A spatiotemporal model of firearm ownership in the United States

Abstract

Firearm injury is a major public health crisis in the United States, where more than 200 people sustain a nonfatal firearm injury and more than 100 people die from it every day. To formulate policy that minimizes firearm-related harms, legislators must have access to spatially resolved firearm possession rates. Here, we create a spatiotemporal econometric model that estimates monthly state-level firearm ownership from two cogent proxies (background checks per capita and fraction of suicides committed with a firearm). From calibration on yearly survey data that assess ownership, we find that both proxies have predictive value in estimation of firearm ownership and that interactions between states cannot be neglected. We demonstrate use of the model in the study of relationships between media coverage, mass shootings, and firearm ownership, uncovering causal associations that are masked by the use of the proxies individually.

Keywords: firearm ownership; firearm violence; spatial econometrics; time series.

Graphical abstract

Figure 1

Time series for proxies of firearm ownership on a national level (A–C) Time series between January 2000 and December 2019 for national-level background checks (A), background checks per capita (B), and fraction of suicides committed with firearms (C).

Figure 2

Predicted fraction of firearm owners in the United States For a Figure360 author presentation of this figure, see https://doi.org/10.1016/j.patter.2022.100546. The plot illustrates the model’s output between January 2000 and December 2019 for the entire country. It is overlaid with GPSS survey annual results, represented by red circles.

Figure 3

Processed time series for computation of transfer entropy (A–F) Nationally aggregated time series between January 2000 and December 2017 for background checks (A), background checks per capita (B), fraction of suicides commited with firearms (C), fraction of firearm owners (D) were seasonally adjusted and detrended. The time series for mass shootings (E) was discretized, and the time series for media output on firearm regulations (F), presented on a logarithmic scale, remained unmodified in the analysis.

Figure 4

Directional interactions in four triads, quantified using transfer entropy (A–F) Causal analysis results for (A) interactions between background checks (BCs), media output (MO) on firearm regulations, and mass shootings (MS); (B) interactions between background checks per capita (BCC), MO, and MS; (C) interactions between the fraction of suicides committed with firearms (SF), MO, and MS; (D) interactions between our model’s firearm ownership (FO), MO, and MS. Dashed arrows reflect non-significant transfer entropy ($0.1<p$), thin solid arrow indicate a trend ($0.05<p<0.1$), and bold solid arrows represent significant transfer entropy ($p<0.05$).

Figure 5

Causal analysis on a state level (A–C) State-level conditional transfer entropy (A) from FO to MS, conditioned on MO; (B) from FO to MO, conditioned on MS; and (C) from MO to FO, conditioned on MS.