NowComment
2-Pane Combined
Comments:
Full Summaries Sorted

A Spatial Analysis of Officer-Involved Shootings in Los Angeles

Author: Debbie Ma; Steven Graves and Jonathan Alvarado

Ma, Debbie, et al. “A Spatial Analysis of Officer-Involved Shootings in Los Angeles.” EBSCO, web.a.ebscohost.com/ehost/detail/detail?vid=2&sid=691cbe7e-648f-4e3c-86d3-710a40c4c7f1@sessionmgr4007&bdata=JnNpdGU9ZWhvc3QtbGl2ZQ==#AN=139644119&db=aph.

Officer-involved shootings have spawned large-scale protest movements, intense media scrutiny, and a measure of political backlash from social conservatives. Over the years, Los Angeles County has suffered several rounds of extreme rioting following controversial episodes of police violence, yet continues to witness hundreds of officer-involved shootings each year. We analyze the spatial distribution of officer-involved shootings in Los Angeles County using ordinary least squares regression and hot-spot mapping techniques. We find the highly uneven distribution of officer-involved shootings across the many enforcement divisions of the LA Police and Sheriff Departments is statistically associated with neighborhood levels of violent crime and the average age of local citizenry. Other common predictors of officer-involved shootings, including race, ethnicity, and neighborhood economic conditions, are discussed, along with implications for policy and practice.

Keywords: Ethnicity; Officer-Involved Shootings; Los Angeles; Police; Regression

In the summer of 2014, protests erupted in Ferguson, Missouri, a small town on the suburban fringe of St. Louis, following the shooting death of Michael Brown. Brown, an eighteen-year-old black male, was unarmed when Darren Wilson, a twenty-eight-year-old white police officer, shot and killed him.[ 1] Demonstrators rioted against the actions of the local police, whom they claimed were racist, and whom they felt routinely violated the civil rights of African-Americans living in Ferguson. The riots in Ferguson were symptomatic of widespread frustration in the United States (especially among urban people of color) over relationships with law enforcement.

Perhaps surprisingly, Los Angeles, a region with a prominent history of race-based, controversial police action, was largely spared from the protests that marked some other urban regions in 2014–15 (Queally 2017), and certainly saw nothing on the scale of the 1992 riots that followed the acquittal of officers involved in the 1991 beating of black motorist Rodney King. Still, Los Angeles, by most metrics, continues to be the epicenter of officer-involved shootings (OIS) in the United States, easily outpacing the number and rates of OIS in other large metropolitan areas (see, e.g. [20]; [ 2]). Although none of the shootings in Los Angeles have gained publicity anywhere near the level garnered by the shooting of Michael Brown, the sheer volume of OIS that occur in Los Angeles cannot be ignored. Understanding the nature of OIS in Los Angeles motivates the current research. We provide an empirical test of how neighborhood characteristics, specifically the racial and ethnic makeup of a neighborhood, age, and income, affect the likelihood of a person being shot by an officer of the law. Fundamentally, we posit that most human behaviors, including life-and-death decisions, are made in a spatial context and thus are inextricably linked to situational variables. As such, we argue that it is possible that the factors that predict OIS in Los Angeles may vary from those identified in the extant research. We begin by reviewing the policies governing the use of a firearm by police, before turning our attention to a survey of the research surrounding OIS as they relate to geographical factors.

Use-of-Force Policies

Policies dictating the use of force by officers vary greatly across the nation, and each individual police department or municipality is responsible for establishing specific protocols. As [31] stated, the use of force by police officers is guided by two entities—the United States Supreme Court, and departmental use of force policies. One similarity in all use-of-force policies is that they indicate that the force an officer utilizes against citizens must be proportional to the resistance displayed by the citizen ([ 3]). It is expected and known that the law-enforcement profession requires the potential for using "reasonable" and "necessary" force in chaotic situations, including the use of deadly force ([ 1]). However, this is a highly subjective process. Faced with similar situations, one individual may perceive a situation as life threatening and react by utilizing deadly force in response. Another individual faced with the same situation may not feel his or her life is in danger and may react with moderation. The spatial context is very likely a factor in the calculation of risk associated with any situation. While departmental policies differ by each individual agency, it has been found that careful implementation of policies dictating when to use varying levels of force are beneficial. Studies analyzing the policies of police agencies have shown that more-restrictive policies on lethal force decreased the number of citizens shot by police without compromising officer safety, crime levels, or arrest behavior ([32]).

The two law enforcement agencies involved in this study are the Los Angeles County Sheriff's Department (LASD) and the Los Angeles Police Department (LAPD). According to the LASD Training and Policy Manual (2012), the use of deadly force policy includes: ( 1) self-defense or defense of others; ( 2) preventing the escape of a fleeing felon when there is probable cause that the suspect represents a significant threat of death or serious physical injury; and ( 3) warning shots and cover fire are to be used only when deadly force is justified and the member can justify the need for utilizing either tactic. Similarly, LAPD's (2015) use-of-force policies state that officers can utilize deadly force to: ( 1) protect themselves or others from what is reasonably believed to be imminent threat of death or serious bodily injury; ( 2) prevent a crime where the suspect's actions place person(s) in imminent threat of death or serious bodily injury; and ( 3) prevent the escape of a fleeing felon when there is probable cause that the suspect represents a significant threat of death or serious physical injury.

Although policies surrounding OIS may vary from department to department, none of them allow for explicit consideration of neighborhood features where the shooting occurs. That said, police officers are aware that certain neighborhoods are more dangerous than others and perhaps enter those neighborhoods with a heightened sense of vigilance ([25]; [30]). And indeed, previous research suggests that neighborhood crime rates and racial composition relate to officers' behavior ([26]). In the next section, we review some of the previous research that has established a link between geographic factors and OIS.

Factors Associated with Officer-Involved Shootings

Researchers studying OIS routinely focus on a common set of explanatory variables when they try to understand and explain OIS. Generally, these variables include some measure of crime rate, racial diversity, and income (e.g. [ 8]; [26]; [29]). The precise operationalization of these variables differs from study to study. For example, some researchers might include a measure of violent crime rates, whereas others might include homicide rates (see [ 4]); some research operationalizes racial diversity as the proportion of non-white residents in an area, whereas others might use the proportion of black residents (e.g. [ 7]; [21]). Finally, some researchers measure income with median household income, while others use proportion of people living below the poverty level (see [29]).

In addition to varying in terms of the predictors that researchers have used, the level at which these analyses have been conducted also varies. [ 9] examined police deadly force at the state level and found that after controlling other variables, economic inequality best predicted police killings. A handful of studies have analyzed OIS at the city level as well. [28] found strong correlations between "Police Homicides" and homicides from all cases. [14] find that percent non-white and levels of segregation positively affects the rate of police homicides. [29] found a strong relationship between one measure of police homicides and economic inequality, and a weaker relationship with percent black, while noting the persistent mediating role of violent crime. [ 8] study of 170 U.S. cities found that racial inequality and the local murder rate accounted for the rates of deadly force used by police. Still others have reported the rate of OIS of black versus white individuals can be predicted by the proportion of blacks in the area ([26]). As [ 7] and [ 9] point out, city-level data can be plagued by extreme variation and poor data quality. [11] detailed some of the additional problems created by aggregating data at the city and state level, focusing on the manner in which inter-city and state-scale analyses can mask very large variances within cities or states.

One recent analysis conducted by [26] examined 213 metropolitan areas over a twenty-one-year period. In their analyses, they regressed the difference in rates of OIS of black versus white individuals on mean-centered proportion black, mean-centered proportion living below the poverty level, mean-centered population, and, critically, the differential arrest rates between blacks and whites. This allowed the researchers to test the intercept of their regression model in a hypothetical metropolitan area that was statistically average in all ways, except that there was no racial difference in crime. Their analysis revealed that police were significantly more likely to shoot black individuals than whites, when statistically removing any differences in criminal activity.

Neighborhood analyses of OIS are few. Analysis at the neighborhood level within large urban regions offers some advantages. Relevant to our current research is what happens within single cities or single police departments. In one of the few analyses that do exist, [11] studied OIS patterns in St. Louis using a block-group-level analysis. Although they found that most OIS involved white police officers and armed African-Americans, neither the racial composition nor the socio-economic status of the neighborhoodshad a significant effect on the rate of OIS. Only local rates of firearm violence were a statistically significant predictor of OIS rates. Their study also uncovered a curvilinear association between local firearm violence and OIS. Neighborhoods with exceptionally high rates of firearm violence actually had a lower OIS rate than those with moderately high levels of firearm violence.

One reason that department or individual city-level analyses might be scarce is that OIS are relatively rare events, so they are difficult to research empirically. For that reason, researchers have sometimes instead examined use of non-deadly force. In one such study, [33] sent researchers into the field for ride-alongs with police officers in Indianapolis, Indiana, and St. Petersburg, Florida. These researchers carefully recorded and coded 3,300 police-civilian interactions and found that police employed greater levels of non-deadly force in neighborhoods with higher rates of minority residents. However, this apparent bias disappeared when the researchers accounted for how disadvantaged a neighborhood was or the rate of homicides in the neighborhood. Imaginably, police officers in this study may have adjusted their behavior because they knew that their behaviors were being observed. In this way, this particular study is vulnerable to a demand interpretation.

Officer-Involved Shootings: Data Considerations

In understanding the factors that relate to OIS, two critical, data-related problems present themselves. First, despite widespread public and academic interest in OIS, there exists a shocking lack of reliable, comprehensive data cataloging OIS, some thirty years after [ 9] lamented the lack of such data. The Centers for Disease Control's National Vital Statistics data and the FBI's Supplemental Homicide report are two of the most often-cited sources for OIS, but recent investigations point to dramatic undercounts (see, e.g. [11]; see also [ 6]). [22] argued that the actual number of OIS is likely more than double the approximately four hundred reported officially. Many law-enforcement agencies either file data selectively or fail to report it at all ([24]). As a result, various grassroots efforts have been launched that offer alternative databases of OIS, curated largely through Internet research and crowd-sourced reporting. Two of the better ones, Fatal Encounters (Fatalencounters.org) and Killed by Police (Killedbypolice.net), both put the number of OIS nationally at over 1,200 in 2015 (see [22]).

The second major limitation of the extant datasets involves what is recorded as an OIS. Specifically, the databases mentioned above include only cases in which an individual was shot and killed by police officers. The federal databases include only instances in which an individual has died. Likewise, as their names all suggest, Fatal Encounters, Killed by Police, and Fatal Force (a database assembled by the Washington Post) consist only of persons who died after officers used a firearm. Critically, there is no accounting for instances in which an officer opens fire but does not kill the intended target. To fully and accurately capture all OIS, one must necessarily have a record of instances in which individuals died and those in which individuals survived. This limitation may be hugely consequential for our understanding of OIS.

The scope of this problem is currently unknown. What is understood is that officers miss their targets a nontrivial amount of the time. In one study conducted by [34], officers' shooting was investigated under simulated pressure. Their study found that officers missed their intended targets between 53.8 percent and 74.6 percent of the time. Given the high miss rate of officer firing, one might intuit that a high proportion of targets may never be struck or suffer non-fatal injuries.

Mapping Los Angeles Crime Data

Crime data were aggregated from two sources: the Los Angeles Police Department (LAPD; City of Los Angeles 2019) and the Los Angeles Sheriff's Department (LASD; [17]) from 2010 to 2015. Both agencies provide comprehensive databases of criminal activity in their respective jurisdictions at the end of each year. Each reports essentially the same data, using nearly identical terminology, and both report the location of each reported crime where possible. The only significant difference in the datasets is the manner in which each agency provides location data. While both agencies provide partly anonymized addresses for crime (e.g. 500 block of Main Street), the LASD, perhaps accidentally, provided presumably accurate address coordinates (in State Plane V format) for all reported crimes. LASD has since removed this type of coordinate data from its public databases. The LAPD provided latitude and longitude coordinates for each reported crime, but LASD anonymized coordinates as well. As a result, maps of the crime data provided by LASD painted a highly accurate picture of the actual location of crimes, while crime in LAPD's jurisdiction appears to occur only at or very near intersections (e.g. at exactly 500 Main Street). To resolve this troublesome characteristic, we modified the LAPD data by adding a random number between two and ninety-eight to each block-level address, in order to provide a more realistic distribution of crimes along each one-block length of street. This technique moved the reported location of most crimes away from a corner address on each block to some random point along the block, without moving the mapped location of the crime off of the block on which the crime was originally reported. LAPD crimes reported as occurring at an intersection (e.g. Main and Maple Street) were mapped at the intersection. ArcMap's geocoding service was used for around ninety-five percent of the addresses, and Google's API was used for the remaining five percent that were unable to be geocoded by ESRI's address finder. Around one percent of crimes were without proper address information; those were removed from the database. The initial map of violent crimes reported by LAPD from 2010 to 2015 included 123,942 homicides, rapes, attempted rapes, robberies, attempted robberies, various felonious assaults, and weapons crimes. The corresponding map for LASD included 71,970 crimes from the corresponding categories. We merged both maps into a single point map displaying nearly 200,000 total crimes. Jurisdictions within Los Angeles but not LAPD or LASD, such as Pasadena Police Department, were not included in this study.

Mapping Officer-Involved Shooting Data

KPCC's news bureau ran a series of stories about OIS in Los Angeles, and in the course of their investigations created a website with links to digital copies of each investigative review of LA County OIS, authored by LA County's District Attorney's Office ([12]). These reports included every instance between 2010 and 2014 in which an officer discharged his/ her weapon and struck an individual. We viewed all 375 reports, extracted data for a number of variables, and entered them into a spreadsheet. Among the key variables extracted from the reports were locations indicating where officers initiated contact with the individual(s) shot by an officer and the locations where officers discharged their weapons. Both fatal and non-fatal shootings were included in the analysis. Each location was geocoded and mapped.

Nearly all of the investigative reviews provided detailed information recounting the series of events that transpired prior to the officer(s) discharging a weapon. In the few instances where the District Attorney's narratives failed to provide adequate location information, we cross-referenced the DA report against media reports, particularly the Los Angeles Times' Homicide Report Web page ([19]) or the crime databases themselves, to obtain the location of the OIS. Each of the OIS occurring between January 1, 2010, and the last day of 2014 were mapped twice; one map featured the point of initial contact, and the other included the location of the actual shooting. Most OIS occur very near the point of initial contact, but occasionally individuals are shot after a lengthy pursuit. Approximately five OIS included pursuits that began or ended outside of Los Angeles County; these were excluded from the analysis. Of the 375 total OIS, 246 occurred within the jurisdictional boundaries of LAPD or LASD. Of those, 114 were attributable to officers serving with LAPD and 132 involved deputies with LASD.

Methods

Once the point maps of crimes and OIS were completed, they were joined to a series of base maps at three scales: Census Tracts, ZIP Codes, and one of the LAPD and LASD law-enforcement divisions, which are rough equivalents to "precincts" in common parlance. Using GIS software, the number of OIS and crime offenses per capita per unit of analyses was calculated to produce a series of crime rate maps for each category of crime at each spatial scale (ZIP Code, Census Tract, and Division). These crime-rate variables were appended to tables of common demographic variables (age, income, ethnicity, unemployment rate, educational attainment, household status, etc.) at each scale. The maps of OIS rates were used as the dependent variable in a series of exploratory regression analyses designed to identify likely causal variables from a list of many dozens of demographic and crime variables, to eliminate variables exhibiting excessive collinearity problems, and to evaluate the stability of various explanatory models.

Finding a suitable model was a lengthy, iterative process. Many dozens of combinations of demographic variables, extracted from ESRI's expansive Business Analyst dataset, were combined with dozens of crime and crime-rate variables, producing a number of stable models, but few combinations of variables created any models with strong predictive power. The scale of the unit of analysis proved important in the stability of the models; Tract- and ZIP Code-level analyses were less stable because of the overabundance of observations without data (zero officer-involved shootings, zero homicides, etc.) . The maps with high levels of spatial clustering of no data observations resulted in excessive data skewing and contributed to unacceptable levels of spatial autocorrelation among model residuals. The enforcement division-level analysis proved most stable among the experimental spatial resolutions. Among the independent crime-rate variables, the homicide and assault with a deadly weapon (ADW) rates were consistent, strong predictors of the number and/or rate of OIS at the division level, even after controlling on a multitude of demographic variables (age, sex, income, race, ethnicity, homeownership, etc.) . An examination of the residual maps for models using the homicide rate variable prompted us to remove some homicides from the dataset because evidence, or logic, suggested the mapped location reflected not where the victim was murdered, but rather where the corpse was found. For example, several of the homicides eliminated from the dataset were reported within the Angeles National Forest. Catalina Island's Sheriff division was also removed from consideration since it was a consistent outlier.

Because our experimental models pointed directly at neighborhood rates of violence as the primary explanatory variable, we entered a second phase of analysis in which we assumed OIS were primarily a response to perceived levels of neighborhoodviolence, rather than neighborhood demographics. In order to better model violent crime as a perceptual variable, we converted both the Homicide and ADW point maps to "hot-spot" maps using a technique called Kernel Density Smoothing. This technique transformed the vector-based point maps of crimes into a raster (image) map, wherein each raster cell (a square pixel) is assigned a value representing proximity to one or more crimes.

The hot-spot raster map represents the spatial density of crimes differently. The software assigns values to each pixel according to its proximity to the mapped coordinates of each crime. We used a search radius of one mile. Therefore, map pixels immediately adjacent to the mapped crime were assigned a value of 1.0; pixels roughly one-half mile from each crime are assigned a value of 0.5; and those beyond roughly one mile are considered too distant from the crime and thus assigned a value of zero. Grid cells that are proximate to multiple crimes are assigned a cumulative value reflecting the sum of overlapping pixel values if they are within roughly one mile of multiple crimes. Pixels near multiple crimes may amass a value far greater than those near a single crime, creating what is popularly known as a crime "hot spot."

Converting the crime data to a raster format presented several advantages. Kernel Density Smoothing allowed us to treat each crime less as a discrete geographic event and more as an event that characterized, or affected the perception of, a small region surrounding the crime. Rasterizing the data also allowed us to map crime in a way that permitted the effects of crimes, including the perception of danger, to cross arbitrary census, ZIP Code, and precinct boundaries. This technique further ameliorates the problems presented by the randomized and anonymized LAPD data. Most importantly, it stands to reason that Kernel Density Smoothing maps function as a far more accurate proxy of officers' mental map of neighborhood violence.

The hot spot maps of homicides (Figure 1) and ADW (Figure 2) show dramatic differences in the level of violent crime across Los Angeles County. Many areas of the county had neither a homicide nor an ADW within a mile during the study period. The maximum value for the spatial density of homicides is over forty-three, occurring in two spots within LAPD's Southeast Division. Both hot spots are within about two miles of the intersection of Interstates 110 and 105. The ADW hot spot map is similar, but the most intense hot spot locations are near MacArthur Park and Skid Row, rather than in South LA.

The kernel density maps subsequently were used to create a series of crime-density maps at the enforcement division level. We used these in lieu of the more common per capita homicide and ADW rate variables for our analysis. Using a GIS tool called "Zonal Statistics," our new crime rate maps depict the mean spatial density of homicides and ADW crimes for each enforcement division (Figures 3–4). The Zonal Statistics tool takes each raster pixel from the hot spot map and identifies its location with respect to the enforcement-division boundaries. Then the tool identifies the minimum and maximum pixel values from the raster map, and calculates the range of values as well as the sum, mean, and standard deviation of the cell values within division boundaries. The sum of values and the pixel mean per division are excellent measures of criminal activity, and function as a superior

Graph: Figure 1. Southern Los Angeles County - Hot Spot Map of Homicides and OIS Locations 2010–2014.

Graph: Figure 2. Southern Los Angeles County - Hot Spot Map of Assault with Deadly Weapon Crimes and OIS Locations 2010–2014.

proxy for perceived levels of violent crime within each enforcement division. The value representing the mean spatial density of homicides and ADW at the enforcement division were included as causal variables in a final round of exploratory regression tests.

The final round of exploratory regression tests again measured the interplay between the number and rate of OIS, crime-rate variables, and a host of demographic variables. Several models emerged that were both robust and stable, but the top-performing models included some measure of the spatial density of ADW or homicide activity, along with an estimation of the age of citizens within each enforcement division. The most robust and stable models predicting the rate of OIS per capita using Ordinary Least Squares regression testing were the mean spatial density of homicides and ADW crimes, paired with the mean age of citizens within each division (see Table 1 for sample regression diagnostics). Model performance was modestly boosted by converting the OIS rate, or crime density measure to a logged scale. Moran's I tests showed neither model suffered from excessive spatial autocorrelation.

Graph: Table 1.

Graph: Figure 3. Southern Los Angeles County – Mean Spatial Density of Homicides and OIS Locations 2010–2014.

Graph: Figure 4. Southern Los Angeles County – Mean Spatial Density of ADW Crimes and OIS Locations 2010–2014.

Graph: Table 2.

General Discussion

The results of our statistical tests suggest that OIS are far more likely to occur in neighborhoods marked by high levels of violence, and that spatial density of both homicides and felonious assaults (ADW) are robust indicators of where OIS will occur. Our results also suggest that the rate of OIS is exacerbated in neighborhoods where the population is younger, even after controlling for the local violent crime rates. These conclusions are consistent with a recent study by [11] analyzing similar data in St. Louis. Our findings suggest that violence within a neighborhood invites or prompts an in-kind response from law enforcement. It is obvious that there are very few OIS in neighborhoods (enforcement divisions) with low levels of violent crime.

These findings do depart somewhat from the dominant media narrative surrounding OIS, which have focused on the racial and ethnic characteristics of the shooting victims and the officers, rather than the characteristics of the neighborhoods that play host to these incidents. However, it must be emphasized that our findings do not indicate that racism and/or ethnic bias are not factors in the rate or pattern of OIS in Los Angeles County. We can say that the pattern of OIS closely mirrors the pattern of violence in the county, and the ethnicity or racial characteristics of the various neighborhoods in which OIS occur does not appear to significantly increase or decrease the OIS rate, after controlling for the median age of the neighborhood.

It must be reiterated that many studies, mostly done within the context of Social Disorganization Theory, have established links between a host of neighborhood demographic characteristics (ethnicity, income, density, etc.) and violent crime (for a review of this literature, see [13]).[ 2] So although there may be a connection between neighborhood demographic characteristics and the rate of crime, we could not establish a direct connection between neighborhood demographics and the rate or number of OIS that was both statistically robust and stable. Certainly, none of the variables or combination of variables we tested predicted the rate of OIS the way the crime variables themselves did.

The difficulty in distinguishing racial or ethnic bias from the effects created by a community's other characteristics has an exceptionally long history. Some of the earliest criminological studies, conducted in the U.S. by scholars from the so-called Chicago School, established that various immigrant groups (Italians, Irish, etc.) were not naturally prone to criminal behavior; rather, their neighborhoods suffered from elevated criminal activity as a result of conditions endemic to the neighborhood (see e.g. Shaw & McKay 2019).

One might imagine that the choice of which operationalization one selects for a given variable might not matter much, given the high degree of collinearity among different measures. For instance, murder rate is likely to correlate positively with weapons violation arrests. However, the way in which a variable such as crime rate is deployed in a given analysis might significantly impact the findings. For example, a recent analysis by [ 4], suggests that the metric a researcher uses to operationalize crime may have significant consequences on the ultimate findings. In their analysis of 2015–16 data taken from The Guardian's online database, they tested for a racial discrepancy in shooting whites versus blacks. They examined all fatal shootings where individuals were unarmed and not actively threatening police, along with deadly shootings in which officers misidentified the target as armed. When they benchmarked their analysis against different metrics of violent crime, their results shifted dramatically—sometimes even reversing entirely.

After testing for the effects of many dozens of demographic characteristics and measures of crime on the rate of OIS, we found that the age profile of people living within the enforcement divisions is a stronger indicator of the rate of OIS than the racial or ethnic makeup of the divisions. This finding may very well be indicative of the likelihood that young people are more prone to do things that require and/or invite a more frequent armed response from law enforcement; perhaps it suggests that enforcement officers are more prone to shoot at young people. This is not an altogether surprising finding, but it does diverge somewhat from the media narrative that focuses on race and ethnicity narratives. We recognize that mean age correlates with a number of measures of race/ethnicity, income, and educational attainment, and indeed we identified models in which race/ethnicity, measures of income, or educational attainment were statistically significant, but they were not as robust as models including mean age, and other variables (e.g. ethnicity) did not remain significant after controlling for neighborhood age variables. This study invites further investigation into the potential effect of age in policing and law enforcement.

Another important finding of our study was the identification of enforcement divisions where the rate of OIS was far above or below model predictions (Figures 5 and 6). For example, we found that in some black neighborhoods, the rate of OIS fell below what one would expect, given the local violent crime rates and local mean age. In fact, the three law-enforcement divisions with populations of more than thirty percent African-American (LAPD Southwest, LAPD 77th Street and LASD Marina Del Ray), each had lower-than-expected rates of OIS, given the local age profile and pattern of crime density.

The role race/ethnicity plays in the rate of OIS in Latino (Hispanic) neighborhoods is less clear. Most of the models we evaluated indicated that percent Hispanic was not a statistically significant predictor of the rate or number of OIS per division, especially after controlling on crime rates and median age. For example, the two most robust models indicated that the three enforcement divisions with more than eighty percent Latino citizens each had higher-than-expected rates of OIS, even after controlling for the violent crime rate and neighborhood age characteristics. However, the same models most overpredicted the OIS rate in LAPD's Newton Division, where about seventy-eight percent of the population is Latino. In other words, the Newton Division had the biggest negative difference in actual rate of OIS versus model predictions.

Graph: Figure 5. Southern Los Angeles County – Model 1 Standardized Residuals (Homicide).

Graph: Figure 6. Southern Los Angeles County – Model 2 Standardized Residuals (ADW) e 6.

The location of model residuals might be the most compelling findings of this study. Some enforcement divisions have far more OIS than the model suggests they ought to, even after accounting for the local violent crime rates and neighborhood demographics. Others, such as LAPD's Newton Division, are just the opposite. The model points to an unevenness in the use of force around LA County that cannot be explained by reference to local violent crime rates or neighborhood demographics. LASD's East LA, Compton, and Lakewood Divisions, along with LAPD's Central Division, have unusually high rates of OIS. The East LA Sheriff's Department OIS rate is more than 2.3 standard deviations above what Model 1 indicated it should be. On the other hand, LAPD's Newton Division OIS rate is more than 2.4 standard deviations below the expected rate; LAPD's Mission and Wilshire Division also have lower-than-expected rates of OIS. These findings again echo the findings of [11] in St. Louis. The statistical anomalies may point to differences in divisional policy, training regimens, or even unevenness in divisional "departmental culture."

The policy and practical implications for leadership at LAPD and LASD are many, but a few points are salient. Clearly, this study invites both the LAPD and the LASD to closely examine divisional differences in the rate at which officers discharge their weapons. Obviously, local conditions are a factor, but a good deal of the OIS rate remains difficult to account for in some enforcement divisions. For reasons that remain unclear, officers in some areas of Los Angeles County are far more restrained in their use of force than they are in other areas, even where local crime rates and demographic profiles are virtually identical. Intra- and inter-agency policy and practice conversations and training on handling high-pressure situations could perhaps lower the rate at which law officers resort to deadly force.

Finally, as described in the introduction to this article, there is a surprising—one might argue unacceptable—lack of data on OIS. We argue for the continued need for local governments to offer transparent and easily accessible data to the public; moreover, we would urge policy makers and law enforcement agencies to consider including all instances in which an officer discharged a weapon with the intent to stop, harm, or kill a suspect. Without having a full view of the problem, it is very difficult to fully account for all the factors that contribute to OIS. Only then will law enforcement and policy-makers be able to offer any meaningful recommendations to a problem needing serious attention.

DMU Timestamp: February 03, 2020 23:30





Image
0 comments, 0 areas
add area
add comment
change display
Video
add comment

Quickstart: Commenting and Sharing

How to Comment
  • Click icons on the left to see existing comments.
  • Desktop/Laptop: double-click any text, highlight a section of an image, or add a comment while a video is playing to start a new conversation.
    Tablet/Phone: single click then click on the "Start One" link (look right or below).
  • Click "Reply" on a comment to join the conversation.
How to Share Documents
  1. "Upload" a new document.
  2. "Invite" others to it.

Logging in, please wait... Blue_on_grey_spinner