Skip to main content
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Jan 1.
Published in final edited form as: Ann Epidemiol. 2020 Aug 22;53:42–49.e3. doi: 10.1016/j.annepidem.2020.08.014

Black and unarmed: Statistical interaction between age, perceived mental illness, and geographic region among males fatally shot by police using case-only design

Marilyn D Thomas a, Nicholas P Jewell b, Amani M Allen b
PMCID: PMC7736192  NIHMSID: NIHMS1623665  PMID: 32835768

Abstract

Purpose:

We examine whether race and armed status interact to modify the risk of being fatally shot by police within categories of civilian age and mental illness status, and US region.

Methods:

Data are from The Washington Post online public-use database of all US police-involved shooting deaths. The sample includes Black and White males with known armed status who were killed 1/1/2015 through 12/31/2019 (N=3,090). A case-only design is employed to assess multiplicative interaction using adjusted logistic regression.

Results:

The fully-adjusted interaction estimate is null (SOR=0.75; 95% CI=0.55–1.04). However, adjusted estimates within strata show that the risk of being armed versus unarmed when fatally shot is smaller for Black compared to White males over age 54 (SOR=0.18; 95% CI=0.06–0.65), those showing mental illness signs (SOR=0.50; 95% CI=0.26–0.98), and those killed in the South (SOR=0.52; 95% CI=0.33–0.83), and that the risk is greater in the Midwest (SOR=2.42; 95% CI=1.11–5.26). Notably, there is no Black-White difference in armed status among younger age groups (SOR≈0.89).

Conclusion:

Race and armed status may interact leaving Black males at higher risk of being unarmed compared to White males when fatally shot by police among those over age 54, mentally impaired, and residing in the South. Causal interaction suggests a lower risk for unarmed Blacks in the Midwest. Researchers should further explore the utility of case-only design to study social-environmental interaction.

INTRODUCTION

Following numerous high-profile police shootings in the United States (US), the killing of unarmed Black men at the hands of law enforcement has become a primary focus of public health and social justice concern.117 Establishing whether true disparities exist by racial and gender group has been challenging using government databases due to the under-reporting of deaths from legal intervention1825 “injuries inflicted by the police or other law-enforcing agents, including military on duty, in the course of arresting or attempting to arrest lawbreakers, suppressing disturbances, maintaining order, and other legal action.” 21 As a result, The Washington Post (WaPo) and The Guardian newspapers began counting the number of people killed by police in 2015, and collecting other concomitant information.26,27 The Guardian’s database is substantially more reliable than government systems21,22 and shows 98% agreement with WaPo.28

Research using more reliable databases reveal that compared to Whites, Blacks are more likely to be stopped,2931 to be arrested,31,32 to experience police use-of-force,30,3335 to experience lethal force,4,12,15,16,3638 and to be fatally shot when unarmed.4,16,39,40 In addition, those killed are overwhelmingly male (96%) and about half are under age 35.37,39 This growing body of evidence suggests that the intersection between a civilian’s perceived race and armed status may modify the risk of being fatally shot by police (i.e., becoming a case), and that age and gender are also key, meaning there may be a statistical interaction in play.

Finding an appropriate catchment population, or controls, to test for statistical interaction between race and armed status is problematic. Population-level measures are the most robust41 but, unlike gun ownership,42 the racial population distribution of armed status—those patrolled by police and perceived to have a weapon—is unknown. Law enforcement agencies are not required to collect or report demographic nor armed status information associated with police–civilian encounters.43 Moreover, because police encounters and arrests are systemically biased against minorities,29,41,4446 data collected by agencies and used in research have underestimated or reported “reversed” racial disparities. 41 This type of bias results from conditioning on police encounter rates which are higher for Blacks (inflated denominator) and lower for Whites (deflated denominator) thereby making the groups appear more statistically similar.41 Knox, Lowe & Mummolo (2019) developed a novel method to account for such statistical bias in estimating racial disparities under a variety of plausible assumptions (e.g., use-of-force reporting requirements); however, estimates remain conditioned on police–civilian encounters and civilian armed status was not considered.45 Consequently, establishing an adequate reference or control group to estimate race and armed status interaction on police killings is fraught with challenges.

In epidemiology, the case-only study design is an efficient and innovative approach to evaluate potential statistical interaction between two risk factors. Commonly used to examine gene-environment interaction (G*E) among cancer patients,4749 case-only studies consist only of people that have the outcome (i.e., cases), as implied. The design is used to identify multiplicative interaction between environmental exposures and inherited traits that are independent in the population.50 Among cases, odds ratios that estimate the odds of G in those with E compared to the odds of G in those without E, and vice versa, reflect interaction on an outcome in the underlying population.47,51,52 Case-only design has been a useful method to study social-environmental interaction, particularly by personal characteristics such as race.5356 Causality can be inferred when the case-only interaction is mechanistic—the estimate is greater than 1 (i.e., positive) and G and E are independent in the population and never preventative.48

Both risk factors (Black and armed) are not likely preventative against being shot by police,4,3740,57 but the absence of information on the racial distribution of armed status among those patrolled by police make G*E independence an untestable assumption. However, given what is known about US racial distributions of weapons carrying, race and armed status appear to be independent. Studies reporting on weapons and firearms carrying show no consistent pattern by race among nationally representative samples of adults,5861 high school students,61,62 and adolescents and young adults.6365 Together, these findings provide some evidence that race and armed status are plausibly independent in the general population albeit more conclusive reporting is needed.

Assuming independence holds, newly reliable data on police killings can be analyzed using case-only series to investigate the modifying effect that armed status may have on the association between one’s race and being killed by police, thereby eliminating the need for controls. However, in this example, prevailing opinion is that one would expect negative-multiplicative interaction between race and armed status in police encounters (i.e., odds ratio<1). That is, the gap between fatalities for armed versus unarmed civilians is less for Blacks than for Whites. Johnson et al. (2019) applied case-only series to identify civilian and officer predictors of the race of a person fatally shot by police using WaPo and The Guardian data and reported no evidence of an anti-black disparity in armed status. Notably, the analysis was restricted to civilian deaths in 2015.56 A major contribution to the literature would be an examination across all police killings since 2015 using the most current WaPo data given that, unlike The Guardian, data collection has continued.

The main objective of this study is to examine statistical interaction between race and armed status on being shot and killed by police among Black and White males using case-only design. Race is conceptualized as a social construct representing the race-based experience of being perceived as Black or White in the US.6668 Being armed inherently increases the probability of being shot for anyone during a police encounter; however, Blacks maintain much higher risk whether armed or not.4,12,15,16,3640 We thereby hypothesize that the gap between being armed versus unarmed is smaller for Black compared to White males. Given that police use-of-force is higher for younger people,34,37,56 those perceived as mentally ill,6971 and in cities with higher Black populations,35,72,73 a secondary objective is to evaluate potential three-way interaction between race and armed status with age, mental illness status, and US region. Armed status disparities are theorized to reduce with age yet increase with signs of mental illness and in regions with larger Black population densities.

MATERIALS AND METHODS

Study Sample

Data are from The Washington Post (WaPo).26 WaPo maintains an online public-use database of all on-duty police-involved shooting deaths since January 1, 2015 which has been utilized in several recent publications.23,31,39,40,74 WaPo regularly collects, updates, and maintains information from every fatal shooting by police. The database includes civilian demographic information, armed status, perceived mental illness status, perceived threat-level, fleeing status, and incident location. WaPo validates the circumstances of the shootings using information from local reporting, police websites, social media, and open-source databases tracking police killings. Cases from January 1, 2015 through December 31, 2019 are analyzed (N=4,938). The study aim is to predict outcomes for unarmed Black males, hence the study sample is restricted to Black and White males with known armed status. The final sample size is comprised of 3,090 cases (Figure 1).

Figure 1.

Figure 1.

The Washington Post (Wapo) study sample selection process.

Outcome Assessment

Using a case-only design, G or E can be modeled as the outcome variable because either orientation will yield the same multiplicative interaction estimate (e.g. 2×3=3×2=6). Because race contributes so strongly to the risk of being killed by police, race is selected as the outcome. The two racial categories identified in WaPo (Non-Hispanic [NH] White, NH Black) are collapsed into a dichotomous race variable which excludes Hispanics (Black=0 [i.e., White], Black=1).

Exposure Assessment

Armed status.

WaPo recorded 86 armed categories ranging from “air conditioner” to “vehicle”. Given that officers have perceived civilian personal items as “life-threatening weapons” (e.g., cell phone),75,76 armed status is assessed dichotomously by grouping all armed categories together, separate from those reportedly “unarmed” (unarmed=0, armed=1). The unarmed category presumably includes those whom officers misperceived as being armed during the event but there is no way to know with certainty.

Covariates

Theoretical and empirical variables considered to confound the association between race and armed status include civilian socioeconomic status, demographic characteristics, and perceived mental health status during the encounter25,36,37,71,77,78 (Figure A.1). Civilian behaviors and situational factors during the encounter are identified as mediators79,80 and therefore are excluded to avoid over-controlling.

Civilian age is coded as a 5-level ordinal variable (<25=1, 25–34=2, 35–44=3, 45–54=4, >54=5) to evaluate potential non-linear or dose-response associations. A civilian showing signs of mental illness (SMI) during the encounter is coded dichotomously (No=0, Yes=1). Apart from race and age, WaPo did not collect civilian socioeconomic and other demographic information. Because US Census municipal-level data is not available for an extensive number of fatalities, county-level data is used to proxy individual-level demographic characteristics using QuickFacts—2014–2018 population estimates. 81 Proportions account for Black population, White population, population >64 years, persons foreign-born, owner-occupied housing units, households with a computer, households with broadband internet, persons with ≥Bachelor’s degree, disabled persons age <65, persons without health insurance (2019 estimate), civilian labor force, persons in poverty, median gross rent, and median household income (2018 dollars). Uniform Crime Reporting Program Data account for county-level violent and property crime rates (per 100,000 people).82 US region (hereafter region) is accounted for categorically (West=1, Midwest=2, South=3, Northeast=4).83

Statistical Analysis

STATA 16 IC is used for all statistical analyses.84 There are no missing data apart from civilian age (n=47), which is negligible (1.5%). Civilian demographic characteristics do not differ between missing and non-missing age groups suggesting that the data are missing at random, though differences are found across regions (Table A.1). Hence missing values are imputed using sample mean replacement (age 37)85 and sensitivity analyses are conducted to assess bias. Model specification includes theory and data-driven approaches.86,87 Using case-only design, logistic regression of adjusted odds ratios yield an interaction estimate between the odds of being armed (vs. unarmed) for Black compared to White males. Put another way, estimates give the odds ratio of being killed for Black armed males compared to White armed males. Since being fatally shot by police is a rare event, odds ratios provide pseudo-risk interaction estimates (SOR; hereafter risk).47,48 The specified reduced form equations that give rise to the interaction term are given below; where D represents the outcome.50

Equation A shows a standard population logistic regression model with the interaction term given by β3:

log(Pr[DG,E, Covariates]/1Pr[DG, E, Covariates])=β0+β1G+β2E+β3G*E+ Covariates 

Equation B shows a case-only logistic regression with the interaction term given by α1 which is equivalent to β3 in Equation A, assuming G and E population independence:50

log(Pr[GE, Covariates, D=1]/1Pr[GE, Covariates, D=1])=α0+α1E+ Covariates 

RESULTS

Sample Characteristics

Sample distributions of civilian and regional characteristics are presented in Table 1. Notably, annual Black-White gaps among those fatally shot by police are decreasing for those armed (Figure 2) and increasing for those unarmed (Figure 3).

Table 1.

WaPo sample characteristics of White and Black men shot and killed by police during 2015–19.

Total White Black

n (%)
Cases 3090 (100) 2016 (65) 1074 (35)
Armed status
Armed 2878 (93) 1924 (94) 992 (91)
Unarmed 212 (7) 114 (6) 99 (9)
Age
mean (sd) 37 (13) 40(13) 32 (11)
>25 512 (17) 221 (11) 291 (27)
25–34 976 (31 ) 581 (29) 395 (37)
35–44 677(22) 464 (23) 213 (20)
44–54 499(16) 403 (20) 96 (9)
>54 426 (14) 347 (17) 79 (7)
Mental Illness Signs
No 2321 (75) 1409 (70) 912 (85)
Yes 769 (25) 607 (30) 162 (15)
US Region
West 788 (25) 610 (30) 178(17)
Midwest 613 (20) 384 (19) 229 (21 )
South 1437 (47) 878 (44) 559 (52)
Northeast 252 (8) 144 (7) 108(10)
mean (sd)
Black Population (%) 15.3 (14.9) 10.7 (12.0) 24.0 (15.9)
Median Gross Rent ($) 992 (280) 954 (273) 1064 (278)
Disabled & age <65 (%) 10.0 (3.0) 10.0 (3.2) 8.9 (2.6)
Violent Crime Rate (100K) 388 (250) 334 (220) 491 (268)

Abbreviations: WaPo=The Washington Post; sd=standard deviation

Note: Those classified as armed include those whom were actually armed, therefore those whom police potentially misperceived as armed are included in the unarmed category.

Figure 2.

Figure 2.

WaPo racial distribution of armed males killed annually by police.

Figure 3.

Figure 3.

WaPo racial distribution of unarmed males killed annually by police.

Regression model specification

Age and SMI status are forced-in covariates. Among those contributing to 100% of the principal components (eigenvalues>1), bivariate analysis identify collinearity between several demographic variables (Table B.1). Hence, collinear covariates explaining the least of the variance—White population, ≥Bachelor’s degree, labor force, and median income—are omitted. All remaining covariates confound the main exposure–outcome association using multivariable regression adjusted for one covariate at a time (P<0.05). Nested modeling and Akaike and Bayesian information criteria determined poverty and broadband internet do not improve model fit, and they are subsequently excluded. Variance inflation factors show no remaining multicollinearity. Therefore, the final model is adjusted for age, SMI, region, median gross rent, violent crime rate, and percent Black population and disabled persons age <65. Sensitivity analyses confirm that our imputation method did not add bias or reduce precision (Table B.2).

Multivariable logistic regression

Table 2 presents nested logistic regression models (crude to fully-adjusted). Model 1 shows a lower predicted crude risk of being armed (vs. unarmed) for Black compared to White males shot and killed by police (SOR=0.60; 95% confidence interval [CI]=0.45–0.79). Model 2 shows this disparity is reduced after accounting for civilian age and SMI status (SOR=0.72; 95% CI=0.54–0.97). Null interactions are estimated in Models 3–5 after further adjustment for (a) region and percent Black population (Model 3), (b) percent Black population, median gross rent, and percent disabled persons age <65 (Model 4), and (c) violent crime rate (Model 5). Since all civilian and regional confounders are significant at P<0.01, potential 3-way interactions are assessed by stratifying adjusted estimates by civilian age, SMI status, and region.

Table 2.

Nested logistic regression of the interaction between being Black and armed among males shot by police during 2015–19 in WaPo (N=3,090)

Model 1 Model 2 Model 3 Model 4 Model 5
SOR LB 95% CI UB 95% CI SOR LB 95% CI UB 95% CI SOR LB 95% CI UB 95% CI SOR LB 95% CI UB 95% CI SOR LB 95% CI UB 95% CI

Black*Armed 0.60** 0.45 0.79 0.72* 0.54 0.97 0.74 0.54 1.02 0.76 0.55 1.05 0.75 0.55 1.04
Age 0.95 0.95 0.96 0.95 0.94 0.96 0.95 0.95 0.96 0.95 0.95 0.96
Mental Illness Signs 0.44 0.36 0.54 0.43 0.34 0.53 0.40 0.32 0.50 0.40 0.32 0.50
US Region 0.98 0.89 1.09 1.12 0.01 1.24 1.19 1.07 1.32
Black Population 688 349 1357 533 267 1065 164 78 348
Median Gross Rent 1.00 1.00 1.00 1.00 1.00 1.00
Disabled & age <65 4e-5 4e-7 4e-3 4e-5 4e-7 4e-3
Violent Crime Rate 1.15 1.11 1.20
Intercept 0.86a 5.03b 1.80c 1.62d 0.85e
BIC 3995 3694 3196 3120 3081

Abbreviations: WaPo=The Washington Post; SOR=odds ratio interaction estimate; LB=lower bound; UB=upper bound; BIC=Bayesian information criterion Interaction reference group=White and Armed

a

95% CI=0.66–1.13

b

95% CI=3.51–7.21

c

95% CI=1.16–2.78

d

95% CI=0.66–4.00

e

95% CI=0.33–2.17

**

P<0.001;

*

P<0.05

Model 1 is unadjusted; Model 2 adjusts for covariates in Model 1 and civilian age and showing mental illness signs; Model 3 adjusts for covariates in Model 2 and US region and percent Black population; Model 4 adjusts for covariates in Model 3 and median gross rent, percent disabled under age 65; and Model 5 adjusts for covariates in Model 4 and violent crime rate.

Figure 4 illustrates the fully-adjusted interaction estimate from Model 5 in Table 2 and interaction estimates adjusted for median gross rent, violent crime rate, and percent Black population and disabled persons age <65, and then stratified by (a) age, further adjusted for SMI and region, (b) SMI, further adjusted for age and region, and (c) region, further adjusted for age and SMI. Compared to White males, the risk of being armed is smaller for Black males over age 54 (SOR=0.18; 95% CI=0.06–0.65), those showing mental illness signs (SOR=0.50; 95% CI=0.26–0.98), and those killed in the South (SOR=0.52; 95% CI=0.33–0.83), and that the risk is greater in the Midwest (SOR=2.42; 95% CI=1.11–5.26) (Table B.3).

Figure 4.

Figure 4.

Fully-adjusted interaction estimate from Model 5 in Table 2 of the risk of being armed versus unarmed for Black compared to White males fatally shot by police 2015–19 in WaPo, and adjusted interaction estimates stratified by age, SMI status, and US region, where entries [CI] are main effects from case-only statistical Equation B which estimate the interaction term in statistical Equation A, assuming independence (N=3,090).

DISCUSSION

This is the first study to report statistical interaction between race and armed status among males fatally shot by police using a case-only study design. The fully-adjusted interaction estimate is null. All but one of the stratified-adjusted estimates predict negative-multiplicative interaction (SOR<1), therefore most study estimates do not meet the full criteria for mechanistic interaction rendering them non-interpretable for causal effects (i.e., SOR>1).47,48 Nevertheless, strong multiplicative interaction does emerge across all stratified models. These results provide evidence of possible true interaction—the combined effect of two exposures differ from the sum of the individual effects—versus effect modification— main effects differs across covariate strata.8890 Our findings suggest that simply adjusting for civilian and regional main effects may mask more nuanced three-way interactions that contribute to the Black-White disparities in police killings. Researchers should consider three-way interaction terms or stratifying police killing cases by specific civilian and regional characteristics when possible.

Interaction with age

Contrary to our hypothesis, we found no racial differences in the risk of being unarmed among younger males killed by police. The literature consistently points to higher fatality rates among males under age 35, along with considerable racial inequities among those unarmed.3,4,9,13,25,36 Thus, unarmed Black (vs. White) males are thereby expected to have an even greater risk when they are young. Aligned with this hypothesis and prior work, increasing age was associated with lower risk of being killed when we simply controlled for (vs. stratified by) age (Table 2). The Black-White disparity in armed status we found among older males is likely because our study is the first to evaluate the intersection of race, armed status, and age. Importantly, unarmed males above age 54 are over 5 times more likely to be Black than White (11% vs. 2%), therefore officer perceptions regarding civilian armed status could vary by civilian age and/or race. Such differential officer engagement can exacerbate inequities in civilian arrests and/or police-related injury, potentially leading to over-representation of older Black men in the criminal and urgent care systems. Future investigations should confirm the strength of the association we find between being Black and unarmed among older males, and assess whether racial bias may play a role.

Interaction with signs of mental illness

Consistent with our hypothesis and prior literature,70,71 the armed status gap is worse for Black males perceived as mentally ill. Officer use-of-force is generally greater against mentally impaired people with disparities for males, non-Whites, and those armed.6971,91 In the current study, White males are more likely to be perceived as mentally ill. This is unexpected given that Blacks are disproportionately diagnosed with psychotic disorders compared to Whites by 3 to 1.92 Selective ascertainment and reporting of mental illness by police may be biased towards Whites and against Blacks. For instance, officers show a higher likelihood to commit a White suspect to an involuntary psychiatric hold compared to a Black suspect taking the same action, which investigators suggest likely leads to a higher prevalence of incarceration and less mental illness treatment for Blacks.93 Further, Black (vs. White) males perceived as mentally ill in the current study are also less likely to be armed (88% vs. 95% respectively). Together, these findings suggest that anti-Black bias may be a stronger contributing factor in officer deadly use-of-force than the armed status of mentally impaired persons. However, potential underreporting of mental illness signs among Black males could bias our estimate away from the null. Further study among Black men with diagnosed mental illness would help disentangle the risk factors that may drive racial disparities in police killings within this highly vulnerable population.

Interaction with US Region

We are the first to report a Black-White armed status disparity by US region where Black males are at higher risk of being armed and fatally shot in the Midwest. Mechanistic interaction criteria is met thus causality can be inferred.48 As expected, Black (vs. White) males are at lower risk of being armed in the Southern region, which has high Black population density according to the US Census.94 Conversely, the White population is highly dense in the Midwest.94 This new evidence suggests that causal factors related to racial population density plays a role in the intersectional risk of being an unarmed Black male when killed by police.

Limitations and strengths

Study findings are generalizable: The sample includes nationally representative males fatally shot during 2015–19. Causal inference can only be inferred for one mechanistic interaction, though temporality is established for all estimates as race and armed status (exposures) precede a fatal shooting (outcome). Our use of case-only design to assess statistical interaction between race and armed status is novel but subject to an unverifiable assumption: the lack of verification on race and armed status independence amongst those patrolled by police is a key limitation in interpreting the reported interaction estimates. The hypothesized and highlighted negative interaction makes it impossible to weaken this assumption. However, if weapons carrying is generally higher among White (vs. Black) males patrolled by police, our results would underestimate racial gaps, and vice-versa if higher among Black males. The same is true for officer misperceptions of armed status: if misperceptions are more likely to be directed at White males (vs. Black males as expected), then our estimates will be biased towards the null, meaning our results are underestimated. Further, our use of county-level measures as proxies for individual-level confounding increase the likelihood of residual confounding. Study validity also largely depends on accurate data collection and proper validation of cases by WaPo. To assess data agreement, the number of 2015–16 WaPo cases (N=1,958) were cross-referenced with those in The Guardian database fatally shot by police during 2015–16 (N=2,027).27 Results show 97% agreement, providing evidence of high repeatability of our study findings, although it is unlikely that all known cases have been accurately reported to any system. Last, the likelihood of a committing a Type II error is higher for interactions with smaller cell sizes.

Conclusion

This study uses case-only design to provide preliminary evidence that race and armed status interact leaving Black males at higher risk of being unarmed compared to White males when fatally shot by police, particularly for those over age 54, perceived as mentally ill, and killed in the South. Three-way causal interaction likely exists between race, armed status, and US region, reducing the risk for unarmed Blacks in the Midwest. We introduce a novel approach to investigating potential 3-way social-environmental statistical interactions that exacerbate racial disparities in deadly police shootings. US law enforcement agency policies and training focused on addressing anti-Black bias in officer deadly use-of-force may buffer racial gaps in police killings. A national registry of law enforcement data would help us to better understand how age, perceived mental illness, and geographic region impacts the use of lethal force against unarmed Black men in the US.

ACKNOWLEDGEMENTS

MD Thomas was partially supported by NIGMS grant UL1GM118985, USA and by a Ford Foundation Predoctoral Fellowship administered by the National Academies of Sciences, Engineering, and Medicine, USA. AM Allen was also partially supported by NIMHD grant P60MD006902, USA. Study sponsors did not participate in study design, data collection, data analysis, interpretation of study results, or drafting of the manuscript. We wish to thank the HEARTs (Health Effects Associated with Racism Threat) Research Group at the [omitted] Center for Social Medicine for comments regarding the refinement of this manuscript, particularly Eli Michaels, Dr. Rachel Berkowitz, Elleni Hailu, Kevin Lee, Dr. Thu Thi Nguyen, and Saba Sohail.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Abbreviations:

CI

confidence intervals

US

Unites States

WaPo

The Washington Post

APPENDICES

Figure A.1.

Figure A.1.

Conceptual framework representing working knowledge of factors linking a civilian’s armed status and being shot and killed by police. Civilian behavior and situational factors were identified as mediators. Therefore, civilian socioeconomic status, demographics, and mental health status were assessed for confounding the association between being armed, shot and killed by police.

Table A.1.

Sensitivity analysis comparing confounder distributions between civilian’s missing and not missing age among White and Black men shot and killed by police during 2015–19 in WaPo (N=3,090)

Not Missing Age Missing Age

n (%)
Cases 3043 (100) 47 (100)
Armed status
Armed 2834 (93) 44 (94)
Unarmed 209 (7) 3 (6)
Mental Illness Signs
No 2286 (75) 35 (75)
Yes 757 (25) 12 (25)
US Region
West 777 (26) 11 (23)
Midwest 609 (20) 4 (9)
South 1405 (46) 32 (68)
Northeast 252 (8) 0 (0)
mean (sd)
Black Population (%) 15 (15) 20 (17)
Median Gross Rent ($) 991 (279) 1029 (330)
Disabled & age <65 (%) 10 (3) 10 (3)
Violent Crime Rate (100K) 388 (249) 429 (283)

Abbreviations: WaPo=The Washington Post; sd=standard deviation

Table B.1.

Pearson correlation results comparing county-level covariates.

2018 pop >age 65 years White Black Foreign-born Owner-occupied housing units Median Gross Rent Households with broadband BA Degree Age >65+ disability Health insurance Labor force Median household income Poverty Violent Poverty rate Property crime rate

2018 population 1.000
>age 65 years −0.258 1.000
White −0.157 0.412 1.000
Black 0.003 −0.293 −0.877 1.000
Foreign-born 0.655 −0.373 −0.307 0.073 1.000
Owner-occupied housing units −0.434 0.535 0.603 −0.454 −0.586 1.000
Median gross rent 0.412 −0.386 −0.379 0.126 0.735 −0.501 1.000
Households with broadband 0.230 −0.339 0.012 −0.147 0.406 −0.207 0.670 1.000
BA Degree 0.191 −0.410 −0.375 0.232 0.419 −0.506 0.720 0.659 1.000
Disabled & age <65 0.405 0.501 0.199 −0.045 −0.608 0.338 −0.680 −0.663 −0.651 1.000
Health insurance 0.064 0.031 0.015 0.047 0.093 0.053 −0.249 −0.305 −0.351 0.114 1.000
Labor force 0.226 −0.669 −0.218 0.123 0.382 −0.391 0.531 0.657 0.698 −0.734 −0.222 1.000
M edian household income 0.188 −0.376 −0.160 −0.052 0.458 −0.170 0.832 0.745 0.753 −0.706 −0.389 0.664 1.000
Poverty −0.053 0.061 −0.310 0.390 −0.169 −0.256 −0.441 −0.716 −0.397 0.567 0.241 0.537 −0.711 1.000
Violent crime rate 0.160 −0.243 −0.485 0.478 0.189 −0.463 0.051 −0.075 0.076 −0.042 0.112 0.111 −0.168 0.324 1.000
Property crime rate 0.024 −0.222 −0.352 0.352 0.053 −0.352 −0.037 0.006 0.039 0.025 0.174 0.116 −0.205 0.252 0.724 1.000

Bolded=P<0.05

Italicized column name=Variable contributed to 100% of the principal components.

Italicized row name=Selected for adjustment in final models post-nested model and best model fit analyses.

Table B.2.

Sensitivity analysis comparing logistic regression using imputed data versus complete case analysis of White and Black men shot and killed by police during 2015–19

Complete Case (N=3,043) Imputed (N=3,090)
SOR LB 95% CI UB 95% CI SOR LB 95% CI UB 95% CI SOR LB 95% CI UB 95% CI SOR LB 95% CI UB 95% CI

Black*Armed *0.613 0.462 0.813 0.772 0.558 1.069 *0.597 0.451 0.790 0.754 0.546 1.040
Age 0.953 0.946 0.961 0.953 0.946 0.961
Mental Illness Signs 0.391 0.312 0.490 0.399 0.319 0.499
US Region 1.193 1.071 1.328 1.190 1.069 1.324
Black Population 147.7 69.3 315.1 164.2 77.5 347.9
Median Gross Rent 1.001 1.000 1.001 1.001 1.000 1.001
Disabled & age <65 4.E-05 4.E-07 0.005 4.E-05 4.E-07 0.004
Violent Crime Rate 1.156 1.109 1.205 1.152 1.105 1.200
Intercept 0.833a 0.784b 0.860c 0.847d
BIC 3918.0 3029.6 3995.2 3081.0

Abbreviations: WaPo=The Washington Post; SOR=odds ratio interaction estimate; LB=lower bound; UB=upper bound; BIC=Bayesian information criterion

Interaction reference group=White and Armed

a

95% CI=0.635–1.094

b

95% CI=0.304–2.020

c

95% CI=0.656–1.126

d

95% CI=0.332–2.166

Bolded=P<0.01;

*

P<0.001

Note: All covariates in the adjusted models had p-values less than 0.01 but no asterisk is shown on estimates as they were not the main associations of interest.

Table B.3.

Logistic regression of the interaction between being Black and armed among males shot by police during 2015–19 in WaPo, stratified by civilian characteristics (N=3,090)

Black*Armed
SOR LB 95% CI UB 95% CI

Agea
>25 0.893 0.464 1.719
25–34 0.884 0.498 1.569
35–44 0.715 0.392 1.305
44–54 0.554 0.161 1.909
>54 *0.186 0.051 0.672
SMIb
No 0.812 0.562 1.173
US Regionc Yes 0.485 0.250 0.938
West 0.541 0.274 1.070
Midwest 2.418 1.111 5.263
South *0.525 0.332 0.831
Northeast 0.805 0.202 3.219

Abbreviations: WaPo=The Washington Post; SMI=signs of mental illness; SOR=odds ratio interaction estimate; LB=lower bound; UB=upper bound

Interaction reference group= White and Armed.

a

Adjusted for SMI, US region, median gross rent, violent crime rate, and percent Black population and persons disabled & age <65.

b

Adjusted for age, US region, median gross rent, violent crime rate, and percent Black population and persons disabled & age <65.

c

Adjusted for age, SMI, median gross rent, violent crime rate, and percent Black population and persons disabled & age <65.

Bolded=P<0.05;

*

P<0.01

Footnotes

MD Thomas, primary investigator, conceptualized the study, coordinated all logistics related to data collection, cleaning, and coding, designed and performed all statistical analyses (and takes responsibility for the integrity of the data analyzed), and took final responsibility for drafting the manuscript; NP Jewell assisted in data analysis and interpretation, and editing and approval of the manuscript; AM Allen assisted with the design of the study, data analysis and interpretation, and editing and approval of the manuscript.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

REFERENCES

  • 1.Alang S, McAlpine D, McCreedy E, Hardeman R. Police brutality and black health: setting the agenda for public health scholars. Am J Public Health. 2017;107(5):662–665. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Bassett MT. # BlackLivesMatter—a challenge to the medical and public health communities. New England Journal of Medicine. 2015;372(12):1085–1087. [DOI] [PubMed] [Google Scholar]
  • 3.Crosby AE, Lyons B. Assessing homicides by and of US law-enforcement officers. The New England journal of medicine. 2016;375(16):1509. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.DeGue S, Fowler KA, Calkins C. Deaths Due to Use of Lethal Force by Law Enforcement: Findings From the National Violent Death Reporting System, 17 US States, 2009–2012. American journal of preventive medicine. 2016;51(5):S173–S187. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Guardian T. The Counted: people killed by police in th US. 2017. [Google Scholar]
  • 6.Hardeman RR, Medina EM, Kozhimannil KB. Structural racism and supporting black lives—the role of health professionals. New England Journal of Medicine. 2016;375(22):2113–2115. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Jee-Lyn García J, Sharif MZ. Black lives matter: a commentary on racism and public health. Am J Public Health. 2015;105(8):e27–e30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Jennings JT, Rubado ME. Preventing the Use of Deadly Force: The Relationship between Police Agency Policies and Rates of Officer-Involved Gun Deaths. Public Adm Rev. 2017;77(2):217–225. [Google Scholar]
  • 9.Chaney C, Robertson RV. Armed and dangerous? An examination of fatal shootings of unarmed black people by police. Journal of Pan African Studies. 2015;8(4):45–78. [Google Scholar]
  • 10.Dunham RG, Petersen N. Making Black Lives Matter: Evidence‐Based Policies for Reducing Police Bias in the Use of Deadly Force. Criminology & Public Policy. 2017;16(1):341–348. [Google Scholar]
  • 11.Hall AV, Hall EV, Perry JL. Black and blue: Exploring racial bias and law enforcement in the killings of unarmed black male civilians. Am Psychol. 2016;71(3):175. [DOI] [PubMed] [Google Scholar]
  • 12.Hehman E, Flake JK, Calanchini J. Disproportionate use of lethal force in policing is associated with regional racial biases of residents. Social psychological and personality science. 2018;9(4):393–401. [Google Scholar]
  • 13.Martinot S. On the epidemic of police killings. Social Justice. 2014:52–75. [Google Scholar]
  • 14.Mesic A, Franklin L, Cansever A, et al. The relationship between structural racism and black-white disparities in fatal police shootings at the state level. J Natl Med Assoc. 2018;110(2):106–116. [DOI] [PubMed] [Google Scholar]
  • 15.Price JH, Payton E. Implicit Racial Bias and Police Use of Lethal Force: Justifiable Homicide or Potential Discrimination? Journal of African American Studies. 2017:1–10. [Google Scholar]
  • 16.Ross CT. A multi-level Bayesian analysis of racial bias in police shootings at the county-level in the United States, 2011–2014. PloS one. 2015;10(11):e0141854. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Smiley C, Fakunle D. From “brute” to “thug:” The demonization and criminalization of unarmed Black male victims in America. Journal of human behavior in the social environment. 2016;26(3–4):350–366. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Barber C, Azrael D, Cohen A, et al. Homicides by police: comparing counts from The National Violent Death Reporting System, vital statistics, and supplementary homicide reports. Journal Information. 2016;106(5). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Barry R, Jones C. Incomplete records: Hundreds of police killings uncounted in federal stats. The Wall Street Journal. 2014:A1. [Google Scholar]
  • 20.Fachner G, White MD, Coldren JR Jr, Stewart JK. Need for a National Center for Police Shootings and Deadly Force Research, Training, and Technical Assistance. 2014. [Google Scholar]
  • 21.Feldman JM, Gruskin S, Coull BA, Krieger N. Quantifying underreporting of law-enforcement-related deaths in United States vital statistics and news-media-based data sources: A capture– recapture analysis. PLoS medicine. 2017;14(10):e1002399. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Krieger N, Chen JT, Waterman PD, Kiang MV, Feldman J. Police killings and police deaths are public health data and can be counted. PLoS medicine. 2015;12(12):e1001915. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Leeper GE. CONDITIONAL SPENDING AND THE NEED FOR DATA ON LETHAL USE OF POLICE FORCE. N Y Univ Law Rev. 2017;92(6):2053–2093. [Google Scholar]
  • 24.Loftin C, Wiersema B, McDowall D, Dobrin A. Underreporting of justifiable homicides committed by police officers in the United States, 1976–1998. Am J Public Health. 2003;93(7):1117–1121. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Shane JM, Lawton B, Swenson Z. The prevalence of fatal police shootings by US police, 2015–2016: Patterns and answers from a new data set. Journal of criminal justice. 2017;52:101–111. [Google Scholar]
  • 26.Fatal Force. The Washington Post. 2020. https://www.washingtonpost.com/graphics/investigations/police-shootings-database/ Accessed on February 9, 2020.
  • 27.The Counted: People killed by police in the US. The Guardian 2016. https://www.theguardian.com/us-news/ng-interactive/2015/jun/01/the-counted-police-killings-us-database Accessed on February 9, 2020.
  • 28.Williams HE, Bowman SW, Jung JT. The limitations of government databases for analyzing fatal officer-involved shootings in the United States. Criminal Justice Policy Review. 2019;30(2):201–222. [Google Scholar]
  • 29.Sewell AA, Jefferson KA. Collateral damage: the health effects of invasive police encounters in New York City. Journal of Urban Health. 2016;93(1):42–67. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Hyland S, Langton L, Davis E. Police Use of Nonfatal Force, 2002–11. Washington DC: US Department of Justice, Office of Justice Programs, Bureau of Justice Statistics; 2015. [Google Scholar]
  • 31.Miller TR, Lawrence BA, Carlson NN, et al. Perils of police action: a cautionary tale from US data sets. Inj Prev. 2017;23(1):27–32. [DOI] [PubMed] [Google Scholar]
  • 32.Barnes J, Jorgensen C, Beaver KM, Boutwell BB, Wright JP. Arrest prevalence in a national sample of adults: The role of sex and race/ethnicity. American Journal of Criminal Justice. 2015;40(3):457–465. [Google Scholar]
  • 33.Feldman JM, Chen JT, Waterman PD, Krieger N. Temporal trends and racial/ethnic inequalities for legal intervention injuries treated in emergency departments: US men and women age 15–34, 2001–2014. Journal of Urban Health. 2016;93(5):797–807. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Feldman JM, Chen JT, Waterman PD, Krieger N. Temporal Trends and Racial/Ethnic Inequalities for Legal Intervention Injuries Treated in Emergency Departments: US Men and Women Age 15–34, 2001–2014. J Urban Health. 2016;93(5):797–807. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Smith BW, Holmes MD. Police use of excessive force in minority communities: A test of the minority threat, place, and community accountability hypotheses. Social Problems. 2014;61(1):83–104. [Google Scholar]
  • 36.Buehler JW. Racial/ethnic disparities in the use of lethal force by US Police, 2010–2014. Am J Public Health. 2017;107(2). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Krieger N, Kiang MV, Chen JT, Waterman PD. Trends in US deaths due to legal intervention among black and white men, age 15–34 years, by county income level: 1960–2010. Harvard Public Health Review. 2015;3:1–5. [Google Scholar]
  • 38.Mesic A, Franklin L, Cansever A, et al. The Relationship Between Structural Racism and Black-White Disparities in Fatal Police Shootings at the State Level. J Natl Med Assoc. 2018. [DOI] [PubMed] [Google Scholar]
  • 39.Nix J, Campbell BA, Byers EH, Alpert GP. A Bird’s Eye View of Civilians Killed by Police in 2015. Criminology & Public Policy. 2017. [Google Scholar]
  • 40.Lott JR, Moody CE. Do white police officers unfairly target black suspects? Available at SSRN 2870189. 2016. [Google Scholar]
  • 41.Ross CT, Winterhalder B, McElreath R. Resolution of apparent paradoxes in the race-specific frequency of use-of-force by police. Palgrave Communications. 2018;4(1):61. [Google Scholar]
  • 42.Azrael D, Hepburn L, Hemenway D, Miller M. The stock and flow of US firearms: results from the 2015 National Firearms Survey. RSF: The Russell Sage Foundation Journal of the Social Sciences. 2017;3(5):38–57. [Google Scholar]
  • 43.US Bureau of Justice Statistics. All Data Collections. 2019. [Google Scholar]
  • 44.Najdowski CJ, Bottoms BL, Goff PA. Stereotype threat and racial differences in citizens’ experiences of police encounters. Law and human behavior. 2015;39(5):463. [DOI] [PubMed] [Google Scholar]
  • 45.Knox D, Lowe W, Mummolo J. The Bias Is Built In: How Administrative Records Mask Racially Biased Policing. Available at SSRN. 2019. [Google Scholar]
  • 46.Knox D, Mummolo J. Making inferences about racial disparities in police violence. Available at SSRN 3431132. 2019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Gatto NM, Campbell UB, Rundle AG, Ahsan H. Further development of the case-only design for assessing gene–environment interaction: evaluation of and adjustment for bias. International journal of epidemiology. 2004;33(5):1014–1024. [DOI] [PubMed] [Google Scholar]
  • 48.VanderWeele TJ, Hernández‐Díaz S, Hernán MA. Case‐only gene‐environment interaction studies: when does association imply mechanistic interaction? Genetic epidemiology. 2010;34(4):327–334. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Thomas D. Gene–environment-wide association studies: emerging approaches. Nature Reviews Genetics. 2010;11(4):259. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.VanderWeele TJ, Knol MJ. A tutorial on interaction. Epidemiologic Methods. 2014;3(1):33–72. [Google Scholar]
  • 51.Prentice RL, Pyke R. Logistic disease incidence models and case-control studies. Biometrika. 1979;66(3):403–411. [Google Scholar]
  • 52.Schmidt S, Schaid DJ. Potential misinterpretation of the case-only study to assess gene-environment interaction. American Journal of Epidemiology. 1999;150(8):878–885. [DOI] [PubMed] [Google Scholar]
  • 53.Frangakis CE, Petridou E. Modelling risk factors for injuries from dog bites in Greece: a case-only design and analysis. Accident Analysis & Prevention. 2003;35(3):435–438. [DOI] [PubMed] [Google Scholar]
  • 54.Qiu H, Tian L, Kin-fai Ho VC, Pun XW, Ignatius TS. Air pollution and mortality: effect modification by personal characteristics and specific cause of death in a case-only study. Environmental pollution. 2015;199:192–197. [DOI] [PubMed] [Google Scholar]
  • 55.Zanobetti A, O’neill MS, Gronlund CJ, Schwartz JD. Susceptibility to mortality in weather extremes: effect modification by personal and small area characteristics in a multi-city case-only analysis. Epidemiology. 2013;24(6):809. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Johnson DJ, Tress T, Burkel N, Taylor C, Cesario J. Officer characteristics and racial disparities in fatal officer-involved shootings. Proceedings of the National Academy of Sciences. 2019;116(32):15877–15882. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
  • 57.Sadler MS, Correll J, Park B, Judd CM. The world is not black and white: Racial bias in the decision to shoot in a multiethnic context. J Soc Issues. 2012;68(2):286–313. [Google Scholar]
  • 58.Felson RB, Pare P-P. Gun cultures or honor cultures? Explaining regional and race differences in weapon carrying. Social forces. 2010;88(3):1357–1378. [Google Scholar]
  • 59.Kleck G, Gertz M. Carrying guns for protection: results from the National Self-Defense Survey. Journal of Research in Crime and Delinquency. 1998;35(2):193–224. [Google Scholar]
  • 60.Rowhani-Rahbar A, Azrael D, Lyons VH, Simonetti JA, Miller M. Loaded handgun carrying among US adults, 2015. Am J Public Health. 2017;107(12):1930–1936. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61.Wallace LN. Armed kids, armed adults? Weapon carrying from adolescence to adulthood. Youth violence and juvenile justice. 2017;15(1):84–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62.Control CsfD Prevention. Weapon-carrying among high school students: United States, 1990. Morbidity and Mortality Weekly Report. 1991;40(40):681–684. [PubMed] [Google Scholar]
  • 63.Mulye TP, Park MJ, Nelson CD, Adams SH, Irwin CE Jr, Brindis CD. Trends in adolescent and young adult health in the United States. Journal of Adolescent Health. 2009;45(1):8–24. [DOI] [PubMed] [Google Scholar]
  • 64.Vaughn MG, Nelson EJ, Salas-Wright CP, DeLisi M, Qian Z. Handgun carrying among White youth increasing in the United States: New evidence from the National Survey on Drug Use and Health 2002–2013. Preventive medicine. 2016;88:127–133. [DOI] [PubMed] [Google Scholar]
  • 65.Vaughn MG, Perron BE, Abdon A, Olate R, Groom R, Wu L-T. Correlates of handgun carrying among adolescents in the United States. Journal of interpersonal violence. 2012;27(10):2003–2021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66.Ford CL, Airhihenbuwa CO. The public health critical race methodology: praxis for antiracism research. Soc Sci Med. 2010;71(8):1390–1398. [DOI] [PubMed] [Google Scholar]
  • 67.Ford CL, Harawa NT. A new conceptualization of ethnicity for social epidemiologic and health equity research. Soc Sci Med. 2010;71(2):251–258. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68.Budget OoM. Race and ethnic standards for federal statistics and administrative reporting (Statistical Directive No. 15). Federal Register. 1978;40:19269–19279. [PubMed] [Google Scholar]
  • 69.Johnson RR. Suspect mental disorder and police use of force. Criminal Justice and Behavior. 2011;38(2):127–145. [Google Scholar]
  • 70.Kaminski RJ, Digiovanni C, Downs R. The use of force between the police and persons with impaired judgment. Police Quarterly. 2004;7(3):311–338. [Google Scholar]
  • 71.Rossler MT, Terrill W. Mental illness, police use of force, and citizen injury. Police quarterly. 2017;20(2):189–212. [Google Scholar]
  • 72.Holmes MD, Painter MA, Smith BW. Race, place, and police-caused homicide in US municipalities. Justice Quarterly. 2019;36(5):751–786. [Google Scholar]
  • 73.Smith BW. Structural and organizational predictors of homicide by police. Policing: An International Journal of Police Strategies & Management. 2004;27(4):539–557. [Google Scholar]
  • 74.Nicholson‐Crotty S, Nicholson‐Crotty J, Fernandez S. Will more black cops matter? Officer race and police‐involved homicides of black citizens. Public Adm Rev. 2017;77(2):206–216. [Google Scholar]
  • 75.Fitzsimmons EG. 12-Year-Old Boy Dies After Police in Cleveland Shoot Him. The New York Times2014. [Google Scholar]
  • 76.Real JAD. No Charges in Sacramento Police Shooting of Stephon Clark. The New York Times. March 2, 2019, 2019;A: 15. [Google Scholar]
  • 77.Smith MR, Kaminski RJ, Alpert GP, Fridell LA, MacDonald J, Kubu B. A multi-method evaluation of police use of force outcomes: Final report to the National Institute of Justice. SURVEY METHODOLOGY. 2009;3:1. [Google Scholar]
  • 78.Fridell LA. Explaining the disparity in results across studies assessing racial disparity in police use of force: a research note. American journal of criminal justice. 2017;42(3):502–513. [Google Scholar]
  • 79.Friedrich RJ. POLICE USE OF FORCE - INDIVIDUALS, SITUATIONS, AND ORGANIZATIONS. Annals of the American Academy of Political and Social Science. 1980;452(NOV):82–97. [Google Scholar]
  • 80.Terrill W, Mastrofski SD. Situational and officer-based determinants of police coercion. Justice quarterly. 2002;19(2):215–248. [Google Scholar]
  • 81.QuickFacts: Statistics for all states and counties, and for cities and towns with a population of 5,000 or more. 2020. https://www.census.gov/quickfacts/fact/table/US/PST045219 Accessed on February 27, 2020. [Google Scholar]
  • 82.Federal Bureau of Investigation Uniform Crime Reporting Program Data: County-Level Detailed Arrest and Offense Data, United States, 2014. . 2014. [Google Scholar]
  • 83.US Census Bureau. Regions. 2019; https://www.census.gov/regions, Accessed May 30, 2019.
  • 84.Soper D A-priori Sample Size Calculator for Multiple Regression. Free Statistics Calculators 2018; https://www.danielsoper.com/statcalc/calculator.aspx?id=1. Accessed August 1, 2018 Accessed November 21, 2018.
  • 85.Hawthorne G, Hawthorne G, Elliott P. Imputing cross-sectional missing data: comparison of common techniques. Australian & New Zealand Journal of Psychiatry. 2005;39(7):583–590. [DOI] [PubMed] [Google Scholar]
  • 86.Fox J Model selection, averaging, and validation In: Applied regression analysis and generalized linear models. Sage Publications; 2015:669–698. [Google Scholar]
  • 87.Collinearity Fox J. and its purported remedies In: Applied regression analysis and generalized linear models. Sage Publications; 2015:341–368. [Google Scholar]
  • 88.VanderWeele TJ. On the distinction between interaction and effect modification. Epidemiology. 2009;20(6):863–871. [DOI] [PubMed] [Google Scholar]
  • 89.Jewell NP. Causal inference and extraneous factors: confounding and interaction In: Statistics for Epidemiology. Chapman and Hall/CRC; 2003:109–138. [Google Scholar]
  • 90.Aschengrau A, Seage G. Effect Measure Modification In: Epidemiology in Public Health. Burlington, MA: Jones & Barlett Learning; 2014:349–362. [Google Scholar]
  • 91.Morabito MS, Kerr AN, Watson A, Draine J, Ottati V, Angell B. Crisis intervention teams and people with mental illness: Exploring the factors that influence the use of force. Crime & delinquency. 2012;58(1):57–77. [Google Scholar]
  • 92.Schwartz RC, Blankenship DM. Racial disparities in psychotic disorder diagnosis: A review of empirical literature. World journal of Psychiatry. 2014;4(4):133. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 93.Cooper VG, Mclearen AM, Zapf PA. Dispositional decisions with the mentally ill: Police perceptions and characteristics. Police Quarterly. 2004;7(3):295–310. [Google Scholar]
  • 94.Census Data Mapper: 2010 US Census. 2010. https://datamapper.geo.census.gov/map.html Accessed on March 16, 2020. [Google Scholar]

RESOURCES