Bus stop balancing: Stop the Stop for buses
Data: MTA bus GTFS (September 2026 schedule), the MTA’s bus speeds between timepoint stops (weekdays, March–August 2026), bus ridership by route, route speeds. Run 2026-10-05.
Many NYC local bus stops sit one or two blocks apart. Each stop costs every rider on board a few seconds of braking, door time and pulling out; closing it costs the people who used it a longer walk. Stop the Stop weighed those two for subway stations. This does it for local buses, route by route, under three spacing rules.
The MTA already does this. Its stated goal for local routes is average spacing of about 1,000–1,320 ft (a quarter mile); in Queens it planned to go from about 950 ft to 1,300 ft, and in the Bronx from 882 ft to 1,092 ft. It put the saving at about 20 s per stop removed. Two corrections to the brief: the Bronx redesign took effect in June 2022, not 2025; the Queens one took effect on 29 June and 31 August 2025. So the check below uses both.
Answer
Section titled “Answer”- A stop costs a local bus about 31 s (95% CI 24–37), measured on 227 routes. That sits at the top of the 20–30 s range in the literature and the MTA’s own 20 s.
- Under a 1,000 ft rule, 5,089 of 17,333 route-direction stops (29%) go. The average local trip is 4.9 min faster (9% of scheduled run time) at 31 s. Riders on board save 30,400 hours per weekday and the walkers lose 13,100 hours, 26,300 hours with walking counted double. Net: +4,100 rider-hours per weekday, about 12 s per boarding.
- That net is small and fragile. It flips sign if the stop costs 20 s instead of 31, or if a rule goes to 1,320 ft. The 750 ft rule holds up best: it nets positive at 2x walking for any stop cost above about 20 s.
- The real redesigns gained far less than the model predicts. Queens routes got 1.4% faster against other boroughs; the model says 7–12%. Read the model’s savings as a ceiling.
- If removed stops are the quiet ones, the case is clear. Boardings per stop aren’t public, so the model removes stops as busy as the average. If the removed stops carry half the average boardings, the 1,000 ft net is +17,300 hours (+6,500 even at 20 s).
Method
Section titled “Method”Stops and spacing. For each route and direction, the most common weekday stop pattern on 16 September 2026 (a Wednesday, school in session). Each stop is projected onto the route’s shape, never moving backwards; spacing is the distance along the shape between neighbours. 345 routes, 21,635 stops; 77 express routes left out, since they skip most local stops. The 268 local, limited and SBS routes carry 1.18 million weekday boardings.
Check on spacing (mean gap, ft): Queens 1,327 (MTA target 1,300), Bronx 1,054 (MTA plan 1,092), Staten Island 1,011, Manhattan 847, Brooklyn 826. The two redesigned boroughs match the MTA’s stated figures; the two with no redesign are the tightest.
Rules. For a minimum spacing of 750, 1,000 or 1,320 ft: take the closest pair of kept stops under the minimum and remove the one that leaves the more even spacing. Repeat until none is left. Never removed: the first and last stop, any stop within 100 m of a subway station (8.8% of stops), any stop where two or more routes cross (13%: others that stop within 60 m and run at 40° or more to this route; routes sharing the street don’t count). 23% of stops are protected. No removal may leave a gap over twice the minimum.
Stop cost. Each timepoint segment’s average time (7 am–7 pm weekdays, weighted by buses) is regressed on the number of stops inside it and its length, with a fixed effect for each route and direction. The stop count comes from the GTFS pattern; a segment is kept only if GTFS and MTA distances agree within 10%. Local routes only. The interval resamples whole routes.
| Hours | Segments | Routes | Seconds per stop | 95% CI | Implied non-stop speed |
|---|---|---|---|---|---|
| 7 am–7 pm | 2,408 | 227 | 30.8 | 23.7–37.4 | 13.5 mph |
| Midday 10 am–3 pm | 2,454 | 226 | 32.1 | 24.0–38.6 | 14.1 mph |
| Peaks 7–10 am, 4–7 pm | 2,407 | 227 | 29.0 | 22.3–35.4 | 13.2 mph |
Without route fixed effects the figure is 35 s. We use 31 s as the central case and 20 s (the MTA’s) as the cautious one.
Riders. Boardings per weekday = March–August 2026 route total ÷ 128 weekdays (federal holidays dropped). Spread evenly over a route’s stops, split between directions by trip count. Riders on board at any point = boardings × trip length ÷ route length, with trip length at 2 miles, capped at the route’s length. We took 2 miles as a round figure for a NYC local bus ride; it is an assumption, not a measured number, and net benefit scales with it directly.
Walking. Users of a removed stop (boardings plus alightings) walk to the nearest kept stop at 3 mph. Two ways to count the extra distance:
- Spread (central): users are spread evenly along the street. For a stop between two kept neighbours the average extra walk is about half the gap.
- Stop (upper bound): everyone stands at the stop, half walk to each neighbour, so the extra walk is the full gap. Your brief gave this rule; it assumes the worst, because only people right at the stop walk the whole gap.
Walking counts double in the net (the planning convention); 1x and 3x are shown too.
Results
Section titled “Results”Stop cost 31 s, walking counted double, spread walking. Per weekday.
| Rule | Stops removed | Spacing, mean gap before → after | Min saved per trip | Rider-hours saved | Walking (2x), hours | Net, hours | Extra walk per user |
|---|---|---|---|---|---|---|---|
| 750 ft | 2,665 (15%) | 1,303 → 1,449 | 2.6 (4.8%) | 15,800 | 10,300 | +5,500 | 1.1 min |
| 1,000 ft | 5,089 (29%) | 1,303 → 1,672 | 4.9 (9.4%) | 30,400 | 26,300 | +4,100 | 1.4 min |
| 1,320 ft | 7,346 (42%) | 1,303 → 2,008 | 7.1 (13.8%) | 44,100 | 51,400 | −7,300 | 1.8 min |
(Mean gaps include a few very long gaps, so they rise more than a typical gap.) The buses also save 2,200 / 4,300 / 6,200 bus-hours per weekday, which the MTA could spend on more service.
How sensitive is the net? Net rider-hours per weekday:
| Rule | 31 s, 1x walk | 31 s, 2x | 20 s, 1x | 20 s, 2x | 31 s, 2x, stop walking |
|---|---|---|---|---|---|
| 750 ft | +10,700 | +5,500 | +5,000 | −100 | −6,200 |
| 1,000 ft | +17,300 | +4,100 | +6,500 | −6,700 | −25,400 |
| 1,320 ft | +18,400 | −7,300 | +2,700 | −23,000 | −64,300 |
At 1,000 ft the net turns positive at 2x walking once a stop costs about 27 s (spread walking).
Routes
Section titled “Routes”By route, at 1,000 ft, the winners are the tight-spaced routes with plenty of riders. 133 of 248 routes that lose a stop net positive at 31 s and 2x; at 20 s, 8 do.
| Route | Borough | Stops cut / total (both ways) | Spacing before → after (ft) | Min saved per trip | Weekday boardings | Net hours (31 s, 2x) |
|---|---|---|---|---|---|---|
| M4 | Manhattan | 72 / 152 | 691 → 1,334 | 18.5 | 11,000 | +229 |
| B1 | Brooklyn | 44 / 98 | 743 → 1,372 | 11.4 | 10,800 | +189 |
| B3 | Brooklyn | 43 / 87 | 662 → 1,337 | 11.1 | 7,050 | +174 |
| M11 | Manhattan | 55 / 116 | 699 → 1,349 | 14.2 | 8,290 | +170 |
| M1 | Manhattan | 60 / 133 | 684 → 1,265 | 15.5 | 7,950 | +167 |
| B8 | Brooklyn | 71 / 151 | 709 → 1,358 | 18.4 | 9,040 | +164 |
| M15 | Manhattan | 52 / 121 | 764 → 1,356 | 13.4 | 13,400 | +162 |
| M3 | Manhattan | 73 / 162 | 704 → 1,296 | 18.8 | 8,520 | +150 |
| B36 | Brooklyn | 37 / 81 | 706 → 1,327 | 9.6 | 6,880 | +146 |
| B44 | Brooklyn | 56 / 128 | 726 → 1,304 | 14.4 | 7,030 | +125 |
| M7 | Manhattan | 49 / 117 | 712 → 1,244 | 12.7 | 7,310 | +124 |
| M102 | Manhattan | 46 / 113 | 734 → 1,253 | 11.9 | 7,540 | +121 |
| M31 | Manhattan | 22 / 57 | 742 → 1,236 | 5.7 | 8,080 | +118 |
| B63 | Brooklyn | 48 / 109 | 744 → 1,351 | 12.4 | 6,930 | +114 |
| B60 | Brooklyn | 70 / 133 | 638 → 1,372 | 18.1 | 3,840 | +109 |
Minutes saved per trip are the route’s average over both directions, at 31 s a stop. Taken at face value, 11–19 min per trip is far too much; see the check below. Routes that lose out: Q32, Q25, Q60, M101 and Q63, routes whose stops are already 1,200–1,600 ft apart, so the few cuts save little time but still send walkers further.
By borough (1,000 ft, 31 s, 2x): Brooklyn +2,900 hours, Manhattan +2,200, Staten Island +70, Bronx −30, Queens −1,000. Queens and the Bronx have already had their redesigns, so the stops left are further apart and harder to justify cutting.
Check against the real redesigns
Section titled “Check against the real redesigns”We don’t have the pre-redesign stop lists, so this tests size, not route by route. Observed = weekday speed of the borough’s non-express routes in January–May of the year after against the year before, minus the same change on other boroughs’ routes. Predicted = the MTA’s stated change in spacing × stop cost, against the borough’s old seconds per mile.
| Bronx (June 2022) | Queens (Jun–Aug 2025) | |
|---|---|---|
| Spacing, MTA’s old → new (ft) | 882 → 1,092 | 950 → 1,300 |
| Stops per mile removed | 1.15 | 1.50 |
| Observed speed gain over controls | +2.5% | +1.4% |
| Predicted at 31 s | +7.4% | +12.4% |
| Predicted at 20 s | +4.7% | +7.6% |
| Stop cost that would match the observed gain | 11 s | 4 s |
The MTA reported Bronx redesign routes beating the systemwide trend by 4%; we get 2.5%. The model’s order of magnitude is right, but it overshoots by 2–9x. So treat 31 s as the cost of an average stop and the saving from the stops actually removed as smaller.
Caveats
Section titled “Caveats”- The model overpredicts, per the check above. The likeliest cause: the stops a redesign removes are the quiet ones, where buses often don’t stop at all. Our regression counts every stop as a stop. A second cause: the regression picks up other things that vary with stop density (signals, busy corridors). Gains from the 20 s column, or less, are the safer claim.
- Equal boardings at every stop. No stop-level data. Real boardings are lopsided toward transfers, schools and hospitals, which this model protects only through the subway and cross-street rules. If removed stops see half the average, the 1,000 ft net is +17,300 hours at 31 s.
- Riders on board rests on a 2-mile average trip we did not measure; riders are not matched to where they board and alight. The model counts every on-board rider at every removed stop; people who board or get off at the stop are counted in walking, not riding.
- Walking is distance along the street, not a street map, and ignores crossings, hills, weather, and riders with disabilities or who carry shopping. Those are the people the MTA tries to protect when it keeps stops. Walking time doesn’t capture a longer wait at a stop in the dark or the cold either.
- Stop cost comes from timepoint segments of 1–2 miles that include signals and traffic; stops in a segment are all counted even if the bus skips some. The segment data is March–August 2026; the stop lists are September 2026.
- Redesign check compares boroughs, not routes, and uses the MTA’s stated spacing rather than the stop lists. The Bronx comparison sits in the pandemic recovery; both include route changes beyond stop removal. Treat it as a rough check.
- Frequency isn’t modelled. The bus-hours saved could run more buses and shorten waits. We leave that out, and any riders drawn by faster service.
- Stop patterns: only each direction’s most common weekday pattern; short-turns and variants are ignored. Shared stops on parallel routes are treated one route at a time.