This study has two components. The first is a descriptive epidemiologic analysis to describe the distribution of assaults and assault characteristics among LEOs. The second is a pooled cross-sectional analysis comparing LEO demographics, and situational and encounter characteristics to predict which factors result in increased odds of lethal assaults.
Data source for law enforcement officer assaults
Data for this study were assembled from the FBI’s Law Enforcement Officers Killed and Assaulted (LEOKA) database (FBI 2012). As part of the Uniform Crime Reporting program, the FBI generates this database from reports of every line-of-duty fatal assault (i.e., homicide), and nonfatal assault committed with a firearm or knife/cutting instrument that result in an injury (FBI 2004). Because the LEOKA database only captures nonfatal assaults committed with a knife/cutting instrument or firearm, this data represents only a subset of LEOs that experience a nonfatal assault. The database also includes LEOs that were off-duty if they were acting in an official capacity at the time of the assault (e.g., they identified themselves as an officer).
The LEOKA database contains a number of variables for each assault including those relevant to this study: suspects’ weapon type (e.g. firearm, blunt instrument, car), LEO assignment (e.g., one-officer vehicle), encounter (e.g., traffic stop, robbery in progress), location of the primary wound, distance from the suspect, whether the LEO fired his/her service weapon, and whether the LEO was wearing body armor when assaulted. Data for nonfatal assaults were not available in the LEOKA prior to 1998. Data for fatal assaults were available to the researchers back through 1984; however, this analysis only included fatal assaults from 1998 to 2013 to make appropriate comparisons between fatal and nonfatal assaults.
Analytic methods
Descriptive statistics were used to describe differences in fatal and nonfatal assaults for encounter, assignment, primary wound location, use of body armor, and suspects’ weapon use. A pooled cross-sectional analysis was conducted to examine which factors (LEO demographics, and situational and encounter characteristics) were associated with the odds of a lethal outcome following an assault against a LEO. Assaults against LEOs were coded as those that resulted in a fatality compared to assaults that were not fatal. Simple logistic regression was used to calculate odds ratios (OR) for the factors hypothesized to be related to whether an assault would result in a fatality: 1) LEO characteristics-age, experience, race (measured as White, Black, Asian, Native American), use of body armor, being disarmed by the suspect, discharging of the service weapon; 2) situational characteristics-type of assignment, distance from the suspect, and type of weapon used by suspect; and 3) encounter characteristics-i.e., the type of call the LEO was on or responding to at the time of the assault (e.g., disturbance call, traffic stop, robbery in progress).
Multiple logistic regression (MLR) was used to evaluate which characteristics increased the odds of an assault resulting in a fatality while controlling for factors that simple logistic regression indicated was also associated with fatal outcomes. There were a number of variables that were significant in the single logistic regression model and considered for the multiple regression model. These variables were excluded from the multiple regression model if they became insignificant (p < 0.05), did not improve model fit (measured by AIC and BIC), or significantly inflated variance (see Additional file 1: Table S1 for the complete results from the simple logistic regression).
In the MLR model for odds of lethality, there was collinearity between a LEO’s age and his/her level of experience. There was also collinearity between whether a LEO was disarmed and assaulted with his/her own gun. Both age and whether a LEO was disarmed were retained in the multiple logistic regression. Age was retained, as LEOs are likely to have increasing experience as they age. Whether the LEO was disarmed was retained as he/she could not be assaulted with his/her own weapon if not first disarmed. The final model included assignment, encounter, primary wound location, suspects’ use of a firearm, the age of the LEO, and whether the LEO wore body armor, was disarmed, or fired his/her weapon.
Analyses were conducted using Stata IC v 13 (StataCorp 2013). This study was deemed to be “not human subjects” research by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board.