Does CBTC make each stop slower?
Data: the full subway feed archive, April 2021 to October 2026. Cutover dates: CBTC cutovers, 2021–2026. Run 2026-10-02.
Why we asked
Section titled “Why we asked”In What a subway stop costs a stop cost about 61 s on the L (CBTC) but 30 s on the J (old signals). One guess: CBTC’s cautious braking adds time at every station. The MTA is spending billions to put CBTC on about 75 more miles, so if that were true it would matter.
Answer: no
Section titled “Answer: no”We find no sign that CBTC makes stops slower. Where it did change anything, it sped up running between stations. The L-vs-J gap must come from something else, such as crowding, platform layout or line geometry.
| Test | Effect per segment, vs. control | Change in stop cost | Change per km moving |
|---|---|---|---|
| Queens Blvd West, automatic operation (Mar 2024) | −8.5 s (−5.1%), CI −13.5 to −4.2 | +2.6 s, CI −2.8 to +7.3 | −9.1 s/km, CI −13.2 to −3.8 |
| Queens Blvd West, CBTC (Apr 2021–Feb 2022) | +3.9 s (+1.3%), CI −0.5 to +8.9 | −0.5 s, CI −6.8 to +5.9 | +3.2 s/km, CI −1.3 to +8.7 |
| Culver CBTC (Feb–Sep 2024) | −4.2 s (−3.5%), CI −9.6 to +1.2 | can’t tell (CI ±30 s) | can’t tell |
| Placebo: Culver a year before CBTC | +2.1 s, CI −0.1 to +4.5 | −0.5 s | +3.7 s/km |
How to read it:
- Automatic operation on Queens Blvd made trains about 5% faster between stations, after removing the citywide trend. That’s about half the MTA’s “up to 10%”. Stops didn’t change.
- Switching Queens Blvd to CBTC in 2021 changed nothing measurable on its own. The gain came two years later, when computers started driving the trains.
- Culver looks about 3.5% faster, but the CI includes zero. Its stations are all similar distances apart (500–950 m), so we can’t separate stop time from running time there.
- The placebo is near zero, which suggests the method isn’t inventing effects.
Method
Section titled “Method”- Segments. For each station-to-station segment, take the trimmed mean (10th–90th percentile) departure-to-departure time on weekday rush hours (7–10 am, 5–7 pm), excluding holidays.
- Windows. Compare October–November before and after, a year apart, so season stays the same. For the 2021 test it’s April–May 2021 vs. April–May 2022.
- Difference-in-differences. Treated = both ends in the cutover section. Control = segments on all lines that had no signal work then. That excludes the L, the 7, Queens Blvd East and 8 Av, which was mid-install. Effect = mean change on treated minus mean change on control. CIs come from bootstrapping over segments.
- Splitting the effect. Regress each treated segment’s net change on its length:
change = stop part + per-km part × km. The intercept is the change in what a stop costs.
Caveats
Section titled “Caveats”- 2021 baseline isn’t clean. Some Queens Blvd sections went live before April 2021, and ridership was still recovering from the pandemic.
- Small samples: 22 Culver segments and 40 Queens Blvd segments.
- One season per test. Rerunning with other months would firm this up.
- Rush hour only. Midday or night effects could differ.
- Fleet changes (R160 refits, R179s on CBTC in 2025) fall mostly outside these windows, but not entirely.