The Camera Promised Fewer Crashes
The logic felt airtight: put a camera on the intersection, photograph every driver who runs the red, mail the fine, and watch violations collapse. Fewer red-light runners means fewer T-bone collisions, fewer ambulances, fewer funerals, which is why more than 400 American communities bought the argument, including 36 of the 50 largest cities. The flaw: it assumed the camera only changes the would-be red-light runner's behavior, when it also changes the behavior of the driver who decides, in the last half-second of yellow, that the ticket is not worth it. That driver stands on the brakes, while the driver behind, watching the light instead of the bumper ahead, does not.
A Meta-Analysis Asks the Rude Question
In 2009, Alena Erke of Norway's Institute of Transport Economics pooled published evaluations of red-light cameras' effect on crashes into a single meta-analysis, and the result was the opposite of the sales pitch. Among the studies with the strongest controls, cameras were associated with about 15 percent more crashes overall, 40 percent more rear-end collisions, and 10 percent fewer right-angle collisions, the T-bones the cameras exist to prevent. None of the three effects reached statistical significance, and her verdict was blunt: on the whole, the cameras "do not seem to be a successful safety measure."
Erke's most damaging finding was the meta-regression: the better a study's methods, the less impressive the cameras looked, with studies controlling for regression to the mean reporting markedly less favorable results. Cities install cameras at their worst intersections after a bad year, so consider an intersection averaging 12 injury crashes a year that spikes to 20, gets a camera, then records 13. The program claims a 35 percent reduction, but the spike was always likely to fall back toward 12; studies that adjusted their baselines stripped away the miracle.
Eleven Years Later, the Trade-Off Refuses to Die
Three researchers affiliated with the Insurance Institute for Highway Safety published a comment warning against pooling such heterogeneous studies, and Høye, Erke's successor at the Norwegian institute, answered by redoing the analysis with a larger sample. The 2013 update landed in nearly the same place: all crashes up 6 percent, injury crashes down 13 percent, neither significant; right-angle collisions down 13 percent, rear-end up 39 percent; for injury crashes, right-angle down 33 percent, rear-end up 19 percent. Warning signs at the entrances to a whole enforcement area also worked better than signs at each intersection.
The most comprehensive synthesis, a 2020 Campbell Collaboration review, pooled 41 analyses from 38 studies, finding total injury crashes fell 20 percent, right-angle crashes 24 percent, right-angle injury crashes 29 percent, and rear-end crashes rose 19 percent, with the injury reduction and the rear-end increase both statistically significant. Three syntheses, eleven years apart, describe the same machine. The table below lines them up, a comparison none of the papers drew.
| Outcome | Erke 2009 | Høye 2013 | Cohn et al. 2020 |
|---|---|---|---|
| All crashes | +15% (n.s.) | +6% (n.s.) | n/a |
| Injury crashes | n/a | -13% (n.s.) | -20% |
| Right-angle crashes | -10% (n.s.) | -13% | -24% |
| Right-angle injury crashes | n/a | -33% | -29% |
| Rear-end crashes | +40% (n.s.) | +39% | +19% |
| Rear-end injury crashes | n/a | +19% | n/a |
n.s. = not statistically significant; n/a = outcome not reported in that synthesis.
The Break-Even Rule Nobody Computed
No synthesis asked the question a city council needs: at which intersections does the trade pay off? Take the Campbell estimates, a 29 percent reduction in right-angle injury crashes against a 19 percent increase in rear-end crashes: for an intersection with R right-angle injury crashes and E rear-end crashes per year, cameras reduce total injuries when 0.29R exceeds 0.19E, which rearranges to R/E > 0.66. An intersection needs roughly two right-angle injury crashes for every three rear-end crashes to break even; below that ratio, the camera is expected to add injuries. Cameras suit some intersections and not others; uniform rollout without screening by crash profile is guessing.
The Strongest Case for the Cameras
The camera advocates are not fools, and their case deserves full strength. First, Erke's alarming 15 percent was not statistically significant, so policy should not be convicted on a confidence interval that comfortably includes zero; second, pooling small before-after studies from different countries, eras, and designs into one number is aggressive, and can manufacture precision that is not there.
Third, the severity argument is their strongest card. Injuries occur in 39 percent of red-light-running crashes: the crashes cameras prevent are the ones that kill, a T-bone at speed, while the crashes they cause are disproportionately low-speed bumper taps. The Federal Highway Administration's crash-modification database credits cameras with a 17 percent reduction in fatal crashes, which means that if the machine converts one fatal T-bone into three dented bumpers, the crash count rises and the body count falls, and only one of those numbers matters. Fourth, an IIHS study reported lower fatal red-light-running crash rates in cities with camera programs; in Houston, installation was associated with 18 percent more non-angle crashes while removal brought 26 percent more angle crashes.
What We Didn't Prove
Several caveats belong on the record: no randomized trial of red-light cameras exists, and every study in all three syntheses is quasi-experimental, with debatable adjustments. Most pooled effects in the first two syntheses are not statistically significant, and Høye flagged publication bias: null results in file drawers would overstate benefits. Programs are heterogeneous: warning signs, yellow timing, fine levels, and enforcement thresholds all vary, so "red-light cameras" is a category, not a treatment. And no crash study measures the revenue motive: most American programs are run by for-profit contractors, a governance problem the literature does not address.
The Bottom Line
Three independent syntheses across eleven years agree on the shape of the trade: red-light cameras reduce the severe right-angle crashes they were built to stop and increase the rear-end crashes nobody campaigned on. Whether the trade is worth making depends on the intersection, the signage, and whether you count crashes or bodies, which is why cameras everywhere is not supported by the evidence; the targeted version is cameras where right-angle injury crashes dominate, area-wide warning signs, and evaluations that control for regression to the mean. The camera is neither a scam nor a savior. It is a tool with a known side effect, and such tools demand prescriptions, not slogans.
What You Can Do
First, drive like the trade-off is real: the rear-end spike comes from hard braking at the yellow, so keep your distance and watch the bumper ahead, not just the light. The driver in front of you is doing ticket math; give them room to be wrong.
Second, if your city operates cameras, ask for the evaluation, not the press release, and ask whether the comparison controlled for regression to the mean and spillover effects, since Erke's meta-regression showed that studies skipping those controls reported rosier results, making an evaluation without them advertising.
Third, if you sit anywhere near the decision, steal the break-even rule: demand intersection-level crash profiles, favor area-wide gateway warning signs, and require a pre-registered evaluation up front, because a program that cannot say which intersections clear the two-to-three ratio is not a safety program but a fundraising program with good lighting.