Waymo study estimates sober drivers face 3.1–3.9 times the fatal-crash risk at night

Being sober does not erase the extra danger of driving after dark. In research described on October 8, 2026, Waymo estimated that sober drivers face 3.1–3.9 times the fatal-crash risk at night compared with daytime. The study covers the 50 most populous U.S. urban areas and offers a stricter benchmark for evaluating autonomous vehicles.

The finding concerns drivers with zero blood alcohol concentration, not roads emptied of impaired drivers. Waymo says part of the remaining nighttime risk comes from sober people having to react to impaired road users. It also points to reduced visibility, inadequate lighting, fatigue and road layouts that encourage speeding as nighttime hazards that sobriety alone cannot fix.

Reconstructing the miles behind the risk

Calculating a sober-driver crash rate requires more than counting crashes. Researchers also need to estimate how many miles sober people drive. National traffic databases do not track mileage by a driver’s blood alcohol concentration, leaving a gap in the information needed to make that comparison.

Waymo’s researchers used an exposure-reconstruction model—a method for estimating those miles. They combined toxicology records from the Fatality Analysis Reporting System, a national fatal-crash database, with real-world vehicle data known as telematics. That let them estimate sober and impaired driving exposure across the urban areas and build city-specific baselines. Waymo says the study has been accepted for publication in the Journal of Safety Research.

The estimated urban baseline without alcohol-involved drivers

Fatal-crash involvement rate

1.42→1.10

incidents per 100 million miles

Waymo estimates that excluding alcohol-involved drivers lowers the average urban fatal-crash involvement rate by 23%. This is a modeled comparison, not the result of removing impaired drivers from actual roads.

The benchmark changes with the city

The estimated effect of excluding impaired drivers varies sharply by location. In Houston, the fatal-crash rate falls by 31%; in Minneapolis, it falls by 13%. Waymo argues that this geographic difference makes local human-driver benchmarks more useful than national averages when assessing road safety. The same adjustment for alcohol does not produce a uniform change across cities.

For autonomous vehicles, excluding impaired drivers sets a higher safety bar than comparison with the full human-driving population. Waymo describes the sober-driver baseline as an additional research lens. It says it continues to measure its overall safety impact against real-world human driving under actual road conditions, rather than replacing that comparison with an idealized driver.

Those are different questions: how an autonomous system compares with the people currently on the road, and how it compares with a sober subset. This study develops the human side of that comparison. Its nighttime finding is not a measurement of Waymo’s own performance, nor does a sober-driver baseline describe a road free of other hazards.