Did the bus lanes NYC added in 2023–2025 make buses faster?
Data: the MTA’s bus speeds between timepoint stops, by month (2023–24, 2025 on), weekdays 6 am–8 pm, January 2023 to August 2026; NYC DOT’s bus lanes with opening dates; MTA bus GTFS for route shapes. Run 2026-10-05.
Short answer
Section titled “Short answer”Yes, by about 3%. Buses ran 3.3% faster (95% interval 1.9 to 4.7) on stretches that got a new lane, compared with similar stretches that never had one, over the same months. At the treated stretches’ pre-lane speed of 6.6 mph, that saves about 18 seconds per bus-mile (interval 9 to 27).
The gain shows up in the first month and holds for a year. Speeds on the treated stretches were not diverging from the controls beforehand. But the estimate rests on 53 stretches on 23 routes, and gains differ a lot by street, from 1.4% (East Gun Hill Rd) to 5.8% (3 Av, Dec 2023 opening) to 8% (East 96 St). Treat 3% as the typical case, and 2–5% as the range to plan around.
| Speed change | 95% interval | Seconds saved per bus-mile | Interval | |
|---|---|---|---|---|
| Bus-mile weighted (headline) | +3.3% | +1.9 to +4.7 | 17.6 | 9 to 27 |
| Each corridor counts once | +3.9% | +2.5 to +5.0 | 23.0 | 14 to 30 |
| Each stretch counts once | +1.5% | −1.8 to +3.9 | 8.6 | −11 to 22 |
| Raw: treated, 12 months after vs 6 before | +2.6% | |||
| Raw: matched controls, same months | −0.7% |
The weighting matters. Counting stretches equally gives more say to low-frequency routes (the express buses on 3 Av, BX28 and BX38 on Gun Hill Rd), which gained less, and widens the interval. Weighting by bus-miles tracks what buses and riders experience, so it is the headline.
Method
Section titled “Method”Which stretches. A stretch is one route and direction between two consecutive timepoint stops. Its path comes from the route’s GTFS shape (as in bus lanes). DOT lists 724 new lane segments, about 30 miles, opened August 2023 to December 2025 (the file has none dated 2026). For each stretch we measure the share of its path within 15 m of those lanes, by opening date.
- Treated: half or more of the path on new lanes, and under a quarter on older all-day lanes. The opening month is the first date the new lanes cover half the path. 137 stretches qualify.
- Control: under 10% of the path near any lane, old or new. 3,384 stretches.
- Everything else (older lanes, mixed) is left out.
- A stretch needs 100+ buses a month and at least 3 months of speeds in the 6 months before and in the 12 after. That leaves 53 treated stretches (23 routes, 8 corridors, 51 route-miles) and 2,570 controls (312 routes). Most of the rest lost their data when the stop pair changed: the Queens bus redesign (July 2025) replaced most Hillside Av pairs, and Q66 pairs on Northern Blvd changed after March 2024.
The comparison. For each opening month, each stretch’s log speed is measured against its own average in the 6 months before (the opening month is dropped). We then take treated minus control over the same calendar months, for 12 months after. Controls are matched within each opening month on region (borough, with Manhattan split at 60 St for the congestion zone), express or local, and third of baseline speed. Within each match group the controls are weighted to the treated stretches’ bus-miles. Intervals come from resampling routes 1,000 times, redoing the matching each time; that allows for buses on one route moving together.
Did treated stretches already differ before?
Section titled “Did treated stretches already differ before?”No. Monthly gap, treated minus control, by months from opening (average of monthly estimates; each month is measured against the 6 months before opening, so the last two rows of the pre-period average to zero by construction):
| Months from opening | −12 to −10 | −9 to −7 | −6 to −4 | −3 to −1 | 1 to 3 | 4 to 6 | 7 to 9 | 10 to 12 |
|---|---|---|---|---|---|---|---|---|
| Gap | +0.6% | 0.0% | −0.4% | +0.4% | +3.6% | +2.8% | +4.4% | +3.1% |
- Trend over months −12 to −1: −0.04 points a month (interval −0.30 to +0.28).
- Months −12 to −7 against the reference: +0.25% (−1.2 to +1.6).
- Single pre-months bounce between −1.1% and +2.2% (month −3 is high); there is no trend.
- After opening: months 1–6 +3.2% (+1.8 to +4.5); months 7–12 +3.5% (+1.7 to +5.2). The gain does not fade within a year.
- Bus service on treated stretches did not fall (buses a weekday, after over before: 1.004; controls 0.988).
By group
Section titled “By group”| Stretches | Routes | Speed change | 95% interval | Seconds saved per bus-mile | |
|---|---|---|---|---|---|
| Lane type | |||||
| Offset | 52 | 22 | +3.5% | +2.1 to +4.8 | 18.4 |
| Curbside, center running, busway | too few | ||||
| Lane before? | |||||
| No lane before | 49 | 22 | +3.2% | +1.6 to +4.7 | 17.3 |
| Peak-hour or shared lane before (Hillside Av) | 4 | 2 | +4.3% | +2.7 to +6.5 | 18.2 |
| Borough | |||||
| Queens | 12 | 4 | +4.7% | +3.1 to +8.1 | 21.4 |
| Manhattan | 26 | 13 | +3.7% | −0.1 to +5.8 | 22.1 |
| Bronx | 13 | 4 | +1.3% | −0.1 to +2.1 | 6.6 |
| Brooklyn | 2 | 2 | too few | ||
| Baseline speed | |||||
| Slowest third (4.7 mph) | 21 | 10 | +3.5% | +0.7 to +5.1 | 26.3 |
| Middle third (6.9 mph) | 16 | 9 | +3.6% | +1.8 to +5.5 | 18.4 |
| Fastest third (9.4 mph) | 10 | 7 | +2.4% | −4.4 to +6.3 | 9.1 |
| Express routes | 6 | 6 | +3.3% | +1.8 to +5.2 | 11.5 |
| Share of stretch on the lane | |||||
| 90–100% | 37 | 20 | +3.6% | +1.9 to +5.1 | 17.6 |
| 75–90% | 7 | 5 | +1.2% | −1.0 to +3.5 | 6.4 |
| 50–75% | 9 | 6 | +4.5% | +1.2 to +7.6 | 28.5 |
- Nearly every new lane is an offset lane, so offset against curbside against busway can’t be told apart here. No busway opened in this window. All 53 stretches have 24-hour lanes.
- Slower streets save more seconds for the same percentage gain. The Bronx gains (Gun Hill Rd, University Av) are small on both counts.
- Stretches only partly on the lane gain no less, so the share of the stretch on the lane doesn’t show a clear dose effect. On average 90% of a treated stretch sits on the lane, so seconds per bus-mile of lane run about 10% higher than the figures above.
Corridors
Section titled “Corridors”Only corridors with 3+ usable stretches are listed. The other new lanes (Pelham Pkwy, Allen St/Pike St, Flatbush Av) enter the totals.
| Corridor (opened) | Route-miles | Routes | Speed change | 95% interval | Seconds saved per bus-mile | Bus-hours saved per weekday |
|---|---|---|---|---|---|---|
| Hillside Av, Queens (Sep 2025) | 10.2 | 3 | +5.7% | +4.3 to +9.4 | 23 | 4.3 |
| East 96 St, Manhattan (Nov 2024) | 1.2 | 1 | +8.3% | +7.0 to +9.3 | 74 | 3.8 |
| 3 Av, Manhattan (Dec 2023) | 6.8 | 4 | +5.8% | +3.3 to +8.2 | 36 | 3.7 |
| Northern Blvd, Queens (Aug 2023) | 3.9 | 1 | +3.0% | +2.4 to +3.7 | 17 | 1.8 |
| West 96 St, Manhattan (Nov 2024) | 1.8 | 2 | +3.3% | +1.9 to +6.2 | 26 | 1.3 |
| 3 Av, Manhattan (Aug 2025) | 14.6 | 9 | +1.8% | −3.4 to +5.4 | 8 | 1.2 |
| East Gun Hill Rd, Bronx (Oct 2023) | 7.6 | 2 | +1.4% | +0.8 to +2.0 | 7 | 0.9 |
| University Av, Bronx (Dec 2023) | 2.5 | 1 | +1.9% | +1.4 to +2.5 | 10 | 0.8 |
Corridors with one or two routes have intervals that are too narrow: the resampling can’t draw on enough routes. Hillside Av is the biggest opening (about 8 miles) but the least clean: the lane opened with the Queens bus redesign, and only 8 of its stretches kept their stop pairs. Taking out any one corridor leaves the answer between 2.7% (without Hillside Av) and 3.7% (without East Gun Hill Rd).
In human units
Section titled “In human units”- Per bus-mile: 18 seconds (9 to 27). A 5-mile ride entirely in lanes saves about 1.5 minutes; a 1-mile lane saves 18 seconds per bus. The rule behind it: the lane cuts about 3.2% of the time a bus now takes, which is 0.032 × 3,600 ÷ mph seconds per mile.
- Bus-hours: the 53 stretches in the study save 16 bus-hours per weekday (6 am–8 pm). All 89 treated stretches with buses in the latest six months carry about 755 bus-hours a weekday; the same gain on them is 25 bus-hours a weekday. That is bus time, not rider time; multiply by riders per bus for the latter.
Using it for proposed lanes
Section titled “Using it for proposed lanes”If a lane on these streets gained the typical 3.3% (1.9 to 4.7), the saving per bus-mile at today’s speeds (from bus lanes) would be:
| Street | Speed now | Seconds saved per bus-mile | Interval |
|---|---|---|---|
| Madison Av, Manhattan | 5.7 mph | 20 | 12 to 28 |
| Broadway, Manhattan | 5.5 mph | 21 | 12 to 29 |
| Queens Blvd | 6.0 mph | 19 | 11 to 27 |
| Flatbush Av, Brooklyn | 6.6 mph | 18 | 10 to 24 |
These assume the new lane works like the ones in the study. They are not forecasts for each street. See the caveats.
Caveats
Section titled “Caveats”- Few treated stretches. 53 stretches on 23 routes, in 8 corridors; 84 more were flagged but lacked a before or after. The interval allows for route-to-route noise but not for how those corridors were picked. Adding stretches with only 2 reference months (82 stretches, mostly Hillside Av) gives +3.1% (+2.2 to +4.0).
- A lane comes with other things. Opening a lane usually brings camera enforcement, signal changes and curb rules. This measures the package, not paint alone.
- Lane dates are DOT’s. 195 of 724 segments share the date 2025-09-15, which looks like a batch date. The opening month is dropped, but a lane marked weeks late or early blurs months 1–2. The estimates for months 1–3 (+3.6%) match later months, so this does not drive the result.
- Controls may be touched. Buses on treated routes run on untreated stretches too, and could gain (or lose) from fewer delays. Using only controls from routes with no treated stretch gives +3.2% (+1.6 to +4.9).
- Congestion pricing started in January 2025, inside the windows of the late 2024 openings and the 2025 ones. Controls are matched within the zone, and dropping Manhattan below 60 St gives +3.5% (+2.1 to +4.8).
- Stretches are long. A timepoint stretch averages a mile and some run on several streets. A lane covering 90% of a stretch is counted as treated; the gain on the lane itself may be a little larger.
- Lanes are on centrelines. The DOT file doesn’t say which direction a lane serves, so a stretch counts as treated if it runs along the lane in either direction.
- No opening dated 2026 is in the DOT file (updated April 2026), so “2023–2026” here means openings from August 2023 to December 2025, followed to August 2026.
- The MTA changed datasets in January 2025. The January step for control stretches (+3.8%) sits between those of other years (+1.6% in 2024, −1.4% in 2026), and every comparison uses the same calendar months, so it should cancel.
- Peak-only lanes, busways, curbside lanes, and lanes in Brooklyn and Staten Island are not covered. Do not read the result across to them.