Analysis of circumstances in police‐related fatalities comparing medical examiner records with a public database

Abstract

Assessing circumstances surrounding police‐related deaths is hampered by the absence of an official national database. The National Vital Statistics System, which compiles death certificates filed by medical examiners and coroners, misses approximately one‐half of deaths due to police use of force. The void has been filled by the emergence of several unofficial databases sourced from public records, which have been used, for example, to estimate the number of fatal police shootings of unarmed individuals. However, these public databases have not been validated against official records. Furthermore, there have been no attempts to explore the complexity of circumstances leading to police‐related fatalities. The present study was conducted to compare information in one public database, Fatal Encounters (FE), with matched records from the King County Medical Examiner's Office (ME) and to measure the role of overlapping circumstances leading to fatal outcomes. For this purpose, the two data sources were linked by names. Circumstantial elements were extracted, categorized, and analyzed according to Manner of Death, Pre‐Custody criteria, and Race assignments. Overlapping circumstances were resolved by the calculation of conditional probabilities. The results found 172 matched records in which agreements between the two sources ranged from 70% to 100%. Disagreements were present in whether the decedent was unarmed and not fleeing. The distribution of deaths according to race was disproportionate relative to the county population. The study demonstrated the importance of data in measuring police‐related fatalities and the need for caution in interpreting data that does not fully encompass the range of complex circumstances.

Highlights.

• A database sourced on public records of police‐related deaths was linked to medical examiner data.

• Agreement between the two data sources was variable and had important differences.

• Analysis of conditional probabilities revealed the role of multiple overlapping circumstances.

• Data sourced from public records needs to be interpreted with attention to all circumstances.

• In the absence of a comprehensive national database, public opinion may rest on incomplete data.

1. INTRODUCTION

Tracking deaths due to police actions is complicated by the absence of a national government‐sponsored data collection system. It is estimated that the National Vital Statistics System misses approximately half of all unnatural deaths, including homicides, caused by police officers [1, 2, 3, 4, 5, 6]. This deficiency is at least partly due to the failure of medical examiners and coroners in providing relevant information on death certificates [4] and has been addressed by the National Association of Medical Examiners (NAME) [7] and by a recent study of death certificates issued by a medical examiner's office over 30 years [8]. Of the studies published about police‐related deaths, few go beyond the numbers and explore the complex circumstances leading to fatalities [8, 9, 10]. In the absence of a national data repository of police‐related deaths, several unofficial, public databases have appeared that contain information collected through public records and the internet [11, 12, 13, 14, 15, 16, 17, 18, 19]. Although these contain a wealth of information, including circumstances, their inclusion criteria and reporting methods are variable [11, 12], and none have been compared to data collected by government agencies. To compare circumstantial information in one of these public databases with medical examiner records, the present study was conducted and designed around linking records in the Fatal Encounters database [16, 18] with those in the database of the King County Medical Examiner's Office (KCMEO) in Seattle, Washington. As of March 2026, Fatal Encounters contained over 31,000 records from across the United States during the years 2000 to 2021. The Fatal Encounters website [16] describes their three main methods of information collection: paid researchers, public records requests, and crowdsourced data. The objective of this study was to compare and analyze circumstantial information contained in the linked Fatal Encounters and medical examiner datasets.

2. METHODS AND MATERIALS

For this study, the Fatal Encounters database (FE) was downloaded through its website [16] in January 2025. Records from the KCMEO database (ME) were linked, by names, to records in FE. Only circumstantial elements of FE were retained. Circumstances in FE were contained in five separate columns: (1) Highest level of force; (2) Armed/Unarmed; (3) Alleged weapon; (4) Aggressive physical movement; and (5) Fleeing/Not Fleeing. Because each of the five columns contained multiple descriptors, FE was further processed into 15 categories by taking each column as a group. From the FE group, Highest level of force, the categories of Gunshot, Vehicle, and Other were extracted. From the Armed/Unarmed group, the categories of Armed, Unarmed, and Other/Uncertain were extracted. From the Alleged weapon group, categories of Firearm, Sharp Object, and Other were extracted. From the FE group, Aggressive physical movement, the categories Brandished weapon, Used weapon, Threatening action, and Other were extracted. Finally, from the Fleeing/Not fleeing group, the categories Fleeing and Not fleeing were extracted. Subcategories captured additional information in the major categories.

From ME records, circumstantial categories were developed and populated by manually abstracting medicolegal death investigator (MDI) narrative reports. The MDI reports came directly from police accounts at the time of the fatalities. Circumstantial details, or elements, were extracted from the MDI reports and placed into one or more of the major categories developed around the following themes, in which the decedent was:

• involved in an altercation or displayed threatening behavior,

• wielding or using a weapon,

• involved in an incident involving a motor vehicle,

• fleeing from police, either on foot or in a motor vehicle,

• suspected of a robbery or other felony,

• involved in a domestic violence incident or domestic dispute,

• in an armed standoff with police or a Special Weapons and Tactics (SWAT) response,

• displaying suicidal, intoxicated, bizarre, or otherwise aberrant behavior,

• subjected to less‐than‐lethal means in attempts to subdue,

Subcategories described how the incidents fit into the major categories.

Following the NAME recommendations [7] that were used in a previous study [8], ME records were assigned into four Pre‐Custody categories: (1) Shooting – deaths from gunshots fired by law enforcement; (2) Pursuit – deaths from collisions following law enforcement initiated vehicular pursuits, including deaths of persons other than the pursued subjects; (3) Restraint – deaths without major physical injury following restraint during apprehension by law enforcement officers; and (4) Other – deaths not fitting into the preceding subcategories, such as those due to suicides or accidents when decedents acted to avoid arrest, blunt force injuries sustained during arrest, or injuries from collisions caused by law enforcement vehicles not in pursuit. ME records also contained Manner of Death and Race assignments used for this study.

Data analysis of the linked data sets was performed with Python/Pandas. From the analyses, tables were constructed to evaluate circumstantial elements in the following ways:

(1) describing circumstantial elements in both sets of records; (2) categorizing circumstantial elements according to medical examiner designations; (3) comparing circumstantial elements found in FE with those in ME; and (4) describing overlapping circumstances in records having more than a single circumstantial element.

3. RESULTS

3.1. Matched data sources

The Fatal Encounters database (FE), downloaded in January 2025, contained 206 records from King County, Washington, covering the years 2000–2021. Twelve names of decedents were listed as “Name withheld by police.” Of the remaining 194 records in FE, 178 (91.8%) were successfully linked by names to ME records. Six records were excluded from this study after the author's review found no evidence of police involvement, leaving 172 FE records matching ME records. These 172 matched records formed the basis of the present study. In contrast to the previous study, which examined both In‐Custody and Pre‐Custody deaths [8], the present study included only Pre‐Custody deaths, which was the sole focus of FE.

3.2. Fatal encounters (FE) records

Table 1 displays the circumstantial elements in the 172 matched FE records in the 15 categories extracted from the five FE groups: Highest level of force, Armed/Unarmed, Alleged weapon, Aggressive physical movement, Fleeing/Not Fleeing. Categories' abbreviations were used to conserve space in subsequent tables. The categories in the first FE group, Highest level of force, refer to the major force causing death: Gunshot (HGS), Vehicle (HVE), and Other (HOT). The Other category (HOT) of the Highest level of force group contained 10 various conditions. It is important to acknowledge that FE included no death certificate information; thus, the entries in the Highest level of force group were based solely on public records. Multiple circumstantial elements were apparent in several FE records, ranging from 1 (85 records) to 5 (64 records), with a mean of 2.8 and a median of 3 circumstances per record. Through the years, 2000–2010, only the categories in the FE group, Highest level of force, had non‐blank entries. Thus, the percentages in the other groups were calculated on 98 completed entries, while those in the Highest level of force group were calculated on 172.

TABLE 1.

Circumstances of police‐related deaths from FE records according to FE circumstance groups and categories.

FE circumstance group (number)	Category	Abbreviation	N (%*)	FE subcategory	n
FE Group: Highest level of force (172)	Gunshot	HGS	141 (82.0)	‐	‐
	Vehicle	HVE	21 (12.2)	‐	‐
	Other	HOT	10 (5.8)
				Tasered	5
				Asphyxiated/Restrained	2
				Medical emergency (1), Beaten/Bludgeoned (1), Drowned (1)	3
FE Group: Armed/Unarmed (87)	Armed	AMY	63 (64.3)	‐	‐
	Unarmed	AMN	22 (22.4)	‐	‐
	Other/Uncertain	AMU	2 (2.0)	‐	‐
FE Group: Alleged weapon (65)	Firearm	WFA	39 (39.8)	‐	‐
	Sharp object	WSH	20 (20.4)	‐	‐
	Other/Uncertain	WOT	6 (6.1)
				Pipe (1), Stun gun (1), Golf club (1), Pen (1), Uncertain (2)	6
FE Group: Aggressive physical movement (73)	Brandished weapon	GBR	23 (23.5)	‐	‐
	Used weapon	GUS	17 (17.4)	‐	‐
	Threatened	GTH	25 (25.5)
				Advanced toward officer(s)	14
				Sudden threatening movement	8
				Threatened with vehicle	2
				Reached to waist	1
	Other	GOT	8 (8.2)
				Struggled	4
				Self‐inflicted injury	2
				Uncertain	2
FE Group: Fleeing/Not Fleeing (87)	Fleeing	FLY	28 (28.6)
				Fleeing/Foot	14
				Fleeing/Vehicle	13
				Fleeing/Motorcycle	1
	Not fleeing	FLN	59 (60.2)	‐	‐‐‐

^*Percents calculated from denominator equal to number of records with completed entries: 172 for FE Group: Highest level of force; 98 for all other FE groups.

Table 2 arranges FE circumstances according to ME designations: Manner of Death, Pre‐Custody category, and Race assignment. This allows interpretations reading across rows, as follows: of 138 deaths certified Homicide, 136 (98.6%) were due to gunshot (HGS); 59 (43%) decedents were armed (AMY); 12 (9%) were unarmed (AMN); 37 (27%) carried a firearm (WFA); 19 (14%) carried a knife or other sharp object (WHS); 19 (14%) were fleeing (FLY); and 53 (38%) were not fleeing (FLN). Reading downward in these columns and using denominators from Table 1 finds that of the 63 deaths of those who were armed (AMY), 59 (94%) were in the Shooting Pre‐Custody category, and the same number were certified Homicide; two (3%) were in each of the Restraint or Other Pre‐Custody categories, and the same four were certified either Suicide (2) or Undetermined (2). In the Race Assignment group, 103 (60%) were assigned White, 46 (27%) Black, and 15 (8.7%) Asian/PI. From these numbers, the ratios of White:Black:Asian/PI in this table were 1.0:0.45:0.14, evidencing disproportionality with respect to the 2010 King County general population [20], in which the White:Black:Asian/PI ratios were 1.0:0.09:0.21.

TABLE 2.

Circumstances from FE records according to ME designations.

ME designation	Attribute	N	FE circumstances n (% of N)
			HGS	HVE	AMY	AMN	WFA	WSH	GBR	GUS	GTH	FLY	FLN
Manner	Homicide	138	136 (98.6)	1 (0.7)	59 (42.8)	12 (8.7)	37 (26.8)	19 (13.8)	21 (15.2)	17 (12.3)	25 (18.1)	19 (13.8)	53 (38.4)
	Accident	24	0 (0.0)	20 (83.3)	0 (0.0)	10 (41.7)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	7 (29.2)	4 (16.7)
	Suicide	5	5 (100.0)	0 (0.0)	2 (40.0)	0 (0.0)	2 (40.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	1 (20.0)	1 (20.0)
	Undetermined	5	0 (0.0)	0 (0.0)	2 (40.0)	0 (0.0)	0 (0.0)	1 (20.0)	2 (40.0)	0 (0.0)	0 (0.0)	1 (20.0)	1 (20.0)
Pre‐custody category	Shooting	137	137 (100.0)	0 (0.0)	59 (43.1)	10 (7.3)	37 (27.0)	19 (13.9)	21 (15.3)	17 (12.4)	25 (18.2)	19 (13.9)	51 (37.2)
	Pursuit	19	0 (0.0)	19 (100.0)	0 (0.0)	9 (47.4)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	7 (36.8)	3 (15.8)
	Restraint	8	0 (0.0)	0 (0.0)	2 (25.0)	0 (0.0)	0 (0.0)	1 (12.5)	2 (25.0)	0 (0.0)	0 (0.0)	1 (12.5)	1 (12.5)
	Other	8	4 (50.0)	2 (25.0)	2 (25.0)	3 (37.5)	2 (25.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	1 (12.5)	4 (50.0)
Race assignment	White	103	84 (81.6)	15 (14.6)	39 (37.9)	15 (14.6)	26 (25.2)	11 (10.7)	11 (10.7)	12 (11.6)	18 (17.5)	17 (16.5)	38 (36.9)
	Black	46	36 (78.3)	6 (13.0)	16 (34.8)	4 (8.7)	10 (21.7)	5 (10.9)	7 (15.2)	5 (10.9)	5 (10.9)	10 (21.7)	11 (23.9)
	Asian/PI	15	13 (86.7)	0 (0.0)	4 (26.7)	3 (20.0)	0 (0.0)	3 (20.0)	2 (13.3)	0 (0.0)	2 (13.3)	1 (6.7)	6 (40.0)
	NatAmer	5	5 (100.0)	0 (0.0)	3 (60.0)	0 (0.0)	2 (40.0)	1 (20.0)	3 (60.0)	0 (0.0)	0 (0.0)	0 (0.0)	3 (60.0)
	Other/NotSpec	3	3 (100.0)	0 (0.0)	1 (33.3)	0 (0.0)	1 (33.3)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	1 (33.3)

Note: Highest level of force – Gunshot (HGS), Vehicle (HVE); Armed (AMY), Unarmed (AMN); Alleged weapon – Firearm (WFA), Sharp object (WSH); Aggressive physical movement – Brandished weapon (GBR), Used weapon (GUS), Threatened (GTH); Fleeing (FLY), Not fleeing (FLN).

3.3. Medical examiner (ME) records

Table 3 details the circumstantial elements of ME records within 10 categories along with their corresponding subcategories. The largest category, Altercation/Threatening (ALT), comprised 109 (64.5%) of the 172 records and described altercations or threatening behaviors involving the decedent with police officer(s), with other(s) besides police officer(s), or with other(s) and police officer(s). The following categories recorded circumstances in a similar fashion: those in which the decedents were Wielding (WDG) or Using weapons (USG); in Traffic/Involving vehicle (TRV) incidents; Fleeing (FLG) from police; involved in a suspected Robbery or other felony (ROF); in a reported Domestic violence/disturbance (DVD) incident; in an armed Standoff (STO) with police or an incident having Special Weapons and Tactical (SWAT) response; or reported or witnessed to display Aberrant behavior (BEH). The BEH subcategory of Suicidal behavior included cases in which the decedent manifested explicit or implicit intent to die in the police encounter, either by a self‐inflicted wound or by forcing police to deliver the fatal wound. The final category comprised deaths in which police attempted unsuccessful Less than lethal (LTL) means to subdue the decedent. Multiple circumstances were apparent in several ME records. The number of circumstantial elements in each of the 172 records ranged from 0 (11 records) to 6 (3 records), with a mean of 2.6 and a median of 3 overlapping circumstances per record. Because of the overlaps, the set of circumstantial elements was not treated as independent occurrences or events. Instead, circumstances required a different level of analysis, as described later in Section 3.5.

TABLE 3.

Circumstances of 172 police‐related deaths from ME records according to ME circumstance categories.

ME circumstance category	Abbreviation	N (% of 172)	ME subcategory	n
Altercation/threatening	ALT	111 (64.5)	Police	72
			Other individual(s)	14
			Police and other(s)	25
Wielding weapon	WDG	93 (54.1)	Firearm	57
			Sharp object	30
			Firearm and sharp object	4
			Other	2
Using weapon	USG	49 (28.5)	Firearm	35
			Sharp object	13
			Firearm and sharp object	1
Traffic/involving vehicle	TRV	47 (27.3)	Traffic stop or violation	17
			Vehicular pursuit	16
			Threatened police with vehicle	4
			Miscellaneous	10
Fleeing, foot or vehicle	FLG	46 (26.7)	Fleeing/foot	14
			Fleeing/vehicle	26
			Fleeing/foot and vehicle	6
Robbery/other felony	ROF	26 (15.1)	Robbery	14
			Stolen vehicle	9
			Other	3
Domestic violence/disturbance	DVD	25 (14.5)	‐	‐
Standoff or SWAT	STO	20 (11.6)	Standoff	12
			SWAT	5
			Standoff and SWAT	3
Aberrant behavior	BEH	18 (10.5)	Suicidal	11
			Bizarre or intoxicated	7
Less than lethal means	LTL	16 (9.3)	Taser	9
			OC spray	1
			Both taser and OC spray	3
			Projectile, “chemical gas”	3

ME circumstances according to attributes in each of the ME designations, Manner of Death, Pre‐Custody Category, and Race Assignment are shown in Table 4. Information in each row can be read as previously described. For example, in the Manner of Death designations, of 138 deaths certified Homicide, the decedent was involved in an altercation or displayed threatening behavior (ALT) in 103 (76%), was wielding a weapon (WDG) in 87 (63%), and using a weapon (USG) in 44 (32%). The rows describing deaths in the Pre‐Custody designations can be read in the same fashion. Of 103 decedents who had a Race assignment of White, 61 (59%) were in an altercation or displayed threatening behavior, 57 (55%) were wielding a weapon, and 29 (28%) were using a weapon. Although the overall racial disproportionalities in Table 4 are the same as in Table 2, there are variable differences in the proportions within the individual ME Circumstance categories. The differences are most evident in the ALT category.

TABLE 4.

Circumstances from ME records according to ME designations.

ME designation	Attribute	N	ME circumstances n (% of N)
			ALT	WDG	USG	TRV	FLG	ROF	DVD	STO	BEH	LTL
Manner	Homicide	138	105 (76.1)	87 (63.0)	44 (31.9)	25 (18.1)	29 (21.0)	22 (15.9)	22 (15.9)	18 (13.0)	12 (8.7)	9 (6.5)
	Accident	24	1 (4.2)	0 (0.0)	0 (0.0)	21 (87.5)	15 (62.5)	4 (16.7)	0 (0.0)	1 (4.2)	1 (4.2)	1 (4.2)
	Suicide	5	2 (40.0)	5 (100.0)	5 (100.0)	1 (20.0)	2 (40.0)	0 (0.0)	1 (20.0)	1 (20.0)	2 (40.0)	1 (20.0)
	Undetermined	5	3 (60.0)	1 (20.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	2 (40.0)	0 (0.0)	3 (60.0)	5 (100.0)
Pre‐custody category	Shooting	137	104 (75.9)	88 (64.2)	45 (32.8)	24 (17.5)	29 (21.2)	22 (16.1)	21 (15.3)	18 (13.1)	12 (8.8)	9 (6.6)
	Pursuit	19	0 (0.0)	0 (0.0)	0 (0.0)	19 (100.0)	14 (73.7)	4 (21.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
	Restraint	8	4 (3.7)	1 (12.5)	0 (0.0)	1 (12.5)	0 (0.0)	0 (0.0)	2 (25.0)	0 (0.0)	4 (50.0)	6 (75.0)
	Other	8	3 (50.0)	4 (50.0)	4 (50.0)	3 (37.5)	3 (37.5)	0 (0.0)	2 (25.0)	2 (25.0)	2 (25.0)	1 (12.5)
Race assignment	White	103	61 (59.2)	57 (55.3)	29 (28.2)	28 (27.2)	30 (29.1)	15 (14.6)	12 (11.6)	12 (11.6)	13 (12.6)	9 (8.7)
	Black	46	32 (69.6)	22 (47.8)	13 (28.3)	13 (28.3)	11 (23.9)	10 (21.7)	6 (13.4)	4 (8.7)	3 (6.5)	4 (8.7)
	Asian/PI	15	14 (93.3)	9 (60.0)	5 (33.3)	4 (26.7)	3 (20.0)	1 (6.7)	5 (33.3)	3 (20.0)	1 (6.7)	1 (6.7)
	NatAmer	5	3 (60.0)	4 (80.0)	1 (20.0)	0 (0.0)	1 (20.0)	0 (0.0)	1 (20.0)	0 (0.0)	1 (20.0)	1 (20.0)
	Other/NotSpec	3	1 (33.3)	1 (33.3)	1 (33.3)	2 (66.7)	1 (33.3)	0 (0.0)	1 (33.3)	1 (33.3)	0 (0.0)	1 (33.3)

Abbreviations: ALT, Altercation or threatening behavior; BEH, Aberrant behavior; DVD, Domestic violence or disturbance; FLG, Fleeing; LTL, Less than lethal means; ROF, Robbery or other felony; STO, Standoff or SWAT response; TRV, Traffic or involving vehicle; USG, Using weapon; WDG, Wielding weapon.

Reading down the N column in Table 4 provides another view of the data. Of 172 deaths, 138 (80%) were certified Homicide, 24 (14%) Accident, 5 (3%) Suicide, and 3 (1.7%) Undetermined. Reading down the other columns in Table 4, but using the corresponding denominators from Table 3, describes the remaining data. For example, of 109 deaths of decedents involved in altercations or displaying threatening behavior (ALT), 103 (94%) were certified Homicide, 1 (0.92%) certified Accident, 2 (1.8%) certified Suicide, and 3 (2.8%) certified Undetermined.

3.4. Comparison of fatal encounters and medical examiner records

Comparing data between FE and ME highlights similarities and differences. In Table 2, the numbers in the FE category, Highest level of force – Gunshot (HGS), corresponded well with those in the ME categories, Homicide and Pre‐Custody – Shooting. The numbers in the FE category, Highest level of force – Vehicle (HVE), corresponded well with the ME categories, Accident and Pre‐Custody– Pursuit. Table 5 allows comparisons in other analogous categories. The agreement between the FE category, Highest level of force – Vehicle (HVE), and the ME category, Traffic or involving vehicle (TRV), was perfect. In those referring to decedents who were armed, agreements were 79% between the FE category, Armed (AMY), and the ME category, Wielding weapon (WDG), 77% between the FE category, Alleged weapon – Firearm (WFA), and the ME category, Wielding weapon (WDG), 90% between the FE category, Alleged weapon – Sharp object (WSH), and the ME category, Wielding weapon (WDG), 74% between the FE category, Aggressive physical movement – Brandished weapon (GBR), and the ME category, Wielding weapon (WDG), and 88% between the FE category, Aggressive physical movement – Used weapon (GUS) and the ME category, Using weapon (USG). Agreement was also 88% between the FE category, Aggressive physical movement – Threatened (GTH), and the ME category, Altercation/Threatening (ALT). In 50 records in which both FE and ME specified weapons, agreement was 96%, with 28 of 30 records matching firearms, 18 of 18 records matching sharp objects, and 2 of 2 records matching other objects.

TABLE 5.

Circumstances in FE records compared to circumstances in ME records.

FE circumstances	N	Circumstances in ME records n (% of N)
		ALT	WDG	USG	TRV	FLG	ROF	DVD	STO	BEH	LTL
HGS	141	106 (75.2)	92 (65.2)	49 (34.8)	25 (17.7)	31 (22.0)	22 (15.6)	22 (15.6)	19 (13.5)	14 (9.9)	10 (7.1)
HVE	21	0 (0.0)	0 (0.0)	0 (0.0)	21 (100.0)	14 (66.7)	4 (19.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
HOT	10	5 (50.0)	1 (10.0)	0 (0.0)	1 (10.0)	1 (10.0)	0 (0.0)	3 (30.0)	1 (10.0)	4 (40.0)	6 (60.0)
AMY	63	52 (82.5)	50 (79.4)	25 (39.7)	8 (12.7)	14 (22.2)	8 (12.7)	11 (17.5)	9 (14.3)	5 (7.9)	7 (11.1)
AMN	22	9 (40.9)	4 (18.2)	2 (9.1)	14 (63.6)	13 (59.1)	3 (13.6)	3 (13.6)	1 (4.6)	0 (0.0)	0 (0.0)
AMU	2	1 (50.0)	0 (0.0)	0 (0.0)	2 (100.0)	1 (50.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
WFA	39	30 (76.9)	30 (76.9)	20 (51.3)	4 (10.3)	9 (23.1)	5 (12.8)	4 (10.3)	9 (23.1)	2 (5.1)	1 (2.6)
WSH	20	18 (90.0)	18 (90.0)	5 (25.0)	4 (20.0)	5 (25.0)	3 (15.0)	4 (20.0)	0 (0.0)	1 (5.0)	5 (25.0)
WOT	6	5 (83.3)	2 (33.3)	0 (0.0)	2 (33.3)	1 (16.7)	0 (0.0)	3 (50.0)	0 (0.0)	2 (33.3)	1 (16.7)
GBR	23	18 (78.3)	17 (73.9)	5 (21.7)	4 (17.4)	3 (13.0)	1 (4.4)	4 (17.4)	1 (4.4)	3 (13.0)	3 (13.0)
GUS	17	15 (88.2)	15 (88.2)	15 (88.2)	1 (5.9)	5 (29.4)	3 (17.6)	3 (17.6)	6 (35.3)	1 (5.9)	0 (0.0)
GTH	25	22 (88.0)	19 (76.0)	5 (20.0)	4 (16.0)	7 (28.0)	6 (24.0)	3 (12.0)	0 (0.0)	1 (4.0)	3 (12.0)
GOT	8	6 (75.0)	3 (37.5)	2 (25.0)	2 (25.0)	2 (25.0)	0 (0.0)	3 (37.5)	2 (25.0)	0 (0.0)	1 (12.5)
FLY	28	17 (60.7)	11 (39.3)	7 (25.0)	12 (42.9)	17 (60.7)	6 (21.4)	2 (7.1)	3 (10.7)	1 (3.6)	1 (3.6)
FLN	59	45 (76.3)	43 (72.9)	20 (33.9)	12 (20.3)	11 (18.6)	5 (8.5)	12 (20.3)	7 (11.9)	4 (6.8)	6 (10.2)

Note: Highest level of force – Gunshot (HGS), Vehicle (HVE), Other (HOT); Armed (AMY), Unarmed (AMN), Uncertain (AMU); Alleged weapon – Firearm (WFA), Sharp object (WSH), Other (WOT); Aggressive physical movement – Brandished weapon (GBR), Used weapon (GUS), Threatened (GTH), Other (WOT); Fleeing (FLY), Not fleeing (FLN).

Abbreviations: ALT, Altercation or threatening behavior; BEH, Aberrant behavior; DVD, Domestic violence or disturbance; FLG, Fleeing; LTL, Less than lethal means; ROF, Robbery or other felony; STO, Standoff or SWAT response; TRV, Traffic or involving vehicle; USG, Using weapon; WDG, Wielding weapon.

Disagreements between FE and ME records were most apparent in the FE negation categories, Unarmed (AMN) and Not fleeing (FLN). In 22 FE records, the decedent was said to have been unarmed. Of these, 10 were in the FE group, indicating they died by gunshot; 10 were in the FE group, indicating they died by vehicle; and two were in the FE group, indicating they died by other means. Of the same 22 deaths FE reported as unarmed, 10 were in the ME Pre‐Custody – Shooting designation, all certified Homicide; nine were in the Pre‐Custody – Pursuit designation, eight certified Accident and one Homicide; and three were in the Pre‐Custody – Other designation, two certified Accident and one Homicide. Of the 10 classified by FE as unarmed, all of which were in the ME Pre‐Custody – Shooting designation, eight were in the ME circumstance category, Altercation/Threatening (ALT), four in the ME circumstance category, Wielding weapon (WDG), two in the ME circumstance category, Using weapon (USG), two in the ME circumstance category, Robbery or other felony (ROF), five in the ME circumstance category, Fleeing (FLG) – four by vehicle and one by foot, two in the ME circumstance category, Domestic violence or disturbance (DVD), and four in the ME circumstance category, Traffic or involving vehicle (TRV). Only one of the 10 in the ME Pre‐Custody – Shooting category fell into none of the ME Circumstance categories. Nine were shot by police in incidents having multiple overlapping circumstances: three fell into 2 categories; three into three categories; and three into four categories. Of 59 FE records indicating the decedent was not fleeing (FLN), the ME records indicated 11 (19%) were fleeing. Further resolution was not possible because the ME records had no explicit negation categories.

In the FE group, Highest level of force, records indicated that “Tasered” was responsible for five deaths, and “Asphyxiated/Restrained” was responsible for two deaths. All seven of these were accounted for in the ME Restraint Pre‐Custody category; two of which were certified by manner, Accident, and five were certified Undetermined. In the ME Circumstance category, indicating police used less than lethal methods (LTL), 12 deaths were associated with taser deployment in attempts to subdue the decedents; six of these were in the ME – Shooting Pre‐Custody category and certified Homicide, and six were in the ME – Restraint Pre‐Custody category with five certified Undetermined and one certified Accident. No death certificates attributed the Cause of Death to tasers.

3.5. Conditional probabilities of circumstances

As mentioned earlier, because of overlapping circumstances, the set of circumstantial elements was not analyzed as if they represented independent occurrences or events. Instead, they were analyzed as conditional probabilities [21]. Table 6 shows the relationships between pairs of circumstantial elements in the ME records. Conditional probabilities are commonly written as P(B|A), meaning that the probability of B occurring, given or knowing A, is expressed as P(B|A). P(B) is the individual, or prior probability of B occurring alone, prior to the condition of A. Following this convention in Table 6, the percentages in brackets express the prior probabilities of the circumstantial elements heading each column. The percentages in parentheses in the column below the bracketed prior probabilities express the conditional probabilities of the same elements, conditioned by the circumstances given by the row labels, in the sense, P(Column|Row). Proof that pairs of circumstances were not independent is given by the inequalities of their prior and conditional probabilities. Two events are independent only if P(A) = P(A|B) and P(A) = P(B|A), which simply means that the occurrence of one has no effect on the other. In this sample set of 172 deaths due to police actions, the prior probability of vehicle involvement without other conditions was P(TRV) = 27%. However, the probability of TRV within the condition of a decedent fleeing was P(TRV|FLG) = 54%. For another example: in the same sample set, the prior probability of the decedent being in a fight or displaying threatening behavior without other conditions was P(ALT) = 64%. However, if the decedent was wielding a weapon, the conditional probability of the decedent being in a fight or displaying threatening behavior was P(ALT|WDG) = 82%. The data can also be configured to express negations (~), as follows: in this group of 172 deaths due to police actions, the prior probability of the decedent wielding any weapon was P(WDG) = 54%. However, if the decedent was not using the weapon and not fighting or displaying threatening behavior, the conditional probability was P(~ALT| ~ USG|WDG) = 18%. Restricting the conditions further to sharp objects, the conditional probability was P(~ALT| ~ USG|WDG = “sharp object”) = 10%. Looking behind the data found this subset to include only two deaths. One was a decedent wielding a knife who was shot while fleeing from a suspected robbery. The other was shot by a police officer while crossing the street, simply waving a knife.

TABLE 6.

Conditional probabilities for pairs of ME circumstantial elements.

Circumstances in ME records	N	Circumstances in ME records [% of 172] n (% of N)
		ALT [64.5]	WDG [54.1]	USG [28.5]	TRV [27.3]	FLG [26.7]	ROF [15.1]	DVD [14.5]	STO [11.6]	BEH [10.5]	LTL [9.3]
ALT	111	111 (100.0)	76 (68.5)	39 (35.1)	21 (18.9)	23 (20.7)	14 (12.6)	22 (19.8)	15 (13.5)	12 (10.8)	13 (11.7)
WDG	93	76 (81.7)	93 (100.0)	49 (52.7)	10 (10.8)	22 (23.7)	13 (14.0)	17 (18.3)	16 (17.2)	13 (14.0)	9 (9.7)
USG	49	39 (79.6)	49 (100.0)	49 (100.0)	4 (8.2)	11 (22.4)	7 (14.3)	6 (12.2)	11 (22.4)	9 (18.4)	3 (6.1)
TRV	47	21 (44.7)	10 (21.3)	4 (8.5)	47 (100.0)	25 (53.2)	7 (14.9)	1 (2.1)	0 (0.0)	2 (4.3)	2 (4.3)
FLG	46	23 (50.0)	22 (47.8)	11 (23.9)	25 (54.4)	46 (100.0)	16 (34.8)	6 (13.0)	4 (8.7)	2 (4.4)	3 (6.5)
ROF	26	14 (53.8)	13 (50.0)	7 (26.9)	7 (26.9)	16 (61.5)	26 (100.0)	0 (0.0)	3 (11.5)	0 (0.0)	3 (11.5)
DVD	25	22 (88.0)	17 (68.0)	6 (24.0)	1 (4.0)	6 (24.0)	0 (0.0)	25 (100.0)	5 (20.0)	5 (20.0)	5 (20.0)
STO	20	15 (75.0)	16 (80.0)	11 (55.0)	0 (0.0)	4 (20.0)	3 (15.0)	5 (25.0)	20 (100.0)	1 (5.0)	2 (10.0)
BEH	18	12 (66.7)	13 (72.2)	9 (50.0)	2 (11.1)	2 (11.1)	0 (0.0)	5 (27.8)	1 (5.6)	18 (100.0)	5 (27.8)
LTL	16	13 (81.2)	9 (56.2)	3 (18.8)	2 (12.5)	3 (18.8)	3 (18.8)	5 (31.2)	2 (12.5)	5 (31.2)	16 (100.0)

Abbreviations: ALT, Altercation or threatening behavior; BEH, Aberrant behavior; DVD, Domestic violence or disturbance; FLG, Fleeing; LTL, Less than lethal means; ROF, Robbery or other felony; STO, Standoff or SWAT response; TRV, Traffic or involving vehicle; USG, Using weapon; WDG, Wielding weapon.

4. DISCUSSION

Beyond the challenge of accurately documenting circumstances, which requires independent and careful investigation, it is necessary to understand the dynamics of circumstances leading to the fatal outcome. Like the present one, previous studies [8, 10] have shown altercations, threatening behavior with and without weapons, and criminal activities to be major circumstances in police‐related fatalities. However, there is rarely a single “circumstance;” nearly always, there is a complex of decisions and actions on the part of both police and decedent. In this context, the present study analyzed records of circumstances from the same collection of police‐related fatalities in linked datasets from two different sources: one from medical examiner records (ME) and the other (FE) assembled from public records. These two data sets reflect different perspectives, one based on public records and the other on medical examiner records, which in turn were based largely on police accounts. Circumstances from each source were grouped into the medical examiner designations of Manner of Death, Pre‐Custody category, and Race assignment. Comparison of the two sources revealed differences between the two perspectives. From the FE perspective, all 172 deaths were presented as due solely to police actions, whether by gunshot, vehicle, or some other means. From the ME perspective of Manner of Death, nearly 80% were certified Homicide, meaning that police actions directly caused the death. However, 14% of deaths were certified Accident, mostly due to collisions following police pursuits; 3% were certified Suicide, all due to self‐inflicted gunshot wounds; and 1.7% were certified Undetermined. Thus, out of 172 deaths in this study, approximately 17% of deaths were due to actions of the decedent, either by self‐inflicted gunshot wounds or leading police in vehicular pursuits. These distinctions were not apparent in the FE records but may be important for public reaction, assessing police actions, liability, and policies. Furthermore, the FE data gave no indication of activities the decedent initiated toward the fatal event, such as fleeing from a robbery or other crime, assaulting or threatening individuals other than police officers, such as in domestic violence, or displaying aberrant and potentially violent behavior. Reconciling the Other category of the FE group, Highest Level of Force, was complicated by having no official Cause of Death from death certificates, which led to misrepresenting tasers in causing death. The FE records obscure the circumstances of decedents said to be unarmed. Of the 22 decedents reported as unarmed in FE, 10 were certified Accident, and nine were in ME Pre‐Custody category, Pursuit. Limiting deaths to the 10 unarmed individuals shot by police, it was found that nine had multiple overlapping circumstances reported in the ME records, including threatening behavior and wielding or using weapons. Furthermore, classifying “self‐inflicted injury” in the FE group, Aggressive physical movement, oversimplifies a complex interaction.

Comparisons between analogous FE and ME circumstance categories found the level of agreement to range from 75% to 100%. The comparisons between ME and FE were complicated for several reasons. As mentioned previously, FE records included neither Manner of Death nor Pre‐Custody categories. FE records had to be processed to extract the individual circumstantial elements. Additional complications in making comparisons were due to differing language in the two data sets and by criteria in making assignments, such as distinctions between armed, wielding, and brandishing a weapon. The word, “altercation,” frequently appeared in ME data but never in FE. Records in FE were primarily focused on whether the decedent was armed, with what type of weapon, unarmed, engaged in threatening behavior, and fleeing or not fleeing. Circumstantial categories in ME records were created in this study to allow comparisons with FE and to extend the range of circumstances. Despite the complications, the ME category, Altercation/Threatening was analogous to FE categories extracted from the FE group, Aggressive physical movement. The ME categories Wielding weapon and Using weapon were directly analogous to FE categories. FE included two negation categories that were not in ME, and five ME categories were not in FE. Both FE and ME data showed disproportionalities in Race assignments relative to the county population. Discussions about race assignments and views on racial bias are outside the scope of the present study but are addressed in a previous one [8].

Not surprisingly, capturing the essence of complex interactions in data was challenging. Table 6 was constructed to show conditional probabilities between pairs of circumstantial elements. For example, in incidents involving an altercation or threatening behavior, probabilities increased under conditions of an armed decedent or domestic violence. Probabilities were low if less than lethal means were used but increased when the decedent displayed aberrant behavior. Of course, the probabilities apply only to the present population of 172 deaths due to police use of force and are not generalizable. More meaningful estimates would need to be based on all incidents of police use of force in which there was bodily injury, both fatal and nonfatal. Nevertheless, the approach advanced in the present study may reveal insights reflecting decedent and police actions leading to a fatal outcome. A singular example was in the final sentence of the Results section, which referenced a low probability in the death of a Native American woodcarver who was shot while waving a knife at a police officer. His death was ruled “not justified” by the police department firearms review board [22] and eventually led to legislated police reform [23] and a consent decree with the US Department of Justice [24, 25].

The present study had several limitations. In addition to the complications already mentioned, this analysis of 172 records from a single medical examiner jurisdiction is not generalizable, and the low number of cases in several categories precluded statistical validation. The two sources of information, ME and FE, were both based on preliminary, partial, and potentially biased information. Social and traditional media are often suspected of being rife with bias and inaccuracies. While the FE website describes efforts by paid researchers to validate data, it is possible that inaccuracies remain. Similar concerns could be expressed about medical examiner reports that rely on police accounts. The terminology used in the two sources often differed and complicated comparisons. FE had many records with missing critical entries. Six complex tables were required to present the results of the study. The circumstantial categories of the ME records were designed to make data analysis manageable, but sacrificed granularity of detail. Finally, it is important to acknowledge that conditional probabilities merely express relationships within the data and should not be interpreted to represent cause and effect without deeper analysis. Conditional probabilities may sometimes seem counterintuitive. For example, in Table 6, the use of less‐than‐lethal force appeared to increase the probability of fatality in an altercation from P(ALT) = 64% to P(ALT|LTL) = 81%. Two explanations may explain this result: either taser deployment failed to subdue a combative or threatening decedent or, alternatively, taser deployment provoked combative or threatening behavior.

5. CONCLUSION

The present study revealed the depth of complexity in analyzing multiple circumstances involving the dynamics of police‐related deaths. Despite the limited scope of the present study, it may be important to consider the potential value of capturing complex circumstantial information in data. It is often taken as a truism that events are so complex that they can only be studied on a “case‐by‐case” basis. Granted, police‐related fatalities are complex events, but if thoroughly investigated, individual circumstantial elements can usually be identified. The unofficial FE database represents an attempt to capture multiple circumstantial elements into a single dataset, which is an important effort, especially in the absence of an official national database. The present study showed that it is possible to subject multiple circumstances to data analysis with the potential to yield insights that might be overlooked by examining events individually or out of a broader context. In analyzing 10 deaths in which the FE records indicated police shot unarmed individuals, comparison with ME records found that the majority had multiple circumstances leading to the fatal event. In four, the ME records indicated the decedent was armed with a firearm, and in two, the decedent was using a firearm. The amount of data analysis in this study was extensive and would have been more complex had it not been for simplifying circumstantial categories. Even with a mere 172 records, the information generated herein was difficult to describe, either in words or in multiple tables. A larger database assembled with well‐defined circumstantial categories and analyzed using modern data science tools would be invaluable, not only as a public information resource but also for examining and setting police and legal policies and standards through precedents and probabilities. For example, it is possible that the “reasonable officer” standard [26] could be defined objectively. Finally, as suggested previously [11], in the absence of a comprehensive national database, one established on thorough investigations and validated through all official records, databases limited to public records may come to serve as the only source of information describing police‐related deaths. This situation may not be optimal in preserving public trust in the nation's law enforcement institutions.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.