Somebody finally found a way to measure distracted driving without asking drivers to admit they were doing it.
Four Harvard-affiliated researchers — Vishal Patel, Christopher Worsham, Michael Liu and Anupam Jena — published a study in JAMA Network Open on August 17 arguing that U.S. traffic deaths jump on the days major music albums drop. The full methodology is laid out in their NBER working paper, and the design is more interesting than the headline number.
The trick
Studying phone distraction is a nightmare because you can’t randomize it. You can’t assign one group of drivers to scroll Spotify on the interstate and see who dies. And survey data is worthless, because nobody checks the box marked “I was picking a track when I rear-ended that Tahoe.”
So the researchers went looking for a moment when millions of Americans reach for their phones simultaneously for reasons that have nothing to do with driving. Album release days fit. Labels schedule releases to maximize streams, not to coincide with heavy travel. That makes the date effectively random with respect to crash risk — a natural experiment in when phones get used rather than who uses them.
Related Articles
- America Didn’t Ban Polestar’s Factory. It Banned Its Owner.
- 31 Felony Charges Later, Georgia’s Tag Office Has One Very Awkward Question: Who Else Has Her Login?
They pulled the ten albums with the highest single-day Spotify stream counts from 2017 through 2022. Taylor Swift’s Midnights topped the list at 184.7 million first-day streams on October 21, 2022, followed by Drake’s Certified Lover Boy and Bad Bunny’s Un Verano Sin Ti. Then they pulled every fatal crash in NHTSA’s Fatality Analysis Reporting System for release day and the ten days on either side.
Streaming confirmed the premise: 123.3 million streams on release days against 86.1 million on surrounding days, a 43% bump. For all ten albums, release day was the single biggest streaming day.
Fatalities followed. Adjusted for holidays, day of week, week of year and calendar year, release days averaged 139.1 traffic deaths versus 120.9 on the surrounding days. That’s 18.2 additional deaths, a 15.1% relative increase, and roughly 182 extra fatalities across those ten dates.
Why it probably isn’t a fluke
Albums drop on Fridays. Fridays are already deadly. The researchers knew you’d say that, so they ran the numbers against 1,000 simulations using randomly chosen Fridays from the same period. Only 20 of those 1,000 produced an effect bigger than what they observed. They also re-ran each album’s calendar date in every other study year — same season, same holiday context, no album — and found nothing. And restricting the analysis to Friday releases compared against the Fridays immediately before and after still returned an odds ratio of 1.10.
The subgroup breakdown is what sells it. If release-day parties and alcohol were driving this, you’d expect the extra deaths to cluster at night, among drinking drivers, in cars full of people. The opposite happened. The increase was concentrated among sober drivers, in single-occupant vehicles, in clear weather. Day versus night showed no meaningful difference at all.
Read that pattern as a mechanism and it describes one thing: a person alone in a car on a nice afternoon, sober, looking at a screen.
The part that should bother you if you spec’d CarPlay
Here’s the finding buried in Figure 4. After restricting the sample to vehicles built in 2016 or later — the point when CarPlay went mainstream — cars equipped with Apple CarPlay showed a larger release-day increase than cars without it.
That looks backwards, because CarPlay is supposed to be the safer option. And by one measure it is. The AAA Foundation’s 2018 work with University of Utah researchers found CarPlay and Android Auto were 24% faster than native systems for placing a call and 31% faster for programming navigation, and generated moderate overall demand where the factory systems generated very high demand. Both projected interfaces beat the OEM software.
But “less demanding per interaction” and “less total distraction” are different claims. Lower the friction on a task and people do it more often. Widen the freeway, get more traffic. If clicking three tiles to queue Her Loss takes four seconds instead of twelve, you don’t spend the saved eight seconds watching the road — you go find the next thing. The study’s authors make essentially this argument: the smartphone-optimized interface may lower the barrier to interacting at all, an offsetting effect that nets out worse.
Worth remembering the benchmark AAA attached to all this: taking your eyes off the road for more than two seconds doubles crash risk. Even CarPlay navigation entry ran up to 33 seconds.
The undercount nobody talks about
Here’s the context that makes this study necessary in the first place. NHTSA’s 2024 research note counts 3,208 people killed in distraction-affected crashes, 8% of the year’s 39,254 traffic deaths. Of the 2,955 fatal distraction crashes, only 404 were coded with a cellphone in use.
NHTSA is refreshingly blunt about why that number is soft. Police crash reports don’t have a consistent distraction field across jurisdictions. Officers often depend on a surviving driver volunteering that they were texting. When the distracted driver dies, there’s frequently nobody left to ask.
For scale: the agency’s own economic impact study used naturalistic observation — cameras in cars watching real drivers — and estimated distraction was involved in 29% of all crashes and 10,546 fatalities in 2019. That’s more than triple what the police-report data captures.
Related Articles
- Florida’s Flashy New ‘Super Speeder’ Law Grabbed the Headlines — a Forgotten 1971 Statute Grabbed the License
- McLaren Wants More US Dealerships. State Franchise Law Says Not So Fast
So when a quasi-experimental design finds an 18-death bump on ten specific Fridays, the honest read isn’t that it discovered a small new problem. It’s that it found a visible edge of a much larger one the official statistics were never built to see.
What to actually do with this
Queue before you roll. The authors’ own advice is to pick the album, hit play, and put the phone down before the car moves. Nobody’s asking you to drive in silence.
Use the wheel, not the screen. Steering wheel controls exist precisely because they don’t require eyes-off-road time. Track-skip on a stalk or a thumb wheel is a fundamentally different act than browsing a grid of album art.
Download it. Offline playback removes the buffering-and-retrying loop that pulls your attention back to the screen on bad cell coverage.
Hand the phone to your passenger. The single-occupant finding suggests a second person in the car meaningfully reduces this risk, probably by taking over DJ duty.
Know that the law and the data both split phone from dashboard. NHTSA codes “adjusting audio or climate controls” and “using other components integral to vehicle” as distraction categories entirely separate from cellphone use, and most state hands-free statutes are written around handheld devices. Fiddling with a factory touchscreen generally isn’t the same offense as holding a phone, even when the eyes-off-road time is identical.
One honest caveat the authors flag: FARS only contains fatal crashes. Every fender-bender, every totaled car, every insurance claim generated by somebody hunting for track seven — none of that is in this dataset. The real number is bigger. We just can’t see it yet.