A new Insurance Institute for Highway Safety study compared smartphone telematics with police reports for almost 17,000 crashes in Connecticut, Illinois, Ohio, and Texas. After excluding passive screen activity such as an untouched navigation display, the data still show a huge gap: phones detected active use before crashes far more often than police documented it.
| Active phone use | Within 30 seconds of crash | Within 10 seconds of crash | Recorded in police reports* |
|---|---|---|---|
| Handheld calls | 22 | 16 | 7 |
| Hands-free calls | 90 | 53 | 1 |
| Handling/using the phone | 639 | 310 | 20 |
| Total | 751 | 379 | 28 |
*Police reports did not specify whether phone use occurred within 10 or 30 seconds of impact. The 28 police-reported cases are a single comparison total for the matched crash sample.
Telematics detected almost 27 times more active phone use within 30 seconds of impact than police reports did. Even inside the final 10 seconds, it found more than 13 times as much.
That sounds outrageous until you consider the mechanics of a car crash (no pun intended).
By the time the squad car rolls up, the phone is probably back in a pocket, cupholder, purse, or somewhere under the passenger seat covered in broken bits of glass. Nobody standing roadside has any particular incentive to volunteer, “Yeah, officer, I was choosing a new Spotify playlist when I rear-ended that Camry.”
That makes distracted driving a particularly tough thing to prove later, which makes a rough pairing with the general disincentive police officers have towards investigating crashes at all if there is no obvious criminal angle.
Your Insurance App May Know More Than the Officer
The telematics data came from Cambridge Mobile Telematics, or CMT, which supplies DriveWell technology used by insurance companies, fleets and various safe-driving programs.
DriveWell isn’t necessarily an app you’d recognize by that name. The technology can live inside an insurer’s branded app and automatically detect trips using the phone’s sensors and location data. Depending on the program, it can record speeding, hard braking, aggressive acceleration, phone interaction and likely crashes.
If you’ve ever agreed to let your insurance company watch you drive in exchange for a discount, there’s a decent chance you’ve met this general species of electronic hall monitor.
For this study, CMT could also separate different kinds of phone activity. A call routed through the handset could register as handheld. Speakerphone, Bluetooth and vehicle infotainment could register as hands-free. Physical manipulation covered the obvious stuff: tapping, scrolling, typing, swiping, gaming and holding the device.
We didn’t dive too deeply into passive usage, and excluded it from our stats up above. CMT defines passive phone use as an unlocked screen without physical interaction. That could be somebody watching YouTube, sure, but it could just as easily be Google Maps sitting untouched in a dashboard mount, doing precisely what Google Maps is supposed to do.
There Are Some Big Caveats
Phone use before a crash does not automatically make the phone user responsible for the crash.
The researchers didn’t have fault determinations. Someone could be sitting at a red light on a hands-free call and get rammed by a pickup from behind. Activity 29 seconds before impact also tells us a lot less than somebody actively tapping the screen one second before impact.
CMT additionally uses machine learning to estimate whether its user was driving or riding as a passenger. That classification has reportedly exceeded 96 percent accuracy, which is impressive, but leaves some room for error.
So I would not take these numbers and proclaim that 751 crashes were “caused by phones.” The study doesn’t support that. What the study does support, very strongly, is that active phone use immediately before crashes happens a hell of a lot more often than police reports manage to capture.
Police reports are good at recording evidence that is still there when the police arrive. A broken control arm will generally remain broken. A smashed bumper is pretty cooperative evidence. Skid marks have the courtesy to stay on the pavement.
A thumb scrolling TikTok leaves basically nothing. No digital finger print, ironically, that the police can readily access.
Unless somebody admits to it, a witness catches it (preferably on camera), or investigators gain access to electronic data with warrants, the evidence is well-hidden before the first responder even gets out of the car.
Take the Official Stats on Distracted Driving with a Grain of Salt
All that is worth remembering whenever you see an official distracted-driving statistic. There may be a very good count of the crashes where phone use could actually be established afterward. The phones themselves suggest there were a whole lot more crashes where nobody ever knew to write it down.
