The ambulance is slower: where the extra time goes

Closely. The dispatch file, summed the way the Mayor’s Management Report counts (final segment 1 to 3, timed from the call’s creation, valid times only), gives 7:29, 8:01, 8:17, 8:52 and 10:37 for fiscal 2022 to 2026 against the report’s 7:26, 7:59, 8:16, 8:49 and 10:32: within one to five seconds every year, and a rise of 1:45 against the report’s 1:43. Counts match too: 610,791 life-threatening calls in fiscal 2026 against the report’s 610,771. Two years differ by about 1,700 calls in opposite directions (fiscal 2023 and 2025), which the published data do not explain. The weekly 911 file reproduces the report’s end-to-end figures exactly: 11:21 and 13:09.

 

Administrative

 

How to read this, and what it cannot show

One bar per fiscal year, in decimal minutes (10.6 minutes is 10:36). Means of the dispatch file’s monthly sums.

Embed this chart

Paste this into any web page. The chart stays interactive and keeps its margins; keep the credit line with it.

<iframe src="https://sparkydata.dev/embed/the-ambulance-is-slower/fiscal_years" title="Ambulance response by fiscal year, 2017 to 2026, in decimal minutes: switch between response, dispatch and travel." width="100%" height="820" style="border:0;max-width:760px" loading="lazy"></iframe> <p>Chart: <a href="https://sparkydata.dev/articles/the-ambulance-is-slower#chart-fiscal_years">SparkyData</a>. Data CC BY 4.0.</p>

To fiscal 1988. The city’s annual Mayor’s Management Report has printed ambulance response to life-threatening calls every year since then, and each report reprints earlier years, so the series can be chained across its changes of name. The fiscal 1999 report charts 1988 to 1999 and says “comparable measurements began in Fiscal 1988”; the fiscal 2004 report carries fiscal 2000 and 2001 unchanged under the newer name, “average response time to life-threatening medical emergencies by ambulance units”. The reports later added an end-to-end figure from the 911 call, starting with fiscal 2013, and now label this one “FDNY dispatch and travel time only”. Where a later report revised a year, it was by a second (fiscal 2003 to 2005), and the later print is used.

Before 2005 there are no public call records to check the figures against. From fiscal 2006 the dispatch file can, and it lands within five seconds of the city’s figure every year. The tools changed over those decades too: the FDNY fitted every ambulance with an automatic vehicle locator between November 2005 and the end of fiscal 2006, for example. So a difference of a few seconds across decades says little; the swings of minutes are what the series shows. The figure was 11:04 in fiscal 1988 and 10:03 in 1989, then between 8:21 and 8:58 from 1990 to 1996, the year of the merger.

Yes. The chart below shows every borough from fiscal 2017, for response, dispatch, travel or the share of calls held, and for all life-threatening calls, those urgent from the start, or those upgraded later.

Nothing here points to a citywide traffic slowdown, but these aggregate data cannot estimate traffic’s contribution. Fire engines answering life-threatening medical calls slowed by 11 seconds in fiscal 2026 (5:52 to 6:03), and at structural fires by three, against 1:43 for ambulances on the same measure. Engines are stationed and sent differently from ambulances, so they are a benchmark rather than a control. Manhattan, which contains the congestion zone, had the smallest rise in ambulance travel of the four large boroughs (40 seconds, against 94 in Queens).

A working paper from the National Bureau of Economic Research (not yet peer-reviewed) compares EMS incidents on either side of the congestion zone’s boundary. It estimates that the toll cut total EMS travel by 63 to 70 seconds, but almost all of that came on the trip from the scene to the hospital (54 to 59 seconds). Its estimate for the trip to the patient, the leg inside the travel time here, is 7 to 9 seconds, and the authors call it imprecise. The dispatch file cannot isolate the zone, so we don’t estimate any effect of the toll here; these borough-level figures give no evidence that the toll slowed ambulances.

Not for ambulances. In the report’s own table, life-threatening incidents fell 1.6% in fiscal 2026 and all ambulance incidents 0.6%, and fiscal 2024 was busier on both, at 633,361 and 1,644,446. Patient transports fell 8%. If “record call volume” refers to all 911 calls or to fire calls, those are other series. The dispatch file’s own counts fall a little more (1.8% and 0.9%), because its fiscal 2025 totals sit above the report’s, by about 1,700 life-threatening calls and about 4,500 calls in all.

All three, for different questions. 10:32 is ambulances alone, timed from the original call. 9:13 is the first unit of either kind, ambulance or fire company, so it is shorter because engines often arrive first. 7:28 is the same combined figure timed from the moment an upgraded call was upgraded rather than from its original creation. We reproduce the ambulance-only figure; the combined ones need fire-company times and upgrade times that the public dispatch file does not carry, so we quote them and don’t recompute them.

Because that is the whole wait from the moment the call was entered, and the report counts it that way. The cost is that it includes time before anyone knew the call was life-threatening, when it was queued behind calls coded as more urgent. Since upgraded calls carry 64 of the 105 seconds in the shift-share decomposition, the choice matters. The report shows how much for ambulances and fire companies together: timed from the call’s creation, that combined figure rose from 7:45 to 9:13; timed from the moment of upgrade, from 6:46 to 7:28. The public dispatch file has no upgrade times, so we cannot make the second measure for ambulances alone.

Unusual. From January 2017 to June 2025, leaving out the Covid months from March 2020 to June 2021, a month’s average response moved on the same month a year earlier by 22 seconds on average, with a standard deviation of 27 and a largest rise of 98. Six of the twelve months of fiscal 2026 rose by more than that largest, starting in December 2025, and the year’s rise of 105 seconds is two and a half times the largest annual rise in the dispatch file back to fiscal 2006 (42 seconds, in fiscal 2022). In the city’s own figures, back to fiscal 1988, the largest rise before it was 40 seconds, also in fiscal 2022. The slowest single week in the 911 file is still the week of 23 March 2020, at the height of the first Covid wave.

Dispatch runs from the creation of the EMS job to the first ambulance assigned to it. For a held call it includes the time until a unit could be assigned; for other calls it is the time to choose and assign one. Travel runs from that first assignment to the first ambulance on scene. Because it starts at the first assignment, travel includes any reassignment (a unit sent, then diverted to a more urgent call, and replaced) and the distance of whichever unit was assigned. A longer travel time can therefore mean slower streets, a unit sent from farther away, or a reassignment; the data cannot tell them apart, which is why the fire-company benchmark matters.

Why ambulance hours fell: the report names staffing and hospital withdrawals, but the dispatch file records calls, not crews. Where each ambulance started, so distance and speed. Patient outcomes. Anything below the borough. Medians for the dispatch file: it holds sums and counts, so its times are averages, and the medians come only from the 911 file’s weekly figures. And July 2026 onward in the dispatch file, which the city had not yet published at the start of October.

The dispatch file. FDNY EMS Incident Dispatch Data (NYC Open Data 76xm-jjuj), every EMS incident since January 2005, summed on the city’s server by month. The main sums start in January 2016, by borough, initial and final severity segment, the validity flags and the held flag; a second query, life-threatening calls with valid times only, goes back to January 2005 for the long view, and the two agree to within a second wherever they overlap. Life-threatening means final segment 1 to 3, the report’s definition; the initial-segment definition gives a smaller rise (57 seconds), because it leaves out upgraded calls, and calls life-threatening by both codings rose 56 seconds. Times use only incidents the file flags as having valid response times; every such incident also carries a valid dispatch time and all three sums, so the split into dispatch and travel needs no further exclusion. Fiscal year t runs from July of year t−1 to June of year t. Every time is a mean: summed seconds over incidents.

The 911 file. 911 End-to-End Data (t7p9-n9dy), weekly from November 2013, averages and medians by agency and incident type; fiscal 2015 is the first full year used. The EMS life-threatening rows give call to arrival, call to job creation, and dispatch and travel in minutes; this file’s dispatch runs from the call-taker’s handoff to EMS until an assignment, so it includes EMS call-taking and runs longer than the dispatch file’s, which starts at job creation. A week’s fiscal year is set by its midpoint. Its medians are weekly medians, so their average approximates the typical call rather than measuring it. The latest week, starting 24 August 2026, holds about 8,600 incidents against about 10,500 in recent weeks and may be incomplete, so the summer comparison stops the week before; counting it, the gap would be 2:21 rather than 2:26.

Decomposition. The fiscal 2025 to 2026 rise splits between calls life-threatening from the start and calls upgraded later by a shift-share: each group’s change in average time at its average share, plus the change in its share at its average time. Calls from the start contribute 41 seconds, upgraded calls 64; the total is the 105 seconds of the headline.

The city’s long series. Fiscal 1988 to 2026 as the Mayor’s Management Reports printed it: the fiscal 1999 report’s chart for 1988 to 1999, the fiscal 2001 and 2004 reports for 2000 and 2001, the fiscal 2006 report for 2002, the city’s indicator table for 2003 to 2012, the fiscal 2016 report for 2013 to 2016, the fiscal 2021 report for 2017 to 2021 and the fiscal 2026 report for 2022 to 2026. Overlapping prints agree within a second, including across the indicator’s change of name. The fiscal 2016, 2021 and 2026 reports’ windows meet end to end rather than overlap; across those two joins, the line rebuilt from the dispatch file stays within three seconds of the city’s figure. Key years: 11:04 in fiscal 1988, 8:58 in 1996, 6:25 in 2012 (the low), 8:49 in 2025 and 10:32 in 2026.

The quoted figures. Every figure and sentence quoted here, from the Mayor’s Management Reports, the city’s indicator table, the working paper and amNY, is looked up in the saved copy of its source on each run of the script, and one that is not found stops the build. The load on ambulance capacity divides the report’s incidents (all, or life-threatening) by the year’s in-service hours: the report’s average hours a day times the days in the fiscal year, 366 in fiscal 2024 and 365 in the others.

The full results. Fiscal 2025 and 2026, life-threatening calls: incidents 622,152 and 610,791; response 8:52 and 10:37; dispatch 1:12 and 2:03; travel 7:39 and 8:34; held 6.9% and 11.5%; upgraded 10.4% and 12.1%. Calls life-threatening from the start 7:50 and 8:46, of which dispatch rose 8 seconds and travel 47. Ordinary calls (from the start, not held) 7:44 and 8:33, travel 7:09 and 7:53. Held calls urgent from the start 10:23 and 11:50; all held calls, including those also upgraded, 21:07 and 24:58. Upgraded calls 17:40 and 24:05, dispatch 6:19 and 11:31. By borough, fiscal 2026 and the change: the Bronx 12:16 (+2:04), Manhattan 10:58 (+1:43), Queens 11:00 (+2:30), Brooklyn 9:02 (+1:05), Staten Island 8:48 (+1:11). From the report: ambulance in-service hours per day 8,572 and 8,109; hospital turnaround 40:40 and 40:25; fire companies at life-threatening calls 5:52 and 6:03. From the 911 file: call to arrival 11:21 and 13:09; call to job creation 2:25 and 2:30; dispatch, from the handoff to EMS, 2:17 and 3:05 on average, 1:22 and 1:28 at the weekly median; travel 7:43 and 8:38 on average, 6:47 and 7:33 at the median.

Rerunning it. The analysis script reads the city files and the saved report and writes every figure, chart and file here. The companion dataset is the dispatch file’s monthly sums from January 2016, by borough and call flags, so any dispatch-file average from 2016 can be rebuilt as summed seconds over incidents. The city’s yearly figures from 1988, the yearly figures rebuilt from fiscal 2006 and every quote, with its page, are in the results file; the monthly series from 2005 is in the chart data, to a tenth of a second. The 911 figures come from the city’s weekly file, which the script reads as published. The build manifest names the page version the run was made for and the checksum of the script and of every file linked here.