Traffic

by Tom Vanderbilt

  So Disney tried a form of congestion pricing. It issued ticket books in which the tickets’ values reflected the capacity of the rides. Popular rides like Space Mountain required E tickets, which were more expensive than A tickets, good for tamer attractions like the Horseless Carriage on Main Street. The idea was not only to prevent people from simply lining up for the top attractions but to spread people out across the park, avoiding traffic jams at places like Space Mountain. “One way of increasing capacity is rerouting demand,” Laval said. This was successful to a point, but the signals that prices send can work in different ways. At Disney World, Laval explained, where 80 percent of park entrants were first-time visitors (Disneyland has more repeat visitors), many of whom had no particular itinerary of which rides to go on first, the E tickets were like a big red flashing sign saying, “Ride me first.” Everyone wanted to get their money’s worth, so they immediately gravitated to the most expensive rides. The rides were not only expensive because they were popular, they were popular because they were expensive.

  Phenomena like this shows up in traffic as well: The HOT lanes in Southern California charge more as more people enter them (in order to help keep them from becoming congested); yet sometimes people enter a tolled lane precisely because it is expensive—they think the toll must be so high because the untolled lanes are really jammed. (This sort of behavior subverts the normal economics principle of “price elasticity,” in which the number of users should drop as the toll goes up.)

  Disney finally hit upon the ultimate solution in 1999, when it introduced the FastPass, the system that gives the customer a ticket telling them when to show up at the ride. What FastPass essentially does is exploit the idea that networks function both in space and in time. Rather than waiting in line, the user waits in a “virtual queue,” in time rather than space, and can in the meantime move on to other, less crowded rides (or buy stuff). People can take a chance on the stand-by line, or they can have an assured short wait if they can simply hold off until their assigned time. Obviously, FastPass could not literally work on the highway. Drivers do not want to pull up to a tollbooth and be told, “Come back at two-thirty p.m.” But in principle, congestion pricing works the same way, by redirecting demand on the network in time.

  Traffic can be made to flow better by redirecting demand in space, of course, if traffic engineers know what the demand and available supply are on a network at any given time—and if they can find a way to get that information to drivers. In the past, this has been a necessarily crude process, hindered by delays in getting and sending the information, and the ability to see the network at once with all its interacting flows. Surely you have had the experience of listening in vain to a rapidly spoken traffic report, hoping against hope to get the details on the jam you’re sitting in (and by some law, you never can). And as we saw in Los Angeles, traffic information often comes too late for us to do anything about it, or is not even accurate.

  Rather than surgical strikes at congestion, one can always try carpet bombing. Sam Schwartz (a.k.a. “Gridlock Sam”), New York City’s former traffic commissioner, claims that by declaring “gridlock alert” days, he could “knock fifty thousand or sixty thousand cars out of traffic” by plastering the airwaves with dire warnings. “The Heisenberg principle exists in traffic. If you look at it and announce and tell people about it, it has an effect.” When he wanted to reduce traffic on one parkway so construction crews could work on an overhead rail link, he rolled out more horror stories. “I was able to scare away forty percent of the vehicles from that corridor,” he says. “We measured it. I was amazed at how effective we were. Sometimes when you hear on the radio, they talk about how terrible the traffic is—it’s really me, like the Wizard of Oz behind the curtain.”

  But human psychology has a way of rearing its complicated head. One problem is that you can never quite know how people will react. In one study, researchers assembled a panel of drivers who regularly commuted on U.S. 101 in California’s Silicon Valley. Following a multiple-vehicle crash that took over a half hour to clear, causing extensive delays, the researchers interviewed the commuters. They found that only half the drivers had heard about the incident, and that even the majority of those drivers simply headed to work at the normal time, the normal way. Many people simply seemed unconvinced that they could save any time by changing their plans.

  We have all had these moments. Do I take the local streets when I see there is a crash ahead? Is it better to leave early on Sunday morning to go back to the city, or will everyone else have that same idea? Do I get in the right lane because it seems empty, or is there a reason no one else is in it? It boils down to how we make decisions when we do not have all the facts. We rely instead on heuristics, those little strategies and mental shortcuts we all have in our head: Well, this road is usually busy for only a few minutes, so I’ll stay on it. Or: I bet that since the radio called for snow there will not be many people at the mall. We use our experience; we make predictions.

  This recalls the famous “El Farol problem,” sketched by the economist W. Brian Arthur, after a bar in Albuquerque, New Mexico. The hypothetical scenario imagines that one hundred people would like to go to the bar to listen to live music, but it seems too crowded if more than sixty show up. How does any one person decide whether or not to go? If they go one night and it’s too crowded, do they return the next night, on the thought that people will have been discouraged—or will others have precisely the same thought? Arthur found, in a simulation, that the mean attendance did indeed hover around sixty, but that the attendance numbers for each night continued to oscillate up and down, for the full one hundred weeks of the trial. Which means that one’s chances of going on the right night are essentially random, as people continue to try to adapt their behavior.

  This kind of equilibrium problem happens frequently in traffic, even when people have some information. In 2006, for example, the Dan Ryan Expressway in Chicago was undergoing massive repairs. The first day that eight express lanes were closed, traffic moved surprisingly well. The recommended detours moved more slowly than the highway. This was reported on the news. You can guess what happened on Tuesday: More people flocked to the highway. We can surmise that the expressway’s traffic went down on Wednesday, though it may equally likely have gone up.

  What happens when we no longer have to guess? We are now just at the beginning stages of a revolution in traffic, as navigation devices, increasingly often equipped with real-time traffic information, enter the market. The navigation part alone has important consequences for traffic. Studies have shown that drivers on unfamiliar roads are roughly 25 percent less efficient than they should be—that is, they are lost—and that their total mileage could be cut by 2 percent if they were always shown the best routes. Logistics software now helps cut delivery times and fuel emissions for UPS and other truck fleets simply by finding ways to avoid, when possible, time-consuming left turns in two-way traffic. But the biggest change will occur when each driver can always know which roads are crowded and what alternate routes would be best—not through guesses but through accurate real-time data.

  In theory, this will help reduce inefficiency in the system. Drivers are told there has been a crash ahead, and their in-car device gives them another route that is estimated to save ten minutes. But nothing in traffic is ever so simple.

  The first problem is that real-time data is not yet what its name promises. At Seattle’s Inrix, for example, one of the key providers of traffic information, traffic-pattern data is gathered from a variety of sources, current and historical—from loops to probes on commercial vehicles to the schedule of conventions in Las Vegas, some five billion “data points”—and weighted according to its perceived accuracy and age. “So a thirteen-minute-old traffic speed estimate from a Caltrans sensor in the Los Angeles market would get a sub-five-percent weighting in our estimate of current conditions,” explains Oliver Downs, Inrix’s principal research scientist. Inrix estimates the current con
ditions every minute, but as Downs notes, “it’s a 3.7-minute-old estimate of the current conditions.” Customers, meanwhile, get a new feed every five minutes. “When we say ‘real-time,’” Downs says, this means “less than five minutes.” That might not seem like much, but, as Downs admits, “it’s long relative to how quickly things can change on the roadway.”

  The other problem comes in how people will use that information, or what you should tell them to do based on the information. Michael Schreckenberg, the German physicist known as the “jam professor,” has worked with officials in North Rhine–Westphalia in Germany to provide real-time information, as well as “predictive” traffic forecasts. Like Inrix, if less extensively, they have assembled some 360,000 “fundamental diagrams,” or precise statistical models of the flow behavior of highway sections. They have a good idea of what happens on not only a “normal” day but on all the strange variations: weeks when a holiday falls on Wednesday, the first day there is ice on the road (most people, he notes, will not have yet put on winter tires), the first day of daylight savings time, when a normally light morning trip may occur in the dark.

  This kind of information, along with the data gathered from various loops and sensors, can be used to make precise forecasts about what traffic will be like not only on “normal” days but when crashes or incidents occur. There is, however, a problem: Does the forecast itself change the way people will behave, thus changing the forecast? As the economist Tim Harford notes about Wall Street forecasting, if everyone knew today that a stock was going to rise tomorrow, everyone would buy the stock today—thus making it so expensive it could no longer rise tomorrow.

  Shreckenberg calls this the “self-destroying prognosis.” In his office at the University of Duisburg-Essen, he points to a highway map with its roads variously lit up in free-flowing green or clogged red. “The prognosis says that this road becomes worse in one hour,” he says. “Many people look at that and say, ‘Oh, don’t use the A3.’ Then they go somewhere else. The jam will not occur since everyone turned to another way. This is a problem.” These sorts of oscillations could happen with even short lags in information, in what Shreckenberg calls the “ping-pong effect.” Imagine there are two routes. Drivers are told that one is five minutes faster. Everyone shifts to that route. By the time the information is updated, the route that everyone got on is now five minutes slower. The other road now becomes faster, but it quickly succumbs to the same problem.

  This raises a question: Has the information provided actually helped drivers or the system as a whole—or has it triggered the “selfish routing” mentioned before? Moshe Ben-Akiva, the director of MIT’s Intelligent Transportation Systems program, has studied such travel behavior issues for decades. He calls traffic predictions a “chicken-and-egg problem.” “The correct prediction must take into account how people are going to respond to the prediction,” he says. “You cannot predict what will happen tomorrow without taking into account how people are going to respond to the prediction once the prediction is broadcast.”

  And so researchers create models that anticipate how people will behave, based on how they have behaved in the past. Shreckenberg, in Germany, wonders if this means, in essence, not giving drivers the whole picture. “You have to structure the information. What you want is for the people to do certain things. Telling them the whole truth is not the best way.” This is something on the minds of the big commercial providers of traffic information. As Howard Hayes, vice president of NAVTEQ, said at the firm’s headquarters in Chicago, “What happens if once this really good predictive traffic information becomes available, everyone starts getting shunted over to a different direction, which itself becomes jammed? Ideally you need something sophisticated, so that a certain number of people get shunted to one route and others to another.”

  Since the information is still so limited, and since so few people actually have access to it, we do not really know how it will all play out once everyone is able to know the traffic conditions on every road in a network. Most simulations have shown that more drivers having more real-time information—the closer to actual real-time, the better—can reduce travel times and congestion. Even drivers without real-time information can benefit, it is argued, because better-informed drivers will exit crowded roads, thus making those roads less crowded for uninformed drivers stuck in traffic. But as you might expect, studies suggest that the benefit for any one driver with access to real-time information drops as more people have it. This is, in essence, the death of the shortcut. The more people know the best routes at all times, the less chance of there being some gloriously underutilized road. This is good for all drivers (i.e., the “system”) but less good, say, for the savvy taxi driver.

  Real-time traffic and routing is most valuable, it has been suggested, during nonrecurring congestion. When a road that is normally not crowded is backed up because of a crash, it’s useful to know of better options. During recurring congestion, however, those peak-hour jams that result from too many people going to the same place at once, the advantage shrivels once the tipping point of congestion has been passed. (It is most effective right on the brink, when alternative routes are on the verge of drying up.) In a traffic system that is always congested, any good alternative routes will have already been discovered by other drivers.

  Another shortcoming of real-time routing is due to a curious fact about urban road networks. As a group of researchers observed after studying traffic patterns and road networks in the twenty largest cities in Germany, roads follow what’s called a “power law”—in other words, a small minority of roads carry a huge majority of the traffic. In Dresden, for example, while 50 percent of the total road length carried hardly any traffic at all (0.2 percent), 80 percent of the total traffic ran on less than 10 percent of the roads. The reason is rather obvious: Most drivers tend to drive on the largest roads, because they are the fastest. Even though they may have slowed due to congestion, they are still fastest. Traffic engineers, having built the roads, are generally aware of this fact, and would rather have you stay on the road that was designed for heavy use, instead of engaging in widespread “rat runs” that play havoc with local roads.

  Both the promise and the limits of real-time traffic and routing information were demonstrated to me one day as I drove on Interstate 95 in Connecticut, using real-time traffic information provided by TeleNav via a Motorola mobile phone. The phone had been cheerily giving directions, even offering an evolving estimated time of arrival. Suddenly, an alert sounded: Congestion ahead. I queried the system for the best alternate route. It quickly drew one up, then delivered the bad news: It would take longer than the route I was on. The road I was on, congested or not, was still the best.

  Real-time traffic and routing information and congestion pricing are two sides of the same coin. One tells drivers how to avoid traffic congestion; the other impels drivers to avoid traffic congestion. When the roads are congested, real-time information does little good, except to tell drivers, like the people in line for Disney World’s Space Mountain, how long they can expect to wait. This alone may be enough of a social good. But real-time congestion information, provided by the very cars generating that congestion, promises something else. It can be used to calculate the exact demand for any stretch of road at any time. With congestion pricing, the traffic on the roads will finally be made to act like the traffic in things, with market prices reflecting and shaping the supply and demand.

  When Dangerous Roads Are Safer

  The Highway Conundrum:

  How Drivers Adapt to the Road They See

  An overturned cart is a warning to oncoming drivers.

  —Chinese proverb

  Just before dawn on Sunday, September 3, 1967, there was an unusually festive air in the streets of Stockholm. Cars honked, passersby cheered, people gave flowers to police officers, pretty girls smiled from the curb. The streets were clogged with cars, many of which had been waiting for hours to participate in a historic traffi
c jam. People stole bicycles simply to be a part of traffic. At the moment the bells chimed for six o’clock, Swedes began driving on the right.

  It had taken years of debate, and much preparation, to get to this point. Motions to switch from left-side driving had been raised in Parliament several times in previous decades, only to be shot down. The issue was put before Swedes in a 1955 referendum, but the measure was overwhelmingly defeated. Undeterred, backers of right-side driving finally got a measure approved by the government in 1963.

  Proponents said that driving on the right, as was the practice in the rest of Scandinavia and the bulk of Europe, would lower the number of accidents in which foreigners were increasingly becoming involved. Most cars in use already had steering wheels on the left side. Those opposed, which was most of Sweden, grumbled about the huge costs of the changeover, and said that accident rates were bound to rise.

  As “H-Day” (after höger, the Swedish word for “right”) approached, the predictions of ensuing chaos and destruction grew dire. “What is going to happen here in September has cast many grotesque shadows all over Sweden,” the New York Times observed darkly. This despite four years of preparation and an especially energetic blitz of public-service announcements in the final year before the changeover. There was even a pop song, titled “Håll dej till Höger, Svensson!” or “Let’s All Drive on the Right, Svensson!” (after a stereotypically common Swedish surname).

  And what happened when Swedes started driving on the other side of the road, many for the first time in their lives? The roads got safer. On the Monday after the change, the traffic commissioner reported a below-average number of accidents. True, this may have been anticipated, despite the gloomy predictions. For one, many Swedes, scared witless of the spectacle, undoubtedly chose not to drive, or drove less. For another, a special speed limit, which had already been in place for some months before the changeover, was enforced: 40 kilometers per hour in towns, 60 on open roads, 90 on highways. Lastly, the whole operation was run with Scandinavian efficiency and respect for the law. This was the country that gave the world Volvo, by God—how could it not be safe?