Predictive Extrinsic Factors in Multiple Victim Shootings

Abstract

Although researchers have found support for a relationship between temperature and violence and evidence of temporal patterns in violent crime, research on homicide shows less consistent results and no research on mass murder has been conducted. We address this by examining predictive factors in multi-victim shootings (those with four or more victims, including injured), a more general crime category than mass murder, but one with likely similar predictive factors. We used data from the Gun Violence Archive to understand the relationship between multi-victim shootings and temperature as well as other extrinsic factors. To avoid the confound between season and temperature, we employed temperature anomaly (the difference between actual and expected temperature) as a predictor of daily shooting rate. Using a generalized linear model for the daily count of multi-victim shootings in the U.S., we found that these events are significantly more frequent on weekends, some major holidays, hotter seasons, and when the temperature is higher than usual. Like other crimes, rates of multi-victim shooting vary systematically.

Keywords: Mass shooting; Routine activities theory; Temperature anomaly; Temperature/aggression theories.

Fig. 1

Multiple victim daily shooting rate averaged by year (points), with model M2 cubic fit to daily data (curve), demonstrating the data epoch’s daily MVS rate variation and its capture by the cubic fit. Relative rate (left scale) is referenced to epoch.time = 0

Fig. 2

Model M2 exponentiated monthly coefficients (box centers) and 95% confidence intervals (vertical lines) plotted by average monthly apparent temperature, demonstrating significant association between monthly average temperature and corresponding weekday MVS rate model coefficients. Vertical axis is logarithmic; fit line is linear regression on semi-log scale

Fig. 3

Geographic weightings of city apparent temperatures as combined to produce the continental U.S. apparent temperature. Circle areas are proportional to weighting. Circles are drawn partially transparent to reveal overlaps

Fig. 4

Intra-month standard deviation (SD) of temperature anomaly by month. Horizontal black line represents mean of intra-month SD across 6 years; bars represent bootstrap estimates of 95% confidence intervals from six data points (mean_cl_boot function from R package Hmisc)

Fig. 5

Conditional distributions of MVS count at four model-predicted MVS rates. Data are pooled within each graph for predicted rates within 0.25 of the stated rate (0.5, 1, 1.5, or 2 MVS events per day). Points represent relative frequencies of MVS counts, and bars represent 95% confidence intervals (binomial distribution). Lines correspond to Poisson probability distributions of the graph’s stated rate