The Cops Say <1% of Drivers Were on Their Phones. The Phones Say 8%.

Here's a fun fact that will ruin your morning commute. Every statistic you have ever read about distracted driving was written by people with a powerful incentive to lie, recorded by officers who cannot see the lie happening, and coded from forms that in 26 states do not even have a checkbox for it. Now, for the first time, we have the phone's side of the story: an IIHS study published August 27 matched nearly 17,000 crashes against telematics data and found drivers were using their phones in the 30 seconds before impact in 7 percent of single-vehicle crashes and 8 percent of two-vehicle crashes, while police reports for the exact same crashes recorded phone use in less than 1 percent.^[1] That is not a rounding error. That is an official measurement apparatus that misses at least seven of every eight phones.

Read that number again and let it land: three. When both sources flagged phone use at all, they described the same activity three times in seventeen thousand crashes, an agreement rate of 0.018 percent, which is what a statistician would call functionally zero and what the rest of us would call the paperwork being fiction.^[1] The study's method was simple and slightly invasive: IIHS researchers took anonymized data from Cambridge Mobile Telematics, whose platform powers the safe-driving apps that log crashes, detect phone calls, and track every tap, swipe, and unlocked screen, then linked those events to police reports from four states between 2021 and 2024, matching crashes in the same 15-minute window within a one-mile radius.^[3] Your phone, it turns out, is a better witness than the officer with the clipboard, and it is also the suspect.

The breakdown is where the cosmic joke gets good. Telematics caught 22 handheld calls in the 30 seconds before crashes; police caught 7, a 3x gap, modest, almost respectable. Telematics caught 90 hands-free calls; police caught one, a single solitary one, which is a 90x gap.^[2] And the main event, the behavior that actually fills crash scenes: drivers tapping, swiping, or otherwise physically manipulating their phones before 639 crashes, of which police noticed 20, a ratio of thirty-two to one that names the real epidemic, which is the thumb. Two seconds of eyes off the road doubles crash risk, IIHS notes, and the average thumb-scroll is a festival of two-second glances.^[1] We banned the 22 and let the 639 drive past the checkpoint waving.

Now for the part that should make you put the phone down, assuming you are reading this in traffic, which the data suggests is a coin flip. This sample was not the general public, which is what makes the next part worse. These were drivers who voluntarily enrolled in safe-driving apps, people who consented to be monitored by their insurer for a discount, which is to say people self-selected for caution, and even they were mid-scroll when the airbags deployed.^[3] If the careful ones are at 7 to 8 percent, the rest of us are worse, and we will never know how much worse, because the only instrument that can see it requires the driver's permission, and permission is the one thing the guilty withhold. Then there is the maps paradox: passive use, screen unlocked but untouched, was about as common as active manipulation, and IIHS concludes navigation apps probably account for much of it.^[1] The most distracting app in your car might be the one telling you where to go. In a separate IIHS survey, nearly 25 percent of drivers admitted to watching a video on their phone while driving at least once in the previous 30 days.^[3] A quarter of the road is a movie theater now, so enjoy the commute.

What to do: understand that the risk is not the call, it is the tap. Set your navigation before the car moves, then put the phone face-down somewhere unreachable, because "I was just checking the map" is the most common last sentence the sensors record. If your state is one of the 21 that now bans holding a phone at all, the ban helps; IIHS research says the broader holding bans outperform narrow texting bans, precisely because an officer cannot tell a text from a playlist from the roadside.^[1] If you are enrolled in a telematics discount program, know the price: CMT says it does not sell driving data and the insurer is generally the data controller, but its own privacy policy allows disclosure when required by law, which means the sensor that earns you 10 percent off is one subpoena away from testifying against you.^[4] And if you make laws, aim at the hand, not the act, because the officer will never catch the swipe, so ban the rectangle in the palm.

What this does not prove: the study covers four states, 2021 through 2024, and a self-selected sample, so it is not a national census. The 15-minute, one-mile match window is fuzzy by design, and telematics proves the phone was active, not that the driver looked at it, and not that it caused the crash.^[3] CarPlay and Android Auto are invisible to this method, since the phone screen stays locked, and the study cannot see which apps were open, so the navigation theory is inference, not measurement. Passive use includes screens unlocked for reasons that may not be distraction at all. And Cambridge Mobile Telematics is a commercial vendor with a thriving business in the surveillance economy; it has opinions about how much monitoring the world needs.^[4] None of that erases the 639 taps the police missed.

The strongest case against this story goes like this: maybe 7 to 8 percent is good news. Even the all-seeing phone says 92 percent of crashes had no phone activity in the 30 seconds before impact, which means the distracted-driving epidemic might be smaller than the panic suggests, and the official numbers are not so much wrong as measuring something narrower. Police reports are enforcement documents, not epidemiological instruments; they exist to record illegal behavior, and most of the phone use IIHS counted is perfectly legal, so the undercount is by design, a feature of the form, not a failure of the officer. Grant all of that, every word, and the problem remains: a measurement system missing seven of eight cases cannot tell you whether the epidemic is growing or shrinking, which means every trend line, every campaign, every "distracted driving is down 4 percent" press release is drawn from an instrument with its lens cap on. Both things are true at once, and that is usually when you should pay the closest attention.

Source: Crash/phone-use figures are IIHS's August 2026 telematics study via its news release and secondary coverage; this article did not access the underlying CMT dataset. Behavior-level counts via Jalopnik's reporting. The 7x illustration applied to NHTSA fatality counts is explicitly a hypothetical, not a finding. See methodology for caveats.