100 Deadliest Days · 2026 vs 2025
Across 357 million vehicle-miles over two matched Memorial-Day-to-mid-July windows, overall safety-critical events fell −5.0% year-over-year. But collisions rose +6.4%. Drivers brake less abruptly, yet impacts aren't being prevented. This is risk redistributing, not declining.
·
Risk Didn't Fall. It Moved.
Nexar's dashcam fleet observed 12.1 million rides across 169.4 million miles in the 2026 elapsed summer window (Memorial Day through July 19). Against the matched-calendar 2025 window (187.6M miles, 12.9M rides), the rate of safety-critical events per million miles improved 5.0%, but the improvement hides a redistribution.
Headline Finding: Overall safety-critical events fell −5.0% (Z = −19.9, p < 0.001), driven by a −5.5% drop in hard braking. Yet collision events rose +6.4% (Z = +5.0, p < 0.001). Drivers brake less abruptly, but when impacts happen, they aren't being prevented.
1. Overall SC rate −5.0% (Z = −19.9), a real, robust improvement. 2. But collisions rose +6.4% (Z = +5.0): less braking, not fewer crashes. 3. The clean May-June comparison actually shows +1.8%, early summer 2026 was modestly worse. 4. Late-night's share of daily risk grew even as counts fell, risk concentrating after midnight. 5. July runs +14.9% above May-June, independently matching NHTSA's ~16% summer teen-fatality premium.
Study Context
The "100 Deadliest Days" is the period between Memorial Day and Labor Day, when teen driver fatalities historically spike ~16% above the rest-of-year average (NHTSA). This study compares the elapsed portion of that window in 2026 against the identical calendar window in 2025, using Nexar's dashcam fleet as the observational lens.
Memorial Day to Labor Day. The deadliest stretch of the American driving year.
| 2025 Window · Comparison Baseline | 2026 Window · Study Period | |
|---|---|---|
| Start | May 26 (Mem Day) | May 25 (Mem Day) |
| End | Jul 19, 2025 | Jul 19, 2026 |
| Days | 55 days | 56 days |
| Total rides | 12,923,497 | 12,110,827 |
| Total miles | 187.6M | 169.4M |
◆ A Real-Observation Read, Ahead of NHTSA: Official NHTSA fatality data for summer 2026 won't publish until 2027. Nexar's fleet provides the real-observation, real-time read on summer safety, a leading indicator built from tens of millions of miles of actual driving, not retrospective crash records.
Nexar's fleet grew between 2025 and 2026. Using raw event counts would overstate any deterioration (or understate any improvement). Normalizing by miles driven, events per million miles, isolates genuine behavioral and road-condition signals from fleet-size effects. Every rate in this report is per-million-miles for exactly this reason.
Methodology
Two event types. Hard Brakes: sudden deceleration detected by onboard AI; a proxy for near-misses and hazards. Collisions: High-G impact signatures; the closest proxy to actual collision incidents. Combined as SC events per million miles.
Baseline construction. 2025 window = comparison baseline; shared Memorial Day anchor & DOW structure; US-only geographic filter both years; July 20 excluded (partial-day ingest).
July 2026 supplement. The primary geo-incident table was unavailable past June 30, 2026 at analysis time. July 2026 was supplemented: miles from the full-fleet rides table (58.1M mi for Jul 1-19); events scaled by a calibration ratio from the Jul 2025 overlap (hard-brake 0.459, collision 0.567) capturing the US-fleet share of global incidents.
∑ Statistical framework · Poisson approximation: SE = √(rate ÷ n_million_miles) · Danger Index = (2026 rate ÷ 2025 rate) × 100. Sample sizes of 169M-188M miles provide exceptional statistical power, making even small rate differences highly significant. This is a strength for confidence, and a caution against over-reading tiny effects (see Limitations).
July 2026 event estimates carry an uncertainty band of ~±3%, reflecting potential drift in the US/global incident-share ratio between years. The core findings (−5.0% SC rate, +6.4% collision rate) remain directionally robust under any plausible calibration assumption.
Traffic Volume & Fleet Exposure
The 2026 window shows lower total mileage, but this is a methodology artifact of the July calibration approach, not fleet shrinkage. Rate normalization removes any bias this introduces.
| Metric | 2025 Window | 2026 Window | Change |
|---|---|---|---|
| Total rides | 12,923,497 | 12,110,827 | −6.3% |
| Total miles | 187.6M | 169.4M | −9.7% |
| Miles per ride (avg) | 14.5 | 14.0 | −3.8% |
| Days in window | 55 days | 56 days | +1 day |
Total vehicle-miles per window. The −9.7% reflects the July data-supplement method for Jul 1-19, 2026, not fleet contraction.
Combined Miles Observed
Across both windows, this study rests on 357 million vehicle-miles of real dashcam driving. The rate-normalized comparison (events per million miles) ensures fleet growth between years does not bias the safety read in either direction, the improvement and the collision rise are both genuine per-mile signals.
National Safety Overview · The Divergence
Three metrics, two directions. Hard braking improved and pulled the combined rate down, but collisions moved the opposite way. This split is the analytical core of the study.
| Safety Metric | 2025 / M mi | 2026 / M mi | YoY | Danger Index | Assessment |
|---|---|---|---|---|---|
| Hard Brake Rate | 1,641.3 | 1,550.5 | −5.5% | 94.5 | Below Baseline |
| Collision Rate ★ | 72.2 | 76.8 | +6.4% | 106.4 | Moderate ↑ |
| Safety-Critical (Combined) | 1,713.5 | 1,627.3 | −5.0% | 95.0 | Below Baseline |
Year-over-year change by metric · teal = improvement, amber = deterioration.
✓ The Headline Number (Shareable): Nexar's US fleet recorded 1,627 safety-critical events per million miles in the 2026 window, down from 1,714 in 2025. A 5% year-over-year improvement, even as summer driving exposure stayed high. The real-observation read on summer safety that NHTSA fatality data won't show until 2027.
Within the improved headline, collision events per million miles rose from 72.2 (2025) to 76.8 (2026), a +6.4% increase. This is statistically significant and directionally consistent with NHTSA trends showing severe-event categories persisting even as overall incident frequency declines. Less braking, but not fewer crashes.
May-June Apples-to-Apples
Because July 2026 required calibration, the cleanest like-for-like comparison uses only May 25/26 through June 30 in both years, both from the identical pipeline, no adjustments. It tells a different story than the headline.
| Metric | 2025 May-Jun | 2026 May-Jun | YoY |
|---|---|---|---|
| Rides | 8,539,889 | 7,333,128 | −14.1% |
| Miles | 124.2M | 111.3M | −10.4% |
| Hard Brake Rate / M mi | 1,570.0 | 1,591.6 | +1.4% |
| Collision Rate / M mi | 61.4 | 69.3 | +12.8% |
| SC Rate / M mi | 1,631.4 | 1,660.9 | +1.8% |
Clean Early-Summer Signal Points Up
The pure May-June comparison shows a +1.8% SC rate increase in 2026, while the full window shows −5.0%. The divergence: July 2025 was far riskier (1,874/M mi) than May-June 2025 (1,631/M mi), so the full-window "improvement" partly reflects a high-risk July 2025 being compared to a calibrated, and likely slightly underestimated, July 2026.
The full-window −5.0% is the headline; the May-June +1.8% is the truer early-summer story. Collisions specifically jumped +12.8% in the clean comparison, an even sharper version of the redistribution signal. Read together: 2026 early summer was modestly more dangerous, and the collision deterioration is not a July-calibration artifact.
Statistical Significance
With 169-188 million vehicle-miles per sample, this analysis has exceptional power. All three rate changes clear p < 0.001. The opposite-direction split, braking down, collisions up, is a genuine signal, not sampling noise.
| Metric | 2025 | 2026 | Delta | Z-Score | 95% CI (2026) | Sig. |
|---|---|---|---|---|---|---|
| SC Rate / M mi | 1,713.5 | 1,627.3 | −86.2 (−5.0%) | −19.9 | [1,621-1,633] | p<0.001 |
| Collision / M mi ★ | 72.2 | 76.8 | +4.6 (+6.4%) | +5.0 | [75.5-78.1] | p<0.001 |
| Hard Brake / M mi | 1,641.3 | 1,550.5 | −90.8 (−5.5%) | −20.7 | [1,544-1,557] | p<0.001 |
This is the statistical heart of the paradox: both the overall improvement (Z = −19.9) and the collision rise (Z = +5.0) are highly significant. They are not noise pulling in different directions, they are two real, independently-confirmed phenomena. The reconciliation is behavioral: how people drive changed, and it split the two metrics.
Hourly Patterns · Late-Night Intensifying
All times Eastern. The hourly hard-brake distribution was remarkably stable year-over-year, peaking in the 3-6 PM window. But the tails shifted: late-night and pre-dawn grew their share of daily risk in 2026.
| Hour (ET) | Window | 2025 Share | 2026 Share | Shift | Direction |
|---|---|---|---|---|---|
| 4-5 AM | Pre-Dawn | 0.96% | 1.07% | +0.105pp | ↑ Intensifying |
| 1-2 AM | Late Night | 1.54% | 1.63% | +0.093pp | ↑ Intensifying |
| 3-4 PM | Afternoon Peak | 6.86% | 7.03% | +0.168pp | ↑ Intensifying |
| 5-6 PM | Peak Hour | 7.18% | 7.16% | −0.023pp | Stable |
| 8-9 PM | Evening | 4.91% | 4.68% | −0.233pp | ↓ Improving |
Hard brake share by hour, % of daily total. The distribution peaks at 5 PM; the late-night and pre-dawn tails grew their share in 2026.
The 12-2 AM band and 4-5 AM pre-dawn slot showed the largest share increases in 2026. Though absolute counts are small, these are the highest-risk per-mile hours: impaired driving, fatigue. The rising share means late-night exposure grew relative to daytime driving. Combined with the collision rise, the after-midnight subset of driving is becoming relatively riskier.
Day-of-Week Analysis
Friday improved most; Sunday was the only day that worsened. 2026 day-of-week rates use May 25-Jun 30 classifier data (July DOW breakdown not available via the calibration method).
| Day | 2025 SC / M mi | 2026 SC / M mi | YoY | DI (2026) |
|---|---|---|---|---|
| Sunday | 1,562.5 | 1,587.1 | +1.6% | 101.6 |
| Monday | 1,666.3 | 1,629.0 | −2.2% | 97.8 |
| Tuesday | 1,743.6 | 1,693.6 | −2.9% | 97.1 |
| Wednesday | 1,744.4 | 1,659.5 | −4.9% | 95.1 |
| Thursday | 1,743.9 | 1,719.6 | −1.4% | 98.6 |
| Friday ★ | 1,803.8 | 1,701.5 | −5.7% | 94.3 |
| Saturday | 1,674.4 | 1,608.0 | −4.0% | 96.0 |
YoY change by day · teal = improvement, amber = deterioration.
Friday showed the largest improvement (−5.7%, 1,803.8 → 1,701.5/M mi), suggesting Friday afternoon/evening driving improved most, possibly commute or remote-work shifts. Sunday was the only day that worsened (+1.6%), consistent with late-Sunday return trips and the known Sunday-evening risk window.
State-Level Analysis
Top states by total miles and their safety profiles (full 2025 window vs 2026 May-June, identical data source). The geographic spread is wide, New York worsened, Illinois improved dramatically.
| State | 2025 SC / M mi | 2026 SC / M mi | YoY | 2025 Miles | Assessment |
|---|---|---|---|---|---|
| New York | 2,488 | 2,638 | +6.0% | 36.7M | Highest ↑ |
| California | 2,172 | 2,086 | −3.9% | 22.3M | Improved |
| Texas | 1,440 | 1,455 | +1.0% | 15.8M | Flat |
| Florida | 1,696 | 1,556 | −8.2% | 10.7M | Improved |
| New Jersey | 1,844 | 1,679 | −9.0% | 10.1M | Improved |
| Pennsylvania | 1,762 | 1,654 | −6.1% | 6.1M | Improved |
| Georgia | 1,272 | 1,336 | +5.0% | 5.9M | Worsened |
| Illinois ★ | 1,536 | 1,180 | −23.2% | 5.8M | Best ↓ |
| Massachusetts | 1,971 | 1,879 | −4.7% | 3.3M | Improved |
| Washington | 1,856 | 2,008 | +8.2% | 2.9M | Worsened ↑ |
New York posted the highest rate both years (2,488 → 2,638, +6.0%), dense urban driving, high incident sensitivity. Illinois showed the most dramatic improvement (−23.2%), likely a Chicago-metro compositional or behavioral shift. Florida & New Jersey improved strongly (−8.2%, −9.0%). Washington deteriorated most (+8.2%), driven by the Seattle metro. The wide spread underlines that the national average hides very different local stories.
Risk Windows · Time × Day Interaction
Hard-brake raw counts by time-window and day-type. Every window fell in absolute terms, but raw counts partly reflect the ~6% smaller 2026 ride sample. The share-based read (Section 08) is the truer measure of where risk concentrated.
| Time Window | Day Type | 2025 HB | 2026 HB | Raw Change |
|---|---|---|---|---|
| Late Night (12-5am) | Weekend | 14,483 | 13,131 | −9.3% |
| Late Night (12-5am) | Weekday | 30,210 | 25,549 | −15.4% |
| Afternoon (3-6pm) | Weekend | 31,862 | 27,616 | −13.3% |
| Afternoon (3-6pm) | Weekday | 108,663 | 89,800 | −17.4% |
| Evening (6-10pm) | Weekend | 33,653 | 27,892 | −17.1% |
| Evening (6-10pm) | Weekday | 107,616 | 87,053 | −19.1% |
| Morning (9am-12pm) | Weekday | 81,432 | 66,196 | −18.7% |
| Midday (12-3pm) | Weekend | 31,827 | 26,290 | −17.4% |
▲ Read Counts With Care: The raw reductions above partly reflect the smaller 2026 ride sample (~6% fewer rides). Normalized by miles, the improvement is smaller (−5% overall). The share-based analysis is more meaningful for whether specific windows intensified relative to total driving, and there, late-night share rose even as counts fell.
Late-night hard brakes fell in absolute terms (fewer night miles), but the proportion of daily events occurring at night increased. Late night now represents a larger slice of the daily risk distribution. Combined with the rising collision rate, the late-night subset of driving is becoming relatively riskier in 2026, the clearest expression of the "risk redistributing, not declining" thesis.
Weekly Trend Within the Window
In 2025, safety-critical rates escalated sharply through July, the defining feature of the 100 Deadliest Days. The deadliest weeks come last.
2025 weekly SC rate per million miles · amber = the July peak, the deadliest weeks.
July Safety Premium vs May-June
In 2025, safety-critical rates rose from 1,551 (Memorial Day week) to 1,958 by mid-July, a +26% escalation, with July running +14.9% above May-June. The July 14 week (1,958/M mi) is the single most dangerous weekly reading in either year. The deadliest weeks genuinely come last, which is why the July 20-Labor Day extension (Recommendation 4) matters most.
External Research Comparison
Nexar's fleet findings validate cleanly against published NHTSA, IIHS, and NSC benchmarks, most strikingly on the summer premium, where two completely different data sources nearly agree.
| Metric | Published Finding | Nexar 2026 Result | Assessment |
|---|---|---|---|
| Summer teen fatality premium | NHTSA: ~16% more May-Aug | +14.9% Jul vs May-Jun | Validated ✓ |
| All-ages crash rate | IIHS: 15-24.7 / M mi | 1,627 SC / M mi | Validated ✓ |
| Long-term improvement | IIHS: ~1-3% / yr decline | −5% YoY | Explainable |
| Collision frequency trend | NHTSA: severe crashes persist | Collision +6.4%, HB −5.5% | Validated ✓ |
| Late-night risk premium | NSC: night = highest-risk | Late-night share rose | Validated ✓ |
▲ Collision Divergence Also Validated: NHTSA's long-term data shows fatal/severe crashes not declining at the same rate as minor crashes. Nexar's split, collisions +6.4% while hard brakes −5.5%, is the fleet-data echo of exactly that national pattern.
NHTSA cites ~16% more teen traffic deaths in summer. Nexar's fleet independently observed a +14.9% safety-critical premium for July vs May-June, nearly identical, despite using entirely different data (dashcam events vs fatality records) and covering all driver ages. Strong evidence that Nexar's fleet captures real-world seasonal safety patterns with high fidelity.
Key Findings
The improvement, the buried collision story, the clean-comparison caveat, and where risk concentrated.
Recommendations
Five actions: lead with the buried story, spotlight the redistribution, and extend the window before it closes.
Communications · Lead with the +6.4% collision rise as the buried story
The narrative "overall risk fell, but crashes aren't declining" is more compelling and more accurate than a simple better/worse framing. It also differentiates Nexar's insight from NHTSA's aggregate statistics.
Content Focus · Spotlight late-night and Sunday
Late-night risk intensity (as a share of daily events) rose in 2026, and Sunday is the only day that worsened YoY. Both are press-friendly angles for targeted safety messaging Nexar's data uniquely enables.
Timeliness · Publish before Labor Day (Sep 1)
This analysis covers the active window. Publishing by early-to-mid August preserves timeliness, the data as of July 19 is complete for the key talking points.
Follow-Up · Extend the analysis post-Labor Day
Re-run to capture the full window (Jul 20-Sep 1). The July 20-Aug 31 portion, including peak summer weekends, is missing from this study and likely contains the highest-risk readings of all.
Geographic Spotlight · New York & Washington worsened, localize it
Both states deteriorated in 2026. Consider state-specific releases with New York media contacts to maximize local pickup, paired with Illinois' dramatic −23.2% improvement as a counterpoint.
Limitations & Appendix
July 2026 calibration
Jul 1-19, 2026 events required a calibration ratio (0.459 HB, 0.567 collision) from the Jul 2025 overlap, introducing ±3% uncertainty in the full-window metrics.
Fleet representativeness
Nexar's fleet is not a representative US sample, it skews toward tech-adopting consumers, safety-program commercial fleets, and metro rideshare. Rates aren't directly comparable to general-population fatality stats.
Event definition (proxy)
"Collision" = High-G impact detected by dashcam, not a verified police-reported crash. Some events may be bumps or potholes. But the definition is consistent across both years.
No driver-age data
Covers all fleet drivers, not specifically teens. The "100 Deadliest Days" framing is borrowed from NHTSA teen research but applied to all-driver fleet data.
Composition & window notes
Geographic fleet-mix shifts could show as rate change rather than behavior, state-by-state is the cleanest separator. 2026 window is 56 vs 55 days; July 20 excluded (207 rides). Rate normalization removes both effects.
Hard Brake Share by Hour · Selected
| Hr ET | 2025 % | 2026 % | Shift |
|---|---|---|---|
| 1 AM | 1.54 | 1.63 | +0.09 |
| 4 AM | 0.96 | 1.07 | +0.11 |
| 10 AM | 5.21 | 5.06 | −0.15 |
| 3 PM | 6.86 | 7.03 | +0.17 |
| 5 PM ★ | 7.18 | 7.16 | −0.02 |
| 8 PM | 4.91 | 4.68 | −0.23 |
| 10 PM | 3.23 | 3.10 | −0.13 |
2025 Weekly SC Rate · Full Window
| Week of | Miles | SC / M mi | vs Avg |
|---|---|---|---|
| May 26 | 23.7M | 1,550.7 | −9.5% |
| Jun 9 | 24.2M | 1,581.7 | −7.7% |
| Jun 23 | 24.3M | 1,740.1 | +1.5% |
| Jun 30 | 22.4M | 1,756.3 | +2.5% |
| Jul 7 | 23.7M | 1,882.5 | +9.9% |
| Jul 14 ★ | 20.8M | 1,957.6 | +14.3% |
Analysis by Nexar BI · July 20, 2026 · Data source · Nexar Dashcam Fleet · United States · Windows · May 26-Jul 19, 2025 & May 25-Jul 19, 2026 · 357M vehicle-miles analyzed