Research · 2 min read
A Response-Time Average Hides Dispatch Delay
Response time is a distribution dominated by dispatch delay for lower priorities, not a single average.
Response time is usually sold as a single average. Seattle's public call data says the useful number is a distribution — and that most of the wait for lower-priority calls is dispatch delay, not driving. Priority-one calls hit seven minutes about half the time; priority-two calls rarely do.
Our analysis of 250,000 sampled calls, january 1–december 31, 2024.
What does Seattle's public call data show about response-time distributions?
Across 250,000 sampled Seattle calls from 2024, priority-1 calls had a 6.7-minute median response and 52.5% arrived within seven minutes, while priority-2 calls had a 21.1-minute median of which 14.8 minutes was dispatch delay. About 8.8% of calls had no recorded response time.
The numbers behind the answer
Selected measures only. Denominators and interpretation stay attached so the headline cannot stand alone.
6.7 min
Priority-1 median response
Across 62,638 priority-1 calls; 52.5% arrived within seven minutes.
14.8 min
Priority-2 median dispatch delay
Of a 21.1-minute median priority-2 response, most is dispatch delay.
8.8%
Calls with no recorded response
21,949 of 250,000 sampled calls lack a usable response time and drop out of percentiles.
Read the middle and the tail, not the mean
Across a 250,000-call sample, priority-1 calls have a median response of 6.7 minutes, but the 90th percentile is 16.6 and the 95th is 22.0. Priority-2 calls sit at a 21.1-minute median with a 185-minute 90th percentile. A single average would erase all of that structure.
Roughly eight percent of sampled calls carry no response time at all. Any scorecard that reports only an average is reporting neither the tail nor the missing rows.
Evidence visual
Median response time by call priority
Minutes from call queued to first unit arrival, 2024 sample.
Most of the wait is dispatch delay
Response time here runs from the call being queued to the first unit arriving, which includes time spent waiting to be dispatched and time spent driving. Splitting them changes the story: priority-1 calls average a 1.5-minute median dispatch delay, but priority-2 calls carry a 14.8-minute median dispatch delay.
For lower priorities, the problem is largely queueing, not distance. That distinction matters for any capacity or staffing question — and it is invisible in a blended average.
A reusable evaluation harness, not yet a verdict
The program also stands up a re-usable causal-evaluation harness for future policy questions — difference-in-differences and event studies on a beat-week panel. No intervention-date registry has been supplied, so it is a documented, importable template rather than a fitted result, and we say so.
Nothing here describes 911 Sentinel's responder model, and none of it is a service-level promise. It is what a public record of police calls actually shows.
- Report a target-hit rate and upper percentiles, not just a mean.
- Separate dispatch delay from travel time.
- Show how many calls have no recorded response.
- Exclude autogenerated traffic and admin priorities.
- Pre-register any causal design before looking at outcomes.
Questions property teams ask
Does this replace a response-time average?
It complements it. A median, a target-hit rate, and upper percentiles describe the distribution that a single mean hides.
Why separate dispatch delay from travel?
They imply different fixes. For lower priorities, most of the wait is time before a unit is dispatched, not driving distance.
What are the missing 8.8%?
Calls with no usable recorded response time. They are counted and disclosed rather than silently dropped.
Does the harness show a policy worked?
No. Without a supplied intervention-date registry it is a documented design template, not a fitted result.
Our methods, limits, and sources
How we calculated this
This is original 911 Sentinel research — we gathered the records, ran every calculation below, and published the aggregate dataset.
We sampled 250,000 Seattle CAD calls from 2024 across priorities 1–6 and computed response-time distributions, dispatch-delay decomposition, and censoring.
- We excluded priorities 7, 9, and blank administrative categories.
- We converted response and dispatch-delay seconds to minutes and computed percentiles by priority.
- We decomposed response into dispatch delay and travel.
- We recorded censoring and documented a reusable causal harness.
What this analysis cannot establish
- Response time is queued-to-first-arrival and includes dispatch delay; decompose before comparing.
- Location is beat-blurred with no unit-level geography.
- About 8.8% of sampled calls lack a response time and are excluded from percentiles.
- Autogenerated traffic and administrative priorities are excluded.
- This is public CAD data, not the internal responder model, and is not a service-level promise.
Sources
The raw records come from the sources below; the study design, analysis, charts, and conclusions are our own.
Related questions and practical guides
New coverage — intake open
Ask for the distribution, not the average
Any monitoring or response proposal should show a target-hit rate, upper percentiles, and how missing calls are handled — not a single blended number.