Quantify uncertainty and bias in state Vehicle Miles Traveled estimates used for benchmarking

Quantify the uncertainty and potential bias in state-reported Vehicle Miles Traveled estimates used to construct human crash benchmarks by validating geographic-specific traffic sampling methodologies against ground-truth data in Arizona, California, and Texas.

Background

Benchmark crash rates rely on state VMT estimates derived from sampling and modeling, which may introduce uncertainty and bias. The authors note the absence of ground-truth validation to calibrate these estimates, affecting the precision of benchmark comparisons.

Addressing this gap would strengthen the foundations of ADS-to-human crash rate analyses by improving exposure measurement reliability across jurisdictions.

References

The mileage estimates are also derived from a variety of geographic-specific traffic sampling methodologies, and, in the absence of some ground truth data to validate, it is not clear how much uncertainty or bias should be attributed to these estimates.

— Comparison of Waymo Rider-Only Crash Rates by Crash Type to Human Benchmarks at 56.7 Million Miles  (2505.01515 - Kusano et al., 2 May 2025) in Limitations