8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
1/78
U.S. Department of Justice
Ofce of Justice Programs
Bureau of Justice Statistics
T E C H N I C A L R E P O R T
requirements ended in 2006, BJS undertook efforts tounderstand the variability between SRCs, and within SRCsover time, in terms o data collection methodologies andavailable resources. Tis variability has led to concerns aboutdefinitions, data quality, and undercoverage error.
Tis report responds to issues with variability and profilesthe data quality o the ARD component o the DCRP.Te report (1) gives an overview o the ARD program,(2) describes the data collection methods used by SRCs, and(3) assesses the quality o the resulting data and coverage oreligible ARD cases that involve law enorcement homicides.
Te variation in methodologies resulted in a significantunderestimate o the annual number o arrest-related deaths,including both homicides by law enorcement officers andother types o civilian deaths. Key findings include—
Nationally, the ARD program captured about 50% o theestimated law enorcement homicides during 2003–09and 2011. Tis assessment did not examine the coverageo other deaths the ARD program is designed to capture,including those due to accidents, drug overdoses, andnatural causes.
Te national coverage o homicides by law enorcementcaptured in the ARD program improved rom 2003through 2011. Tis was partially due to the use o opensource searches to identiy potential cases that wereeligible or inclusion. Even with this improvement,assessments indicate that rom 31% to 41% o theestimated homicides by law enorcement personnel werenot captured in the 2011 ARD data collection.
Significant challenges exist because o the lack ostandardized modes or data collection, definitions, scope,participation, and the availability o resources.
Executive summary
Te Arrest-Related Deaths (ARD) program is an annualnational census o persons who die either during theprocess o arrest or while in the custody o state or local
law enorcement personnel. Te Bureau o Justice Statistics(BJS) implemented the ARD program in 2003 as part o theDeaths in Custody Reporting Program (DCRP). Te DCRPwas initiated to ulfill the data collection requirement o theDeaths in Custody Reporting Act o 2000 (DICRA, P.L. 106-247). It collects in-depth inormation on deaths during arrestand incarceration, and it provides national-level inormationon the deaths o suspects and offenders rom their initialcontact with law enorcement personnel through the timethey are incarcerated in a jail or prison.
ARD data are collected to quantiy and describe thecircumstances surrounding civilian deaths that take place
during an arrest or while in the custody o law enorcement.Tese data describe the prevalence and incidence o arrest-related deaths across the nation, identiy the circumstancesor activities that contribute to these deaths, and revealtrends in the causes and circumstances o these deaths incustody at national and state levels. Tese data can be usedto inorm specific policies that may increase the saety olaw enorcement officers and citizens, identiy trainingneeds in law enorcement agencies, and assist in developingprevention strategies.
Te current ARD program relies on state reportingcoordinators (SRCs) in each o the 50 states and the District
o Columbia to identiy and report on all eligible cases oarrest-related deaths. BJS compiles data rom the states toproduce national-level statistics on deaths that occur in theprocess o arrest by, or while in the custody o, state and locallaw enorcement personnel. When the DICRA reporting
March 2015, NCJ 248544
Arrest-Related Deaths Program:
Data Quality ProfileMichael Planty and Andrea M. Burch, Bureau of Justice StatisticsDuren Banks, Lance Couzens, Caroline Blanton, and Devon Cribb, RTI International
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
2/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 2
correctional acilities (2002). In 2003, BJS expanded theDeaths in Custody Reporting Program (DCRP) to includedeaths that occur in the process o arrest by state and local lawenorcement agencies.
Te DCRP provides data that describe the prevalence andincidence o deaths that occurred while the decedent was inthe custody o law enorcement personnel, local jails, or stateprisons. Tese data collections provide policymakers, public
health officials, law enorcement officials, and correctionaladministrators with a comprehensive system or monitoringdeaths within three major components o the criminal justicesystem: the police, local jails, and corrections. Te DCRPcollections are the only source o this type o inormation atnational, state, and local levels.
Prior to the DCRP, prison mortality data were collected at thenational level on a limited number o deaths (i.e., homicide,suicide, natural causes) and at levels o aggregation thatprecluded analysis o subpopulations. Jail mortality data werecollected at inrequent intervals through BJS’s Census of Jails series; however, like prison mortality data, these data were
limited to broad categories o causes and provided limitedinormation on the demographic characteristics o decedents.Deaths in juvenile correctional acilities were collectedperiodically by the Office o Juvenile Justice and DelinquencyPrevention (https://www.ncjrs.gov/pdffiles1/ojjdp/228128.pd).No national collection o arrest-related deaths, as describedby the DICRA legislation, had ever been completed.1 Arrest-related mortality data were restricted to small local studies,mostly at the agency level, and ocused predominantly onofficer-involved homicides (Klinger, 2012; Lofin, Wiersema,McDowall, & Dobrin, 2003).
With inormation collected through the DCRP, BJS can trackchanges in the mortality rates o persons who have contactwith the police or are in custody in jail or prison. Te DCRPdata have improved our understanding o civilian deaths whilein the criminal justice system. For example, through the DCRP,BJS has shown that almost 40% o deaths that occur in theprocess o arrest arise rom causes other than officer-involvedhomicides o suspects (Mumola, 2007; Burch, 2011); that inany given year, about 80% o the roughly 3,000 jails in theUnited States have no deaths and, among those jails reportingdeaths, the modal count is one death; that the leading cause odeath in prisons is cancer, ollowed by heart disease; and thatmost inmates who died in custody had medical conditionsthat predated their arrival to prison or jail, as opposed to
1wo national-level programs, the Uniorm Crime Reports–Supplementary Homicide Reports (SHR) and National Vital StatisticsSystem (NVSS), collect inormation on homicides by law enorcement,but neither the SHR nor the NVSS captures the ull scope o deathsspecified by the DICRA legislation. Te SHR data are limited to justifiable homicides by law enorcement. Te NVSS data identiy deathsattributed to legal intervention. Deaths that do not directly result romlaw enorcement action but still occur during the process o arrest (e.g., vehicular deaths resulting rom pursuits, suspects dying rom medicalconditions, and suicides during standoff situations) are not identified inthe NVSS as legal intervention deaths.
1. Program overview
1.1 Program purpose and justification
Te Deaths in Custody Reporting Act o 2000 (DICRA; P.L.106-297), signed into law on October 13, 2000, required thatstates applying or Violent Offender Incarceration/ruth-in-Sentencing (VOI/IS) grants ollow the U.S. Department oJustice guidelines or a quarterly data collection o inmate
deaths rom all state prisons (public and private), local jails, juvenile detention acilities, and police lockups, as well asdeaths occurring in transit and in the process o arrest. UnderDICRA, state applicants or VOI/IS grants were required toprovide assurance o
“…reporting, on a quarterly basis, inormation regardingthe death o any person who is in the process o arrest, is enroute to be incarcerated, or is incarcerated at a municipal orcounty jail, State prison, or other local or State correctionalacility (including any juvenile acility) that, at minimum,includes—
the name, sex, race, ethnicity, and age o the deceased the date, time, and location o death
a brie description o the circumstances surroundingthe death.”
Te Bureau o Justice Statistics (BJS) implemented thedata collection requirement under DICRA legislation. Incompliance with the specifications outlined in P.L. 106-297,BJS began quarterly data collections to cover all inmatedeaths in local jails (2000), state prisons (2001), and juvenile
HIGHLIGHTS
The Arrest-Related Deaths (ARD) program collectsinformation on deaths occurring in the process of arrest.
An arrest-related death is defined as any death (e.g., gunshotwound, cardiac arrest, or drowning) that occurs during an
interaction with state or local law enforcement personnel,including those that occur—
• during an attempted arrest or in the process of arrest
• while the person is in law enforcement custody (before
transfer to jail)
• shortly after the person’s freedom to leave is restricted.
Exclusions include—
• deaths of bystanders, hostages, and law enforcement
personnel
• deaths occurring during an interaction with federal law
enforcement agents
• deaths of wanted criminal suspects before police contact
• deaths by vehicular pursuits without any direct
police action.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
3/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 3
program when the deaths are attributed to events that occurwhile the person is engaged with law enorcement.
Te most common types o noncriminal deaths reportableto the ARD program occur during law enorcement’sresponse to service calls or assistance or transport. Ofen,these service calls are civilian requests or mental health ormedical assistance. Tese incidents may involve individualswho are suicidal or displaying erratic behavior or may occur
during transport to a mental health acility or hospital. Lawenorcement may have engaged with the decedent to acilitatecare during an emergency, assist with involuntary commitmentto a medical acility, or support emergency medical technicianswhile transporting an individual or continued care.
During an attempted arrest or in the process o arrest—Arrest-related deaths include all deaths that occur during aninteraction with law enorcement personnel in the processo arrest or attempted arrest, regardless o whether physicalcustody was established beore the death. Deaths occurringbeore arrest include those that are attributed to events thattranspired either during apprehension or while the decedent
was detained or questioning or investigation.Deaths that occur while law enorcement personnel attemptto apprehend or arrest an individual (e.g., those that occurduring oot or vehicle pursuits o criminal suspects and duringstandoffs or barricade situations) are reportable. Commonexamples o deaths that occur during apprehension includeofficer-involved shootings; deaths related to the use o orceor law enorcement compliance weapons3 or tactics; vehicleaccidents and collisions caused by either the decedent,intervening law enorcement personnel, or unrelated civilians;other types o atal accidental injuries sustained whileattempting to elude law enorcement personnel, such as allsrom heights and drowning; and suicides committed duringstandoffs and barricade situations.
Deaths that occur during interviews and interrogations,or while a criminal suspect is detained or questioning orinvestigation, are also within the scope o the ARD programand should be reported. Tese arrest-related deaths may takeplace at a law enorcement acility or in the field. Examples othese types o deaths include those attributed to alcohol anddrug intoxications, sudden medical conditions (e.g., cardiacarrest, asthma, stroke, or seizure), choking on ingested objectsor other orms o asphyxiation, and suicides.
In addition to deaths that occur beore a physical arrest, deaths
that occur while law enorcement attempts to establish physicalcustody o a suspect are also reportable to the ARD program.All deaths that are attributed to weapon use by state or locallaw enorcement personnel are considered arrest relatedand are reported. Common examples o deaths attributed tolaw enorcement’s use o weapons include officer-involvedshootings, complications related to the use o conducted
3Compliance or less-than-lethal weapons are designed to induce offendercompliance while minimizing accidental, incidental, and correlativecasualties associated with weapon use.
contracting a atal disease while in custody (Noonan & Ginder,2013). Since its initiation in 2003, the ARD program hasreleased two reports on the number and characteristics oarrest-related deaths (Burch, 2011; Mumola, 2007).
In 2006, the DICRA reporting requirements expired. BJScontinued to collect, analyze, and report on deaths in custody.In December 2014, the Death in Custody Reporting Act o2013 was signed into law (P.L. 113-242, see https://www.
congress.gov/bill/113th-congress/house-bill/1447/text).
1.2 Definitions
Te DICRA legislation required states to submit quarterly dataon deaths that occur in the process o arrest. BJS designedthe ARD program’s scope to include homicides, suicides,accidental deaths, and deaths attributed to intoxication andmedical conditions that occur during a civilian interactionwith state or local law enorcement.
BJS operationalized the DICRA term “arrest-related” todescribe the range o circumstances pertaining to an arrest,including those that occur rom the process o apprehensionto detention. BJS defined a death as arrest related when theevent causing the death (e.g., gunshot wound, cardiac arrest,or drowning) occurs during an interaction with state or locallaw enorcement personnel.2 Tese events can be groupedinto three categories: deaths that occur (1) while a person isdetained or restricted and shortly afer the person’s reedomto leave is restricted, (2) during an attempted arrest or in theprocess o arrest, or (3) while the person is in custody (beoretranser to jail). Exclusions to this definition are described insection 1.3.
Shortly afer reedom to leave is restricted—All deaths that
occur shortly afer a person’s reedom to leave is restricted bystate or local law enorcement personnel should be reported tothe ARD program i the circumstances causing the death occurduring the interaction with law enorcement. For example, i adetained individual sustained an injury during an interactionwith law enorcement personnel and died later as a result othose injuries, the death should be reported to the program.
Although the majority o deaths reported to the ARD programinvolve criminal suspects, individuals not considered subjectso arrest can be detained by law enorcement personnel,such as during a stop-and-risk search or during a vehiclestop. In addition, individuals may be in the custody o law
enorcement or medical or mental health assistance. Deaths ononcriminal persons that occur in the custody o state or locallaw enorcement personnel should be reported to the ARD
2BJS collaborated with state and local law enorcement agencies andlaw enorcement proessional organizations, such as the InternationalAssociation o Chies o Police (IACP) and the National Sheriffs’Association (NSA), to establish the specific criteria or determining whichdeaths to include within the ARD program’s scope. Tese callibrationsaddressed topics regarding which deaths should be eligible or reporting,the number and type o sources o existing data available in each state, andexpectations or how states should identiy within-scope deaths.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
4/78
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
5/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 5
and to complete a CJ-11A Incident Report (appendix A). Asa result, SRCs have developed strategies or identiying andcollecting data that vary by the resources available to them.Consequently, the current ARD program design reflects manystate-specific methodologies that are integrated into a single,national-level program.
2.1 Data providers
Inormation about data providers participating in the ARDprogram is available or calendar years 2003–11.6 During 2011,47 reporting offices submitted ARD data (table 2-1). MostStates have maintained participating SRCs over time, with36 states submitting ARD data every year o data collectionsince 2003 (states are not shown in table 2-1). Although a statemay not have participated in the ARD program in a given year,arrest-related deaths may have been identified and reported byBJS staff in these states.
During 2011, a state criminal justice agency was the mostcommon data reporting contact (16 states), ollowed by adepartment o public saety or other state law enorcementagency (14 states) (table 2-2). In three states, the department ocorrections took a lead role in compiling ARD records.
Participation in the ARD program is voluntary without directpayment or compensation. Only SRCs employed by state
justice statistics analysis centers (SACs) have received ederalunds to support ARD-related work. SACs are units or agenciesat the state government level that use inormation rom allcomponents o the criminal justice system to analyze statewidepolicy issues (see www.jrsa.org). SACs are maintained in
various entities, such as state planning or coordinatingagencies, governors’ offices, criminal justice agencies,and universities.
For 32 states in 2011 (68%), the SRC office reporting ARD dataalso served as the SAC.
2.2 Data collection process
Te ARD program data collection process includes threemain tasks:
(1) identiy arrest-related deaths
(2) acquire inormation about arrest-related deaths
(3) submit CJ-11summary o incidents and CJ-11A
incident reports.
SRCs are responsible or identiying all arrest-related deathswithin their states. Arrest-related deaths may be identifieddirectly by the SRC or indirectly through assistance romanother entity involved with the death (e.g., law enorcementagency) or death investigation (e.g., medical examiner’s orcoroner’s office, prosecutor’s office). SRCs use a variety o
6Although ARD data were accepted afer 2011, program staff did notactively collect data or calendar years 2012 or 2013.
2. Methodology
o implement the ARD collection, BJS developed a programmethodology that relied on establishing centralized SRCs andusing them to operate state-level ARD data collections (seeappendix B or a description o how the current ARD programmethodology was developed).5
Since 2003, SRCs have been responsible or identiying all
ARD reportable deaths in their states and or obtaininginormation about these deaths. BJS provided guidance in howto collect ARD data but did not require SRCs to use a specified,standard methodology. BJS recommends that states use oneor more o the ollowing procedures to identiy arrest-relateddeaths: implementing legislation; contacting state medicalexaminer and county coroner officers; involving state and localprosecutors’ offices; contacting the state police; expanding therole o the state UCR reporters; or conducting open-sourcesearches.
Once a reportable death is identified, SRCs are directed towork with the law enorcement agency involved with the
death to obtain the required data elements about the incident.When the SRC is unable to obtain ull reporting rom the lawenorcement agency involved in the death, the SRC is askedto compile details o the event rom official source documents
5When the DICRA legislation was enacted, only Caliornia and exas hadstatutes requiring that inormation on arrest-related deaths be collectedand reported at the state level. For the remaining 48 states and the Districto Columbia, the ARD program was the first attempt to collect statewidecounts o all deaths that occur in the process o arrest. In developing theprogram, BJS contacted multiple offices in each state to determine anappropriate reporting agent and data provider.
HIGHLIGHTS
Centralized state reporting coordinators (SRCs) operate thestate-level ARD data collection.
The ARD program has three main tasks:
•
identify arrest-related deaths • acquire information about arrest-related deaths
• submit the CJ-11 form, Quarterly Summary of Arrest-
Related Deaths (or summary of incidents) and the CJ-11A
form, Arrest-Related Death Report (or incident report).
SRCs use one or more of the following sources to identifyeligible ARD cases:
• law enforcement
• medical examiner or coroner
• prosecutor’s office
•Uniform Crime Reports (UCRs)
• National Violent Deaths Reporting System (NVDRS)
• open-source media search.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
6/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 6
TABLE 21States or jurisdictions that did not report to the Arrest-Related Deaths Program, 2003–11
State or jurisdiction 2003 2004 2005 2006 2007 2008 2009 2010 2011
Total reporting 44 46 46 44 47 43 42 47 47
Arkansas / / / / /
District of Columbia / /
Georgia / / / / / / / / /
Louisiana /
Maryland / / / / / / /
Mississippi /
Montana / / / / / / /
Nebraska /
Nevada / / / /
New Mexico / /
North Carolina /
Oklahoma /
Tennessee /
Wisconsin / / / /
Wyoming / / / / / / / /
Note: States not listed reported to the ARD program in each year from 2003 through 2011.
Reported.
/ Not reported.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
TABLE 22.Number of state reporting coordinators, by state, Statistical Analysis Center (SAC) status, and type of reporting office,2011
Located in SACs Not located in SACs
Type of reporting office Total Number State Number State
Number of all reporting offices 47 32 15
State criminal justice agency 16 14 AL, AZ, CT, DE, IL, IA, KS, KY,MT, NE, NY, RI, VA, WV
2 PA, VT
Department of public safety/law enforcement 14 9 CO, FL, MO, NJ, OH,OK, SC, SD, TN
5 DC, HI, MI, MN, WA
University 4 4 AK, MS, NV, NM 0
Attorney general’s office 4 2 CA, NH 2 ND, TXGovernor’s office 3 3 MD, MA, UT 0
State department of correction 3 0 3 ID, IN, LA
Office of the chief medical examiner 2 0 2 ME, NC
Office of public health 1 0 1 OR
Note: Four states did not report data in 2011.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
7/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 7
2.2.2 Number and type of case identification strategies
O the 41 SRCs with known active case identification strategies,17 mentioned using multiple strategies or identiying arrest-related deaths, while 24 (59%) said that they used a singlemethod or identiying cases. Among those 24 SRCs, open-source searches were the most common (16 SRCs), ollowedby direct reporting by law enorcement agencies (3), a query omedical examiners’ or coroners’ databases (3), a query o theNVDRS (1), and officer-involved death reports submitted tothe office o the attorney general (1).
Te use o open-source searches as a method o identiyingarrest-related deaths was also common among SRCs usingmultiple case identification strategies. Among the SRCsimplementing multiple strategies or identiying deaths,16 in 17 used open-source searches combined with othermethods, such as surveys o law enorcement (7 SRCs) andqueries o medical examiners’ or coroners’ databases (4 SRCs),NVDRS (3 SRCs), and the National Incident-Based ReportingSystem (NIBRS) or UCR (2 SRCs). One other SRC searchedmedical examiners’ databases and conducted statewide surveyso prosecutors’ reports.
techniques to identiy deaths that are consistent with ARDprogram definitions. BJS encourages SRCs to take an activeapproach in identiying deaths, rather than relying solelyon a system o voluntary reporting by law enorcement ornongovernmental entities. Active strategies are those inwhich the SRC proactively searches or arrest-related deaths,such as methods that include surveying law enorcementagencies, mining existing data collections, and conductingregular open-source searches, in addition to possibly receiving
voluntary submissions rom law enorcement agencies andmedical offices.
SRCs are responsible or obtaining complete incident reportsor each arrest-related death in the state and or transmittinginstruments to ARD program staff or data entry, analysis,and dissemination at the national level. Although all three othese components are incorporated into the national programdesign, the processes by which data are collected vary by SRC.
2.2.1 Active and passive approaches to case identification
In 2011, most SRCs employed active strategies or identiying
deaths. O the 45 SRCs reporting inormation about theirdata collection procedures, 41 reported using a proactivestrategy to identiy arrest-related deaths. Te most commonactive identification strategy involved conducting open-sourcesearches or reportable deaths. In total, 32 SRCs reportedconducting web-based searches as a means or identiyingreportable deaths: 16 conducted solely open-source searches,and 16 conducted open-source searches in addition to anotherstrategy (table 2-3).
SRCs using passive methodologies rely on other agencies toidentiy arrest-related deaths. Passive strategies are those inwhich data are collected through voluntary submissions ocompleted CJ11A incident reports rom other entities, such aslaw enorcement agencies and medical examiners’ or coroners’offices. SRCs employing passive methodologies do not conductindependent searches or arrest-related deaths. SRCs in ourstates said that deaths were identified solely through voluntaryreporting by law enorcement agencies.
TABLE 23Number of states or jurisdictions using active caseidentification strategies, by data source, 2011
Source Number
All states or jurisdictions 51
Active case identification strategies 41
Single-method identification 24
Open-source search 16
Law enforcement survey/referral 3
Medical examiner data request 3
Othera 2
Multiple-method identification 17
Open source and law enforcement survey/referral 7
Open source and database searchb 5
Open source and medical examiner data request 4
Medical examiner and prosecutor survey 1
Passive case identification strategies 4
Unknown case identification strategies 2
Nonparticipating states 4aIncludes sole use of the National Violent Deaths Reporting System (one state) orreports to the office of the attorney general (one state).bIncludes National Incident-Based Reporting System (NIBRS), National ViolentDeaths Reporting System (NVDRS), and Uniform Crime Reports (UCR).
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
8/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 8
the ARD scope or the entire state, while another problemrelates to whether this type o monitoring is conductedsystematically. It was not clear rom discussions with SRCs thatroutine procedures were employed when monitoring newsoutlets (e.g., daily monitoring or monitoring the same outlets).It was also unclear how SRCs determined which outlets tomonitor and whether any investigation was conducted toassess the appropriateness or accuracy o the source content.
Besides traditional media, many SRCs also used open-sourceInternet searches to identiy deaths. Although some specificallymentioned monitoring the web page o the state police orincidents o arrest-related deaths, others had a more difficulttime speciying their procedures. Some SRCs also reported thatthey conducted searches or terms relating to arrest-relateddeaths, but it was unclear whether searches were conductedad hoc or routinely. It was also unclear i the same terms weresearched consistently or i SRCs varied their terms whensearching or arrest-related deaths. Without inormation aboutspecific search terms used by SRCs, the variation o searchterms are unknown across SRCs. Overall, although 26 statesreported using media searches as their primary mode or
identiying arrest-related deaths, no two SRCs used the exactsame procedures.
2.2.4 Types of state reporting coordinators and primarymethods for identifying arrest-related deaths
SRCs were categorized by the location o their offices in anattempt to better understand whether there is a relationshipbetween the type o reporting office and the primary methodor identiying arrest-related deaths. Among the SRCs locatedin state criminal justice agencies, 10 in 14 relied on open-source searches as their primary method or identiying arrest-related deaths. Hal (7) o the SRCs located in departments o
public saety or another state law enorcement agency also usedopen-source searches as their primary means o identiyingdeaths. Among the 14 SRCs located in departments o publicsaety or another state law enorcement agency, three identifiedarrest-related deaths primarily through direct reporting by lawenorcement agencies (table 2-5).
2.2.3 Primary identification methodology
O the 45 participating states with known case identificationmethodologies in 2011, 26 SRCs used the media and open-source searches as their primary manner o identiyingarrest-related deaths (table 2-4).7 Eight SRCs identifieddeaths primarily through direct reporting by law enorcementagencies, five through medical examiners’ or coroners’ offices,three through the NVDRS programs, two through the state
attorney general’s or prosecutors’ offices, and one through theUCR program.
Although SRCs used similar strategies or identiying deaths,differences were ound in how methodologies are applied.SRCs reported using media accounts as their primary caseidentification methodology more requently than any othermethod; however, this strategy varied the most across states.For example, many SRCs stated that their methodology wasto watch the local television news, read the newspaper, orboth. Although some arrest-related deaths will be identified bymonitoring local news, this method is not likely to be sufficientor identiying the ull universe o reportable deaths. Oneproblem relates to whether local outlets are capable o covering
7SRCs using multiple strategies or identiying arrest-related deaths wereable to indicate which method was their primary method or collectingdata and which strategy served as a secondary method or collecting data.
TABLE 24.Primary case identification methodologies, 2011
Source Number Percent
All participating states or jurisdictions 45 100%
Open-source media search 26 58
Law enforcement agency report 8 18
Medical examiner database 5 11
National Violent Deaths Reporting System 3 7
State attorney general’s or prosecutor’s office 2 4
Uniform Crime Reports 1 2
Note: The primary source for identifying arrest-related deaths was unknown in twostates. Four states did not report data in 2011.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
TABLE 25.Primary information source for identifying arrest-related deaths, by location of state reporting coordinator office,2011
Location of office All officesOpen sourcesearch
Direct reporting bylaw enforce-ment
Data request tomedical examineror coroner
National ViolentDeath ReportingSystem
Investigativereports
Uniform CrimeReports
Total 45 26 8 5 3 2 1State criminal justice agency 14 10 2 0 1 0 1
Dept. of public safety/law enforcement 14 7 3 3 0 1 0
University 4 3 1 0 0 0 0
Attorney general's office 4 1 2 0 0 1 0
Governor's office 3 1 0 1 1 0 0
Department of correction 3 3 0 0 0 0 0
Medical examiner's office 2 1 0 1 0 0 0
Office of public health 1 0 0 0 1 0 0
Note: The primary source for identifying arrest-related deaths was unknown in two states. Four states did not report data in 2011.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
9/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 9
least one additional source to obtain the incident data or
completing the CJ-11A. Tirty-six percent o SRCs used amultistep data collection process in which one source was usedto identiy deaths and a different source was used to obtaininormation about the death. Te most common example oa multisource process involved using open-source searches toidentiy an arrest-related death and a police report to completethe CJ-11A.
During 2011, about two-thirds o SRCs used only officialsources o inormation or obtaining data regarding incidentso arrest-related deaths (table 2-6). Law enorcement reportswere the most common source o incident-level inormationabout arrest-related deaths. A small number o SRCs (4%)relied on unofficial sources o inormation, such as journalistic
accounts o events, to obtain details about reportable incidents.
2.2.5 Source of case information
Once an arrest-related death is identified, the SRC isresponsible or either obtaining inormation about the deathdirectly or coordinating efforts by other entities to obtain case-level inormation. In states in which SRCs identiy reportabledeaths, the SRC may take responsibility or acquiringinormation about the incident or may request that anotheragency involved with the death (e.g., law enorcement agency,
medical examiner’s office, or prosecutor’s office) provideinormation about the circumstances surrounding the death.In states in which deaths are identified by other agencies,the agency that identified the death takes responsibility orgathering inormation.
SRCs responsible or directly obtaining inormation aboutthe death acquire case inormation rom a variety o sources,including police reports or press releases, death records orautopsy reports, legal accounts, and the media. Non-SRCentities responsible or gathering case inormation typically usethe same data sources as SRCs because these agencies are ofenthose responsible or initially collecting the inormation. Forexample, law enorcement agencies completing the CJ-11A willlikely use inormation rom the police reports they produced.
2.2.6 Overlap in sources used to identify deaths and toobtain case information
SRCs vary in the number and types o sources they use toobtain inormation about arrest-related death incidents. SomeSRCs rely on a single source o inormation to both identiydeaths and gather inormation about incidents. Other SRCsuse a multistep data collection process in which inormationused to complete the CJ-11A comes rom a source other thanthe one used to identiy the death. For instance, a single-source data collection process could involve using a media
account both to identiy the arrest-related death and to obtaininormation about it. A multisource process involves using onesource (e.g., media report) to identiy a reportable death andone or more additional sources o inormation (e.g., a policereport or autopsy report) to obtain case inormation.
Eleven SRCs (24%) used a single source o inormation oridentiying arrest-related deaths and completing CJ-11Aincident reports (figure 2-1). O those, the most commonsource o inormation was law enorcement reports (six SRCs),ollowed by media reports (two SRCs), medical examinerrecords (two SRCs), and UCR data (one SRC).
Most SRCs (72%) relied on multiple sources o inormationto identiy deaths and to obtain incident data during 2011.Tirty-six percent o SRCs relied on both the source oinormation used to identiy arrest-related deaths and at
0 5 10 15 20
11
17
17
2
Same source used toidentify deaths
Different sources used
to identify deaths
Identification source andadditional sources used
Missing
FIGURE 21Overlap in sources used to identify deaths and tocomplete CJ-11A incident reports, 2011
Note: Four states did not report data in 2011.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
TABLE 26.Types of data sources used to obtain arrest-related deathcase information, 2011
Source Number Percent
Total state reporting coordinators 45 100%
Official sources data 30 67%
Law enforcement report only 19 42
Medical examiner report only 4 9
Multiple official sources 7 16
Combination of official and unofficial sources 13 29%
Unofficial sources of information 2 4%
Note: The primary source for identifying arrest-related deaths was unknown in twostates. Four states did not report data in 2011.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
10/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 10
2.3.1 Data collection cycle
Te ARD data collection is conducted on a calendar-year cycle.Te program operates continuously and accepts submissionsthroughout the year. Te ARD program also accepts reports oarrest-related deaths that occurred beore the current calendaryear and updates to previously reported data.
In January, BJS notifies SRCs that it is beginning a new yearo ARD data collection and provides them with key program
materials, including (1) the CJ-11 summary o incidents orm,(2) the CJ-11A incident report, and (3) a question-by-questionguide to aid in completing the CJ-11A (see appendix A orsurvey instruments and revisions to the CJ-11A incidentreport, 2003–13). In addition, BJS provides the SRCs withgeneral inormation about the program, such as key dates,methods or submitting data, and contact inormation or ARDprogram staff.
When the DICRA legislation was in effect, SRCs were requiredto submit ARD data quarterly. Tey were asked to submita CJ-11 Summary o incidents and corresponding CJ11Aincident reports within 60 days o the end o the quarter. Afer
DICRA expired in 2006, BJS relaxed the quarterly reportingrequirement and gave SRCs the option o submitting their dataeither quarterly, annually, or on a schedule o their choosing.O SRCs reporting ARD data in 2011, about hal (49%)submitted data annually, while 28% continued to collect andreport data quarterly (figure 2-3). Data can be submitted anytime during the year, but SRCs must submit all annual CJ-11and CJ-11A reports no later than March 1 o the ollowingcalendar year.
2.3.2 Data submission
Afer the SRCs complete their annual data collections, theytransmit this state-level data to BJS or analysis. Historically,
the ARD data collection was paper based. Data providers
2.3 Completion of CJ-11 and CJ-11A instruments
Te ARD program uses two data collection instruments, theCJ-11 Quarterly Summary o Arrest-Related Deaths (summaryo incidents) and the CJ-11A Arrest-Related Death Report(incident report). Te CJ-11 orm is completed by SRCs and isa state-level count o arrest-related deaths. SRCs are directedto submit a CJ-11 orm or each reporting period and to show
a count o zero i no arrest-related deaths are identified withinthe reporting period. BJS requests that SRCs submit zerocounts to distinguish between the measured absence o anidentified death and missing data.
Te CJ-11A orm is a detailed incident report that describesthe circumstances surrounding each reportable death. SRCsare asked to submit a complete CJ-11A incident report oreach death recorded on the state-level CJ-11 summary orm.Depending on the method used to identiy the death andthe agency responsible or acquiring inormation about theincident, the SRC may take responsibility or completing theCJ-11A orms or may request that another agency completethe orms.
Inormation about the entity responsible or completingCJ-11A orms during 2011 was available in 45 states. Te lawenorcement agency involved with the death was primarilyresponsible or completing CJ-11A incident reports in 20 states(44%) (figure 2-2). SRCs obtained case inormation andcompleted CJ-11A reports in 16 states (36%), ollowed by staffin medical examiners’ offices (2 states) and prosecutors’ offices(1 state). Multiple entities took responsibility or completingsections o the CJ11A incident report in six states, with jointefforts by the SRC and law enorcement agency in three states,the SRC and medical examiner’s or coroner’s office in twostates, and the law enorcement agency and medical examiner’s
office in one state.
FIGURE 22.Entity responsible for completing CJ-11A incidentreports, 2011
0
5
10
15
20
25
Multipleentities
Prosecutors'office
Medicalexaminers'office (ME)
Statereporting
coordinator (SRC)
Law enforcementagency (LEA)
LEA and ME
SRC and ME
LEA and SRC
Number of states
Note: The primary source for identifying arrest-related deaths was unknown in twostates. Four states did not report data in 2011.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
0
10
20
30
40
5049%
28%
9% 9%
Annual SemiannualQuarterly Ad hoc Monthly
6%
Percent
FIGURE 23.Arrest-related deaths data collection schedule, 2011
Note: Detail may not sum to total due to rounding. Four states did not report datain 2011.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
11/78
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
12/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 12
statistically sound, national-level data because o the variabilityin data quality both within and across states. BJS has developedstandardized data collection instruments (CJ-11A orm),training materials, working definitions, and a web tool toassist SRCs in gathering and submitting the ARD inormation(see Appendix B). However, significant challenges within andbetween states exist in terms o mode, source, and quality.
3.2.2 Variation in program methodologies across states
SRCs developed their data collection methodologies inrelation to their preerences and resources. As a result, theARD program operations varied among participating states(see chapter 2: Methodology ). Deaths may be identified romone or more resources rom among media, open sources, lawenorcement agencies, and medical examiners’ or coroners’offices. Te inormation sources consulted or completing eachCJ-11A orm also differed, and included media reports, policereports, medical examiner or coroner reports, case reportsrom the division o criminal inormation, or any combinationo these sources. Data collection methodologies also differedby the agency responsible or completing the CJ-11A orm,
which can include the SRC, a law enorcement agency, amedical examiner’s or coroner’s office, or a combination othose offices.
BJS assessed the SRCs’ primary method or identiying deaths,the sources o inormation used to complete the CJ-11Aincident report, and the entity completing the CJ-11A. BJSdetermined that SRCs used a wide range o methods orcollecting ARD data in 2011. Te most common method(nine SRCs) or collecting ARD data was or the SRC toidentiy cases through open-source searches.8 Ten, the SRCasked the law enorcement agency involved with the death tocomplete a CJ-11A incident report using the agency’s official
data about the case. Te second most common method (sixSRCs) or collecting ARD data was direct reporting by thelaw enorcement agency involved with the death. In thesestates, the law enorcement agency involved with the arrest-related death took responsibility or identiying it, obtainingand completing a CJ-11A incident report using the agency’sofficial data, and submitting the orm to the SRC. In threestates, SRCs identified deaths through media searches andcompleted CJ-11A instruments by obtaining the manner andcause o death rom a medical examiner’s or coroner’s office.Te rest o the inormation requested on the CJ-11A orm wasobtained through the same open-source record used to identiythe death, namely, media accounts o events. wo SRCs
identified arrest-related deaths through open-source searchesand completed the corresponding CJ-11A instruments withinormation obtained in the media account used to identiy thedeath. Te remaining 24 states reporting inormation abouttheir 2011 ARD data identification and collection strategiesused unique methodologies with varying combinations o thesource used to identiy the arrest-related death, the agency
8Te open-source search methodology is a general description or reviewo publicly available data. Tis term does not indicate a specific or singleprocedure or strategy or identiying and collecting data.
homicides by law enorcement, which may be more easilyunderstood as eligible incidents. Te breadth o the program’sdefinition o “arrest-related” creates problems in identiyingeligible cases, likely contributing to a significant undercounto incidents.
3.1.2 Custody
Te term “custody” includes physically touching or putting
hands on an arrestee. For the ARD program, it also includesany act that indicates an intention to gain control o an arrestsubject under the authority o the law or any situation inwhich a person’s reedom to leave is restricted by state or locallaw enorcement personnel. Tis includes deaths occurringwhile law enorcement officers actively pursue or attempt toapprehend persons o interest or criminal suspects, regardlesso whether physical custody is established. For example, manyARD cases are officer-involved shootings, and virtually noneo these incidents involved subjects who were already in thephysical custody o law enorcement personnel.
Custody includes individuals in physical custody o law
enorcement without being in the process o arrest. o gaincontrol o a situation, law enorcement personnel sometimeshandcuff or detain persons against whom they do not plan tofile charges (e.g., mentally ill or intoxicated persons, or otherswho pose a danger to their own saety).
3.1.3 Reporting deaths in jails
Te goal o the DCRP is to identiy and record all deaths ocivilians in contact with law enorcement and correctionalpersonnel. BJS implemented three separate data collectionsystems to capture deaths occurring in the process o arrest andthose that occur while the person is detained in jail or prison.Deaths reportable to the ARD program are limited to thosethat occurred while the decedents were in the custody o lawenorcement. Tereore, deaths that occur afer an individualhas been processed and custody has been transerred to aholding acility are outside the ARD scope. Given that theunctions o some law enorcement agencies include detention,some SRCs reported difficulty in determining whether thedecedent died in the custody o law enorcement or thecustody o a correctional acility.
3.2 Assessment of data collection modes
3.2.1 Standardization
Te ARD program has implemented variable methodologiessince its first collection in 2003. As a result, data providers havedesigned and developed state-level data collections accordingto their preerences, their capabilities, and the resourcesavailable to them. Te flexibility o the ARD data collectionallows state-level data providers to use a variety o data sourcesand to rely on those that are most reliable in their particularstates, but it presents significant challenges or collecting
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
13/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 13
included media (16 SRCs), law enorcement agencies (7),medical examiners’ or coroners’ offices (3), the state attorneygeneral or prosecutor (1), and the state NVDRS reporter (1).wenty o the 46 SRCs reporting data indicated that they usedmultiple sources o inormation to complete CJ-11A ormsor identified deaths (16 used two sources, and 4 used three ormore sources). As a result o the variability both within andbetween states, records were ofen identified through a varietyo means and comprised both official and unofficial accountso the events.
3.3 Assessment of data quality and coverage
Given the concern about the lack o standardization bothacross and within SRCs in terms o data collection modes,definitions, scope, and availability o the necessary resources,BJS assessed the quality o the program’s coverage—that is,how well the program perorms in capturing the ull universeo eligible arrest-related deaths.
3.3.1 2003–05 national and state assessment
Initial efforts began in 2007 with a ocus on assessing thecoverage o the 2003–05 ARD program by comparinghomicides reported to ARD with those reported in theSupplementary Homicide Report (SHR) (Mumola, 2007). TeSHR is a component o the FBI’s Uniorm Crime Reporting(UCR) program, which compiles inormation on crimesknown to state and local law enorcement agencies in theUnited States. While the UCR provides aggregate annualcounts o the number o homicides occurring in the UnitedStates, the SHR data provide additional inormation abouteach homicide incident, including the jurisdiction, month,year, and circumstances associated with the incident. Like the
ARD, the SHR data have limitations. Most notably, the SHRprogram is voluntary, so a number o law enorcement agencieseither do not consistently send their data to the SHR or do notsubmit it at all.9 Te SHR, however, represents the only othernational database that can provide sufficient inormation onthe circumstances surrounding law enorcement homicides toallow comparison with the ARD program.10
Te ARD-SHR comparisons were limited to law enorcementhomicides because, unlike the ARD program, the SHR doesnot record law enorcement-involved deaths that are attributedto suicide, intoxications, accidental injury, or natural causes.Also unlike the ARD program, the SHR includes only those
homicides by law enorcement in which the use o orce wasruled justifiable. Deaths due to unjustified use o lethal orceby law enorcement personnel are counted with other murders
9See Te Nation’s wo Measures of Homicide. NCJ 246832, BJS web,September 2014.
10Te NVSS is a national database that captures homicides occurring inthe United States, but it does not link homicides that are arrest-relatedwith specific law enorcement agencies as the SHR program does. Tislink to a law enorcement agency is critical or capturing-recapturing theanalytic approach used to assess the ARD coverage.
responsible or completing the CJ-11A orm, and the source oinormation or completing the CJ-11A orm.
Most states (26) relied solely on open-source searches toidentiy ARD-eligible cases. Although about hal o theparticipating states said that they used open-source searchesas their primary strategy or identiying deaths, no standardmethodology exists or this type o identification procedure.Tereore, each o these 26 states was likely to use different
mechanisms, searches, and monitoring plans or identiyingarrest-related deaths.
3.2.3 Variation in program methodologies within states
Some SRCs noted that they did not have a standard operatingprocedure or identiying or collecting inormation on arrest-related deaths. Tey could speciy the primary strategiesthey used to identiy cases (e.g., media or law enorcement)and whether or not they also used alternative methodologiesor identiying deaths. However, the SRCs suggested thatit was more difficult to describe their exact data collectionprocedures because they ofen varied rom one case to another.
For example, SRCs relying on media or open-source searchesdescribed identiying cases by a variety o means. Teseincluded watching television, reading local newspapers, usingGoogle Alerts to conduct daily web-based searches o keywords or phrases, monitoring the state police web page, andsearching other websites. In discussions with SRCs, it appearedto be more common or deaths to be identified ad hoc ratherthan through specified, standard procedures conductedroutinely. O the SRCs reporting more standardized proceduresor identiying cases, some said that their methods o obtainingthe data needed to complete the CJ-11A varied rom case tocase. For instance, some states that relied on law enorcementagencies to complete the CJ11A orms had contingency plans
or collecting data i the law enorcement agency involved withthe death reused to supply the inormation. One SRC reportedthat only about 50% o law enorcement agencies return thecompleted CJ-11A orm. When the law enorcement agencyailed to provide the completed orm, the SRC attemptedto obtain a report rom the medical examiner’s office orthe inormation they needed to complete the orm. I theinormation required to complete the CJ-11A was unavailablerom the medical examiner’s office, the SRC used mediareports to complete the report.
Data collection methodolgies or SRCs participating inthe ARD program differed over time. A survey o SRCs
conducted in 2007 revealed that 42 o 47 that reported dataused more than one source o data to identiy deaths andcomplete CJ-11A orms. Commonly reerenced sourcesincluded local law enorcement agencies (43 SRCs), media(30), medical examiners’ or coroners’ offices (23), statepolice (19), state UCR/SHR reporters (9), and state attorneysgeneral or prosecutors (6). A ollow-up discussion conductedin 2012 suggested that 28 o the 46 states reporting dataused a single source o inormation or identiying cases. Othese, commonly reerenced single sources o inormation
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
14/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 14
equal 0 and the SHR overage variable would equal 2. Althoughthis analysis was limited in that it did not determine whetherthe one case reported to ARD was also one o the three casescaptured in the SHR, it was able to provide evidence o atleast two homicides by the Springfield Police Departmentthat were not captured in ARD. Te SHR overage homicideswere summed as an indication o the minimum number ohomicides by law enorcement missing rom ARD (identifiedas ARD missing). Similarly, the ARD overage cases weretransormed into SHR missing. For the Springfield PoliceDepartment, the ARD missing equaled 2.
Tis procedure was conducted or every law enorcementagency that reported a homicide by law enorcement to eitherARD or the SHR rom 2006 through 2009. Te analysisresulted in a minimum o 701 cases reported to the SHR thatwere not recorded in ARD, and a minimum o 935 cases thatwere recorded in ARD, that were not captured in the SHR.Although this analysis was limited by several actors—namely,that the data were not disaggregated by year o death orcharacteristics o the decedent—it demonstrated coverageerror at the agency level.
3.3.3 2003–11 individual case assessment
o better understand the source o discrepancies in reportingacross the ARD program and the SHR, BJS comparedindividual-level homicide records in each collection. For thisassessment, BJS examined law enorcement homicides inARD and compared them with justifiable homicides in SHRusing the data available rom the entire ARD program anda capture-recapture analysis.11 Tis analysis addresses a keyobstacle to understanding the coverage o the ARD program.Because the ARD program includes only state and local lawenorcement agencies that have been involved in one or more
identified arrest-related deaths and does not include agenciesthat did not have an identified arrest-related death, it is difficultto make inerences about coverage using ARD data alone.Because the SHR data also provide incident-level inormationon homicides, they can be used to help assess the coverage othe ARD data collection. Te assessment consisted primarilyo matching records rom the two programs and determiningthe amount o overlap between collections to estimate the totaluniverse o reporting deaths. Similar to the other assessmentsusing SHR data, this assessment is limited to justifiablehomicides by law enorcement (see Arrest-Related Deaths:echnical Report , NCJ 248544, BJS web, March 2015).
Under the current ARD collection method, SRCs identiydeaths, then usually contact the local agencies in which thedeaths occurred to get the inormation they need to complete aCJ-11A or each death. By this method, BJS can affirm that anidentified death occurred and met the criteria to be includedin ARD, but it cannot confirm that there were in act no other
11ARD data are available or program years 2003–09 and 2011. However,the ARD program is designed to capture all arrest-related deaths,including those due to accidents, drug overdoses, and natural causes.Tese incidents are not assessed in this analysis.
and cannot be disaggregated or comparison to ARD. Te ARDcounts o homicide by law enorcement include all deaths thatresult rom the use o orce.
Results o this assessment indicate that, nationally, a similarnumber o officer-involved homicides were recorded in theARD and SHR programs. In the aggregate, the two programscollected very similar counts rom 2003 through 2005, with1,095 homicides recorded in ARD and 1,082 justifiable
homicides recorded in the SHR. However, the aggregated,national-level comparison o the ARD and SHR data maskeddifferences in reporting at the state level.
Similarly, the aggregate analysis at the state level likely alsomasked differences within states at the agency level. In moststates, the ARD and SHR counts o law enorcement homicidesshowed small differences in the number o deaths reportedto each program, but because arrest-related deaths are rareevents, even small differences in state-level counts couldresult in a substantial change in the proportion o reporteddeaths. In 33 states, the two measures differed by ewer than 10deaths rom 2003 through 2005. Nine states reported counts
that differed by at least 20 deaths, with the difference rangingrom 22 to 194 homicides. Caliornia—one o 2 states with astatute requiring that all arrest-related deaths be reported tothe state attorney general’s office—had the largest reporting
variation, with 354 justifiable homicides by law enorcementreported to the SHR rom 2003 through 2005 but less than halas many (160) reported to the ARD program. Over the sameperiod, Florida reported 98 law enorcement homicides to theARD program but did not report any data to the SHR. akingthe higher count reported by each state or each year yields atotal o 1,489 reported law enorcement homicides rom 2003through 2005—more than a third as many as those reported toeither program.
3.3.2 2006–09 agency assessment
BJS conducted an agency-level analysis o homicides by lawenorcement reported to the ARD program and to the SHRor 2006–09 to determine the number o cases recordedin the SHR that were not recorded in ARD and vice versa.Homicides were disaggregated by agency or this comparison,using the agency’s originating identification (ORI) number.Law enorcement agencies were matched across data sources,and the numbers o recorded homicides were compared tobetter understand how the coverage issues identified in thestate-level analysis were affected by variation in reporting by
law enorcement agencies. BJS was interested in determiningwhether the differences in reporting were due to (1) differentagencies reporting to each system or (2) the same agenciesreporting different counts to both systems.
BJS conducted this analysis by creating overage variables torepresent the number o additional cases reported to the ARDprogram and to the SHR by each law enorcement agency. Forexample, i the Springfield Police Department reported onehomicide by law enorcement to ARD and three homicides bylaw enorcement to the SHR, the ARD overage variable would
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
15/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 15
3.3.5 Capture-recapture analytic approach
Once decedent records have been linked across the ARD andSHR, capture-recapture methods can be used to estimate thetotal law enorcement homicide population and the proportiono that population covered by the ARD program. Tis analysismakes use o the commonly applied Lincoln-Petersen (LP)estimator, with modifications or bias amelioration necessitatedby the characteristics o the ARD program (Seber 1982).
Te simple LP method assumes that law enorcement agenciesthat were not recorded in the ARD data did not have anyofficer-involved homicides during the observation period.Tis approach provides the highest possible ARD coverageestimate and the lowest estimate o the total population o lawenorcement homicides because it assumes that the data arerepresentative o agencies with true zero counts rather thanmissing responses. Essentially, the simple LP approach presentsa best case scenario or estimating the ARD coverage rate.
Te adjusted LP estimate assumes that a portion o the lawenorcement agencies that did not report to the ARD programdid have homicides and that they had, on average, the same
number o officer-involved homicides as law enorcementagencies that reported data to the program. Te adjustedapproach assumes that some law enorcement agencies withhomicides are unrecorded in the ARD data and that theseagencies represent a random sample o the overall set oagencies with homicides.
3.3.6 Results
Following the simple LP approach, the estimated ARDcoverage o all law enorcement homicides is 49%, with a95% confidence interval ranging rom 47% to 50%. In otherwords, the ARD program captured only about hal o all
law enorcement homicides in the United States during theobservation period. Te adjusted LP approach, the ARDprogram is estimated to capture 36% o all law enorcementhomicides, with a 95% confidence interval ranging rom 35%to 38%. Tereore, the coverage o law enorcement homicidescaptured in ARD could be as low as 35% or as high as 50%.Tese results indicate that the ARD program may provide aslightly higher coverage o homicides by law enorcement thanthe SHR, which is estimated to cover 46% o officer-involvedhomicides at best. Further analysis o the data suggests thatabout 28% o homicides by U.S. law enorcement agencies arenot captured by either the ARD or SHR programs. Overall,these findings indicate serious deficiencies in the ability o the
ARD program, as well as the SHR, to capture a universe oreportable homicides by law enorcement.
arrest-related deaths. Tis is because BJS does not survey eacho the approximately 18,000 state and local law enorcementagencies in the United States to determine whether arrest-related deaths occurred.
By linking records submitted to the ARD and SHR collections,it is possible to make inerences about the total size o theofficer-involved homicide population and, thereore, aboutthe population coverage achieved in the ARD program. Tis
analytical technique, known as capture-recapture, is typifiedby studies o wild animal populations, although it is widelyused in the study o human populations as well, particularlyin coverage evaluations or censuses. Te ARD coverageassessment is based on traditional capture-recapture methods,but unique characteristics o ARD data necessitate specialtechniques, including probabilistic record linkage between datasources and adjustments or unobserved agencies.
3.3.4 Probabilistic case-level linking
Capture-recapture analysis depends on the ability to linkrecords across data sources. In wildlie applications, this
requirement is easily met as captured animals are identifiedthrough tagging. In the context o a coverage assessment usingan auxiliary data source not collected as part o the study,this requirement becomes more difficult. Because decedentsare not uniquely identified across data sources (e.g., byname or Social Security number), record linkage must bebased on comparable decedent demographic and incidentcharacteristics. As linkages made without unique identifiersare inherently uncertain, decedents are matched between ARDand SHR sources probabilistically on the basis o measuredsimilarity between records, within a monte carlo replicationramework. Tis means that although a given ARD decedentwill be linked to either no SHR decedents or a single SHR
decedent in a particular replicate, the nature o the link or thatARD decedent (i.e., whether it is linked and, i so, to whichSHR decedent) is allowed to vary across replicates accordingto the probabilities associated with particular record pairings,on the basis o measured similarity between cases. Probabilisticlinking is random in nature and incorporates additionaluncertainty into the process o estimating population coverage.Monte carlo replication allows or this additional uncertaintyto be captured in variance estimates.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
16/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 16
Te assessment is not without limitations. Conducting capture-recapture approaches at the state level was determined to benot statistically sound, given the relatively low number o casesreported to the ARD and SHR programs during 2011. Forexample, our states (Delaware, Rhode Island, Vermont, andWyoming) had zero homicides by law enorcement recorded ineither ARD or the SHR. As such, it was not possible to estimatethe coverage rate o homicides by law enorcement usingsimple and adjusted LP approaches at the state level.
4. Conclusion
Te ARD program is designed to be a national census ocivilian deaths when the event causing the death occursduring an interaction with state or local law enorcement.Understanding when law enorcement officers are involvedin a situation in which a community member dies duringan arrest is critical to a democratic society (Fye 1981, 2002;McEwen, 1996; Klinger, 2008). BJS designed the ARD datacollection methodology to rely on reporting rom independentstate reporting coordinators (SRCs). Te manner in whichARD data are collected varies rom state to state and ofendepends on the resources available to SRCs. Tis variabilityin approaches has led to questions about whether these data
collection methodologies are capable o capturing the universeo arrest-related deaths and o law enorcement homicidesin particular.
Previous research has explored the utility o the SupplementaryHomicide Report (SHR), the National Vital Statistics System(NVSS), law enorcement agency records, media reports,and the ARD program or providing reliable statistics on thenumber o law enorcement homicides in the United States.Tis body o literature has generally concluded that there isno single, reliable system either or reporting or accessing
HIGHLIGHTS
Significant challenges exist with the lack of a standardizedmode for data collection, definitions, scope, agencyparticipation, and the availability of resources.
The current findings are consistent with prior studies thathave shown considerable variation from state to state andover time in reporting to one national system or another.
BJS may provide more prescriptive guidance or requirementsregarding identifying cases and collecting data, but shouldalso incorporate the differences between states in theavailability of reliable sources for information on arrest-related deaths.
When choosing solutions for a complete and accurate datacollection, BJS should consider ways to standardize the datacollection process, implement mandatory reporting by lawenforcement agencies, dedicate resources and funding tosupport data collection efforts at the state and local levels,and focus on a more restricted definition and scope of
eligible cases.
Despite these deficiencies, coverage o these deaths hasimproved since the inception o the program. Differencesin state participation (see table 2-1) and changes in state-level methodologies contribute to national variation inannual coverage rates. Despite dips in coverage in 2005 and2008, results rom both the simple LP and the adjusted LPapproaches indicate that ARD coverage o law enorcementhomicides has increased over time, with the most pronouncedimprovement occurring in the most recent years o datacollection (figure 3-1).
During 2003, the first year o data collection, the estimatedcoverage o homicides by law enorcement was rom about27% (adjusted LP) to 42% (simple LP). Except or datacollected in 2008, the coverage o law enorcement homicidescaptured in ARD improved rom 2006 through 2011.12 Usingthe simple LP approach, the ARD coverage o officer-involvedhomicides grew rom 51% in 2007 to 55% in 2009 and 69% in2011. Te adjusted LP estimate (worst case coverage scenario)or 2011 (59%) indicated better coverage than the simple LPestimate (best case coverage scenario) or 2009 (55%).
Tese data indicate that the data collection methods used inmore recent years are associated with better coverage o law
enorcement homicides. Tis conclusion is supported by theact that the adjusted LP approach also demonstrates improvedcoverage over time, growing rom 40% in 2007 and 2009 to59% in 2011. Te gap between the adjusted and unadjustedcoverage estimates narrowed by 2011, a urther indicationo improved methodologies. Even with this improvement,between 31% and 41% o actual homicides by law enorcementpersonnel were not captured in the 2011 ARD data.
12Te 2010 ARD data were not included in the capture-recapture work.
FIGURE 31.Proportion of law enforcement homicide universecovered by ARD, by estimation method, 2003–09 and
2011
0
20
40
60
80
100
Agency adjustment
No agency adjustment
20112009200820072006200520042003
Percent
2010*
*Data are not shown for 2010 due to an incomplete data collection.
Source: Bureau of Justice Statistics, Arrest-Related Deaths Program, 2011.
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
17/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 17
timeliness and availability are an issue. Many datasets are notavailable in the current year and may lag multiple years behindother sources. Relying on multiple systems would also requirean SRC who has access to and who systematically reviews allthe relevant systems.
A survey that collects inormation about arrest-related deathsdirectly rom local law enorcement agencies (rather thanthrough the SRCs) provides an additional option or improving
ARD program coverage. However, this method would requiresignificant resources.
o achieve any improvements in coverage and consistency inreporting, the ARD program needs more prescriptive guidanceor requirements on how arrest-related deaths are identifiedand report them to BJS. Currently, BJS provides definitions,examples, and general guidance on how SRCs might go aboutidentiying arrest-related deaths and reporting inormationabout them, but it does not require that SRCs ollow a specificmethodology. Furthermore, case identification and datacollection methodology in each state is largely determined anddriven by the individual SRC’s own experience and available
resources. Applying consistent definitions and systematicmeans o identiying deaths and collecting inormation aboutthem rom all appropriate sources in the jurisdiction is criticalto improving ARD program coverage.
Te scope o the current ARD program may also contributeto underreporting. Te analyses reported in the Program
Assessment (chapter 3) and cases ound in the SHR and NVSSare limited to law enorcement homicides. Te ARD programis designed to capture all arrest-related deaths, includingthose due to accidents, drug overdoses, and natural causes.It is the only system that attempts to collect these data, and itcould provide critical inormation on cases that are missingor underreported in other sources, such as deaths ollowingthe use o tactics commonly reerred to as less than lethal,including asers and restraint procedures. However, arrest-related deaths other than law enorcement homicides canbe difficult or SRCs to define and identiy rom the sourcesavailable to them. Furthermore, the more clearly defined lawenorcement homicide type o arrest-related death still suffersrom significant undercoverage, as shown in the capture-recapture analyses presented in chapter 3. Improvements to theARD program might be realized by first ocusing solely on lawenorcement homicides. Afer the data collection methodologyhas been defined, made routine, and shown to result insufficient coverage or law enorcement homicides, a more
expansive definition o arrest-related deaths might be explored.BJS should consider ways to standardize the data collectionmethodologies to improve the reliability and validity ocapturing eligible cases and overall data quality.
data about arrest-related deaths; no one system consistentlyoutperorms others. Furthermore, comparison o various datasources at the national level masks significant variations inreporting to one source or another at the state and jurisdictionlevels (Lofin et al., 2003; Mumola, 2007; Sherman &Langworthy, 1979).
Te current ARD methodology summarized in the Methodology section (chapter 2) and the results o the capture-
recapture analysis described in the Program Assessment section(chapter 3) echo the conclusions o this body o research andhighlight the continuing obstacles to collecting valid andreliable inormation about arrest-related deaths. At best, theARD program captured approximately hal o the estimatedlaw enorcement homicides in the United States during 2003–09 and 2011. Consistent with other research on the utility oexisting sources o data regarding arrest-related deaths, weound that the current data collection process results in asignificant underestimation and potentially a biased picture oarrest-related deaths in the United States (Borrego, 2011; Ho etal., 2009). Some evidence indicates that the system’s coveragehas improved over time. Considering the 2011 data alone, the
ARD program was estimated to capture as many as 69% oall law enorcement homicides in the United States, up romcapturing 42% in 2003.
Although no current national-level system sufficientlymeasures all arrest-related deaths in the United States, previousresearch has shown that one system captures more casesthan another in some states but not in others (e.g., Lofin etal., 2003; Mumola, 2007). Some evidence also indicates thatusing data rom a combination o systems will result in bettercoverage. Te current analysis ound that the ARD programcovered approximately hal o the estimated law enorcementhomicides in the United States during 2003–09 and 2011, but
combining data rom the SHR and the ARD increased coverageto 72%. o improve ARD coverage o arrest-related deaths,the ARD program might thereore build on two aspects othe current design to (1) allow some variation in the methodsused to identiy cases rom state to state and (2) require thatmultiple data sources be used to compile inormation aboutarrest-related deaths within each state.
First, variation in the methods used to identiy arrest-relateddeaths and collect inormation about them might be warrantedrom state to state, depending on the resources available withinthat state. Te source that captures more arrest-related deaths
varies rom state to state, which should not be ignored when
developing a more robust ARD data collection system.Second, most SRCs employed a methodology that combinesinormation rom multiple sources. Given the limitationso existing national-, state-, and local-level data systems,combining data rom all available systems should providea more comprehensive picture o arrest-related deaths.Identiying cases rom multiple systems will require significantresources to ensure that reports are not duplicated. Further,
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
18/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 18
Lofin, C., Wiersema, B., McDowall, D., & Dobrin, A. (2003).Underreporting o justifiable homicides committed by policeofficers in the United States, 1976–1998. American Journal ofPublic Health, 93(7), 1117–1121.
McEwen, . (1996). National data collection on police useo orce. (NCJ 160113). Washington, DC: Bureau o JusticeStatistics. Retrieved rom http://www.bjs.gov/content/pub/pd/ndcopuo.pd
Mumola, C.J. (2007). Arrest-related deaths in the UnitedStates, 2003–2005. (NCJ 219534). Washington, DC: Bureau oJustice Statistics. Retrieved rom http://bjs.ojp.usdoj.gov/index.cm?ty=pbdetail&iid=379
Noonan. M.E., & Ginder, S. (2013). Mortality in local jailsand state prisons, 2000–2011–statistical tables. (NCJ 242186).Washington, DC: Bureau o Justice Statistics. Retrieved romhttp://www.bjs.gov/content/pub/pd/mljsp0011.pd
Seber, G.A.F. 1982. Te estimation o animal abundance andrelated parameters. Macmillan Publishing, New York, NewYork, USA.
Sherman, L.W., & Langworthy, R.H. (1979). Measuringhomicide by police officers. Journal of Criminal Law andCriminology, 70, 546–560.
References
Borrego, A. (2011). Arrest-related deaths in the United States:An assessment o the current measurement. Master’s degreethesis, Arizona State University. http://repository.asu.edu/attachments/56537/content/Borrego_asu_0010N_10572.pd
Burch, A. (2011). Arrest-related deaths, 2003–2009–statisticaltables. (NCJ 235385). Washington, DC: Bureau o JusticeStatistics. Retrieved rom http://www.bjs.gov/content/pub/pd/ard0309st.pd
Fye, J. (1981). Observations on police deadly orce. Crime &Delinquency, 27 , 376–389.
Fye, J. (2002). oo many missing cases: Holes in ourknowledge about police use o orce. Justice Research and Policy ,4, 87–102.
Ho, J.D., Heegaard, W.G., Dawes, D.M., Natarajan, S., Reardon,R.F., & Miner, J.R. (2009). Unexpected arrest-related deathsin America: 12 months o open source surveillance. Western
Journal of Emergency Medicine, 10(2), 68–73.
Klinger, D.A. (2008). On the importance o sound measureso orceul police actions. Criminology and Public Policy, 7 ,605–617.
Klinger, D.A. (2012). On the problems and promise o researchon lethal police violence: A research note. Homicide Studies, 16 ,78–96.
http://www.bjs.gov/content/pub/pdf/ndcopuof.pdfhttp://www.bjs.gov/content/pub/pdf/ndcopuof.pdfhttp://bjs.ojp.usdoj.gov/index.cfm?ty=pbdetail&iid=379http://bjs.ojp.usdoj.gov/index.cfm?ty=pbdetail&iid=379http://www.bjs.gov/content/pub/pdf/mljsp0011.pdfhttp://repository.asu.edu/attachments/56537/content/Borrego_asu_0010N_10572.pdfhttp://repository.asu.edu/attachments/56537/content/Borrego_asu_0010N_10572.pdfhttp://www.bjs.gov/content/pub/pdf/ard0309st.pdfhttp://www.bjs.gov/content/pub/pdf/ard0309st.pdfhttp://www.bjs.gov/content/pub/pdf/ard0309st.pdfhttp://www.bjs.gov/content/pub/pdf/ard0309st.pdfhttp://repository.asu.edu/attachments/56537/content/Borrego_asu_0010N_10572.pdfhttp://repository.asu.edu/attachments/56537/content/Borrego_asu_0010N_10572.pdfhttp://www.bjs.gov/content/pub/pdf/mljsp0011.pdfhttp://bjs.ojp.usdoj.gov/index.cfm?ty=pbdetail&iid=379http://bjs.ojp.usdoj.gov/index.cfm?ty=pbdetail&iid=379http://www.bjs.gov/content/pub/pdf/ndcopuof.pdfhttp://www.bjs.gov/content/pub/pdf/ndcopuof.pdf
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
19/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 19
sustained atal injuries during the event, and i so, how theinjuries were sustained; whether the decedent was restrained inthe time leading up to the death, and i so, the type o restraintsdevices were used by law enorcement; and the decedent’sbehavior during the interaction with law enorcement (i.e.,appeared intoxicated, threatened officers, resisted beinghandcuffed or arrested, tried to escape or flee, physically fightwith the officers involved, or used a weapon to threaten orassault officers). Section A also included items regarding thetype o weapon that caused the death and the type o locationwhere the decedent expired (e.g., crime or arrest scene,medical acility, or en route to booking center).
SRCs were instructed to complete Section B o the CJ-11Aorm or instances in which the death occurred afer bookingwhile the decedent was being held in a temporary holdingacility (but beore arraignment transer to a long-termcorrectional acility). Section B included items on when thedecedent was booked into the law enorcement acility (i.e.,date and time o entry) and the decedent’s condition at thetime o booking (e.g., appeared intoxicated, exhibited mentalhealth problems, or exhibited medical problems). Section B
also included inormation about who caused the death (i.e.,deceased, other detainees, or law enorcement or correctionalstaff) or deaths attributed to homicide and accidental injuryand the means o death (e.g., firearm, blunt instrument,cutting instrument, strangulation, or intoxication) ornonnatural deaths.
Revisions to the instruments
BJS has continued to evaluate the CJ-11 and CJ-11A sincethe instruments were implemented in 2003. Te content othe CJ-11 summary orm has remained relatively stable. A
national listing o SRCs was added to the CJ-11 in 2005 andremained until 2013, when it was transerred to the CJ-11Aincident report. Te CJ-11 summary report was updated in2013 to include revised reporting instructions and the optionto complete the orm as an annual summary (in addition to thecontinued use o reporting quarterly inormation). In addition,SRCs were no longer required to disaggregate state-levelsummary counts by sex.
Te CJ-11A incident report has undergone several revisionssince the original version. Te first revision to the instrumentwas in 2007, which were minor and included (a) removing “stungun or aser” as a response to “What type o weapon(s) caused
the death?” and (b) adding “conducted energy device (e.g.,aser or stun-gun)” as a response to the type o restraint deviceused in the time leading up to the death or the events causingthe death. In 2008, a response o “No – medical/mental healthassistance call” was added as a response to “Had charges beenfiled against the deceased at the time o death?” In addition, aresponse o “Pepper spray, mace” was added as a response tothe type o restraint device used in the time leading up to thedeath or the events causing the death. Finally, “conducted energydevice” was added back to the responses categories or “Whattype o weapon(s) caused the death?”
Appendix A
CJ-11 and CJ-11A data collection instruments andrevisions, 2003–13
racking deaths in “the process o arrest” by state and locallaw enorcement presented a wide array o circumstances thatare difficult to capture. o establish a standardized reporting
orm and process, in 2002 the Bureau o Justice Statistics (BJS)began working with states and law enorcement proessionalorganizations to develop preliminary questionnaires. Teprocess yielded two data collection orms, the CJ-11 QuarterlySummary o Deaths in Law Enorcement Custody (commonlycalled the summary o incidents orm) and the CJ-11AAddendum, Law Enorcement Custodial Death Report(commonly called the incident report) (http://www.bjs.gov/index.cm?ty=dcdetail&iid=428#Questionnaires).
Te CJ-11 summary o incidents orm was to serve as aquarterly summary o arrest-related deaths occurring in eachstate disaggregated by sex. In addition, the CJ-11 providedinstructions or submitting data, examples o reportable deaths,and descriptions o the types o deaths that should be excludedrom the data collection. State reporting coordinators (SRCs)were asked to submit a completed CJ-11 orm indicatingtheir statewide count o reportable deaths or each quarter othe calendar year. SRCs were instructed to complete a CJ-11orm with a count o zero i no arrest-related deaths wereidentified in the state. BJS used these orms to determine stateparticipation in the Arrest-Related Deaths (ARD) program.SRCs were instructed to report zero counts to distinguishbetween the measured absence o an identified death andmissing data.
On the CJ-11A incident report, SRCs were asked to describe
the circumstances surrounding each arrest-related death theylisted on the summary o incidents orm, detailing one deathper CJ-11A orm. Te 2003 CJ-11A orm comprised 13 itemsto be completed or each death and an additional set o 5 itemsor those deaths that occurred beore booking (A1–A5) ora set o 4 items or deaths that occurred afer booking (B1–B4). It included items about the decedent, such as his or hername, date o birth, sex, and race/Hispanic origin. It eaturedquestions regarding the manner and cause o death and askedwhether the cause o death had been determined by a medicalexaminer or coroner. Te CJ-11A also captured inormationabout the decedent’s interaction with law enorcement (i.e., the
name and originating identification [ORI] number o the lawenorcement agency involved, location o the event, whethercharges had been filed against the decedent, and the mostserious offense with which the decedent was being charged atthe time o death) and whether the death occurred beore orafer booking.
For arrest-related deaths that occur beore booking, SRCs wereinstructed to complete Section A o the CJ-11A orm. SectionA included additional items about the incident between lawenorcement and the decedent, including whether the decedent
8/9/2019 BJS Arrest-Related Deaths Program Report (March 2015)
20/78
Arrest-Related Deaths Program: Data Quality Profile | March 2015 20
personnel, medical examiners and coroners, SRCs, andresearchers in the law enorcement and public health fields.Te ARD CJ-11A incident report was revised in 2012 in lighto comments received rom the meeting. Te revised 2013CJ-11A instrument was pilot tested in June 2012 with nine othe program’s current SRCs.
In June 2012, BJS asked nine o the program’s current SRCs toassess the revised instrument and related instructions in terms
o substance, breadth, and clarity. Specifically, BJS asked theSRCs to evaluate whether the questionnaire items enhancedthe quality, utility, and precision o the inormation to becollected and to note any items that were unclear, concernswith respect to the availability o the inormation requested,and response burden. Lastly, SRCs were asked to complete therevised version o the CJ-11A orm or a death record that waspreviously submitted to the ARD program.
BJS removed items that required respondents to speculate aboutoutcomes or that relied on respondents’ opinions or perceptions.For example, BJS removed items related to alleged criminalinvolvement in the events leading up to the death, removed
an item that had asked respondents to indicate whether thedecedent “appeared intoxicated” or “exhibited mental healthproblems.” BJS removed other items that were repetitive or couldbe determined rom responses to other items. For example,Item 13 i