Seattle Police
Call volumeOutcomesHow long it tookHow long it took, in full

Predictive Policing

Predictive Policing calls over time: volume, outcomes, and response speed

Predictive Policing: Call Volume

Call volume is the number of events logged in the Seattle Police Department’s Computer-Aided Dispatch (CAD) system. It counts calls from the public (by 911, phone, text or alarm) as well as officer-initiated activity, once per event, grouped by the call type assigned when the call was dispatched, before any investigation. It is not the same as the number of crimes reported.

Included: Every event logged in CAD: calls from the public and events officers logged themselves (100% of the total). Excluded: Nothing.

Swipe sideways to see the full chart →
020,00040,0002013: 27,901 predictive policing calls logged in CAD2014: 5,492 predictive policing calls logged in CAD2015: 383 predictive policing calls logged in CAD2016: 792 predictive policing calls logged in CAD2013n=27,9012015n=3832016n=792
predictive policing calls logged in CAD
View as table
Call volume — same numbers as the chart above
PeriodPredictive policing calls logged in CAD
201327,901
20145,492
2015383
2016792
Predictive Policing Calls Are Down 97% From Their 2013 PeakSeattle Police logged 27,901 predictive policing calls in 2013 and 792 in 2016, the most recent full year.

Predictive Policing: What Happened When Officers Responded

Outcomes show how each call ended, based on the disposition officers recorded when the call was closed: an arrest or citation, a report taken with no arrest, resolved on scene, unable to locate (officers could not find the incident or the person who called), or other (unfounded or cancelled). Each bar shows the share of that period’s calls.

Included: Every event, called in or officer-initiated, using the disposition recorded when the event was closed. Excluded: Nothing.

Swipe sideways to see the full chart →
100%50%0%2013 · Report taken, no arrest: 0.1%2013 · Resolved on scene: 98.2%2013 · Unable to locate: 1.3%2013 · Other (unfounded / cancelled): 0.3%2014 · Report taken, no arrest: 0.1%2014 · Resolved on scene: 98.4%2014 · Unable to locate: 1.1%2014 · Other (unfounded / cancelled): 0.4%2015 · Report taken, no arrest: 0.3%2015 · Resolved on scene: 99.0%2015 · Unable to locate: 0.8%2016 · Arrest / citation: 0.3%2016 · Report taken, no arrest: 2.5%2016 · Resolved on scene: 96.5%2016 · Unable to locate: 0.8%2013n=27,9012015n=3832016n=792
Arrest / citationReport taken, no arrestResolved on sceneUnable to locateOther (unfounded / cancelled)
View as table
What happened when officers responded — same numbers as the chart above
PeriodEventsReport taken, no arrestResolved on sceneUnable to locateOther (unfounded / cancelled)Arrest / citation
201327,9010.1%98.2%1.3%0.3%—
20145,4920.1%98.4%1.1%0.4%—
20153830.3%99.0%0.8%——
20167922.5%96.5%0.8%—0.3%
“Unable to Locate” Outcomes Are Holding Near 1% of Predictive Policing CallsReport-taken outcomes moved from 0% of calls in 2013 to 2% in 2016. “Unable to locate” means officers could not find the incident or the person who called.

Predictive Policing: How Long It Took

Response time is how long it took, from the moment a called-in event was queued with dispatch, until the first officer arrived on scene. Dispatch delay is the first part of that wait: the time before an officer was assigned and sent. Both lines show the average, which a small number of very long waits pull up, so it runs higher than the typical (median) call.

Included: Only events called in by the public and dispatched to an officer. The averages use those with a recorded officer arrival. The n= under each bar is the number of called-in events. Excluded: Events officers logged themselves (on view), which are 100% of predictive policing events, because they have no wait. Called-in events with no recorded arrival (for example, cancelled before an officer arrived) are left out of the averages. Because most predictive policing events are logged by officers, these charts describe only the small share that were called in. When a call priority is selected, only events of that priority are shown.

Call priority
Swipe sideways to see the full chart →
0 min10 min20 min2013: 10.7 min2014: 7.6 min2015: 3.3 min2016: 0.4 min2013: 16.7 min2014: 10.6 min2015: 7.8 min2013n=332015n=32016n=1
Average response timeAverage dispatch delay
Small sample: Only a small share of predictive policing events are called in. The year with the fewest had only 1, so averages can swing widely; the n= under each year shows its count.
View as table
How long it took — same numbers as the chart above
PeriodCalled-in eventsAverage dispatch delayAverage response time
20133310.7 min16.7 min
2014127.6 min10.6 min
201533.3 min7.8 min
201610.4 min—
No Response Times Were Recorded in 2016None of the 1 called-in events had a recorded officer arrival.
Call priority (SPD Policy 4.065)
Priority 1 · Immediate
Poses a threat to life.
Priority 2 · Urgent
Situations that could escalate if help does not arrive soon.
Priority 3 · Prompt
Response time is not critical; usually a victim waiting to speak with officers.
All calls
Also includes priorities 4 to 9: lower-priority requests, callback reports, and automatic entries such as traffic stops and administrative codes.

Predictive Policing: How Long It Took, in Full

The same response times, broken into bands so the slowest calls are visible. Each bar shows the share of called-in events by how long the first officer took to arrive after the event was queued, including the ones that took hours. “No recorded response” means no arrival was ever logged.

Included: Only events called in by the public and dispatched to an officer, including those with no recorded arrival (shown as “No recorded response”). Excluded: Events officers logged themselves (on view), which are 100% of predictive policing events, because they have no wait. Because most predictive policing events are logged by officers, these charts describe only the small share that were called in. When a call priority is selected, only events of that priority are shown.

Call priority
Swipe sideways to see the full chart →
100%50%0%2013 · No recorded response: 33.3%2013 · 0–5 min: 39.4%2013 · 5–30 min: 12.1%2013 · 30–60 min: 6.1%2013 · 1–4 hr: 9.1%2014 · No recorded response: 41.7%2014 · 0–5 min: 41.7%2014 · 5–30 min: 8.3%2014 · 1–4 hr: 8.3%2015 · No recorded response: 66.7%2015 · 5–30 min: 33.3%2016 · No recorded response: 100.0%2013n=332015n=32016n=1
No recorded response0–5 min5–30 min30–60 min1–4 hr4–8 hrOver 8 hr
Small sample: Only a small share of predictive policing events are called in. The year with the fewest had only 1, so shares can swing widely; the n= under each year shows its count.
View as table
How long it took, in full — including the calls that took hours — same numbers as the chart above
PeriodCalled-in eventsNo recorded response0–5 min5–30 min30–60 min1–4 hr
20133333.3%39.4%12.1%6.1%9.1%
20141241.7%41.7%8.3%—8.3%
2015366.7%—33.3%——
20161100.0%————
0% of Called-In Events Took Over an Hour in 2016That is 0% of 1 called-in events. Counts this small can swing a lot from year to year.
Call priority (SPD Policy 4.065)
Priority 1 · Immediate
Poses a threat to life.
Priority 2 · Urgent
Situations that could escalate if help does not arrive soon.
Priority 3 · Prompt
Response time is not critical; usually a victim waiting to speak with officers.
All calls
Also includes priorities 4 to 9: lower-priority requests, callback reports, and automatic entries such as traffic stops and administrative codes.

“CAD” is Computer-Aided Dispatch, the system Seattle Police use to log and time every call for service — every number on this page comes from it. 2013 and 2016 may be partial years (data begins mid-2009; the latest year runs through the site's last data pull). “n=” under each year is the number of events logged that year in the volume and outcomes charts, and the number of called-in events dispatched to an officer in the response-time charts. Events officers log themselves (on view) are included in volume and outcomes and left out of response times, because they have no wait. The response-time bands measure time from call queued to first unit on scene; “no recorded response” means no arrival was ever logged. Hover any point for its exact value.