← All projects

Seattle · 2020–2025 · Independent project

Who Seattle police stop, and where

Seattle police were under federal oversight for 13 years, and now that it's over, public data is one of the few ways left for residents to check how police use their power. I wanted to see who Seattle police stop, and where, and whether the data can answer the questions people actually ask, so I took six years of stop records and asked them three questions. The short answers are right here, and the maps underneath show how I got them.

Recorded stops
24,897
Patrol areas
51
Years
2020–2025
Built with
R, D3
How we got here
  1. July 2012

    Seattle agrees to a federal consent decree. The Justice Department had found a pattern of excessive force in December 2011.

  2. Sept 2023

    The court orders SPD to record race in stops better and make the data usable by the public.

  3. Sept 2025

    The consent decree ends, and federal oversight is over.

  4. Aug 2026

    The city's Inspector General finds SPD's dashboard still only shows race inside gender groups, a fix it first asked for in 2023.

Terry stop
A short detention based on reasonable suspicion. You aren't free to leave, but you aren't necessarily arrested. A frisk (pat-down) is a separate step.
Patrol area
One of the small districts SPD organizes patrols around ("beats" in the records). Names like O2 and Q1 are just labels, not rankings.
Recorded race
The race the officer wrote down for the person they stopped. It isn't necessarily how that person identifies.

Question 1

Where are stops concentrated?

Ten of the 51 patrol areas account for 42% of recorded stops, but which areas come out on top changes depending on what you compare them against.

How sure
Clear Stops are clearly concentrated, although only 5 areas stay in the top 10 under all three comparisons.
What to keep in mind
A raw count ignores how busy a place is. Comparing stops with residents makes downtown look extreme, because very few people live there, and comparing them with calls to police depends on who picks up the phone.

Annual average, 2020–2025

Where stops were recorded

Stops per year

Seattle police patrol area mapA map showing how Seattle's patrol areas compare when Terry stops are counted, compared with the number of residents, or compared with calls from the public.
O2
Q1

Darker means higher compared with the other patrol areas.

Map 1 · Count

I started by counting the stops.

Patrol area O2 recorded 287 Terry stops over six years, and Q1 recorded 383, so on this map Q1 looks a lot busier.

A count only tells you where reports piled up, though, and it says nothing about how many people were around.

Map 2 · Compared with residents

Then I compared stops with how many people live there.

This is where O2 jumps out, because only 357 people live there, which works out to 1,340 stops a year for every 10,000 residents.

The math is right, but O2 is downtown, where most of the people on a given day are working, shopping or visiting, and none of them count as residents.

Map 3 · Compared with calls

So I compared stops with calls from the public instead.

O2 also logged 16,627 calls from the public, and next to all of those calls its stops look ordinary, so O2 drops toward the middle and Q1 climbs.

Calls have their own problem, because they only show what people report to police, which leaves out a lot of who's there and what happens.

What changed

The "hot spot" depends on the comparison.

Nothing in the records changed between these maps. I only changed what I compared stops with, and ten of the 51 areas moved ten or more places in the ranking.

SPD's dashboard only shows counts, which is why I wanted all three side by side, each with its blind spot written down.

Question 2

Do neighborhoods with more Black residents see more stops?

One model finds a 55% higher stop rate for every 10-point higher share of Black residents, after accounting for everything I measured about each neighborhood. The other three stop-rate models, and a fifth that counts total stops, can't rule out no difference.

How sure
Depends on the model The answer changes with what you account for, so I'm showing all five.
What it can't show
These are patterns across neighborhoods, and we can't establish from them that race caused any particular stop.

First, who lives where

Seattle's neighborhoods aren't evenly mixed, because racial covenants, discriminatory lending and redlining decided where people could live, and those patterns outlasted the laws. Pick a group to see where it makes up more or less of the population, then click a patrol area to see everyone who lives there.

2020 Census residents

American Indian / Alaska Native residents

Share of residents

How the five models compare

Each model accounts for a different mix of things: recorded crime, calls from the public, population, zoning, poverty and a few other neighborhood conditions.

All five modelsEach compares areas with a 10-point higher Black share of residents. Models 1–4 compare stops per resident; model 5 compares total stops. A 1 means no difference.
The model with every neighborhood conditionExpected stops per year as the Black share of residents rises, with a 95% range showing how uncertain that estimate is.

Question 3

Once someone is stopped, are they treated differently?

Yes. When officers recorded the person as Black, 31.6% of stops included a frisk and 26.9% ended in arrest. When they recorded the person as White, it was 24.3% and 22.8%.

How sure
Clear difference The 95% ranges don't overlap (Black: 30.6–32.7% frisked, 25.9–27.9% arrested; White: 23.6–25.1% and 22.1–23.6%). That's clear from the records, but we can't establish with certainty why that is the case.
What it can't show
These numbers start after a stop has already happened, so they say nothing about anyone's chance of being stopped.
How race ends up in the data
  1. 1The officer writes down a race

    It's the race the officer perceived, entered on the stop report.

  2. 2SPD publishes it

    The public file calls this field "Subject Perceived Race."

  3. 3Everyone uses that judgment

    Dashboards, studies and this project all rely on the officer's call, which may not match how the person identifies.

Recorded stops by race, 2020–2025Race shown on its own, not split by gender. 24,897 stops in all; one record listing several people isn't shown.
  • White12,112
  • Black7,627
  • Unknown2,900
  • Asian1,010
  • American Indian or Alaska Native546
  • Not recorded532
  • Native Hawaiian / Pacific Islander169
Frisks and arrests by recorded raceShare of usable stop records, 2020–2025. Lines show 95% ranges.

Where frisks and arrests were recorded

Among recorded stops, 2020–2025

Recorded stops that included a frisk

Percent of usable records

Gray means an area had fewer than 20 usable stop records, so I didn't show a rate.

Explore the records

Try it yourself

Change the year, the comparison, or how the stop started, then click any patrol area to see the numbers behind its color.

One thing I didn't expect: recorded stops fell from 5,769 in 2020 to 3,426 in 2025, but the share that started with an officer's own observation, rather than a call, rose from 31% to 45% (among stops with a recorded call type).

Period
Compare stops by
How the stop started

"Officer saw it" means the record was coded on-view. "Sent after a call" covers 911, phone, alarm, text and later calls. Records without a usable call type only show up under "All recorded."

Annual average, 2020–2025

Where stops were recorded

Stops per year

What I changed

Fixing the dashboard for the people using it

Say you want to know whether Black residents get stopped more. On SPD's dashboard, you have to split stops by gender before you can count them by race. The Inspector General flagged that in 2023, and it was still the same when the office checked again in February 2026. I wasn't happy with that, so I rebuilt the parts that get in a resident's way.

If you want to…On SPD's dashboardIn my version
Count stops by raceRace only appears inside gender groups.Stops by recorded race, on their own.
Read group sizesSizes are labeled with a bare "n."Counts labeled as stops, patrol areas or people.
Read the frisk-rate chartNo axes, and an unexplained white line.Labeled charts, with 95% ranges on every comparison by race.
Get an answerSet the filters and work it out yourself.Answers to three questions up front, with their limits.
Know who decided the raceA filter labeled "perceived" demographics.The steps showing how an officer's judgment becomes data.
Compare areasCounts only.Counts, per resident and per public call, each with its limits.

I worked from the Inspector General's August 2026 review and checked the dashboard myself in October 2026.

See every patrol area in a table
Patrol areaStops per yearPer 10,000 residentsPer 1,000 callsFrisk rateArrest rate
B158.730.69.927%24.7%
B272.23512.922.9%15.5%
B340.317.58.429.8%22.3%
C156.340.11427.2%27.5%
C227.216.38.928.8%28.8%
C340.326.111.630.6%30.6%
D1122.2103.321.827%26.7%
D2119.369.321.220.4%31.8%
D363.742.212.331.2%30.6%
E112484.618.523.1%23%
E2120.7171.226.925.8%20.4%
E387.574.313.827.4%26.3%
F174.555.319.334%30.2%
F290.892.122.527.9%27.2%
F364.362.317.930.6%21.5%
G157.278.414.735%21.6%
G263.255.314.729.3%24.3%
G359.870.718.430.4%25.6%
J130.219.79.632%21%
J220.78.26.728.2%20.2%
J337.322.510.232.6%25.4%
K1103.5459.823.424.5%24.3%
K2118470.525.928.5%24%
K3483.2843.75117.7%16.8%
L145.724.310.233.9%20.4%
L256.230.911.435%24.3%
L35129.811.335%24.2%
M1114.2208.827.618.7%22.5%
M279.2228.825.821.1%26.7%
M3303.83,36170.413.4%16.2%
N12211.4735.6%12.1%
N2104.3103.817.635.1%15.2%
N310881.618.237.3%18.4%
O160.3977.82144.2%27.3%
O247.81,339.917.335.5%25.4%
O364.8246.722.543.4%28.5%
Q163.82720.328.2%34.2%
Q272.2251625.2%27.3%
Q3128.388.12125.7%30.9%
R159.231.912.438.6%20.8%
R2124.285.817.638.3%28.5%
R353.330.912.939.1%27.2%
S144.726.812.340.7%28.4%
S251.839.613.537%24.8%
S345.733.91342.7%21.2%
U157.721.48.530.9%23.1%
U264.731.411.132.5%21.6%
U332.88.76.137.6%25.9%
W152.327.711.529.6%22.6%
W268.727.613.625%27.9%
W337.81712.432.6%26.4%