wasteful excess capacity unless the peak users are charged the full cost of the expansion. Long-distance telephone service allocates the demand with a market mechanism by charging a premium for a call during the peak and by offering discounts during the off peak. By shifting a small portion of the peak demand into the off peak, excess capital investments are avoided and society receives the service it requires at a lower cost. Consumers appear to accept a market-based system for allocating demand in long-distance telephone service with some equanimity, and experience with peak-period pricing in air travel suggests that these policies work and that they are accepted by the populace (Gillen, Vol. 2).

In the case of roads, however, demand is allocated by congestion. Everyone who wants to travel in the peak tries to do so, and the excess demand causes congestion. Economists and travel demand forecasters believe that travel decisions are governed by the “full price of travel,” which includes the out-of-pocket expense plus the value of the time it requires. Given that even so-called free road use comes with significant congestion during the peak, it appears reasonable to ask why motorists willingly endure the delays that occur when they travel at the most congested times. Indeed, the available evidence suggests that when congestion gets bad enough, some travelers shift the timing of their trips. These shifts, however, are not enough to alleviate the crush. The reason that shifts in behavior by drivers confronted with congestion will never be sufficient to reduce congestion (without some form of pricing or rationing) is that motorists using the highways during the peak do not have to pay for the delays they cause each other.

To illustrate this problem, imagine an 8-km (5-mi) long, high-speed segment of highway with a flow rate of 1,999 vehicles per hour in each lane. Such a flow rate would be quite high and nearing the point when the design capacity of the facility would be exceeded. As demand approaches capacity, the throughput of the facility begins dropping disproportionately, which causes delay to mount rapidly. Standard assumptions about vehicle speed under these conditions imply that the trip for the average vehicle will require 25 min (Wohl and Hendrickson 1984, 290). The next vehicle to enter the traffic stream will have a slightly longer trip than 25 min but would add about half an hour of delay to be shared by all the following motorists. This aggregate half-hour of added delay turns out to be fairly small when spread over the thousands of following motorists in the following hours (just 1 second each in this example), but these increases mount up as additional drivers enter the traffic stream. If 400 motorists were added to this traffic stream, the first few additional motorists would