This report examines fatalities associated with vehicle pursuits by
law enforcement in WA State since 2015.
Pursuits are unique among police tactics in their risk of collateral
damage. Nationally, about one-third of pursuits lead to vehicular
accidents, and a substantial fraction (over a third) of those injured or
killed in these accidents are uninvolved bystanders, passengers and
officers.
Here in Washington state, recent legislative changes provide an
opportunity to observe whether pursuit policies can help to reduce these
fatalities.
What this report covers
The report examines the range of fatalities that are associated with
police vehicular pursuits:
“Pursuit vehicular fatalities” - an active or recently terminated
vehicular pursuit led to a vehicular homicide caused by one of the cars
involved in the pursuit.
“After pursuit fatalities” – a pursuit was involved earlier in
the incident, but had ended by the time the fatality occurred. Typically
the person killed in these cases is the fleeing subject who is shot by
police at the end of the chase.
“Attempted stop fatalities” – a subject fled an attempted stop by
police and committed a vehicular homicide in the process, but the police
vehicle did not engage in pursuit.
The majority of the report focuses on the first two types: pursuit
vehicular fatalities and after pursuit fatalities, examined separately.
The attempted stop fatalities are different for several reasons, so we
address them in a section of their
own.
How the report is organized
The table of contents on the left shows the outline of the
information. All of the entries are linked and can be used to navigate
through the report. Note that clicking on an entry of the TOC will cause
the sub-headings to display if there are any.
Within each section of the report listed in the TOC, you may find
“tabs” on the page that contain additional information. The active tab
is highlighted with a blue filled box, and you can click the inactive
tabs on either side of the active tab to access their information.
The graphics use a consistent color coding to help with comparisons,
and the color scheme is linked to the type of data in the graph: pursuit
vehicular fatalities (color codes for victim type), after pursuit
fatalities (color codes for homicide vs. suicide), and incidents (color
codes for pursuit vehicular fatalities and after pursuit
fatalities).
Legislative history
Pursuit policy has been an active area of legislation in WA State,
with parallels to the policy changes that took place in Milwaukee,
WI. The changes define four policy periods in WA state useful for
comparative analysis:
Pre-reform
before July 25, 2021
Pursuit policies during this period were decentralized, and
established by each agency, so there was a patchwork of different
policies across the state. Some policies were quite restrictive, others
left decisions to the discretion of officers. In this report, we look
back to Jan 1, 2015 as the start of this observation period.
HB1054
July 25, 2021 - May 2, 2023
In 2021 the legislature enacted a statewide policy to provide a
uniform standard and to improve public safety by reducing the growing
number of people killed during these pursuits (House
Bill 1054, effective July 25, 2021). At the time, pursuits were
responsible for 10-20% of the fatalities from police activities each
year. About half of those killed were passengers, officers or uninvolved
bystanders.
HB 1054 established a clear standard for balancing the danger posed
to the public by the pursuit against the danger posed by a fleeing
subject. It prioritized public safety: police are allowed to engage in
pursuits when there is a well established threat to public safety –
violent offenses (like carjackings, armed robberies), sex offenses,
DUIs, and prison escapes – but not for misdemeanors or property crimes.
It required that the evidence of the threat meets the “probable cause”
standard – the same standard required to make an arrest. In effect the
law said: If you don’t have enough evidence to arrest the person, then
you can not justify the risk to public safety for this pursuit.
SB5253
May 3, 2023 - June 5, 2024
In 2023 the legislature modified
the policy, lowering the standard of evidence required from probable
cause to reasonable suspicion, adding vehicular and domestic assaults
explicitly to the list of offenses eligible for pursuit, and removing
the requirement for supervisory authorization. These changes took effect
May 3, 2023.
I2113
after June 5, 2024 (Current)
In response to a successful initiative campaign to place a pursuit
policy rollback on the November 2024 ballot, the legislature took the
option of passing the
initiative themselves in March 2024. The initiative removed all of
the explicit public safety requirements for allowable pursuits, and
replaced them with a simple discretionary requirement that “the threat
to the safety of others and the safety risks of failing to apprehend or
identify the person are considered to be greater than the safety risks
of the vehicular pursuit under the circumstances.”
The full texts of the bills passed and the current law (RCW) on
vehicle pursuits can be found in the Pursuit legislation section at the end
of this report.
Impact of policy changes
There are now 4 different policy periods observed in the decade of
data we are tracking. We compare them below using an estimated
annualized rate of pursuit-related fatalities for each period.
In the short time since I-2113 rolled back the pursuit safety
restrictions, the data show that the pursuit-related fatality rate has
risen dramatically.
It is now 6.1 times as high as it was during the HB1054 period,
when pursuits were restricted to violent felonies and DUIs.
In a typical year, this means 14 more people will be killed as a
result of police pursuits.
While the exact value of the policy effects will depend on the metric
used, the pattern observed in the data is clear:
The WA statewide pursuit safeguards adopted in 2021 were working as
intended to improve public safety: they were saving lives. The cost of
weaking those safeguards has been a sharp rise in pursuit-related
fatalities.
The pattern here in WA mirrors what happened in Milwaukee, WI before,
during and after the 5 years that their policy restricted pursuits to
violent offenses (see the official City of Milwaukee pursuit reports here).
Table
annual_rate_df %>%
kbl(caption = "Annualized pursuit fatality rates by policy period*",
align = "lrrrr") %>%
kable_styling(bootstrap_options = "striped") %>%
add_footnote("Includes fatalities from active and recently terminated pursuits only.",
notation = "symbol")
Annualized pursuit fatality rates by policy period*
Policy period
Fatalities
Years
Annualized rate
Rate relative to HB1054
pre-reform
25
6.6
3.8
1.4
HB1054
5
1.8
2.8
1.0
SB5352
8
1.1
7.3
2.6
I2113
24
1.4
17.1
6.1
* Includes fatalities from active and recently terminated
pursuits only.
Comments
There are different methods one can use to assess the impacts of the
sequence of pursuit policy changes in WA. But it is complicated by the
fact that the lengths of the policy periods are quite different, the
dates the policies became effective do not line up with any type of
standardized year (calendar or legislative), and the fatality counts for
the shorter policy periods are too small to support a robust statistical
analysis. We previously reported a fatality count comparison with a
standardized time window, pre and post reform. But with 4, different
length, policy periods, that approach no longer works.
For this policy impact summary section we have decided to report
annualized fatality rates for each policy period. This metric has both
benefits and drawbacks:
Benefits – It provides a single summary number that can
be compared across all periods, regardless of their length, and it
allows us to use the exact dates when legislation became effective to
define the periods.
Drawbacks – In contrast to the number of persons killed
in a specific legislative year (or other standardized time frame), the
annualized fatality rate is an estimate of the average rate of
fatalities per year. The reliability of that estimate depends on the
number of years observed. In our case the policies changed so frequently
that several periods are quite short, which means the annualized
estimate we get is not very reliable.
In later sections of this report we present fatalities broken down by
calender and approximate legislative year, showing the breakdown of the
victims (officer, bystanders, passengers, subjects) in each
timeframe.
Interactive Maps
The first map is restricted to pursuit vehicular fatalities only.
The second map adds in after pursuit fatalities.
Pursuit vehicular fatalities
Each marker represents an active vehicle pursuit that ended with a
vehicular homicide.
Red markers are pursuits that resulted in bystanders, passengers
and/or officers (B/P/O) being killed or injured.
Blue markers are pursuits with subject fatalities only
The icon inside the marker shows whether this was a single or
multiple fatality incident.
This map is also interactive:
Hovering over the pointer brings up the number of people
killed and injured, the law enforcement agency involved, and the date of
the incident;
Clicking the pointer will bring up a url to a news
article on the incident.
Since more than one person may be killed in an incident the number of
incidents will generally be smaller than the number of people killed.
Fatalities are a better metric for understanding who is at risk from
pursuits. Incidents are a better metric for understanding the agencies
involved. We look at both below.
People killed
Summary
How often are pursuits involved?
Pursuits are involved in 29% of fatal encounters with police.
The table below breaks down all fatalities from police activities in
WA since 2015 – noting if and how a pursuit was involved. The categories
are:
Vehicle not involved – No indication that a
vehicle-related fatality was involved in the incident that led to this
fatality.
Pursuit vehicular fatality – A pursuit led directly to a
vehicular fatality (e.g., a crash or someone getting run over). This
includes active and recently terminated pursuits (see Coding section for more details)
After pursuit fatality – A vehicle pursuit occurred
during the incident, but was not the immediate cause of the fatality.
Most of these cases are gun-related fatalities (79% are either shot by
police or die of a self-inflicted gunshot, the homicides and suicides
are reported separately in the table below) the rest are fatalities from
vehicle accidents.
Attempted stop – The subject fled a traffic stop and a
vehicular fatality resulted, but police report they did not engage in
pursuit.
Vehicle accident – An officer commits a vehicular
homicide (on- or off-duty), but there is no pursuit involved.
* Data include all fatalities during a police encounter in WA
since 2015.
How many people are killed in pursuit-related accidents?
Pursuit-related accidents account for about 12% of all persons killed
during police encounters. They are the second leading cause of death
from police activities.
The table below breaks down the cause of death for all persons killed
during an encounter with law enforcement since 2015. After pursuit
vehicular fatalities, all causes comprising less than 2% of fatalities
are grouped together. Note that some of the gunshot fatalities involve
pursuits, but the cause of death is not vehicular homicide.
tab <- fatalities.all %>%
mutate(cod.cat = factor(case_when(
incident.type == "Pursuit vehicular fatality" ~ "Pursuit vehicular fatality",
cod == "Gunshot" ~ "Gunshot",
cod == "Taser" ~ "Taser",
grepl("Asphyx", cod) ~ "Asphyxiated/Restrained",
TRUE ~ "All other causes"),
levels = c("Gunshot", "Pursuit vehicular fatality",
"Taser", "Asphyxiated/Restrained",
"All other causes"))
) %>%
group_by(cod.cat) %>%
summarize(Number = n()) %>%
mutate(Percent = Number/sum(Number)) %>%
bind_rows(data.frame(cod.cat ="Total",
Number = sum(.$Number),
Percent = sum(.$Percent))) %>%
mutate(Percent = scales::percent(Percent, acc = 0.1)) %>%
rename(`Cause of Death` = cod.cat)
tab %>%
kbl(caption = "Persons killed by police in WA since 2015: By cause of death",
align = "lrr") %>%
kable_styling(bootstrap_options = "striped") %>%
row_spec(which(tab$"Cause of Death" == "Total"), bold=T) %>%
footnote("All other causes includes beaten, tasered, accident etc.")
Persons killed by police in WA since 2015: By cause of death
Cause of Death
Number
Percent
Gunshot
352
70.8%
Pursuit vehicular fatality
62
12.5%
Taser
12
2.4%
Asphyxiated/Restrained
10
2.0%
All other causes
61
12.3%
Total
497
100.0%
Note:
All other causes includes beaten, tasered, accident etc.
Who gets killed?
53% of the persons killed during pursuits since 2015 in WA have been
passengers, officers or uninvolved bystanders.
tab <- fatalities.pvf %>%
group_by(Victim=victim) %>%
summarize(Number = n()) %>%
mutate(Percent = Number/sum(Number)) %>%
arrange(desc(Number)) %>%
bind_rows(data.frame(Victim ="Total",
Number = sum(.$Number),
Percent = sum(.$Percent))) %>%
mutate(Percent = scales::percent(Percent),
Victim = factor(Victim,
levels = c("Subject", "Bystander", "Passenger",
"Officer", "Total")))
tab %>%
arrange(Victim) %>%
kbl(caption = "Victims of vehicular pursuits in WA since 2015",
digits = c(0,0,1),
align = "lrr") %>%
kable_styling(bootstrap_options = "striped") %>%
row_spec(which(tab$Victim == "Total"), bold=T) %>%
footnote("See Data section for information on data sources and coding.")
Victims of vehicular pursuits in WA since 2015
Victim
Number
Percent
Subject
29
46.8%
Bystander
14
22.6%
Passenger
17
27.4%
Officer
2
3.2%
Total
62
100.0%
Note:
See Data section for information on data sources and coding.
Trend over time
There are several ways we can examine the data over time. Some are
better suited to revealing the impact of the policy change than
others.
Policy period: This is a simple count of fatalities in
each period, so is easy to understand and plot, but because the periods
represent different lengths of time, direct comparisons are
misleading.
Calendar year: This is a common way to aggregate the
fatality counts into standardized time units in a single graphic, but it
does not give a clear picture of the policy impacts because the laws
changed in the middle of 2021, the middle of 2023, and the middle of
2024.
Legislative year: Aggregating the fatality counts into
standardized legislative years does a somewhat better job of showing the
policy impacts, but is still not accurate because the standardized years
don’t perfectly line up with when legislation becomes
effective.
For both Calendar and Legislative year plots, the current year shows
the year to date (YTD) count, so it is not strictly comparable to
previous years. The YTD count may rise by the end of the year but it
will not decline. As a result increases from previous years to the
current YTD are meaningful but potentially under-estimate the
difference, while declines observed from previous years to the current
YTD are not meaningful since this year’s total is not complete.
Policy period
The key changes to statewide pursuit law went into effect July 25,
2021 (HB1054, the original statewide policy with the strongest public
safeguards), May 03, 2023 (SB5352, the bill that lowered the threshold
from probable cause to reasonable suspicion) and June 05, 2024 (I2113,
the bill that removed explicit guidelines and replaced them with officer
discretion). This divides our observation period into four parts.
The lengths of the periods are quite different, so direct comparisons
of the counts are misleading.
Pursuit vehicular fatalities
The plot below displays the number of people killed in each period,
broken down by the status of the victim.
Hovering over the bars will bring up number of persons
killed for each segment of the bar.
This graph shows only the “after pursuit” fatalities: incidents where
there has been a pursuit, but the death occurs after the pursuit ended.
These include both homicides and suicides following a pursuit, and the
number of each is shown in the bars. Most of these deaths are caused by
gunshot.
Hovering over the bars will bring up number of persons
killed for each bar segment.
This is a common way to aggregate the fatality counts into
standardized time units, but it does not give a clear picture of the
policy impacts because the laws changed in the middle of 2021, 2023 and
2024.
Note that the current year shows the count for the year to date (YTD)
– January 1 - November 11, 2025 – so it is not strictly comparable to
the other years. The YTD count may rise, but it will not decline. As a
result increases from previous years to the current YTD are meaningful
but potentially under-estimate the difference, while declines observed
from previous years to the current YTD are not meaningful since this
year’s total is not complete.
Pursuit vehicular fatalities
The plot below displays the number of people killed in each calendar
year, broken down by the status of the victim.
Hovering over the bars will bring up number of persons
killed for each bar segment.
This graph shows only the “after pursuit” fatalities: incidents where
there has been a pursuit, but the death occurs after the pursuit ended.
These include both homicides and suicides following a pursuit, and the
number of each is shown in the bars. Most of these deaths are caused by
gunshot.
Hovering over the bars will bring up number of persons
killed for each bar segment.
Aggregating the fatality counts into standardized legislative years
does a better job of showing the policy impacts, but because the
standardized years don’t exactly line up with when legislation becomes
effective the timing is still a bit inacurate.
Note that the current legislative year is year to date – August 1 -
November 11, 2025– so it is not strictly comparable to the other years.
The YTD count may rise but it will not decline. As a result increases
from previous years to the current YTD are meaningful but potentially
under-estimate the difference, while declines observed from previous
years to the current YTD are not meaningful since this year’s total is
not complete.
We exclude data from the 2014-2015 legislative year since our data
observation period starts in January 2015 which does not cover that
whole legislative year.
Pursuit vehicular fatalities
The plot below displays the number of people killed in each
legislative year, broken down by the status of the victim.
Hovering over the bars will bring up the number of persons
killed.
df <- fatalities.pvf %>%
filter(date > as.Date("2015-07-30")) %>% # to get the first complete year
group_by(leg.year, Victim=victim) %>%
summarize(Number = n()) %>%
left_join(allrows.legyr.pvf, .) %>%
mutate(Number = tidyr::replace_na(Number,0)) %>%
group_by(leg.year) %>%
mutate(Total = sum(Number),
leg.year = gsub("Aug|Jul", "", leg.year))
ymax <- max(df$Total)
hb1054 <- max(which(df$leg.year == "2020-2021"))/4
sb5352 <- max(which(df$leg.year == "2022-2023"))/4
i2113 <- max(which(df$leg.year == "2023-2024"))/4
p <- ggplot(df,
aes(x=leg.year, y=Number,
fill=Victim)) +
geom_bar(stat = "identity", color="grey", alpha = 0.8) +
# policy period separators
geom_vline(xintercept = hb1054 + .5, color="grey") +
geom_vline(xintercept = sb5352 + .5, color="grey") +
geom_vline(xintercept = i2113 + .5, color="grey") +
annotate("text",
x = "2017-2018",
y = ymax + 1.5,
label = "Pre-reform", size = 3) +
annotate("text",
x = "2021-2022",
y = ymax + 1.5,
label = "HB1054", size = 3) +
annotate("text",
x = "2023-2024",
y = ymax + 1.5,
label = "SB5253", size = 3) +
annotate("text",
x = "2024-2025",
y = ymax + 1.5,
label = "I2113", size = 3) +
annotate("text",
x = max(df$leg.year), # is a factor
y = df$Total[nrow(df)] + 0.5,
label = "LYTD", size = 3) +
labs(title = "Pursuit vehicular fatalities by legislative year",
x = "Legislative Year",
y = "Number") +
scale_fill_manual(values = cols.pvf.4) +
scale_y_continuous(limits = c(0, ymax+2),
breaks= 1:(ymax + 1)) +
theme(axis.text.x = element_text(size = rel(0.7)))
ggplotly(p) #%>% reverse_legend_labels()
After pursuit fatalities
This graph shows only the “after pursuit” fatalities: incidents where
there has been a pursuit, but the death occurs after the pursuit ended.
These include both homicides and suicides following a pursuit, with the
number of each is shown in the bars. Most of these deaths are caused by
gunshot.
Hovering over the bars will bring up number of persons
killed for each bar segment.
Agencies introduce another counting metric, since multiple agencies
may be involved in a single incident.
In this section we identify every agency involved in at least one
pursuit-related fatality, along with the number and type of fatalities
from the incidents involved (“Each agency count”). This means that
fatalities will be multiply counted for incidents with multiple agency
involvement, and the totals cannot be compared to those in previous
sections. For cross-reference, we also include a graph with the
multi-agency fatalities grouped together (“Multi-agency count”).
By type
Pursuit vehicular fatalities
Each agency count
df <- agency.fatality.p.long %>%
filter(grepl("vehicular", incident.type)) %>%
group_by(agency.type, Victim=victim) %>%
summarise(Number = n()) %>%
group_by(agency.type) %>%
mutate(Total = sum(Number)) %>%
mutate(Percent = Total/nrow(fatalities.pvf)) # percent of fatalities, sum > 100
p <- ggplot(df,
aes(x = reorder(agency.type, Total), y = Number,
fill = Victim,
text = paste("Victim: ", Victim, "<br>",
"Number: ", Number))) +
geom_bar(stat="identity", color = "grey", alpha = 0.8) +
geom_text(aes(y = Total,
label = scales::percent(Percent, acc=1)),
size = 3, nudge_y = 1) +
labs(title = "Pursuit Vehicular Fatalities by Agency Type: Each agency count*",
caption = "WA State since 2015; y-axis=pct, bar label=count",
x = "Agency type",
y = "Number") +
scale_fill_manual(values = cols.pvf.4)
ggplotly(p, tooltip = "text") %>% reverse_legend_labels()
Fatalities are counted more than once if multiple agencies are
involved.
Multi-agency count
df <- fatalities.pvf %>%
group_by(agency.type, Victim=victim) %>%
summarise(Number = n()) %>%
group_by(agency.type) %>%
mutate(Total = sum(Number)) %>%
mutate(Percent = Total/nrow(fatalities.pvf))
p <- ggplot(df,
aes(x = reorder(agency.type, Total), y = Number,
fill = Victim,
text = paste("Victim: ", Victim, "<br>",
"Number: ", Number))) +
geom_bar(stat="identity", color = "grey", alpha = 0.8) +
geom_text(aes(y = Total,
label = scales::percent(Percent, acc=1)),
size = 3, nudge_y = 1) +
labs(title = "Pursuit Vehicular Fatalities by Agency Type: Multi-agency count*",
caption = "WA State since 2015; y-axis=pct, bar label=count",
x = "Agency type",
y = "Number") +
scale_fill_manual(values = cols.pvf.4)
ggplotly(p, tooltip = "text") %>% reverse_legend_labels()
Each fatality only counts once
After pursuit fatalities
These graphs show only the “after pursuit” fatalities: incidents
where there has been a pursuit, but the death occurs after the pursuit
ended. These include both homicides and suicides following a pursuit,
and the number of each is shown in the bars. Most of these deaths are
caused by gunshot.
Fatalities are counted more than once if multiple agencies are
involved.
After pursuit fatalities
This graph shows only the “after pursuit” fatalities: incidents where
there has been a pursuit, but the death occurs after the pursuit ended.
These include both homicides and suicides following a pursuit, and the
number of each is shown in the bars. Most of these deaths are caused by
gunshot.
Fatalities are counted more than once if multiple agencies are
involved.
Geography
By County
Pursuit vehicular fatalities
The plot below displays the number of people killed in each period,
broken down by the status of the victim.
df <- fatalities.pvf %>%
group_by(County=county, Victim=victim) %>%
summarise(Number = n()) %>%
group_by(County) %>%
mutate(Total = sum(Number)) %>%
mutate(Percent = Total/nrow(fatalities.pvf))
p <- ggplot(df,
aes(x = reorder(County, Total), y = Number,
fill = Victim,
text = paste("Victim: ", Victim, "<br>",
"Number: ", Number))) +
geom_bar(stat="identity", color = "grey", alpha = 0.8) +
geom_text(aes(y = Total,
label = scales::percent(Percent, acc=1)),
size = 3, nudge_y = 0.5) +
labs(title = "Pursuit Vehicular Fatalities by County",
caption = "WA State since 2015; y-axis=pct, bar label=count",
x = "County",
y = "Number") +
scale_fill_manual(values = cols.pvf.4) +
scale_y_continuous(breaks = 0:max(df$Total))+
coord_flip()
ggplotly(p, tooltip = "text") %>% reverse_legend_labels()
After pursuit fatalities
This graph shows only the “after pursuit” fatalities: incidents where
there has been a pursuit, but the death occurs after the pursuit ended.
These include both homicides and suicides following a pursuit, and the
number of each is shown in the bars. Most of these deaths are caused by
gunshot.
The plot below displays the number of people killed in each period,
broken down by the status of the victim.
df <- fatalities.pvf %>%
group_by(City=city, Victim=victim) %>%
summarise(Number = n()) %>%
group_by(City) %>%
mutate(Total = sum(Number)) %>%
mutate(Percent = Total/nrow(fatalities.pvf))
p <- ggplot(df,
aes(x = reorder(City, Total), y = Number,
fill = Victim,
text = paste("Victim: ", Victim, "<br>",
"Number: ", Number))) +
geom_bar(stat="identity", color = "grey", alpha = 0.8) +
geom_text(aes(y = Total,
label = scales::percent(Percent, acc=1)),
size = 3, nudge_y = 0.5) +
labs(title = "Pursuit Vehicular Fatalities by City",
caption = "WA State since 2015; y-axis=pct, bar label=count",
x = "City",
y = "Number") +
scale_fill_manual(values = cols.pvf.4) +
scale_y_continuous(breaks = 0:max(df$Total))+
coord_flip()
ggplotly(p, tooltip = "text") %>% reverse_legend_labels()
After pursuit fatalities
This graph shows only the “after pursuit” fatalities: incidents where
there has been a pursuit, but the death occurs after the pursuit ended.
These include both homicides and suicides following a pursuit, and the
number of each is shown in the bars. Most of these deaths are caused by
gunshot.
There are too many cities to be legible if plotted, so we restrict
the cities to those with 2 or more after pursuit fatalities.
The plot below displays the number of people killed in each period,
broken down by the status of the victim.
df <- fatalities.pvf %>%
filter(!is.na(WA_District)) %>%
group_by(WA_District, Victim=victim) %>%
summarise(Number = n()) %>%
group_by(WA_District) %>%
mutate(Total = sum(Number)) %>%
mutate(Percent = Total/nrow(fatalities.pvf))
p <- ggplot(df,
aes(x = reorder(WA_District, Total), y = Number,
fill = Victim,
text = paste("Victim: ", Victim, "<br>",
"Number: ", Number))) +
geom_bar(stat="identity", color = "grey", alpha = 0.8) +
geom_text(aes(y = Total,
label = scales::percent(Percent, acc=1)),
size = 3, nudge_y = 0.5) +
labs(title = "Pursuit Vehicular Fatalities by WA Legislative District",
caption = "WA State since 2015; y-axis=pct, bar label=count",
x = "District",
y = "Number") +
scale_fill_manual(values = cols.pvf.4) +
scale_y_continuous(breaks = 0:max(df$Total))+
coord_flip()
ggplotly(p, tooltip = "text") %>% reverse_legend_labels()
After pursuit fatalities
This graph shows only the “after pursuit” fatalities: incidents where
there has been a pursuit, but the death occurs after the pursuit ended.
These include both homicides and suicides following a pursuit, and the
number of each is shown in the bars. Most of these deaths are caused by
gunshot.
These tables and graphs break down the number of incidents, rather
than persons killed, and we restrict the focus to incidents with active
pursuit fatalities. In active pursuit incidents more than one person may
be killed, so there are fewer incidents than fatalities. The same is not
true for the after pursuit fatalities in our dataset: each of these only
has one fatality. So an “incident-based” analysis is the same as a
“fatality-based” analysis for those incidents, and the results would be
the same as in the previous section.
Recall that we do not have data on all vehicular pursuit incidents –
we only have data on the incidents that result in a fatality.
Fatalities per incident
We only observe multi-fatality incidents for the pursuit vehicular
homicides. There may have been accidents and injuries during the after
pursuit fatalities but we are not capturing those in our data.
* Incidents with pursuit fatalities in WA since 2015
Agencies
Agencies introduce another counting metric, since multiple agencies
may be involved in a single incident. In this section we identify every
agency involved in at least one pursuit-related fatality, along with the
number and type of incidents involved (“Each agency count”). This means
that incidents will be multiply counted when incidents have multiple
agencies involved, and the totals cannot be compared to those in
previous sections. For cross-reference, we also include a graph with the
multi-agency fatalities grouped together (“Multi-agency count”).
By type
Each agency count
df <- agency.incident.p.long %>%
group_by(agency.type, Incident=incident.type) %>%
summarise(Number = n()) %>%
group_by(agency.type) %>%
mutate(Total = sum(Number)) %>%
mutate(Percent = Total/nrow(incidents.p))
p <- ggplot(df,
aes(x = reorder(agency.type, Total), y = Number,
fill = Incident,
text = paste("Incident: ", Incident, "<br>",
"Number: ", Number))) +
geom_bar(stat="identity", color = "grey", alpha = 0.8) +
geom_text(aes(y = Total,
label = scales::percent(Percent, acc=1)),
size = 3, nudge_y = 1) +
labs(title = "Pursuit Incidents by Agency Type: Each agency count*",
caption = "WA State since 2015; y-axis=pct, bar label=count",
x = "Agency type",
y = "Number") +
scale_fill_manual(values = cols.inci)
ggplotly(p, tooltip = "text") %>% reverse_legend_labels()
Incidents are counted more than once if multiple agencies are
involved.
Multi-agency count
df <- incidents.p %>%
group_by(agency.type, Incident=incident.type) %>%
summarise(Number = n()) %>%
group_by(agency.type) %>%
mutate(Total = sum(Number)) %>%
mutate(Percent = Total/nrow(incidents.p))
p <- ggplot(df,
aes(x = reorder(agency.type, Total), y = Number,
fill = Incident,
text = paste("Incident: ", Incident, "<br>",
"Number: ", Number))) +
geom_bar(stat="identity", color = "grey", alpha = 0.8) +
geom_text(aes(y = Total,
label = scales::percent(Percent, acc=1)),
size = 3, nudge_y = 1) +
labs(title = "Pursuit Incidents by Agency Type: Multi-agency count*",
caption = "WA State since 2015; y-axis=pct, bar label=count",
x = "Agency type",
y = "Number") +
scale_fill_manual(values = cols.inci)
ggplotly(p, tooltip = "text") %>% reverse_legend_labels()
Each incident only counts once
By name
For this graph, we report only the “each agency count”
statistics.
For incident counts, in contrast to fatality counts, we present only
calendar and legislative year graphics. Neither provides an accurate
“apples-to-apples” comparison.
Calendar year
This is a common way to aggregate the fatality counts into
standardized time units in a single graphic, but it does not give a
clear picture of the policy impacts because the laws changed in the
middle of 2021 and the middle of 2024.
Aggregating the fatality counts into standardized legislative years
does a somewhat better job of showing the policy impacts, but is still
not accurate because the standardized years don’t exactly line up with
when legislation becomes effective. Note also that the current
legislative year is typically not yet complete, so should be interpreted
with care because the total may rise (it will not decline). Increases
from previous years to the current year are meaningful but potentially
under-estimate the difference, declines observed from previous years to
the current year are not meaningful since this year’s total is not
complete.
df <- incidents.p %>%
filter(date > as.Date("2015-07-30")) %>% # to get the first complete year
group_by(leg.year, incident.type) %>%
summarize(Number = n()) %>%
left_join(allrows.legyr.inci, .) %>%
mutate(Number = tidyr::replace_na(Number,0)) %>%
group_by(leg.year) %>%
mutate(Total = sum(Number),
leg.year = gsub("Aug|Jul", "", leg.year))
ymax <- max(df$Total)
hb1054 <- max(which(df$leg.year == "2020-2021"))/3
sb5352 <- max(which(df$leg.year == "2022-2023"))/3
i2113 <- max(which(df$leg.year == "2023-2024"))/3
p <- ggplot(df,
aes(x=leg.year, y=Number,
fill=incident.type)) +
geom_bar(stat = "identity", color="grey", alpha = 0.8) +
# policy period separators
geom_vline(xintercept = hb1054 + .5, color="grey") +
geom_vline(xintercept = sb5352 + .5, color="grey") +
geom_vline(xintercept = i2113 + .5, color="grey") +
annotate("text",
x = "2017-2018",
y = ymax + 1.5,
label = "Pre-reform", size = 3) +
annotate("text",
x = "2021-2022",
y = ymax + 1.5,
label = "HB1054", size = 3) +
annotate("text",
x = "2023-2024",
y = ymax + 1.5,
label = "SB5253", size = 3) +
annotate("text",
x = "2024-2025",
y = ymax + 1.5,
label = "I2113", size = 3) +
annotate("text",
x = max(df$leg.year), # is a factor
y = df$Total[nrow(df)] + 0.5,
label = "LYTD", size = 3) +
labs(title = "Pursuit vehicular fatalities by legislative year",
x = "Legislative Year",
y = "Number") +
scale_fill_manual(values = cols.inci) +
scale_y_continuous(limits = c(0, ymax+2),
breaks= 1:(ymax + 1)) +
theme(axis.text.x = element_text(size = rel(0.7)))
ggplotly(p) #%>% reverse_legend_labels()
Attempted stops
Attempted stops are defined as incidents when an officer signals a
motorist to stop by activating their lights and/or siren, but the
motorist flees and the officer does not pursue. Whether these incidents
should be considered vehicular pursuits is unclear.
The definition of a pursuit in Washington was also codified into law
by HB 1054 in RCW
10.116.060:
For purposes of this section, “vehicular pursuit” means an attempt
by a uniformed peace officer in a vehicle equipped with emergency lights
and a siren to stop a moving vehicle where the operator of the moving
vehicle appears to be aware that the officer is signaling the operator
to stop the vehicle and the operator of the moving vehicle appears to be
willfully resisting or ignoring the officer’s attempt to stop the
vehicle by increasing vehicle speed, making evasive maneuvers, or
operating the vehicle in a reckless manner that endangers the safety of
the community or the officer.
Note this definition does not require that the officer pursue the
fleeing vehicle. Attempted stops would appear to meet this definition of
pursuit.
There are incidents in our dataset that appear to be attempted stops.
Our ability to identify these depends on what is reported in the
supporting document. This is typically a news report, which in turn
often relies on official police press releases or statements, so there
are multiple opportunities for mis-characterization. It also seems
likely that such incidents are more likely to be ignored or misreported
by the media, so we miss them altogether.
Given the ambiguous status of these “attempted stops”, and the
greater likelihood that they will not be captured accurately in our
data, we have chosen to exclude them from the pursuit analysis
above.
But these incidents are dangerous; they trigger the same fear and
flight reactions in the subject, setting in motion a sequence of events
that can lead to accidents, injuries and fatalities – just like active
pursuits.
One example is an incident in Marysville, WA on August 29, 2020.
According to the
news report, an officer was on her way to another call around 2am
when she saw a motorist doing donuts in the roadway. The officer
activated her emergency lights to signal the motorist to stop, and
allegedly he sped off but she did not pursue him. Within a minute he
killed a man on a bicycle, and then crashed through the bedroom of a
duplex killing a 97 year old woman asleep in her bed.
In this section we present information on the attempted stop
incidents we do observe in our data, noting again the caveat that we may
be missing many of these incidents.
Of the 12 attempted stop incidents we observe in our data, 4 involved
multiple fatalities.
These incidents are shown on the map below. More information on each
case can be obtained by hovering over, or clicking, the incident icon on
the map. The list of fatalities is in the “Table of fatalities” tab,
which can be downloaded.
This report has documented some of the collateral damage caused by
police pursuits. No pursuit is begun with the intention of killing
uninvolved bystanders, but every pursuit brings that risk. Is the risk
worth it? Consider the incidents below.
Kimberly
Winslow was killed when she crashed into a tree while being pursued
by Tacoma police for an alleged speeding violation. Her 9 year old son
was in the car with her at the time, he was injured but
survived.
State
Trooper Justin Robert Schaffer was killed while laying down spike
strips to disable a fleeing vehicle during a pursuit. He was hit by the
fleeing vehicle. The pursuit started when a deputy with the Thurston
County Sheriff’s Office identified a suspect in a shoplifting incident
from the day before and initiated a traffic stop.
Kent
Officer Diego Moreno was also killed laying down spike strips, but
he was hit by the patrol car that was chasing the subject. The pursuit
began when the subjects allegedly fired a gun in the air during a
dispute in a parking lot at a bar and restaurant, and fled when they saw
the police car.
Delilah
Minshew, 6 yrs old, and Timothy Escamilla, 8 yrs old A speeding car
going the wrong way on I-82 caused a head-on collision that killed two
children in the other car. Multiple officers, from both the Washington
State Patrol and the Sunnyside police department had begun pursuits of
this vehicle in the previous hour, but terminated them due to safety
concerns. The driver was charged with DUI, and under the WA state
pursuit policy, reasonable suspicion of DUI allowed for pursuit.
At the time of the crash, the final pursuit (4 in total) had been
terminated 10 minutes earlier. For that reason, this case is classified
as “pursuit-involved” rather than “active pursuit,” and is not included
in the active pursuit vehicular fatality count in the graphic above (see
the next section and the Data section of the report for more details on
definitions). This case is included in the analysis above where
pursuit-involved incidents are examined.
A similar
incident involving a wrong way driver on an interstate highway was
handled differently, and resulted in no fatalities.
Data
Sources
The data we do use
This report is based on a locally maintained dataset focused on WA
State. The data are updated about once each week, pulling, cleaning and
merging data from two active national projects:
In addition we conduct independent searches for incidents in WA state
that these national projects may have missed, using a mix of google
alerts and searches of WA-specific sources like medical examiner
reports, independent investigation reports, social media, obituaries and
public disclosure requests.
The primary data source is MPV, but due to coverage limitations in
the MPV project (primarily the exclusion of most pursuit-related
fatalities) we augment their data with cases from IncarcerNation, legacy
data from Fatal Encounters and locally identified incidents. Cases are
linked and de-duplicated across source datasets, and all records in the
final merged dataset include the original source case IDs.
For the purposes of this report, we call the resulting merged dataset
the “WA local” dataset.
The data used across the different sections of this report includes
all fatal encounters with police that have some vehicle involvement:
pursuits, attempted stops and accidental vehicular homicides on and off
duty (the latter we note but exclude from analysis) . These comprise 34%
of all fatalities in the WA local dataset.
The breakdown of these data by source is shown in the tabs below.
Many incidents are included in more than one dataset, some are included
only in one, and there are multiple ways to present the coverage of each
source and pattern of overlap.
Source coverage
The coverage of each data source can be assessed by the presence of
the source ID tags by year, shown below. The columns for each data
source show the number of cases with that source ID tag. The “Total
Cases” column in the table represents the total cases in that year, not
the sum across the data sources. Comparing the number of cases in each
of the Data Source columns to each other and the total gives a sense of
what fraction of that year’s cases were covered by each source, the
overlap across the sources, and how that varies by year.
fatalities.vehicle.related <- fatalities.all %>%
filter(!grepl("other", incident.type))
tab <- fatalities.vehicle.related %>%
group_by(Year = as.character(year)) %>%
summarize(FE = sum(!is.na(feID)),
IN = sum(!is.na(inID)),
MPV = sum(!is.na(mpvID)),
WaPo = sum(!is.na(wapoID)),
WaLocal = sum(!is.na(waID)),
Total.Cases = n()
) %>%
mutate(Percent = round(100*Total.Cases/sum(Total.Cases), 1)) %>%
add_row(Year = "Total", summarise(., across(where(is.numeric), sum)))
tab %>%
kable(caption = "Source Coverage by Year") %>%
kable_styling(bootstrap_options = c("striped","hover")) %>%
add_header_above(header = c(" " = 1, "Fatalities covered by each data source" = 5, "De-duplicated yearly totals"=2)) %>%
row_spec(dim(tab)[1], bold = T) %>%
column_spec(which(names(tab)=="Cases"), bold=T) %>%
column_spec(which(names(tab)=="Percent"), italic=T) %>%
add_footnote(label = "Percents may not sum to 100 due to rounding",
notation = "symbol")
Source Coverage by Year
Fatalities covered by each data source
De-duplicated yearly totals
Year
FE
IN
MPV
WaPo
WaLocal
Total.Cases
Percent
2015
3
0
3
2
1
4
2.4
2016
7
0
4
4
1
8
4.7
2017
13
0
6
6
0
13
7.7
2018
10
0
5
4
0
10
5.9
2019
15
0
5
4
0
15
8.9
2020
16
0
6
5
1
17
10.1
2021
12
12
3
3
0
12
7.1
2022
0
15
10
10
21
21
12.4
2023
0
17
5
5
19
19
11.2
2024
0
24
3
2
26
26
15.4
2025
0
20
6
0
24
24
14.2
Total
76
88
56
45
93
169
100.0
* Percents may not sum to 100 due to rounding
Source overlap
Each case in the dataset can also be identified by the number of
sources that include it, a direct measure of the overlapping sources of
information. The breakdown of source overlap by year is shown below.
tab <- fatalities.vehicle.related %>%
count(Year = as.character(year), source.overlap) %>%
pivot_wider(names_from=source.overlap, values_from=n) %>%
mutate(across(where(is.numeric), ~replace_na(.,0))) %>%
rowwise() %>%
mutate(Total = sum(c_across(where(is.numeric)))) %>%
ungroup() %>%
add_row(Year = "Total", summarise(., across(where(is.numeric), sum)))
tab %>%
kableExtra::kbl(caption = "Breakdown of fatalities by number of data sources and year") %>%
kable_styling(bootstrap_options = c("striped", "hover")) %>%
add_header_above(header = c(" " = 1, "Number of data source(s) for each fatality *" = 4, " " = 1)) %>%
row_spec(which(tab$Year == "Total"), bold=T) %>%
add_footnote(label = "Most records in MPV and WA can be found in multiple sources, see other tabs for details on the overlap",
notation = "symbol")
Breakdown of fatalities by number of data sources and year
Number of data source(s) for each fatality *
Year
1
2
3
4
Total
2015
1
1
2
0
4
2016
1
0
7
0
8
2017
0
0
13
0
13
2018
0
1
9
0
10
2019
0
1
14
0
15
2020
1
1
15
0
17
2021
0
0
0
12
12
2022
7
4
1
9
21
2023
4
10
1
4
19
2024
8
15
1
2
26
2025
9
9
6
0
24
Total
31
42
69
27
169
* Most records in MPV and WA can be found in multiple
sources, see other tabs for details on the overlap
Primary source
When MPV includes an incident, that is our starting point for that
case. Incidents from the other datasets are incorporated only when they
fall outside of MPV’s inclusion criteria. In the table below, we show
the primary source of cases by year: all cases covered by MPV
are listed in that column, the additional cases contributed from FE
(2015-21), IN (2021+) and WA local (2022+) are shown in the other
columns.
tab <- fatalities.vehicle.related %>%
mutate(Source = factor(
case_when(
!is.na(mpvID) ~ "MPV",
!is.na(inSource) ~ "IN",
!is.na(feID) ~ "FE",
!is.na(waID) ~ "WA local",
TRUE ~ "other"),
levels = c("MPV", "FE", "IN", "WA local"))
) %>%
count(Year = as.character(year), Source) %>%
pivot_wider(names_from=Source, values_from=n) %>%
mutate(across(where(is.numeric), ~replace_na(.,0))) %>%
rowwise() %>%
mutate(Total = sum(c_across(where(is.numeric)))) %>%
ungroup() %>%
add_row(Year = "Total", summarise(., across(where(is.numeric), sum)))
tab %>%
kableExtra::kbl(caption = "Primary data source by year") %>%
kable_styling(bootstrap_options = c("striped", "hover")) %>%
add_header_above(header = c(" " = 1, "Primary data source for each fatality *" = 4, " " = 1)) %>%
row_spec(which(tab$Year == "Total"), bold=T) %>%
add_footnote(label = "Most records in MPV and WA can be found in multiple sources, see other tabs for the overlap",
notation = "symbol")
Primary data source by year
Primary data source for each fatality *
Year
MPV
WA local
FE
IN
Total
2015
3
1
0
0
4
2016
4
1
3
0
8
2017
6
0
7
0
13
2018
5
0
5
0
10
2019
5
0
10
0
15
2020
6
1
10
0
17
2021
3
0
9
0
12
2022
10
9
0
2
21
2023
5
11
0
3
19
2024
3
17
0
6
26
2025
6
13
0
5
24
Total
56
53
44
16
169
* Most records in MPV and WA can be found in multiple
sources, see other tabs for the overlap
Source dataset descriptions
Mapping Police Violence
Data: 2013+
Mapping Police Violence
(MPV) aims to establish a comprehensive national open-source dataset
of homicides by police, on and off duty. It includes fatal shootings and
fatalities caused by tasers, physical beatings, restraint holds,
chemical spray, police dog attacks, less-lethal force and vehicular
homicides. Vehicular homicides are included in MPV only when
the fatalities are directly caused by a police action: intentionally
running someone over, and intentional collisions and vehicle disabling
during a pursuit that leads to a fatal accident. As they note on their
methodology
page, these criteria result in the exclusion of most vehicle
pursuit-related fatalities (however they also note that they have begun
to track these pursuits internally, so their coverage may change in the
future). Given this restriction, MPV coverage of the cases in this
report is largely restricted to after-pursuit-homicides.
MPV has incorporated historical data from two legacy projects:
MPV merged and deduplicated the incidents from these two datasets,
removed incidents that did not meet their inclusion criteria, added
eligible incidents that were missed by both projects, kept the
information from both datasets on the included incidents, and is now
augmenting all incidents (back through 2013) with additional
information, as well as updating with new incidents.
MPV is a project of the Campaign
Zero organization. They follow the principles of transparent
reproducible research, with a systematic, multi-level review and coding
process that is published on their methodology
page (recommended reading), and datasets provided in downloadable
form to the public.
IncarcerNation (IN) maintains a comprehensive, searchable website
listing every fatality that has occurred during any police encounter in
the United States, regardless of type, and without judgment. This
includes deaths from direct police use of force [gunshots, tasers, K9,
and other immediate lethal police action], as well as fatalities linked
to police activities and interactions controlled by police that lead to
death. These encompass high-speed pursuits (including deaths of drivers,
passengers, bystanders, and pedestrians), collisions involving police
vehicles, drownings during police chases, falls from heights during
pursuit, asphyxiation during restraint, and suicides occurring in the
context of police operations, often during SWAT incidents. (source:
IncarcerNation’s “About
Us” page).
IN’s inclusion criteria are therefore much broader than MPV’s, and
they are the only active project with coverage equivalent to the legacy
Fatal Encounters project. While most of their incidents overlap with
MPV, they are the unique source for pursuit-related fatalities.
IN does not make their dataset downloadable for the public, so we
manually review and transcribe information for the cases they list in
Washington state. During transcription, duplicate cases, cases from
Washington DC and cases that turn out to be non-police related homicides
are excluded. Many of the IN cases are also covered by MPV and the local
WA project, but when IN is the primary source of information, we note
that.
Fatal Encounters (FE) is a legacy project that sought to establish a
comprehensive national open-source dataset of all deaths during
encounters with police. This project was active through December 31,
2021, includes records going back to 2000, and produced the largest
compilation of persons killed during encounters with police that is
publicly available: 31,641 fatalities at the time the project ended. It
covers deaths by all causes during a police encounter, including
suicides, accidents, killings by off-duty police and vehicular pursuit
related fatalities. The vehicular pursuit coverage in FE is broader than
MPV, as it includes all fatalities from accidents during a pursuit, so
it captures bystanders killed by the fleeing subject.
Legacy FE data are captured by the dataset used in this report in two
ways:
Cases included in the MPV dataset.
Cases excluded by MPV are recovered and included by us, and are
used in this report.
We made many corrections to the FE data and harmonized the variables
while FE was still updating. The scripts for that can be found on our
legacy GitHub repository: https://github.com/nextstepswa/fewapo. We now just read
in the cleaned, harmonized data (included in our current GitHub
repository).
Washington Post
Data: 2015-2025
The Washington Post Fatal Force (WaPo) dataset is a legacy project
that sought to create a national database of fatal shootings by
on-duty police. It was actively updating from January 1, 2015 through
December 31, 2024. There were 10,429 fatalities recorded at the time the
project ended. Given the restriction to fatal shootings, WaPo coverage
of the cases in this report is limited to after-pursuit homicides.
WaPo historical data has been incorporated by the MPV project into
their database, and is integrated into the WA local dataset via
MPV.
For all of our data sources, it is worth noting that:
None include deaths in custody after booking – so all deaths in
jails and prisons are excluded.
All sources may be missing incidents.
A comparison of the cases found in each dataset is presented in the
next section of this report.
There has typically been a 1-2 week lag in MPV updates, but it can
sometimes be longer. The date shown in the “most recent data update” tab
above is an indication of the current lag. As of this report, the last
data updates were:
October 31 2025 for Mapping Police Violence
November 04 2025 for IncarcerNation
November 04 2025 for the WA local data
The data we don’t use
The official government source of data for police pursuit fatalities
is the Fatality
Analysis Reporting System (FARS) maintained by the National Highway
Traffic Safety Agency (NHTSA). Locally, the Washington State Traffic
Safety Commission (WTSC) collects and codes these data and sends them to
the national program. We do not use these data for several reasons:
The official data have a long publication
lag.
Both the WTSC and FARS datasets are only updated once a year, and
those updates can take several months to a year to be released. This
time lag makes it impossible to use these datasets to assess impacts of
recent policy changes. Note that in Washington, real time police traffic
collision reports (PTCRs) can be accessed through the Washington State
Patrol Collision Analysis Tool here, but
these do not identify pursuit-related incidents.
The official data do not include after-pursuit
fatalities.
FARS is a program designed to provide data on fatalities from vehicle
accidents only. After-pursuit fatalities are not captured by this
system.
The official data likely undercount pursuit vehicular
fatalities.
The WTSC and FARS were not originally designed to provide accurate
data on pursuit fatalities. The “police pursuit involved” tag was added
in 1994 to FARS (p. 91,
FARS Users Manual). The way in which this is captured likely varies
across the states. Here in WA, the state WTSC tags an incident as
pursuit-related through a labor-intensive manual coding of the law
enforcement narrative documents submitted with the Police Traffic
Collision Report (PTCR). If the pursuit is not mentioned in the
narrative, if there is no narrative, or if the coder fails to tag a
case, a pursuit incident will fail to be identified. The likelihood that
this results in under-identification of pursuit-related fatalties was
acknowledged by the Research Director of the WTSC in
a recent interview.
Incomplete coverage of fatalities related to police activities is
unfortunately the norm in official datasets. Homicides by police have
been found to be undercounted by about 50% in official government data
sources like the Arrest
Related Death Program, the National
Vital Statistics System and the Uniform
Crime Reports. The most recent analysis to have replicated this
finding is from the Institute for Health Metrics and Evaluation at the
UW in 2021, published in the Lancet, one of the most respected
peer-reviewed science publications in the world. You can access that
paper online here.
It is worth noting that the undercounts were documented and
quantified by comparing official data to open-source data like MPV and
Fatal Encounters. The open source methodology is now recommended
even by federal agencies as a way to improve official data
collection.
The official data do not provide key information on the
individual incidents.
FARS also does not include information on the law enforcement
agenc(ies) involved in the pursuit. The local WTSC data has information
on the agency that reported the incident and the agency that is
investigating the incident, which may or may not be the agency involved
in the pursuit.
By contrast, detailed incident-based information is
available in the data from Fatal Encounters, MPV and Incarcernation. The
document link provided for each case in these datasets contains a wealth
of additional contextual information, and serves as a springboard for
further research on individual cases. These media reports also help to
remind us that each of these “cases” is a person, someone from our
community, and not just a number.
The data sources we rely on are not perfect. They also likely
undercount the true number of fatalities associated with pursuits, as
the search methods used by the projects rely on the incidents leaving a
digital signature online, and not all pursuit fatalities will be
reported this way. Our data sources do, however, provide more complete
information, on a wider range of cases, and in a more timely manner,
than the official data.
The number of fatalities we report here gives a lower bound
on the true number vehicle pursuit-related fatalities. There may be
more, but there will not be less.
Coding
The original data sources are not designed to facilitate direct
analysis of vehicular pursuits. So we have identified and coded these
cases by hand. We used two independent coders for all records in
Washington State from January 2015 to December 2023 to establish a
reliable rubric. Cases since then are typically coded by one person.
Data cleaning
We begin with a dataset that merges information on fatalities in WA
state from the MPV, FE, WaPo, IN and the WA local datasets. The data are
cleaned and new variables constructed to facilitate analysis. Any errors
found are corrected. Updates to the cleaned dataset are made a few times
a month.
More information can be found in this report, which also
has links to the GitHub repository that hosts all of the basic data
cleaning and analysis scripts.
Case inclusion/exclusion in this report
The source dataset keeps track of all persons killed during police
encounters, regardless of the cause of death. For this pursuit report we
include/exclude the following types of cases:
Inclusion
We retain vehicular homicides committed by the driver fleeing a
police pursuit. This is a key contributor to the risks that police
pursuits pose to the general public.
We retain after-pursuit suicides. This term refers to a person
taking their own life with a gun, not to “suicide by cop.” The
classification is based on the media report, which typically relies on a
coroner’s or medical examiner’s determination. These cases are
identified separately in the report tables and graphics.
Exclusion
We exclude fatalities from non-pursuit related on- or off-duty
officer accidents (n = 8). It is worth noting that some of these have
led to substantial public controversy and dismissal of the officer
involved.
We exclude fatalities resulting from encounters with police that
have no pursuit involvement (n = 328).
Pursuit review
All cases in the merged WA local dataset were reviewed for evidence
of pursuit involvement and coded into categories.
We reviewed all evidence available in the original source datasets,
including variables coded by FE, IN, MPV and WaPo and the link to the
original verification document (typically a news article). We conducted
additional online searches to find other media reports, coroner/ME
reports, social media posts, and Law Enforcement Agency announcements.
Using this information we classified cases into the following
categories:
Active Pursuit – vehicular homicides from an active
pursuit (e.g., a crash-related fatality, or someone getting run
over).
Terminated pursuit – vehicular homicides from an active
pursuit that was reportedly terminated just before the accident
occurred. “Just before” means within a minute or a mile.
Involved pursuit – a vehicle pursuit occurred during
the incident, but it had finished before the person was killed.
Typically the person killed in these cases is a fleeing subject who is
either shot by police at the end of the chase, or their death is ruled a
suicide due to a self-inflicted gunshot. Less commonly, if police
terminate a pursuit the subject may continue to flee for some time, and
cause a vehicular homicide that happens well after the end of the
pursuit. In these cases the fatality may be an uninvolved bystander or
passenger.
Attempted stop – vehicular homicides that occur when the
subject flees a traffic stop but is not pursued. We often have to rely
on a media report of the law enforcement statement that there was no
pursuit, but the patrol car lights and/or siren may have been
activated.
Vehicle accident – vehicular homicides by police that
are not related in any way to a pursuit (e.g., an on- or off-duty
officer ran someone over and killed them).
Not pursuit related – all other incidents we reviewed
and determined were not related to vehicle
stops/accidents/pursuits.
Final categories
We grouped the active and terminated pursuits together into the
“Pursuit vehicular homicide” category, and split the involved pursuits
into “After pursuit homicides” and “After pursuit suicides”.
* Table data include all fatalities during a police
encounter.
The majority of the report focuses on pursuit vehicular fatalities
and after-pursuit fatalities. Fatalities from attempted stops are
examined in a separate section. Vehicular accidents and other deadly
force fatalities are excluded.
Coding vehicular fatality victim types
Vehicular fatalities are defined as cases where the cause of death is
a vehicle. This includes fatalities from active and terminated pursuits,
attempted stops and vehicle accidents. There are also a few “Involved
pursuit” incidents where a vehicle is the cause of death. In one the
subject’s car was stopped after a pursuit and he fled by foot onto
Interstate 5 where he was hit and killed. In another the officer was
making a u-turn at the start of the pursuit, and the subject crashed
into them and died.
For each vehicular fatality we coded
The status of the person(s) killed: subject (the driver of the
vehicle being pursued), passenger (in the fleeing vehicle), bystander or
officer.
Whether there were any additional injuries noted in the
supporting documents (again coding status of the injured
person).
What we miss
Our ability to identify and track these incidents is not perfect. The
two key limitations are:
This is a fatality-based dataset, not a dataset of pursuits.
The data we have only includes pursuits from incidents that
result in at least one fatality – so:
We miss pursuits that result in accidents that cause injuries
and/or property damage only
We miss pursuits that end without accidents or
fatalities
For this reason, we are not able to estimate the total number of
pursuits, or the fraction of pursuits that lead to fatalities, injuries
or property damage.
We rely largely on online documents to find and code these
cases.
We miss incidents that do not leave an online trace, or if our
search methods do not capture that trace
We miss incidents if the document reports the fatality but fails
to report on the pursuit
We may misclassify a case if the document has not included the
necessary information
We rely on the level of detail reported in the online documents to
find these cases. Those documents, in turn, typically rely on a press
release or social media post from the involved law enforcement agency.
The law enforcement description of the event is rarely verified with
independent sources, unless it becomes a high profile case and comes
under scrutiny.
Ambiguous cases
In 2 incidents, the information available in the media report left it
somewhat unclear whether the incident should be classified as a pursuit
or as an attempted stop: Robert Bray (9/25/2022) and Stephanie Laguardia
(3/16/2022). We coded these as pursuits, given the balance of the
evidence. If these cases were instead coded as attempted stops, they
would be removed from the pursuit vehicular fatality count. Since both
of these cases occurred during the HB1054 period, this would reduce both
the number of fatalities and the annualized fatality rate during that
period, and would mean that we have underestimated the protective effect
of the pursuit safety policy.
In 1 incident, the events lay on the boundary of the pursuit
category: Sergey Pavlovich (6/28/2019). Here the officer reportedly was
making a U-turn in response to a 911 call for speeding motorcycles, and
saw the motorcycles approaching him at the time. Pavlovich crashed his
motorcycle into the patrol car and was killed. It is unclear whether the
officer intended to give pursuit. We coded this case as
“pursuit-involved”, but it could also be considered a “Vehicular
accident”. Since neither classification is included in the cases used to
estimate policy impact the decision has no effect on those
estimates.
In 1 incident, an unusual set of events could be construed as either
“pursuit-involved” or “terminated pursuit”. This incident involved a
speeding motorist who was actively pursued multiple times over the
course of an hour by 4 different state patrol troopers and by local
police. In each instance, the pursuit was terminated for safety reasons,
as the motorist exceeded 100 mph and was weaving in and out of traffic
on both highways and city streets. The last pursuit was terminated 10
minutes before the motorist, still travelling at high speed and headed
the wrong way down a highway exit ramp, crashed into another vehicle,
killing 2 children in that vehicle (Delilah Minshew and Timothy
Escamilla, 2/28/2023). Since the pursuit was terminated 10 minutes
before the crash, and our definition of terminated pursuits requires
termination within a minute or a mile of the vehicular homicide, we
classified this incident as “pursuit-involved”. If this were classified
instead as a “terminated pursuit”, the definition change would require
reclassifying other cases for consistency, and without doing that the
change in the estimated policy impacts can not be calculated.
Download the data
Persons killed
This table lists all fatalities in the dataset, with some basic
information, the pursuit related coding for this incident, and a link to
a verification document for further details. This document can be used
as a starting place for research on this case.
“LegDist” is the WA legislative district. “Type” is the detailed
pursuit classification for each case.
Buttons for downloading the data are at the top of the table. The
table columns can be filtered and sorted as well.
This table lists the pursuit-related incidents that resulted in
fatalities, with information on the agency (or agencies) involved,
fatalities and injuries. Pursuit-related incidents include pursuit
vehicular fatalities from active and terminated pursuits and
after-pursuit homicides and suicides.
Buttons for downloading the data are at the top of the table, and the
columns can be filtered and sorted.
This report follows the principles of transparent, reproducible
science. All of the scripts, data and software are open source and
freely available to the public, online.
The software used to produce this report comes from the open-source
R project. You can view the
code in the report itself by clicking the Code buttons in
the right margin, or you can download all of the scripts from the
repository.
R is free and runs on most Unix, Windows and MacOS
operating systems. The report script uses rmarkdown and knitr.
If you find a bug in our code, please let us know by
posting it as an issue in the repository.
If you have a question about the methodology, or the
data, or the topic in general, please start a thread on the repository
discussions page (note: you’ll need a GitHub account – they’re free,
just follow the instructions at the link).
____
Comments
There are different methods one can use to assess the impacts of the sequence of pursuit policy changes in WA. But it is complicated by the fact that the lengths of the policy periods are quite different, the dates the policies became effective do not line up with any type of standardized year (calendar or legislative), and the fatality counts for the shorter policy periods are too small to support a robust statistical analysis. We previously reported a fatality count comparison with a standardized time window, pre and post reform. But with 4, different length, policy periods, that approach no longer works.
For this policy impact summary section we have decided to report annualized fatality rates for each policy period. This metric has both benefits and drawbacks:
Benefits – It provides a single summary number that can be compared across all periods, regardless of their length, and it allows us to use the exact dates when legislation became effective to define the periods.
Drawbacks – In contrast to the number of persons killed in a specific legislative year (or other standardized time frame), the annualized fatality rate is an estimate of the average rate of fatalities per year. The reliability of that estimate depends on the number of years observed. In our case the policies changed so frequently that several periods are quite short, which means the annualized estimate we get is not very reliable.
In later sections of this report we present fatalities broken down by calender and approximate legislative year, showing the breakdown of the victims (officer, bystanders, passengers, subjects) in each timeframe.