Your Phone Says You Were Using It. The Police Report Says You Weren't. Your Phone Is Right.

Here's a fun fact that will ruin your morning commute: the official count of phone-involved crashes is a work of fiction, not a lie exactly, because a lie requires someone who knows the truth and writes down something else, and this is purer than that. Nobody knows. We have simply been counting the phones we can see, and it turns out we can barely see any of them.

The Insurance Institute for Highway Safety just ran the experiment safety researchers have wanted for years: take nearly 17,000 real crashes and ask both the police and the phone what happened.^[1] Researchers matched crash records from four states (2021-2024) against telemetry from Cambridge Mobile Telematics, whose platform powers those voluntary safe-driving apps that watch how you drive in exchange for an insurance discount. Recording ran the whole time, and a sensor has no incentive to lie.

Police reports recorded cellphone use in under 1 percent of the crashes. The telemetry showed phone activity within 30 seconds of impact in 7 percent of single-vehicle crashes and 8 percent of two-vehicle crashes, which is at least seven times the official number.^[2] Now look at the categories, because the categories are where the comedy curdles. Phone manipulation ran 590 to 17, a 35-fold undercount; passive screen use 534 to 15; hand-held calls 19 to 6; hands-free calls 87 to 1. Nobody overhears a hands-free call at a crash scene and nobody subpoenas a screen in the rain, so the gap is widest precisely for the phone use no officer can witness. And here is the part that bent my sense of humor sideways: of the nearly 17,000 crashes, both systems flagged phone use in the same crash often enough to compare, and they agreed on the specific type of activity in three cases out of seventeen thousand. Even when both witnesses are in the room, they disagree 99.98 percent of the time, which means the "confirmed" phone crashes in our national statistics are not just undercounted, they are miscategorized.

Scale it up and the official number stops looking like an estimate and starts looking like a rumor. America has roughly 6.7 million police-reported crashes a year. Apply 8 percent and you get something on the order of half a million crashes annually with a phone active in the half-minute before impact, versus the tens of thousands the checkbox on the crash form produces. Every texting-ban debate, every distracted-driving awareness campaign, every insurer's risk model has been arguing over a numerator that was never more than a sample of the samples.

Full honesty requires the counterargument, and it is a good one: a glowing screen is not a distracted driver. The telemetry cannot tell navigation from Netflix, a podcast tap from a doomscroll, and the study cannot say the phone caused any given crash or who was at fault. Every driver had volunteered for a driving-monitoring app, which means the sample skews toward people who are either unusually cautious or unusually young, and CarPlay or Android Auto use never unlocks the screen at all, so some phone distraction is invisible even to the phones. Research scientist Aimee Cox, who led the study, puts it more carefully than I do: any phone use that pulls a driver's eyes off the road can raise crash risk, regardless of whether it fits the old legal categories of texting or talking.^[1] IIHS president David Harkey calls the whole disconnect an open secret in safety research, and he is right. It was an open secret because everyone suspected it and nobody could prove it, and now a phone company proved it with the phones themselves.

Which brings us to the second witness in this story, the one nobody invited. Everyone in this study had volunteered; their data was anonymized, and Cambridge Mobile Telematics says it does not sell driving data. But in insurance programs, the insurer is generally the controller of that data, and the company's privacy policy allows disclosure when required by law or legal process.^[2] Your phone already testifies about your crashes more accurately than the police report does. It knows where you were, whether you were handling it, whether the screen was unlocked. The discount was never free; it was a witness fee.

So here is what you do with this. If you are ever in a crash and suspect the other driver was on their phone, do not accept the police report as proof they were not. Phone records, subpoenaed early, are the evidence the checkbox was never designed to collect. If you drive with a telematics app, understand that your insurer now possesses a pre-crash diary the responding officer will never see, and decide whether the discount is worth the testimony. If you make laws, notice that texting bans were written for 2010 phones while manipulation and passive screen time dominate the undercount; the 21 states that simply prohibit holding a phone at all are the only legal framework aimed at the thing the data actually measures.^[1] And if you just drive: the two-second rule is the actionable piece of all of this. Eyes off the road for two seconds doubles your crash risk. The phone does not have to be "in use" to be the last thing you looked at before everything went sideways.