The plain version: when a highway lane closes, the fastest thing everyone can do is use both lanes to the very end and take turns at the merge point. Almost nobody does it. The reason isn't ignorance. It's that we don't trust each other, and the road has no way to fix that.
I live in Austin. If you've ever crawled up I-35 through the middle of this city, you have personally sat inside the problem this paper is about.
Engineers call the right behavior the zipper merge, and they've measured it: field data from Minnesota shows backups roughly forty percent shorter. One Michigan work zone went from six miles of congestion to three. This isn't a folk remedy. It's settled traffic engineering.
So why does everyone merge early and glare at the people who don't? Because most drivers privately suspect the zipper is correct, but they believe everyone else thinks late merging is cheating. So they get over a mile early to be polite, and a few drivers appoint themselves enforcers and straddle the open lane to block the "cheaters." The blocking is what makes late merging genuinely dangerous, which then proves to everyone that late mergers are reckless. The people punishing the behavior are manufacturing the evidence for punishing it. If that loop sounds like it might apply to more than traffic, you're reading the right website.
Here's what surprised me most when I dug in. Transportation economists have already built most of the machine that would tell us what this costs. The cost of congestion exposure: computable today. The dollar value of delay, fuel, and crashes: computable today, down to $24.01 per person-hour. The chain is missing exactly one number, the middle term, which requires actually observing how far from the merge point real drivers get over. Measuring it costs about forty thousand dollars. The bracketed loss it would let us calculate, and start recovering, is between $1.2 billion and $8.3 billion a year.
I know how that sounds. A consultant telling you billions are lying on the ground is a genre, and not a trustworthy one. That's exactly why the paper publishes the entire chain, every source and every multiplication, so you can check the arithmetic yourself and catch me if I'm wrong.
One more thing, because it's my favorite part. To fund the study, we're sending the identical proposal to four different kinds of funders on the same day and timing the responses. The funding process is itself an experiment about how fast our institutions can move on a cheap, obvious, public-interest question. Either someone funds the measurement, or the silence becomes data. We win either way, which is how I prefer to design things.
The merge is the cheapest possible test of a much bigger idea: that many "behavior problems" are really assurance problems, fixable by repairing what people believe about each other instead of lecturing them. If the fix works on a highway, the template travels to every queue, policy, and public that runs on mutual suspicion. That is why a $40,000 study is worth doing carefully in public.
- Drivers' private preference for the zipper merge exceeds what they believe other drivers approve of by more than fifteen points. That hidden gap is the mechanism.
- If the gap comes in under fifteen points, the mechanism is wrong, and we publish the null result with the same prominence as we would have published the win.
The full chain, the survey instrument, and the protocol are in the paper.