How accurate are bus countdowns?
Data: our own archive of the MTA’s bus predictions (bus_predictions, the feed behind countdown clocks and apps) and bus positions (bus_vehicle_positions). First run: Sunday 2026-10-04, midnight to 2 pm only. To be redone on full weekdays once they’re compacted.
Method
Section titled “Method”- Predictions: every predicted arrival of a trip at a stop, stored whenever it moved 30 s or more. Each counts for as long as it stood, until the next prediction, until the stop dropped out of the feed, or until the bus arrived. So results are weighted like a rider glancing at a sign at a random moment (“glance-hours”).
- Actual arrival: when the bus passed the stop’s position along its route, from its GPS reports snapped to the route shape, interpolated between the reports either side (only when they’re under 3 minutes apart). First stops are left out (a waiting bus “passes” its first stop before it leaves).
- Error = actual minus predicted. Positive: the bus came later than the sign said.
Check on the arrival times. The MTA drops a stop from a trip’s predictions once the bus has passed it. The last time each stop was dropped came a median 86 s after our measured arrival, and 79% fell between 30 s before and 2 minutes after, consistent with the feed’s refresh. (Stops also flicker out and back in: half of trip-stops drop out at least once before the bus arrives. Those gaps count as no prediction shown.)
Results (Sunday morning, 17,381 trips on 268 routes)
Section titled “Results (Sunday morning, 17,381 trips on 268 routes)”| Bus predicted in | Median error | Within 1 min | Bus 1+ min early | Bus 2+ min late | Bus 5+ min late |
|---|---|---|---|---|---|
| 0–2 min | 0.4 min | 83% | 1% | 7% | 3% |
| 2–5 min | 0.6 min | 74% | 8% | 7% | 2% |
| 5–10 min | 0.9 min | 56% | 14% | 14% | 3% |
| 10–20 min | 1.5 min | 36% | 18% | 30% | 8% |
| 20–30 min | 2.3 min | 24% | 19% | 44% | 18% |
- Short countdowns are good; anything over 10 minutes is a rough guess. And errors aren’t one-sided: at 10–20 minutes the bus comes 1+ minute early 18% of the time, and an early bus is the one you miss.
- By route (predictions 5–20 min ahead, within 2 min): worst are Manhattan routes, the M55 (38%), M1 (42%), M11 (43%), M31, M57, M103, plus the B31, B32 and B45; on the M55, 35% of the time the bus comes 5+ minutes later than promised. Best are Queens limited and express-style routes (Q51, Q76, Q80, Q82, Q86: 84–88%). Long, congested Manhattan routes are hard to predict, which is also where riders most need the countdown.
Caveats
Section titled “Caveats”- One Sunday morning. Weekday traffic will make predictions worse; rerun on a full week before quoting numbers.
- Predictions are only stored when they move 30 s or more, so the countdown on a sign can differ from ours by up to 30 s.
- Bus Time apps can show distance (“2 stops away”) instead of minutes; this audits the minutes.