In this issue
Issue #24 August 29, 2013 Aug 29, 2013 Aug 29
Issue #9 January 31, 2013 Jan 31, 2013 Jan 31
Issue #1 October 11, 2012 Oct 11, 2012 Oct 11
From Issue #16 May 9, 2013

Red Light, Green Light

Using cameras to capture red-light infractions may increase accidents.

By Mark Harris Twitter icon 

Barbara Orban makes an unlikely campaigner against red-light cameras. A professor of public health at the University of South Florida in Tampa, she spent much of her career researching improvements in emergency medicine. Then, shortly before her eldest son’s high school graduation in 2003, his car was struck by a driver running a stop light. The impact caused his vehicle to roll over, shattering its windows and buckling the roof. He sustained a major concussion that disrupted his education for the next four years.

Orban’s son was lucky. Around 800 people lose their lives each year in the United States as a result of drivers running red lights; those drivers also injure over 150,000 motorists, cyclists, and pedestrians. Ninety-seven percent of Americans surveyed agree that red-light running represents a serious threat to road safety. Red-light cameras (RLCs) would seem to be the perfect solution. Installed at little cost, they punish red-light runners with traffic tickets, decrease the violation rate, and, usually, provide welcome income for cash-strapped local government. Across the country, about 530 communities in 24 states (plus DC) currently operate red-light cameras.

There’s only one problem, says Orban: “Research that’s correctly designed and constructed has concluded that red-light cameras increase crashes and injuries. They are actually adverse interventions.” Orban believes that the very technology that some claim could have prevented her son’s accident is not only ineffective at saving lives but might even be making our roads more dangerous.

Light work

Back in 2004, while her son was still recovering, Orban was asked to assess the effectiveness of red-light camera systems. Florida’s lawmakers wanted to boost revenue for the state’s trauma centers and were considering traffic tickets as an option. As Orban set about reading the scientific literature, she was surprised to see little agreement on the cameras’ usefulness.

“If some research is finding that cameras are highly effective and others are finding an increase in crashes and injuries, somebody’s methods have to be faulty,” says Orban. “We found that many studies were not measuring the effects of red-light cameras properly, as a public-health researcher testing a new drug would.”

Measuring the effectiveness of red-light cameras is, if anything, trickier than assessing a new pharmaceutical. Red-light cameras are generally (and understandably) deployed at intersections that have been prone to accidents. Statistically, such locations could be expected to experience fewer collisions in the future, regardless of having a camera or not, due to a phenomenon known as regression to the mean. Another factor to consider is the “spillover effect” of a publicity campaign and warning signs. These could lead to all drivers in that area driving more thoughtfully.

Choosing the correct control intersections becomes a critical part of measuring whether the traffic cameras are effective at reducing crashes and injuries. From her public-health perspective, Orban would prefer that researchers choose “placebo” intersections that have characteristics close to those getting cameras: the same sorts of signals, neighborhood, and existing accident rate.

Not everyone agrees. The Insurance Institute of Highway Safety (IIHS), a leading proponent of RLCs, worries so much about spillover effects that it chooses comparison intersections that are quite different from those getting cameras. “We believe that there need to be control intersections in an appropriate other city or, [if] in the same city, we also look at non-signalized intersections,” says Anne McCartt, senior vice president for research at IIHS.

In a paper published in 2011, the IIHS compared 14 large US cities that had installed RLCs with 48 that had not, over two distinct time periods (1992–6 and 2004-8). It found that the cities with cameras experienced 24% fewer fatal red-light-running crashes in the latter period, whereas the camera-less communities saw only a 14% drop. “We would argue that the evidence from strong studies shows that RLCs reduce fatal injuries,” says McCartt.

Orban says not so. “The IIHS chose groups that are dissimilar, and dissimilar in a way that finds cameras to be favorable,” she explains. “Cities that used RLCs had a substantially higher rate of red-light-running fatalities and a higher fatality rate at signalized intersections, in both time periods, than cities that did not use them. Cities starting from a higher base can show greater relative improvement than those already performing well, even in the absence of an intervention. Two of the no-camera cities even had a red-light-running fatality rate of zero, making a reduction impossible.”

Stop already

Such problems with methodology, as well as disagreements over statistical and economic analyses, have plagued RLC research. British road safety researchers at the Cochrane Collaboration conducted a meta-analysis that found no randomized controlled trials at all, and which rejected two-thirds of published studies for failing to meet quality standards. In 2005, the Federal Highway Administration concluded that almost all studies until then had failed to properly account for regression to the mean or spillover effects, and “many had both flaws, in addition to others.”

If that weren’t bad enough, even gathering the raw data is problematic. Ideally, dozens of variables would be considered over a period of many years, before and after installation. These include traffic volume, speed limits, junction configuration, warning signs, weather, and seasonal variability. Just getting the most significant events – the crashes themselves – properly recorded can be a headache, according to Allen Shutt, a vice president at Xerox, a technology services company that has about 1,100 photo enforcement cameras in North America. “Crash data is only as good as the person who wrote it down. Consistency among the individual police officers reporting crashes is very hard to find in places,” he says. “It’s very difficult to get crash data before we implement a red-light program.”

It would be fair to say, then, that many RLC systems are installed blind. Communities are sold cameras with neither a proven justification for needing them nor specific targets to reduce crashes. For all its flurry of pro-camera press releases, the IIHS has carried out only a handful of small studies, covering a dozen intersections in California and Virginia. Barbara Orban has done no original research at all, relying instead on critiquing work done elsewhere. There was simply no good, large-scale research available.

That was about to change. In 2007, the Virginia Transportation Research Council issued one of the biggest and certainly the most credible study of RLCs to date. It documented the safety impacts of cameras at 28 intersections in six jurisdictions of Virginia from 1998 to 2004, and the data analysis took five researchers the best part of two years to carry out. The study accounted for regression to the mean and spillover effects, and used sophisticated empirical Bayes statistical analysis. One of its authors, John Miller, is still a principal research scientist at Virginia’s DOT. “Going into it, we didn’t know what to expect,” he remembers. “There was variability in the literature: some studies had found positive impacts, some negatives. Gradually, as we went through the analysis we noticed differences emerging.”

At first glance, camera fans should have been delighted. The study found that after cameras were installed, red-light-running crashes decreased in all six jurisdictions, in a statistically significant way. In fact, crashes due to late intersection entry declined by an impressive 42% for the entire area.

But that was not the whole story. Once crashes had been controlled for time and traffic volume, red-light accidents actually went up at four intersections. More worrying still, rear-end crashes increased by an average of 27% across the board. And because rear-end crashes are much more common than other accidents, the overall crash rate was 12% higher after RLCs were switched on.

It appears that red-light cameras come with unintended consequences. Many studies have shown that red-light violations decline after cameras are installed, as drivers tire of receiving tickets for hundreds of dollars in the mail. Some motorists start driving more sensibly. But camera makers noticed that other drivers cause what Allen Shutt calls “false triggers.” He says, “People will approach the intersection speeding, then slam on their brakes so they end up halfway across the stop line,” triggering the camera but avoiding an infraction. For anyone following too closely, the chances of a rear-end collision shoot up.

The increase in crashes at some intersections is more difficult to explain. Barbara Orban believes that many of the worst accidents do not happen from someone jumping a light by a fraction of a second. “Even if you’ve got a green light, you’re not going to drive into a red-light runner just because it’s your turn to go,” she says.

She thinks most killer crashes happen when a motorist is DUI, suffering a medical issue, or just the victim of poor engineering. “The accident that injured my son happened at a defective junction,” she says. “You have to bend forward to look up and see the traffic signal. When a car in front turned right, the other driver assumed the light had turned green, accelerated into the intersection, and hit my son.” After a long campaign by Orban, the signals at that intersection were eventually upgraded in 2007.

The Virginia report gave no clear answers. “We couldn’t find a set of geometric or operational characteristics that you could say, ‘this means the cameras are definitely going to have a positive or negative impact at a particular intersection,’” says Miller. In the end, the report recommended only that red-light cameras should not be implemented without a study of intersection-specific crash patterns and characteristics, and should be studied carefully after installation.

Throw it in reverse

Predictably, perhaps, the IIHS took issue with aspects of the Virginia report. While praising its scope and use of empirical Bayes methodology, researchers questioned its models and conclusions. “It would be fair to say we haven’t done a study this size but I don’t think the bigger the study, the better the study,” says Anne McCartt. “We found their study had significant limitations.” The Institute published a point-by-point critique (it was the only organization to do so), which landed on John Miller’s desk.

“They said we should use a different type of model and alter how the model is calibrated,” Miller says. “We did that. The results came out with somewhat different numbers but still materially the same. In fact, when we excluded some data as they suggested, the cameras looked worse than in the original report. The Institute chose not to publish our response.”

Even if crashes increased, RLCs might still be a worthwhile safety intervention. That’s because rear-end crashes tend to be less severe than angled or T-bone impacts. The effect of this can be measured using an economic analysis. Fatal crashes have an economic cost at least an order of magnitude higher than injury crashes, which themselves have a much higher price tag than mere fender benders.

The Virginia report found that cameras were associated with a net negative impact overall, although there were positive benefits for some jurisdictions, and especially for a few accident-prone intersections. When the Federal Highway Administration did a similar analysis for 132 sites, it found RLCs resulted in a ‘modest to moderate’ economic benefit, quantified at $39,000 per intersection per year.

There can also be a direct financial benefit for communities that install RLCs. While ticket prices vary nationwide, from a low of $50 in New York to a hefty $490 in California, most programs do make money. Take Seattle, for instance: the city’s $124 tickets are slightly less than the national average. Two years into its RLC program, its 30 cameras had issued over 27,000 infraction notices. After allowing for installation, running, and judicial costs, the cameras brought over $1 million into the city’s coffers.

Insurers also like RLCs. More cameras lead to more tickets, and more tickets mean higher premiums, even as accident rates decline. While many states do not record photo-enforced infractions, and some expressly prohibit insurers from using them, anti-camera activists like Orban link the IIHS’s pro-camera stance to it being wholly funded by insurance companies. The IIHS’s McCartt dismisses such accusations as not credible. “We describe ourselves as an independent research communications organization,” she says.

Seeing green in red-light tickets

Over the years, red-light cameras have evolved from using labor-intensive photographic film and expensive induction, laser, and radar detectors into small, easily installed, Ethernet-connected cameras that record infractions digitally. “Most importantly, cameras have become less expensive,” says Allen Shutt. “You only need one or two tickets a day at an intersection to have a self-funding program.” Most cameras catch many more.

It is this revenue-raising prowess of cameras that makes some drivers see red. A single camera in Texas, for example, is estimated to have raised $2.5 million over four years. For law-abiding motorists, it can be tempting to view cameras as simply a tax on dangerous driving. But as tax-raising goes, red-light cameras are remarkably inefficient. Consider Seattle again. While the city earned $1 million from its cameras, the next biggest beneficiary was camera maker American Traffic Solutions, which pocketed over $500,000. The program also cost Seattle’s police department over $200,000 in the time it took officers to review and issue infraction notices, it cost the Seattle Department of Transportation over $100,000 to oversee the program, and the courts had $80,000 in expenses to hold hearings for drivers contesting tickets.

If RLCs are not actually reducing crashes, all of these people are wasting their time. Other cities have earned an even smaller slice of the pie, with Dallas, Tallahassee, and San Diego sharing just 10–13% of fines. Some cities, San Diego among them, have now switched off their RLCs, and nine states ban them altogether.

Alternatives to RLCs do exist. Adjusting the yellow-light interval is favored by many road-safety campaigners, including Barbara Orban and the IIHS. Simply extending the yellow-light duration by one second can reduce red-light running violations by up to 50%. Accidents also fall, by as much as 15%, although congestion may increase.

But researchers at Morgan State University have estimated that 72 vehicles per hour are delayed at each RLC intersection anyway, as nervous drivers stop prematurely during yellow lights. The Federal Highway Administration also notes that roundabouts can reduce fatal and injury crashes by up to 87%, and need replacing only half as often as traffic signals. “We think building roundabouts would be a better solution at a lot of intersections,” says McCartt. “But it’s not feasible everywhere, and we couldn’t afford to do it anyway.”

In jurisdictions that permit them, red-light cameras look increasingly irresistible. For the cost of building a single roundabout (average price $350,000), a community can fund a one-year pilot RLC program and guarantee itself a revenue stream in perpetuity, all while claiming to crack down on dangerous drivers.

Cameras leveraging cloud-based software are also becoming increasingly capable. Xerox’s red-light cameras can already simultaneously capture the speed of approaching vehicles, check license plates for stolen cars or uninsured drivers, and determine occupancy rates to enforce HOV laws. A few communities are now trialing photo enforcement on school buses and even humble stop signs. Each infraction comes with a cash penalty, naturally.

Allen Shutt couldn’t be happier: “The way we’re headed is that we hope someday communities can afford a camera on almost every intersection.” Barbara Orban is less convinced, noting, “Motor vehicle safety was improving long before cameras came along. The question is, is there economic value to the community? There’s economic value to the government and the vendors providing the service, because you’re transferring wealth from the public to them. But does the public ultimately benefit at some higher level? That has not been measured.”

Flashing hazards

There is little appetite to measure the public benefit. Amy Aeron-Thomas, the author of the 2005 Cochrane Collaboration meta-analysis, says, “We were not able to get funding to update this study, as red-light cameras are taken as a given in Britain and there is no concern about their effectiveness.”

John Miller is not so sure. “It would be interesting and appropriate to look at data from a wide variety of geographies and a longer data set,” he says. “Getting 24 or 36 months of after-data is one thing, but if you had after-data for 10 years, that would be interesting. It would be important to control for travel volume, yellow-light timing, and other factors that might vary with driver behavior or from one region to another. It would be another big study.”

The prospects for such wide-ranging research are slim. The Department of Transportation took a $2 billion hit in the sequester, and communities with cameras have little incentive to carry out expensive, time-consuming studies that could kill their golden geese. This leaves jurisdictions to decide whether to install their own cameras while relying on aging, limited, and sometimes discredited research.

That red-light cameras can be and have been effective at reducing crashes, injuries, and deaths at certain intersections seems extremely probable. But it remains far from clear whether the scientific evidence justifies their universal use, especially where simpler, cheaper, or more-proven safety measures have not been tried.

Photo by Paul Sableman. Used under Creative Commons CC BY 2.0 license. This issue’s cover photo by Michael Hitoshi via Getty Images.

Mark Harris averages 7.18 minute miles running, 15.8 mph on long distance bike rides, has 6.2% body fat and has telomeres that are 8.08 kilobasepairs long and shrinking. He has written 104 stories for The Sunday Times in London, 31 features for The Economist and 1 for The Magazine. He lives in 98103. His website is shamefully neglected.

You can purchase our complete archives, almost 300 articles, as a DRM-free ebook in PDF, EPUB, and MOBI formats. We ceased publication of new work on December 18, 2014.
You can purchase our complete archives, almost 300 articles, as a DRM-free ebook in PDF, EPUB, and MOBI formats.
©2017 Aperiodical LLC. The Magazine's online ISSN: 2334-4970. We ceased publication on December 18, 2014. You can purchase our complete archives, almost 300 articles, as a DRM-free ebook in PDF, EPUB, and MOBI formats. Read our privacy policy. Learn more about us. Billing troubles? Email us. Talk with us on Facebook and Twitter. Consult our FAQ for more answers. iPhone, iPad, and iPod touch are trademarks of Apple Inc., registered in the U.S. and other countries. App Store is a service mark of Apple Inc.