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. Specifically here: the decision to drive. But the mechanism the researchers are tracking is broader 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 completed surveys about how many standard drinks they thought they'd consumed, what they estimated their BAC to be, and how confident they were in that estimate. They also rated their perceived level of intoxication and how capable they felt of driving.

Nobody knew which condition they were in. The blinding matters: it removed the anchor of "I had three drinks, I'm probably at X." Participants had to go on feeling alone.

What they found

  • Estimation. The mid group (~0.055%) was most accurate in counting how many drinks they'd consumed. 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, people randomly assigned to the mid and high conditions were already more confident in their ability to drive over the legal limit than the low group. A baseline difference in attitude, present before any alcohol at all.

  • Mellanby effect. As BAC fell, the high group reported better driving capability than the mid group, despite still having a higher actual BAC. The study names this the Mellanby effect: subjective intoxication feels stronger when BAC is rising than when it's at the same level on the way down. Someone past their peak feels less impaired than someone still climbing at the exact same BAC.

  • Drive decision. Logistic regression showed that estimated BAC and confidence in that estimate - not felt impairment - significantly predicted whether someone said they'd drive. 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 this. The predictor of driving intention wasn't the felt sense of impairment but the estimated BAC and confidence in that number. 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.

A couple of things to note: participants were low-risk drinkers in a controlled lab, not habitual heavy drinkers with built-up tolerance, so the numbers won't transfer directly. And the outcome is stated intentions, not actual behavior behind a wheel. The pattern holds. Transfer to real roads 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 external, not the feeling. On a night where your BAC climbed past what you planned, that internal "I'm probably okay" is precisely what the data shows breaks first. AlcoBalance gives you the actual number: 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