Heat doesn’t make people drive worse. It shrinks the margin for error.
We measured 1.5 million rides across three July heat waves. Drivers reacted more. They avoided less.
+26.0%
High-G impact rate
2,602 vs 2,066 per million rides
+6.8%
Hard brake rate
50,950 vs 47,720 per million rides
Heat wave days vs May baseline. Same-weekday comparisons.
1.5M
rides measured
8
heat wave days
3
July heat windows
Impacts rose four times faster.
Three heat waves crossed the country in twelve days: July 2 to 4, July 6 to 8, and July 12 to 13. We measured 1,526,252 rides across those eight days and compared each day against the same weekday in May, a baseline with 6.4 million rides.
Change in event rate vs May
8 of 8
Heat wave days above their May baseline on impacts. Range: +7.2% to +46.5%.
+46.5%
Wednesday July 8 was the highest day, in the second heat window.
In May, the fleet recorded one high-G impact for every 23 hard-braking events. Across the heat wave days, one for every 20.
Fewer trips. A narrower margin.
Both event rates went up. They did not go up together. Impacts rose roughly four times as much as hard braking, while the fleet made fewer trips.
−7.0%
Rides per day
190,782 vs 205,241 in May
+2.0%
Average speed
19.77 vs 19.39 miles per hour
Who stayed on the road
Consumer vehicles showed the sharpest pullback, with time behind the wheel down 6.7%. Trips also got marginally shorter in time.
The pattern is consistent with fewer optional trips and a greater share of work travel. The data does not identify why each trip was made.
What does not fit
On July 6 and 7, hard braking fell slightly while impacts rose. Drivers may have been more cautious, or the travel mix may have shifted after Independence Day.
We cannot separate those explanations here. On the other six days, braking and impacts moved in the same direction.
The signal is clear. Its limits matter.
Heat is hard to study behind the wheel. It does not leave a mark on a driving record. This study compares fleet events on designated heat wave days with a mild-month baseline.
What was measured
1,526,252 rides across eight heat wave days, compared with the same weekdays in May. The fleet was 93.7% rideshare and taxi, concentrated in dense urban states.
No road or in-vehicle temperatures were measured. The study cannot establish whether a particular driver was hot or whether their air conditioning worked.
A proxy, not a crash count
High-G events capture sudden deceleration consistent with a collision. They are not confirmed crashes and cannot be compared directly with police-reported crash figures.
One signal held loosely
Attention lapses ran 3.1% above their comparison period. The highest day was July 7.
Distraction records extend back only 90 days, so May was unavailable. The reference was June 23 to July 1, a warm period without weekday matching. This is supporting evidence, not a primary finding.
Seasonality remains in view
May is a useful but imperfect baseline. Summer travel patterns and the driver mix may explain part of the difference. This comparison alone does not isolate heat as the cause.
Independent measurement means showing what moved, and what we cannot yet explain.
Source: Empiric Earth Heat Wave July 2026 data pull, High-G version. High-G impact events are a collision proxy, not confirmed crashes. Interpretation is observational.