Measuring an Ambulance Response From Dispatch, Not From the Moment Someone Dials

Someone dials 911. Some number of minutes later a unit stops at a curb. Between those two moments sits a figure that gets read into council minutes, printed in state reports, and compared across counties: the response time. That figure does not cover the whole gap between the dial and the curb, and the documents that define it say so in plain type.

The national EMS data standard gives every moment in a call its own numbered timestamp. A published sheet of interval definitions states which two of those timestamps the response figure subtracts. The public research dataset then performs that subtraction across tens of millions of records and, in the same appendix, states the value above which the result is discarded. None of it is concealed; the manuals are open.

What follows is that chain in order, with the governing text at each step and, at each step, a note on what the next link no longer carries.

Span B is the only one the response figure uses. Spans A and C are separately defined intervals with their own names; neither is inside B.

Ten Named Timestamps, and the Two That Make the Number

NEMSIS, the National Emergency Medical Services Information System, is the data standard that EMS charting software writes to. NHTSA's EMS.gov page on nationwide EMS incident data gives its size in one line: "The EMS Data Standard includes 596 data elements that can be implemented by an EMS system. 165 of those are national elements." The elements that hold clock readings sit in a section called eTimes, and each one is numbered.

A single call moves through them in order. eTimes.01 is PSAP Call Date/Time, the moment a call reaches a public safety answering point. eTimes.02 is Dispatch Notified Date/Time. eTimes.03 is Unit Notified by Dispatch Date/Time. eTimes.05 is Unit En Route Date/Time, eTimes.06 is Unit Arrived on Scene Date/Time, and eTimes.07 is Arrived at Patient Date/Time. After that come eTimes.09, Unit Left Scene Date/Time; eTimes.11, Patient Arrived at Destination Date/Time; eTimes.12, Destination Patient Transfer of Care Date/Time; and eTimes.13, Unit Back in Service Date/Time.

Two of those deserve a second look, because ordinary speech merges them. Arriving on scene and arriving at the patient are separate elements with separate numbers. A rig parked outside a twelve-story building has satisfied one of them and not the other, and the standard keeps a place for the difference.

The arithmetic comes from a Data Managers Council sheet of EMS time interval definitions posted by NEMSIS. The headline interval appears there as "Total Unit Response Time: Unit Arrived on Scene (eTimes.06) – Unit Notified (eTimes.03) = Minutes/Seconds". The span that comes before it has its own separate line: "PSAP to Unit Notified Time: Unit Notified by Dispatch (eTimes.03) – PSAP Call (eTimes.01) = Minutes/Seconds".

So the response figure begins when a unit is told, not when a phone rings, and it ends when a vehicle has stopped at the scene, not when a responder has reached a patient. Both excluded spans are measured. They are simply measured under other names, and those other names are not the one that gets quoted.

The Chain

1. A call reaches a public safety answering point. The standard records that instant as eTimes.01.
2. Dispatch is notified, which is eTimes.02, and dispatch notifies a unit, which is eTimes.03. The interval definitions sheet names that span Dispatch Notified to Unit Notified Time and does not break it down further.
3. The unit goes en route at eTimes.05, arrives on scene at eTimes.06, and arrives at the patient at eTimes.07.
4. The interval named Total Unit Response Time subtracts eTimes.03 from eTimes.06. Step one, step two, and the tail end of step three all fall outside it.
5. In the public research dataset that subtraction is stored as a derived variable, EMSSystemResponseTimeMin. The earlier span is a different variable, EMSDispatchCenterTimeSec. The curb-to-patient span is a third, EMSSceneToPatientTimeMin.
6. Each derived variable carries its own ceiling. A result above that ceiling is not stored as a large number. It is set to null.
7. What survives lands in a file the manual describes as a convenience sample, not a census, with missing data that the manual says are not missing at random.
8. Averages computed over that file are quoted as response times, and the word "response" carries every edge listed above.

What Counts as Filled In

The NEMSIS data dictionary overview defines four levels of obligation for any element. "Mandatory" means "Must be completed and does not allow for NOT values". "Required" means "Must be completed and allows NOT values". "Recommended" means "Does not need to be completed and allows NOT values". "Optional" means "Does not need to be completed and does not allow for NOT values".

The phrase "NOT values" is the hinge. The same overview defines three of them. "Not Applicable" is "The data element is not applicable or pertinent to the EMS event." "Not Recorded" carries an instruction to the software: "If a data element was unintentionally left blank the EMS software should auto-populate it with 'Not Recorded'." And "Not Reporting" is "The data element is not collected by the EMS agency or state. This NOT value does not apply to National elements where 'Usage = Required'."

Read that against the four usage levels and a consequence falls out. Except at Mandatory and Optional, an element can be present, valid, and schema-complete while holding no clock reading. A record can pass validation with a NOT value sitting where a timestamp would go. Such a record contributes nothing to any interval that needed it, without ever registering as an error.

The overview also defines what national status means: "National = Yes, is an indication that the data element is required to be collected: at the local EMS agency level and submitted to the state". That is the route every record takes. Agency software writes it, the state receives it, and only then does anything reach the national file.

Bar lengths are linear in minutes, so the two sub-hour ceilings are almost invisible beside the sub-day ones. That gap is the point: two intervals from the same call are cut off at heights that differ by roughly twenty-four times.

The Ceilings Are Not All the Same Height

The 2024 public-release research dataset user manual lists the computed elapsed-time variables and, beside each one, a note. For the interval that becomes the response figure, the note reads: "EMSSystemResponseTimeMin Time difference in minutes between eTimes.03 and eTimes.06. Note: if > 1,439 minutes, set to null." For the span before dispatch reaches the unit: "EMSDispatchCenterTimeSec Time difference in minutes between eTimes.03 and eTimes.01. Note: if > 3,599 seconds, set to null."

Both cutoffs are worth reading precisely. A value of exactly 3,599 seconds is kept, because the rule discards only what is greater than 3,599 seconds. Exactly 1,439 minutes is kept for the same reason. The cut falls one unit above the stated number, not at it.

The heights differ enormously. The dispatch-center interval is discarded above 3,599 seconds, which is one second short of an hour. Chute time carries its own note: "EMSChuteTimeMin Time difference in minutes between eTimes.03 and eTimes.05. Note: if > 59 minutes, set to null." The response interval, by contrast, survives up to 1,439 minutes, which is one minute short of a full day. The curb-to-patient interval sits between them at 719 minutes.

There is also a blanket rule: "For all elapsed times, negative elapsed time values are set to null". Negative intervals are not exotic: a dispatch clock and a tablet clock that disagree by a few seconds produce one on a short call. Those records lose the interval entirely rather than pulling an average down.

A ceiling that removes a twenty-six-hour response is almost certainly removing a typing error rather than a real wait. That is a defensible design. It is still a rule about which observations reach the average, and rules of that kind move published numbers. This site has traced the same mechanism in electric utility reporting, where removing major event days takes the 2024 outage average from 662.6 minutes to 131.6. The exclusion is documented in both cases. So is the size of the gap it opens.

The Denominator Is Not the Whole File

The scale is real. The 2024 public-release dataset "includes 60,298,684 Version 3.5.0 EMS activations collected from 14,756 agencies located in 54 states and territories." That is a very large file. It is not a count of American EMS responses.

The manual is direct about this, in capital letters: "PLEASE NOTE THAT THE PUBLIC RELEASE RESEARCH DATASET IS NOT A POPULATION BASED DATA SET." It adds that "The National EMS Database is subject to the limitations of any 'convenience sample'" and that "The data may not be representative of all states or EMS agencies in the nation." It also warns that records are not people: "The data are 'event-based' and not 'patient-based'. That is, a single patient may be represented in more than one record."

On the gaps, it is equally direct. "The proportion of missing data varies across data elements in National EMS Database, and it is important to decide how to deal with missing data when doing analyses." And then the sentence that matters most for any published average: "In most cases, NEMSIS data are not missing at random and analyses, therefore, are subject to bias if missing data are ignored."

The annual national EMS data report for 2024 shows how that plays out on a single chart. Under the heading "URBANICITY OF EMS INCIDENT LOCATION" it prints four categories and their counts: Urban, 39,579,519; Rural, 2,652,293; Suburban, 2,526,946; and Frontier, 452,100. Those four counts sum to 45,210,858. The same report's total activation count is 60,298,684. The difference, 15,087,826 activations, does not appear as a category on that chart, so any share read off it rests on the smaller base.

That is the same shape as a reporting boundary that quietly sets a denominator. It appeared here before in wireless outage reporting, where the percentage of cell sites out of service moves when the reporting area does. Neither case involves a wrong number. Both involve a number whose base is smaller than the headline suggests.

The four printed categories and the report's own total activation count, drawn on the same scale. The dashed segment is the arithmetic difference between them.

No Federal Rule Grades the Response

A reasonable assumption at this point is that a federal threshold sits behind this, a number an ambulance service must beat. There is none. The National Advisory Committee on Rural Health and Human Services, in its November 2022 policy brief Access to Emergency Medical Services in Rural Communities, places the authority with the states: "State governments and their offices of EMS are responsible for setting EMS training standards and certification levels." Response-time targets in the United States are written into local contracts and state rules, not into a federal ceiling.

The same brief also supplies a widely quoted response-time figure, and it supplies it with an attribution. It reports that a 2017 article in JAMA Surgery found the "national average from the time of a 911 call to arrival on scene was 7 minutes", that the same figure "increases to more than 14 minutes in rural settings", and that this comes "with nearly 1 of 10 encounters waiting almost 30 minutes for the arrival of EMS personnel."

Read that against the chain above. From the time of a 911 call to arrival on scene is eTimes.01 to eTimes.06. It is not eTimes.03 to eTimes.06. That figure spans a different interval than the derived variable the national dataset labels a system response time, and the difference between the two is exactly the dispatch-center span that the dataset stores separately and discards above 3,599 seconds.

The Interval That Decides the Next Call

The clock does not stop when a patient reaches a hospital. The interval definitions sheet names the wait that follows: "Destination Transfer of Care (Wall Time): Destination Patient Transfer of Care (eTimes.12) – Patient Arrived at Destination (eTimes.11) = Minutes/Seconds". Until that transfer happens, the crew is standing in a corridor with the patient still theirs.

The unit is not available again until eTimes.13. The sheet defines that span too: "Total Unit Call Time (Time on Task): Unit Back in Service (eTimes.13) – Unit Notified by Dispatch (eTimes.03) = Minutes/Seconds". Wall time sits inside it. So an emergency department's ability to take a patient off a stretcher becomes, one link later, a constraint on how many ambulances a county has in service. This site has followed the upstream half of that already, where an admitted patient's wait is set by inpatient beds, not the emergency room.

One more line describes work that produces no patient at all: "Canceled Response Time: Unit Canceled (eTimes.14) – Unit Notified by Dispatch (eTimes.03) = Minutes/Seconds". A unit that rolls and is turned around consumed minutes of availability and generated a record with no arrival timestamp in it.

Where This Doesn't Apply

None of this says a local system cannot measure the full span. Every timestamp involved is in the standard, and an agency or a state that wants the interval from eTimes.01 to eTimes.06 can compute it from its own records. Many do. The point is narrower: the derived variable distributed in the national public research file is not that interval.

The ceilings described here are properties of the derived variables in the public-release research dataset. They are not properties of the raw records an agency holds, and they are not a claim that any underlying timestamp was deleted. A computer-aided dispatch system can carry sub-second resolution on events the national file summarizes.

The urbanicity arithmetic is an observation about one chart in one annual report. It is not a general statement about how complete NEMSIS is, and the report does not describe the difference as missing data.

The 7-minute, 14-minute, and near-30-minute figures are not federal measurements. They are findings from a 2017 journal article, quoted inside a 2022 federal advisory committee brief, and they carry that article's own methods and vintage rather than the dataset's.

Finally, nothing above argues that a ceiling or an exclusion is a mistake. Discarding a negative elapsed time is obviously correct. Discarding a chute time above 59 minutes very probably removes a stale record rather than a crew that took an hour to leave the bay. These are reasonable rules. They are also rules, and they run before any average is taken.

So the published response number is what it says it is, once the definitions are read: an interval that opens when a unit is notified and closes when a vehicle has arrived, computed over records that cleared a set of ceilings, drawn from a file that describes itself as a convenience sample. The seconds spent on the phone and the minutes spent crossing a lobby are both measured, both stored, and both somewhere else.