Critical Context — Read First: The Nexar fleet is approximately 85% Taxi/Rideshare vehicles operated by professional drivers who do not make personal modal-shift choices. This analysis measures commercial vehicle activity, not personal commuter car trips. Bike to Work Day primarily affects personal commuters - a segment representing ~3% of the Nexar fleet (Consumer vehicles). The absence of a large national fleet signal is expected and informative, not a null result. Any effects detected are directional indicators only.
Executive Summary
On May 15, 2025, an estimated 100,000+ Americans participated in the national Bike to Work Day - the largest annual event promoting bicycle commuting. Nexar recorded 358,419 rides on that day across its connected-camera fleet. This report examines whether that event left a measurable signature in commercial driving patterns, safety metrics, and cyclist detection rates.
The Headline Finding: No statistically significant fleet-wide impact was detected. This is the expected result. Rideshare and taxi drivers do not choose to bike to work — their vehicle operations are demand-driven, not commuter-driven. The more informative signals lie in the daytime hourly pattern and state-level variation. Using the stricter 6-date pre-event baseline: High G Impact reaches significance (+17.3%, Z=+2.10) but is likely a seasonal confound, not an event signal.
Baseline: 6-Date Pre-Event Window: All comparisons in this report use the 6-date pre-event baseline (Apr 3, Apr 10, Apr 17, Apr 24, May 1, May 8 - the six Thursdays immediately before the event), avg: 367,090 rides/day. This is the closest temporal match to May 15 and avoids post-event seasonality. The full data table below shows all 12 Thursdays for context, but all statistics are computed against the pre-event 6-date window only. Note: Taxi/Rideshare shows Z=−4.75 (−2.3%) which reflects the natural April→May seasonal decline in rideshare volumes, not event displacement.
Event Background & Study Context
Bike to Work Day (National Bike Month, third Friday of May) is the most widely observed cycling advocacy event in the United States. In 2025, the event fell on May 15.
Scale of the Event
- MTC Bay Area 2025: 16,000 Energizer Station visitors; logged 159,204 miles — 33% above the 120K-mile goal.
- Approximately 100,000 historic annual participants nationally.
- MWCOG DC Metro 2025: Event confirmed for May 15 with an 18,000 T-shirt registration cap and 100+ pit stops across the region.
- Cycling context: PeopleForBikes (2024) reports 112M Americans rode bikes in 2024 — a 10-year high. 67% of micromobility commuters report they would otherwise have driven.
- Habitual bike commuters: ACS 2022 counts 731,272 habitual commuters (0.54% of US workers). High-cycling cities: Davis CA (16.6%), Boulder CO (10.4%), Portland (6.5%), DC & SF (~4%).
Why This Event Matters for Fleet Analysis
Bike to Work Day is an intentional, geographically concentrated perturbation to commuter modal share. If Nexar's fleet reflected personal commuter behavior, we would expect a small but detectable reduction in morning peak riding volume in cycling-positive metros. The critical challenge is that Nexar's fleet composition does not reflect personal commuter behavior.
Fleet Composition Context ~85% Taxi/Rideshare (professional, demand-driven) · ~15% Consumer + Other vehicle types. Of Consumer vehicles (~10,000–10,600/day), some fraction represents personal commuter trips — but even this segment is geographically dispersed and not exclusively AM-commute behavior. The effective personal commuter segment exposed to Bike to Work Day effects represents well under 3% of total fleet rides.
Methodology
Study Design
This is a quasi-experimental before/after study comparing May 15, 2025 to a matched set of Thursdays drawn from a ±6-week window. Thursdays only were used to control for day-of-week variation — Bike to Work Day always falls on a Friday, but the closest available commercial dataset used Thursday as the representative weekday anchor.
All comparisons use ride-level aggregate metrics from Nexar's MT_MOD_RISK and MT_MLP BigQuery fact tables, queried at national scope.
Baseline Windows
Two baseline windows are used throughout this report, for different analytical purposes:
⚡ Statistical Power Analysis
With 6 baseline dates (N=6), the minimum detectable effect at 80% power is approximately ±9% of the mean ride count (~33,000 rides). Real Bike to Work Day effects in a commercial rideshare fleet — if any exist — are expected to be well under 1% nationally (~3,600 rides). This study is fundamentally underpowered to detect the true underlying effect.
The null result does not mean the effect is zero; it means the effect, if it exists, is below our detection threshold. More baseline dates, or city-level analysis, would be required to reach adequate power.
Incident Rate Methodology
Incident rates are expressed per million rides (e.g., "35,177 hard brake events per million rides") to normalize for day-to-day variation in total fleet size. This dual-pipeline approach (raw event counts divided by fleet-normalized ride volume) is standard for Nexar safety analytics.
Cyclist Detection Pipeline
Cyclist and pedestrian detections come from the MT_MLP Multi-Label Perception pipeline. This pipeline was in early rollout during the study window. April–May 8 baseline dates show near-zero detections, confirming the pipeline was not yet deployed at scale. Only May 15 and post-event dates can be compared on this metric, limiting statistical inference.
Geographic Scope
All primary analyses are conducted at national scope. City-level and state-level daily analysis could not be completed due to database scan cost constraints. See Limitations for discussion of why this is a material gap.
Overall Driving Volume
The core question: did the Nexar fleet record fewer rides on Bike to Work Day than on comparable Thursdays? The short answer: not significantly. Both baselines tell a consistent story — slight underperformance relative to the recent pre-event window, essentially flat relative to the full seasonal window.
Pre-Event 6-Date Baseline
Baseline avg: 367,090 (6 Thursdays: Apr 3, Apr 10, Apr 17, Apr 24, May 1, May 8)
May 15: 358,419
Delta: −2.4%
Directionally consistent wit h Nordback benchmark · N O T S I G N I F I C A N T ( z = − 1 . 5 0 )
Seasonal Trend Note: The data shows a clear seasonal decline in ride volume: April averages ~373K rides/Thursday, declining to ~337K by late June. May 15 sits at 358K — consistent with the seasonal trajectory between April's higher volumes and June's lower ones. The pre-event 6-date baseline of 367K captures the most proximate comparator window, avoiding the downward seasonal drag of post-event dates.
Hourly Traffic Patterns
Querying IT_TRX_RIDE_LOCATIONS_CLASSIFIER across all rides (no country filter) in UTC time, the event day shows a consistent slight decline in the overnight/early-morning UTC window (UTC 03–08: −4.6% to −6.3%) and is essentially flat during afternoon and evening hours (UTC 13–23: −1.7% to +0.4%). This is fully consistent with the overall −1.0% metric for this dataset.
The steepest drop is at UTC 05:00 (−6.3%) — US late-evening/overnight in all time zones. Only four hours show marginal positive readings (+0.1% to +0.4%), all within noise.
Dataset & Methodology Note
Hourly data uses IT_TRX_RIDE_LOCATIONS_CLASSIFIER with no country or state filter (all rides, UTC timestamps). Event-day total in this table: 283,074 rides vs baseline avg 285,860 rides/day = −1.0%. All three metrics (rides, hours, km) are compared against the same 6-date pre-event baseline. Times are UTC — US Eastern is UTC−4, Central UTC−5, Mountain UTC−6, Pacific UTC−7.
Safety Metrics
Four incident types are tracked as rates per million rides, normalized for daily fleet volume variation. On Bike to Work Day, all four metrics trend in a favorable direction — though none cross the threshold of statistical significance.
Safety Signal Summary (6-Date Pre-Event Baseline) With the 6-date pre-event baseline (Apr 3 – May 8), hard braking (−0.6%) and cornering (−0.8%) are essentially flat. Harsh acceleration is slightly up (+3.8%). High G Impact is the only statistically significant finding at +17.3% (Z = +2.10) — though this appears to reflect the lower April baseline rates rather than a genuine event day elevation. All three behavioral metrics (braking, cornering, acceleration) are directionally consistent with "Safety in Numbers" literature (NHTSA 2023) — trending flat to slightly lower.
★ Significant with 6-date pre-event baseline (z=+2.10, p<0.05). Treat with caution — April baseline dates have structurally lower High G rates (~771–919/M) vs. post-May dates (~1,039–1,305/M), suggesting seasonal confound rather than event-driven signal.
Caveat on hight G impact finding
The High G Impact significance (Z=+ 2.10, p <0.05) with the pre-event baseline is driven primarily by the April baseline having lower inherent High G rates (~ 7 71 –9 19/M) compared to post - May dates (~ 1,039– 1,305 /M). This is a hallmark of seasonal confound rather than event - driven signal. Do not report this as a causal finding without further segmentation.
State-Level Analysis
Geographic Analysis Methodology
State-level ride counts for May 15 are compared directly against the 6-date pre-event baseline average (Apr 3, Apr 10, Apr 17, Apr 24, May 1, May 8) using IT_TRX_RIDE_LOCATIONS_CLASSIFIER (partitioned by ride_start_date, US rides with GPS state classification). States with ★ show ≥5% deviation from baseline.
State Distribution — Top 25 States by Volume + % Extremes
(May 15 vs 6-date pre-event baseline average. Top 3 gainers / decliners by ride % highlighted (≥500 rides threshold).)
Notable State Patterns
Wisconsin (+12.9% rides, +29.0% hours, +23.0% km) and Georgia (+11.4% rides) lead all high-volume states — Milwaukee and Atlanta both host well-organized Bike to Work Day events. Minnesota (+9.8% rides, +21.3% hours, +27.9% km) rounds out the top-3 gainers, consistent with strong Twin Cities cycling advocacy.
Rhode Island (−9.4% rides, −6.6% hours) is the sharpest decliner, with both ride count and duration falling simultaneously. Massachusetts (−6.8% rides) shows a curious split: fewer rides but +5.1% hours and +5.4% km, suggesting longer-trip vehicles remained on road while short-trip commuters shifted modes.
New York, Texas, and California — the three largest markets — cluster close to zero (+1.6%, +0.5%, +3.1%), showing no dramatic event effect in the largest metro fleets.
Oregon (+5.3% rides, +28.1% hours) is a notable outlier: rides rose modestly but hours surged, possibly reflecting longer-duration fleet operations in Portland on a high-cycling-culture day.
External Research Comparison
Nordback et al. (2014) — The Academic Benchmark
The most rigorous peer-reviewed study of Bike to Work Day's effect on motor vehicle volumes is Nordback et al. (2014), conducted in Boulder, Colorado. Their finding: motor vehicle volume was lower on Bike to Work Day in 88% of corridor-day observations vs. comparable weekdays (statistically significant, p<0.05). This is the only study of its kind.
Reconciling Nordback with Our National Null Result
Our national result (no significant fleet-wide effect) is consistent with, not contradictory to, the Nordback finding. Four reasons:
- Fleet composition: Nexar's fleet is 85% rideshare/commercial — not personal commuters. Nordback studied personal vehicle corridors. These are different populations.
- Geography: Boulder, CO has approximately 10.4% bike commute rate — 19 times the US national average of 0.54%. Effects that are significant in Boulder would be diluted to undetectable at national scale.
- Statistical power: This study is underpowered to detect sub-1% national effects. The minimum detectable effect is ±6%; real Bike to Work Day effects in a commercial fleet are expected to be well under 1% nationally.
- Post-2020 work patterns: Hybrid work has raised baseline variance in commuter volumes substantially since Nordback's 2014 study, reducing effective statistical power for this type of analysis.
Other Research Benchmarks
- PeopleForBikes 2024: 112M Americans Rode Bikes — 10-year high in cycling participation. 67% of micromobility commuters report they would otherwise have driven — suggesting meaningful modal shift potential in cyclist-positive demographics.
- MTC Bay Area 2025: 159,204 Miles Logged — 33% above the 120K-mile goal. 16,000 Energizer Station visitors. Demonstrates that participation events drive real behavior change in high-cycling metros.
- NHTSA 2023 — Safety in Numbers: 1,166 cyclist fatalities nationally. Academic literature documents a "Safety in Numbers" effect: more cyclists = drivers more alert. Consistent with our mild (NS) safety metric improvement.
- ACS 2022 — 0.54% Habitual Bike Commute Rate: 731,272 habitual bicycle commuters nationally. High-cycling cities: Davis CA (16.6%), Boulder CO (10.4%), Portland (6.5%), DC & SF (~4%). Effect is highly geographically concentrated.
Key Findings
Primary Finding · Baseline / No Change
No Fleet-Wide Impact Detected
Ride volume: +0.25% vs. full baseline (Z=+0.07, NS) or −2.4% vs. pre-event baseline (Z=−1.50, NS). Expected result given the fleet is 85% commercial rideshare. The null result is informative: rideshare demand is not displaced by cycling events at the national scale.
Secondary Finding · Intraday Pattern
7 AM Commute Hour: +3.78% (Local Time)
The largest single-hour deviation in the dataset, measured in each ride's local timezone. Morning hours (6 AM–6 PM local) are consistently above baseline; overnight hours below. This intraday shift pattern is consistent with event-related activity concentration in commute hours.
Statistical Finding · Pre-Event Baseline Only
High G Impact: +17.3% (Z=+2.10, p<0.05)
Only metric reaching significance with the 6-date pre-event baseline. Likely reflects April vs. seasonal structural difference in High G rates rather than a true event signal. Treat as hypothesis-generating only.
Safety Finding · Directional / Not Significant
Hard Brake & Cornering: Essentially Flat
With 6-date pre-event baseline: hard brake −0.6%, cornering −0.8%, both near zero and not statistically significant. The directional trend toward slightly safer driving is not confirmed.
Exploratory Finding · Low Confidence
Cyclist Detections: +55% (34 vs 22)*
Directionally consistent with more cyclists on the road. However, 95% CI intervals overlap, pipeline was in early rollout, and pedestrian detections moved in the opposite direction. This is hypothesis-generating, not a confirmed finding.
Geographic Signal · Monthly Proxy
DC −11.4%, Pennsylvania −10.2%
Washington DC and Pennsylvania show the largest month-over-month improvement in May vs. April/June average — both major Bike to Work Day markets with strong cycling culture. Monthly aggregate only; not event-day specific. Virginia, Massachusetts, and Connecticut show similar −3% to −4% patterns.
Limitations
About This Section
Limitations are presented as a first-class section — not a footnote. These are not excuses for null results; they define the scope within which any finding should be interpreted. A result that doesn't acknowledge limitations is not a result — it's advocacy.
- Fleet Composition — The Fundamental Mismatch: 85% of the Nexar fleet is professional rideshare/taxi operators. These drivers do not make personal modal-shift choices. Any Bike to Work Day effect would manifest primarily in the small personal/consumer segment, but that segment's small size severely limits statistical power. This study cannot answer "did Americans drive less" — it can only answer "did the Nexar commercial fleet operate differently."
- Underpowered Study Design: With N=11 baseline dates, minimum detectable effect (80% power) is approximately ±6%. Expected real-world Bike to Work Day effects are sub-1%. Even if the true effect exists, this study cannot detect it. Conclusions are limited to "no large effect" — not "no effect."
- No City-Level or Corridor-Level Analysis: Database scan cost constraints prevented daily city/state disaggregation. Bike to Work Day effects are geographically concentrated (Boulder CO, DC, Portland, SF). A national aggregate study cannot detect effects that are real at the city level. This is the most material methodological gap.
- MLP Pipeline Ramp-Up Confound: Cyclist and pedestrian detections cannot be compared to April–May 8 dates because the MLP pipeline was not yet deployed at scale. The only valid comparison is May 15 vs. post-rollout dates (May 29 onward). With n=2–3 post-rollout observations, no baseline variance estimate is possible.
- Seasonal Volume Decline: Fleet ride volumes show a clear seasonal decline (~373K in April → ~332K in late June). The pre-event 6-date baseline (Apr 3 – May 8) is the most proximate comparator window but introduces asymmetry (no post-event symmetric window), placing all comparators in the higher-volume April–early May period.
- Thursday vs. Friday Mismatch: Bike to Work Day falls on a Friday. Available Thursday comparators are used throughout this analysis. Friday and Thursday driving profiles differ in rideshare demand, commute timing, and professional driver availability — introducing a systematic day-of-week confound that cannot be fully corrected with a Thursday baseline.
- Vehicle Classification Artifact ("Unknown" Category): The Unknown vehicle type grows monotonically from April through June due to MLP classification pipeline rollout. This is a data artifact, not a behavioral signal. Any analysis including Unknown in totals will be confounded. Vehicle type analysis uses the pre-event 6-date baseline to minimize this confound.
- No Weather or External Event Controls: Day-to-day ride volume variation is also driven by weather, local events, holidays, and economic shocks. No weather controls were applied in this analysis. If May 15 experienced unusually good or bad weather in major metro areas, this would confound the comparison. May 15 weather data was not incorporated into the study design.
Methodology Appendix
Incident Rate Calculation
Incident rates are computed as:
Rate = (Total incident events on day D) / (Total rides on day D) × 1,000,000
This normalization controls for day-to-day variation in total fleet size and operational hours. The dual-pipeline assumption means incident events from both the MT_MOD_RISK pipeline (hard brake, cornering, harsh acceleration, high-G impact) are sourced from camera telemetry — not self-reported data.
Vehicle Classification Approach
Nexar vehicles are classified into Consumer, Taxi/Rideshare, Bus, Truck, and Unknown categories at time of device registration and updated via the MLP pipeline as classification models improve. The Unknown category reflects vehicles not yet classified by the MLP system and grows during rollout periods — it is not a stable vehicle type.
Poisson Confidence Intervals for Detection Counts
For low-count Poisson observations (cyclist detections: n=22–34), 95% confidence intervals are computed using the exact Poisson method:
CI = [χ²(2k, α/2)/2, χ²(2k+2, 1−α/2)/2] where k = observed count, α = 0.05
Z-Score Computation
Z-scores for day-level metrics use:
z = (x_event − μ_baseline) / σ_baseline
where μ and σ are computed from the 6-date pre-event baseline window (Apr 3, Apr 10, Apr 17, Apr 24, May 1, May 8). No multiple-comparison correction was applied (Bonferroni or similar) — applying such correction would make all results even more decidedly non-significant.
Want to dive deeper?
Nexar helps industry leaders turn vision into value. Dive into real-world success stories where our data redefined the road ahead — and see what it can do for you.
