I read this study because it measures something I keep coming back to: not how drunk someone is, but how drunk they think they are. The gap between those two numbers is where decisions get made. Here it's one decision - whether to drive. But the mechanism the researchers are tracking runs wider than traffic safety. The catch: the same impairment that changes your state also changes how you read it.

What they did

Researchers ran an RCT (randomized controlled trial) with 160 low-risk drinkers. Each participant was randomly assigned, blinded, to one of three target BAC (blood alcohol concentration) conditions: low (~0.025%), mid (~0.055%), or high (~0.085%). At multiple time points - while BAC was rising, and again as it fell - participants filled out surveys: how many standard drinks they thought they'd had, what they estimated their BAC to be, how confident they were in that estimate. They also rated how intoxicated they felt and how capable of driving.

Nobody knew which condition they were in. That blinding is the point. It removed the usual anchor - "I had three drinks, so I'm probably around X" - and left people with nothing but feel.

What they found

  • Estimation. The mid group (~0.055%) counted their drinks most accurately. The low group overestimated, the high group underestimated. When the buzz is faint, you're still tracking. When you're genuinely impaired, the accounting slips.

  • Pre-drink. Before anyone had a single drink, the people randomly assigned to the mid and high conditions were already more confident they could drive over the legal limit than the low group. A baseline difference in attitude, sitting there before any alcohol at all.

  • Mellanby effect. As BAC fell, the high group reported better driving capability than the mid group, while their actual BAC was still higher. The study names this the Mellanby effect: subjective intoxication feels stronger on the way up than at the same level on the way down. Someone past their peak feels less impaired than someone still climbing at the exact same number.

  • Drive decision. Logistic regression showed that estimated BAC and confidence in that estimate - not felt impairment - significantly predicted whether someone said they'd drive. So the variable that drives the decision is the one that breaks down most at high intoxication.

What it means

The most impaired people have the worst read on how impaired they are. And the decision to drive runs on that read, not on how the impairment actually feels.

The regression result sharpens it. What predicted driving intention wasn't the felt sense of impairment but the estimated number and the confidence attached to it. The logic runs: "I think I'm at X, I trust that estimate, so I'll drive." When the estimate is wrong - which is exactly what the high-BAC data shows - that logic fails quietly.

Two limits. Participants were low-risk drinkers in a controlled lab, not habitual heavy drinkers with built-up tolerance, so the numbers won't transfer one to one. And the outcome is stated intention, not what someone does behind a wheel. The pattern holds. The road is still an open question.

Self-assessment under alcohol breaks down exactly when intoxication is highest. To know where you actually are - your peak, when it fades - you need something outside your own head. On a night where your BAC climbed past what you planned, that internal "I'm probably okay" is the first thing the data shows breaking. AlcoBalance gives you the number instead: where you are on the curve right now, when you peaked, when you'll clear. Something to steer by when the internal read is least trustworthy.

Source: Journal of Safety Research, doi:10.1016/j.jsr.2026.05.009