for off-peak motorists who arrive after rush hour. With peak-period highway pricing, a variable toll dissuades some motorists from entering freeways at those access points where traffic demand is high, and where such surges in demand may push the freeway over the critical threshold at which traffic flow collapses. Pricing prevents a breakdown of traffic flow in the first instance, and thus maintains a high level of vehicle throughput throughout the rush hours. As shown in the graph above, each variably priced lane in the median of State Route 91 in Orange County, California, carries twice as many vehicles per lane as the free lanes during the hour with heaviest traffic. Pricing has allowed twice as many vehicles to be served per lane at three to four times the speed on the free lanes. ”

As a transportation engineer, I can see several flaws with this argument. The SR91 toll lanes are a separate facility than the SR91 free lanes. If the tolls were removed from the SR91 toll lanes, would the throughput be as high? If so, the gain cannot be attributed to the tolls. This is an easy experiment to do (set the tolls to zero one day). It should be noted, the throughput at a particular cross-section might be lower on the untolled lanes, because of queue spillbacks. Because the toll lanes are separated from the untolled lanes, they are able to bypass the series of bottlenecks that are SR91. Throughput is most appropriately measured at active bottlenecks. The discharge rate at bottlenecks when demand exceeds capacity is nearly constant (well within 10% of the pre-queue formation maximum rate) according to much research in the area: e.g. Michael J. Cassidy and Robert L. Bertini (1999) Some traffic features at freeway bottlenecks. Transportation Research Part B: Methodological. Volume 33, Issue 1, February 1999, Pages 25-42 or: Zhang, Lei and David Levinson Some Properties of Flows at Freeway Bottlenecks. Journal of the Transportation Research Board No. 1883 122-131. so one has to assume that if the toll lane outperforms the untolled lane, the problem is that the throughput is not in fact measured at the active bottleneck, but rather upstream of the bottleneck. It is natural that this might have a lower per-lane throughput than the bottleneck itself.

Think about it this way: A bottleneck has 2 lanes, assume a capacity of 2000 vehicles per lane per hour, so it can handle 4000 vehicles per hour. Upstream there are 3 lanes (or 2 lanes, but they are joined by an on-ramp, so effectively 3 lanes). At the peak those three lanes each carry 2000 vehicles per lane per hour for a short period of time. However once those vehicles reach a bottleneck that can only accommodate 4000 vehicles, a queue forms (at the rate of 2000 vehicles per hour spread across the the three lanes). Eventually that queue reaches the measurement point, and flow there drops from 2000 vehicles per lane per hour to 1333 vehicles per lane per hour. To the untrained eye, this may appear as a loss of throughput due to being untolled, but in fact it is a loss of throughput due to a bottleneck spilling backwards. The 2000 vplph was unsustainable because of the downstream bottleneck, and would only be observable for a short period of time. Pricing cannot solve that problem except by reducing demand.

The comparison made in the FHWA report is fundamentally misleading, and in the long run does no service to road pricing. The evidence on I-394 for instance suggests that at bottleneck points the throughput on the free lanes exceeds that on the tolled lanes, because the toll lanes have prices that prevent demand from approaching capacity (to ensure reliability). As readers of this blog will note, I am of course supportive of congestion pricing in general (if the transaction costs can be brought down), but making unsupportable and misleading claims should be discouraged.