Investigating the association between household firearm ownership and suicide rates in the United States using spatial regression models
Introduction
Suicide is a complex mental health issue with devastating consequences on individuals, families, and communities worldwide. Nearly 45,000 suicide deaths were reported in the U.S. in 2016 alone, which exceeded total deaths from road accidents, or was more than 2.5 times the casualties from homicides (Our World in Data, 2019). Furthermore, among population aged 18 years and older, 9.8 million had serious suicidal thoughts, and 1.4 million had attempted suicide in the same year. Besides all the other loss, suicide also brings tremendous economic burdens. The average economic loss of one suicide death was estimated to be over $1.3 million, and the estimated annual total economic loss due to suicides and suicide attempts was between $50.8 billion and $93.5 billion in 2013 (Florence et al., 2015; Shepard et al., 2016).
The overall age-adjusted suicide rate (SR) in the United States has increased by 25.4%, from 10.5 to 13.4 per 100,000 between 1999 and 2016 (Stone et al., 2018). In addition, SRs have increased in all the 50 states except Nevada. The net increases ranged from 0.8 per 100,000 (Delaware) to 8.1 (Wyoming), while the relative increase was between 5.9% (Delaware) and 57.6% (North Dakota). Also, the SR rate increased by more than 30% in half of the states. Nevada was the only state that experienced a declined rate (1%) but the SR in the state also has been consistently higher than the national average throughout the study period (Stone et al., 2018). Moreover, SR in the western and northwestern states has been consistently higher than that in the rest of the country (Ha & Tu, 2018; Kim et al., 2011; Rossen et al., 2018; Trgovac, Kedron, and Bagchi-Sen 2015). Further, SR was not only overall higher but also has been rising faster in rural areas (Rossen et al., 2018; Singh & Siahpush, 2002).
In the United States, firearms are the most common and lethal means of suicide. In 2016, less than 5% of self-inflicted firearm shot in suicide attempts resulted in close to 50% of all suicide deaths (USDHHS 2019). The disproportionally high fatality rate of firearm suicide and the fact that the U.S. has the highest household firearm ownership in the world have led to an extensive investigation on the association between firearm ownership and SRs in the country. The case-control studies and the ecological studies are the two most common approaches used in past studies to examine such relationships (Miller & Hemenway, 1999). The former approach overall produces stronger evidence, but it usually analyzes only a small sample size and is also more difficult and costly to perform (Brent et al., 1993). The latter approach, more suitable for making large scale comparisons and generating hypotheses, has been adopted by most of the existing research. However, conclusions based on this approach is generally less convincing because the analysis is based on aggregated data. Ecologic studies can further be divided into two types, geographical and longitudinal (Levin, 2006). Compared with the longitudinal studies, the geographic studies require less data and it has been more commonly used. More longitudinal studies have been conducted in recent years as time series suicide data became more available (Miller et al., 2006; Siegel & Rothman, 2016). Regardless of the adopted method, the published studies have rather consistently shown significant and positive association between SRs, particularly firearm SRs, and the level of the household firearm ownership in the United States.
Further, most U.S.-based geographical ecological studies investigated either the association between SRs and firearm density or the effects of firearm regulations on SRs, and these studies relied mainly on correlation analysis and ordinary least squared (OLS) regression analysis to examine the relationship. One gap in the literature is that, though the spatial dependency in SRs has long been recognized (Balint et al., 2014; Helbich et al., 2012; Wasserma and Stack 1995), the effect has yet to be explicitly modeled in the studies of the relationship between SRs and household firearm ownership. Thus, the major objective of this study is to examine whether spatially-informed regression models can provide more robust model estimation. We collected and used the most recent SR data that were available to us when conducting this study. We controlled the major covariates in our regression models based on literature review and data availability. We intended to answer two research questions: 1) Do spatially-informed regression models perform better than the OLS models? 2) To what extent do county-level SRs (all, firearm and non-firearm) relate to state-level firearm ownership, after controlling other covariates in the models?
In the following sections, we first conduct a brief literature review, which is followed by an introduction of our data, modeling strategy, and models. We then present, interpret and compare the model results. We finally summarize the major findings, discuss the limitations, and propose future directions of this study before ending the paper with a short conclusion.
Section snippets
The suicide-firearm connection
Many suicide behaviors are impulsive by nature. The impulsiveness of many firearm related suicide attempts means that the risk period is likely be short-lived and many suicidal individuals are not determined to kill themselves. This fact provides the basis for the so-called means reduction for suicide prevention, that is, suicide can be prevented by reducing the availability and accessibility of the most lethal means (Peterson, Peterson, O’Shanick, & Swann, 1985; Chapdelaine et al., 1991;
The outcome variable: suicide rates
The SR data were collected from the Web-based Injury Statistics Query and Reporting System (WISQARS, 2019) maintained by the National Center for Health Statistics (NCHS), part of the U.S. Centers for Disease Control and Prevention (USCDC). For this study, we obtained age-adjusted and smoothed SR data (all, firearm, and non-firearm, aSR, fSR, and nfSR afterward) at the county (or equivalent) level in the 48 conterminous states between 2008 and 2014 (USCDC, 2019). The District of Columbia was
Methods
In this study, we began with building basic models and gradually moved to more complicated ones. So, we built OLS models first, and two forms of SAR models next, and HSAR models the last. Model diagnostic statistics were used to compare and evaluate these models so that the best modeling approach could be determined. In addition, separate individual models were built for the three outcomes variables, aSR, fSR, and nfSR.
A SAR model takes two general forms: spatial lag (SARlag) and spatial error
Exploratory spatial data analysis (ESDA)
The county-level SRs (aSR, fSR, and nfSR) were mapped and represented in Fig. 1. Overall, they were notably higher in the West and Midwest (except for most of California and Washington), and along the Appalachian Mountains. In addition, the spatial disparity of the SRs was more pronounced for fSR than for nfSR. Descriptive statistics of all the variables were summarized in Table 1.
The histograms of the raw SRs indicated that the data were right-skewed. To reduce the skewness, the Box-Cox
Discussion
The major findings of this study are summarized as follows. First, by explicitly accounting for the spatial effect in SR data, the SARlag model was superior to the classic OLS model. In addition, the HSAR model was found not to be significantly better than the SARlag model despite its designed conceptual advantages. One possible explanation is that most of the state-level spatial dependence has already been explained by the county-level effect. Second, the state-level firearm ownership was
CRediT authorship contribution statement
Wei Tu: Supervision, Conceptualization, Methodology, Writing - original draft. Hoehun Ha: Data curation, Investigation, Writing - review & editing. Weifeng Wang: Methodology, Formal analysis, Software. Liang Liu: Conceptualization, Methodology, Writing - review & editing.
Acknowledgments
We would like to thank Lingling Chen for providing research assistance for the manuscript. We would also like to thank the editor of Applied Geography and the two anonymous referees for their insightful comments and constructive criticism. The analyses and opinions in this article are entirely the responsibility of the authors.
References (71)
Under the hood: Issues in the specification and interpretation of spatial regression models
Agricultural Economics
(2002)- et al.
Area variations in health: A spatial multilevel modeling approach
Health & Place
(2012) - et al.
Investigation of the marked and long-standing spatial inhomogeneity of the Hungarian suicide rate: A spatial regression approach
Journal of Affective Disorders
(2014) - et al.
A school-based multilevel study of adolescent suicide ideation in California high schools
The Journal of Pediatrics
(2018) Suicide prevention through means restriction: Assessing the risk of substitution: A critical review and synthesis
Accident Analysis & Prevention
(2005)- et al.
Space in its place: Developing the link in medical geography
Social Science & Medicine
(1993) - et al.
The relationship between firearms and suicide: A review of the literature
Aggression and Violent Behavior
(1999) - et al.
County-level trends in suicide rates in the US, 2005–2015
American Journal of Preventive Medicine
(2018) - et al.
Geographic variation in male suicide rates in the United States
Applied Geography
(2015) - et al.
Suicide prevention strategies for improving population health
Child and Adolescent Psychiatric Clinics
(2016)
Suicide prevention strategies revisited: 10-year systematic review
The Lancet Psychiatry
Spatial weights matrix selection and model averaging for spatial autoregressive models
Journal of Econometrics
Project 2025: Firearms & suicide prevention
The association between gun ownership and statewide overall suicide rates
Suicide and Life-Threatening Behavior
Guns and homicides: A multiscale geographically weighted instrumental variables approach
Geographical Analysis
Comparing estimation methods for spatial econometrics techniques using R
An analysis of transformations
Journal of the Royal Statistical Society: Series B (Methodological)
Firearms and adolescent suicide: A community case-control study
American Journal of Diseases of Children
Firearm-related injuries in Canada: Issues for prevention
Canadian Medical Association Journal: Canadian Medical Association Journal
The Second amendment and firearms regulation: A venerable tradition regulating liberty while securing public safety
Multilevel analysis in public health research
Annual Review of Public Health
Spatial autoregressive models for geographically hierarchical data structures
Geographical Analysis
Spatial random slope multilevel modeling using multivariate conditional autoregressive models: A case study of subjective travel satisfaction in Beijing
Annals of the Association of American Geographers
Applied spatial econometrics: Raising the bar
Spatial Economic Analysis
Ecological fallacy, statistics of
Estimated lifetime medical and work-loss costs of emergency department–treated nonfatal injuries — United States, 2013
Risk factors for suicidal thoughts and behaviors: A meta-analysis of 50 years of research
Psychological Bulletin
Geography: Cartographic boundary shapefiles - counties
Multilevel statistical models
An ecological study on the spatially varying relationship between county-level suicide rates and altitude in the United States
International Journal of Environmental Research and Public Health
Geospatial examination of lithium in drinking water and suicide mortality
International Journal of Health Geographics
Medical geography: Taking space seriously
Progress in Human Geography
Gun ownership and social gun culture
Injury Prevention
State-specific, racial and ethnic heterogeneity in trends of firearm-related fatality rates in the USA from 2000 to 2010
BMJ Open
Firearm legislation reform in the European union: Impact on firearm availability, firearm suicide and homicide rates in Austria
The British Journal of Psychiatry
Cited by (7)
Spatial analysis of multivariate factors influencing suicide hotspots in Urban Tamil Nadu
2024, Journal of Affective Disorders ReportsThe effect of sociodemographic factors on the risk of poor mental health in Akron (Ohio): A Bayesian hierarchical spatial analysis
2021, Spatial and Spatio-temporal EpidemiologyCitation Excerpt :First, few US-based studies examine the spatial effect on mental health risk. These include the works of Fortney et al. (2007), Tran and Morrison (2020), and Tu et al. (2020). The majority of the literature that looks into the spatial effect of mental health risk is outside the US, such as Law and Perlman (2018), which is focused on Canada; Helbich et al. (2017) in Germany, Takahasi et al. (2017) in Japan, and Persad (2020) in Canada.
Modelling and mapping the poverty levels with applied spatial regression model in South Sulawesi province of Indonesia
2024, Journal of Social Economics ResearchThe COVID-19 Epidemic Spreading Effects
2022, Sustainability (Switzerland)