How have firearm laws changed in states with unexpected decreases or increases in firearm homicide, 1990–2019?

Abstract

Background

Firearm violence is one of the leading preventable causes of death and injury in the United States and is on the rise. While policies regulating access to firearms offer opportunities to prevent firearm-related deaths, an understanding of the holistic impact of changing state firearm policies on firearm homicide rates over the last 30 years is limited.

Objectives

To identify US states that showed unexpected decreases and increases in firearm homicide rates and summarise their firearm policy changes in the last three decades.

Methods

We analysed changes in firearm homicide rates by US state and county from 1990 to 2019. We triangulated across three estimation approaches to derive state rankings and identify the top and bottom three states which consistently showed unexpected decreases (low outliers) and increases (high outliers) in firearm homicide rates. We summarised firearm policy changes in state outliers using the RAND State Firearm Law Database.

Results

We identified New York, District of Columbia, and Hawaii as low state outliers and Delaware, New Jersey, and Missouri as high state outliers. Low state outliers made more restrictive firearm policy changes than high state outliers, which covered a wider range of policy types. Restrictive changes in high state outliers primarily targeted high-risk populations (e.g., prohibited possessors, safe storage). Specific legislative details, such as the age threshold (18 vs 21 years old) for firearm minimum age requirements, also emerged as important for differentiating low from high state outliers.

Conclusions

While no firearm law change emerged as necessary or sufficient, an accumulation of diverse restrictive firearm policies may be key to alleviating the death toll from firearm homicide.

Highlights

• •We use an inductive approach to investigate the role of Firearm polices in shaping Firearm homicide.
• •Firearm homicide in New York, District of Columbia and Hawaii unexpectedly improved.
• •Firearm homicide in Delaware, New Jersey and Missouri unexpectedly deteriorated.
• •No one Firearm law was necessary or sufficient in shaping Firearm homicide.
• •Instead, an accumulation of a wide range of restrictive Firearm policies may be key.

1. Introduction

Firearm violence is one of the leading preventable causes of death and injury in the United States (US) and continues to rise (Sauaia et al., 2022). Despite promising improvements in the 1990s and early 2000s, firearm-related deaths in 2020 exceeded any year on record with more than 45,000 people losing their lives to firearms (WISQARS Fatal Injury Reports, 2021; Centers for Disease Control and Prevention, 2010; A Public Health Crisis Decades, 2019; Galea & Abdalla, 2022). In 2020, homicide accounted for around 40% of all firearm deaths – over 19,000 fatalities – due to an unprecedented 35% increase in firearm homicide from 2019 to 2020. This increase was disproportionally concentrated geographically and demographically among men and young people (aged 15–34 years) (Degli Esposti et al., 2022a, 2022b; Firearm, 2022; Smart et al., 2022). Policies regulating access to firearms have a role in preventing firearm-related deaths, but the impact of changing policies over time is not clearly established.

Federal firearm policy has been limited in the US, with federal regulation primarily restricted to dealer licenses and background checks (Law Center, 2023). Instead, most firearm regulation occurs at the state level, making the landscape of firearm policy in the US complex and uneven (Siegel et al., 2017). While state comparisons find that stronger firearm restrictions are generally associated with lower rates of firearm death (Fleegler et al., 2013; Kaufman et al., 2018), many of these studies are cross-sectional and do not speak to the causal role of changing firearm policies in driving state firearm deaths (Smart et al., 2020). Quasi-experimental examinations of specific legal changes can estimate the impact of a given law in isolation but struggle to account for the broader policy context. For example, there is growing evidence that Stand Your Ground (SYG) laws are linked to increases in firearm homicide rates (Degli Esposti et al., 2022a, 2022b; Yakubovich et al., 2020); that laws regulating firearm dealers and purchasing licences (Permit To Purchase, PTP) are associated with decreases in firearm homicide rates; and that Child Access Prevention (CAP) laws are effective in reducing unintentional firearm deaths among children (Degli Esposti et al., 2022a, 2022c; Irvin et al., 2014; Lee et al., 2017; Santaella-Tenorio et al., 2016; Schell et al., 2020; Zeoli et al., 2019). These studies also highlight variation in state responses to the same law (Degli Esposti et al., 2022a, 2022d), and such evaluations are ill-equipped to investigate the entire policy context and how different state firearm laws may be combining and interacting to shape differential impacts (Degli Esposti et al., 2022a, 2022b; Rich et al., 2022).

To advance understanding of the relationship between changing firearm policies and increases or decreases in firearm-related harms, researchers need to draw upon different approaches to investigate this complex issue. Traditional deductive analyses, as described above, test existing theories using empirical observations. However, relying solely on this approach loses critical information around the wider policy context, and has fallen short in explaining within- and between-state response heterogeneity to the same firearm law. An inductive approach, on the other hand, uses observations as the starting point of analysis to explore and develop new theories. It can thus provide a complementary method to further a more holistic understanding of state firearm policy and their impacts. Previous work has indicated that firearm homicide research could benefit from inductive approaches since homicide is more geographically concentrated to hotspots than other types of intent (e.g., suicide) (Degli Esposti et al., 2022a, 2022b; Smart et al., 2022). Inductive approaches have shed critical insights and identified key drivers elsewhere in the literature (e.g., the cost of healthcare) (The Cost Conundrum | The, 2009), and anecdotally, California is often held as a potential template for modelling nationwide firearm policies due to argued success in reducing firearm homicides (California's Answer to Gun Violence, 1977). However, this approach has not yet been systematically applied to investigate changing firearm policies and how they relate to unexpected changes in firearm homicide rates over time.

In this study, we build on prior work (Degli Esposti et al., 2022a, 2022b), to implement a novel approach to examining state firearm policies. By modelling expected changes in firearm homicides, we triangulate across model estimates to rank and identify states that showed unexpected decreases and increases in firearm homicide from 1990 to 2019. We provide detailed summaries of firearm policies in these state outliers to gain new insights on the role of specific and wider policy changes in shaping trends in firearm homicide in the last three decades.

2. Methods

This descriptive ecological analysis used firearm mortality data from 49 states and the District of Columbia from January 1, 1990, to December 31, 2019. The research received institutional review board approval from the University of Pennsylvania, US.

2.1. Data

We analysed restricted access mortality data from the National Vital Statistics System (National Center for Health Statistics, 1989–2019). These data are based on deidentified death records of underlying cause of death for >99% of all deaths in the US, defined by the International Classification of Diseases (ICD). We extracted deaths by firearm homicide and non-firearm homicide (eTable 1 for ICD codes). Information on time of death and county of residence at time of death as a proxy for county of death, as well as corresponding population estimates. We created a panel dataset of space-time observations, with each observation representing 1 year of data per county for a total of 96,441 county-years (3,111 counties x 31 years). Alaska, Bedford City (Virginia), and Broomfield County (Colorado) were excluded due to definitional inconsistences during the study period (Dorn, 2021). All non-firearm homicides deaths in 2001 were coded as missing and then imputed using last observation carried forward to guard against inflation from the deaths caused by the 9/11 terrorist attacks.

We condensed the panel dataset to six 5-year intervals, starting with January 1, 1990, to December 31, 1994, and ending with January 1, 2015, to December 31, 2019 (see eTable 2 for all 5-year intervals). Five-year intervals were used to minimize zero and low counts, in line with previous methodology (Degli Esposti et al., 2022a, 2022b). Summed totals (homicide divided by population totals) were used to derive county-level rates. We further derived an aggregated state-level dataset for the state-level analyses (see Approaches 1 and 2 in Statistical Analysis) by summing the total number of firearm homicides and the population size of all counties in each state.

2.2. Firearm laws

To identify changes in state firearm laws from 1990 to 2019, we extracted information from RAND Corporation's longitudinal database of firearm laws: the RAND State Firearm Law Database (version 4.0) (Cherney et al., 2022). RAND's database includes complete information on all state laws from 1979 through June 2022 that correspond to the primary firearm law categories (e.g., background checks, firearm sales restrictions, see Table 1) grouped into the policy target: ownership, sales, and use. Full details of the database, including its development, methodology, and comprehensive list of laws are specified elsewhere (Cherney et al., 2022). We extracted information on change in state firearm laws during our study period (January 1, 1990, to Decemeber 31, 2019), which corresponded to the main policy types listed in Table 1. These policies were adapted from RAND's original law types to create a condensed – thus simplified and more interpretable – categorisation system. Nevertheless, the main policies and laws are consistent with RAND's original formulation. Change was defined as an implementation, modification, and/or a repeal that occurred between January 1, 1990, to Decemeber 31, 2019. We grouped law changes that referred to the same sub-type of law and experienced the same type of change on the same date.

Table 1.

Main types of firearm policies adapted from the RAND State Firearm Law Database.

Target of the policy	Firearm policy	Description	Specific laws
Policy	Local laws	Laws prohibiting local laws to overwrite state laws, these laws can cover all firearm laws broadly or specific classes of firearm laws (e.g., registration, concealed carry)	n/a
Ownership	Purchasing licences	Laws that determine whether a license or permit – issued by a government authority – is required for an individual to buy and own a firearm, and laws that establish the maximum amount of time law enforcement may wait before issuing a permit or denying the application	Permit to purchase (PTP); maximum waiting period
	Prohibited possessors	Laws prohibiting the purchasing or possession (via seizures) of firearms for high-risk individuals, including felons, those with a history of violence, or individuals with mental conditions or an intellectual disability. Includes laws removing firearms where there is reasonable concern of violence, or the individual is a known risk to themselves or others	Firearm seizures; misdemeanour violence prohibition (MVP); domestic violence restraining order (DVRO); extreme risk protection order (ERPO)
	Minimum age	Laws establishing a minimum age for firearm possession, and laws prohibiting the sale to and/or purchase of firearms under a specified age	Youth possession; minimum age to purchase and sale
	Registration	Laws that require individuals to record their ownership of a firearm with a designated government agency (typically a law enforcement agency) and requires that these records are updated when transferred to a new owner (with a few exceptions)	n/a
	Reporting lost and stolen firearms	Laws that require firearm owners to notify law enforcement about the loss or theft of a firearm to deter gun trafficking, straw purchasing, and illegal possession	n/a
	Gun trafficking	Laws that prohibit the transfer of a firearm to a person who is ineligible to possess a firearm, as well as the purchase of or furnishing of a firearm for a person while knowing that the person is prohibited from possessing a firearm under state or federal law	Straw purchase
	Safety training	Laws that require individuals to undergo firearm safety training prior to being able to purchase and/or carry a firearm and/or to obtain a permit for carrying a concealed weapon	n/a
	Safe storage	Laws that require gun owners to store their firearms unloaded and locked when unattended in order to help prevent unauthorized users (e.g., children) from accessing and using firearms, and prevents individuals from furnishing guns to minors	Child access prevention (CAP): negligent storage; intentional, knowing, or reckless provision
Sales	Dealer licenses and inspections	Laws that regulate firearm dealers, including requiring dealers to obtain a license from the state and permit or require inspections of dealers	n/a
	Background checks	Laws that identify individuals who are ineligible to purchase firearms and prevent those persons from obtaining them. Background checks most commonly apply to sales from dealers but can also cover private sales	Comprehensive background checks (CBC)
	Waiting period	Laws that prevent gun purchases from taking possession of their firearm immediately upon purchase and/or completion of a background check. These laws impose delays of days or weeks between the purchase and the date which the buyer may take possession of the firearm	n/a
	Firearm sales restrictions	Laws that determine specific conditions of firearm sales; both in terms of the quantity of sales and the type of firearm sold	Handgun bans; assault weapons ban (AWB); Saturday Night Specials or junk gun ban; one handgun a month; untraceable firearms or firearm parts without a unique serial number
Use	Firearm carry	Laws that allow individuals to carry firearms in public places. Includes laws specifying the carrying of concealed weapons (CCW), as well as open carry where the firearm is visible (open carry)	Shall-issue right to carry (RTC); may issue; open carry
	Self-defense	Laws that determine individual rights on the use of lethal violence, including the use of firearms, in self-defense; from public places to an individual's property (home, vehicle, workplace)	Stand your ground (SYG); castle doctrine
	Location restrictions	Laws that prohibit the possession of firearms in specific locations, such as airplanes, post offices, government buildings, and public schools/colleagues/universities	Gun-free zones; gun-free school zones act (GFSZA)
	Firearm use in educational settings	Laws that allow school personnel, in addition to their other official duties, to carry firearms on school property (not including guards or police officers). Staff authorized to carry firearms may be required to take training in use of force, weapons proficiency, the law, etc.	n/a

The details of the RAND State Firearm Law Database are detailed elsewhere (Cherney et al., 2022).

2.3. Statistical analysis

We used observed-to-expected (O/E) ratios to rank and identify states that showed the largest changes (increases and decreases) in firearm homicide rates from 1990–1994 to 2015–2019. Observed and expected values were estimated using three approaches (see below).

Approach 1: Relative change in state firearm homicide. We derived relative (crude) change in state firearm homicide rates from 1990–1994 to 2015–2019. This was estimated by dividing observed firearm homicide rates in 1990–1994 (i.e., the expected values for 2015–2019) by observed homicide rates in 2015–2019 for each state to derive O/E ratios. O/E ratios greater than 1 indicated increases, while O/E ratios smaller than 1 indicated decreases in firearm homicide rates from 1990–1994 to 2015–2019.

Approach 2: Discrepancy between state firearm and non-firearm homicide. We estimated the expected values for firearm homicide rates in 2015–2019 using previous firearm homicide rates and non-firearm homicides rates for 5-year intervals throughout the full study period (1990–2019). We modelled firearm homicide rates by entering time (as 5-year intervals, see eTable 2) and non-firearm homicide rates as predictors in a regression model. This model was then used to estimate expected rates of firearm homicide in 2015–2019. We compared these expected values with the observed values for firearm homicide in 2015–2019 to derive O/E ratios. Observed values greater than expected values represented those states that had higher-than-expected homicide rates based on trends in firearm and non-firearm homicide rates.

Approach 3: Unexpected county-level changes in firearm homicide. Bayesian spatially explicit regression models were fitted for firearm homicide rates using a negative binomial model with the population (logged) as an offset variable (Moraga, 2021; eMethods). This modelling approach can explicitly account for spatial autocorrelation via the Besag-York-Mollié (BYM) model and by incorporating a spatially structured random effect, specifically queen adjacency. Extending on previous methodology (Degli Esposti et al., 2022a, 2022b), firearm homicide rates for 1990–1994 and county characteristics (geography, education, economics, politics, health, and firearm availability) were entered as predictors (see eTable 3 and eFig. 1 for details of county characteristics). Observed values represented firearm homicide rates in 2015–2019 while expected values were estimated by our Bayesian models. We derived O/E ratios by collapsing county-level observed and estimated rates to the state level. Discrepancies between observed and expected rates indicated whether a state showed lower- or higher-than-expected county changes in firearm homicide rates from 1990–1994 to 2015–2019. Technical details for the model and information on the county characteristics are specified in eMethods, eTable 3, and eFig. 1 in the Supplement. Models were fitted using the Integrated Nested Laplace Approximation method in R (version 4.1.0) (R-INLA) (Rue et al., 2009).

Overall ranked scores. For each approach, we ranked states from the smallest to largest O/E ratio (range: 1–50). To overcome biases specific to any one approach, we triangulated across all three estimation approaches to identify states that consistently emerged as showing unexpected changes. We generated an average overall rank (composite) score so that states were then assigned a new ranking based on their average ranking when looking across all three estimation approaches. The states which had the bottom (lowest) and top (highest) composite scores were identified as state outliers. The bottom three states represent states with the most robust unexpected decreases over time (i.e., low state outliers), whereas the top three states represented the states with the most robust unexpected increases over time (i.e., high state outliers).

3. Results

From January 1, 1990, through December 31, 1994, there were 87,438 firearm homicides which fell to 70,302 from January 1, 2015, through December 31, 2019, (eTable 4). California accounted for the largest share of firearm homicides during this early period (16.8%) but only 14% of the US population, followed by Texas (9.9% firearm homicides; 8% US population) and New York (9.7% firearm homicides; 7.3% US population). While California and Texas continued to account for the highest absolute burden of homicides more recently (2015–2019), New York only accounted for 2.5% of firearm homicides.

Supplemental figures illustrate the information used for the three estimation approaches (eFigs. 2–4). Resultant O/E ratios and state rankings obtained are summarised in eTable 5 and are visualised in eFigs. 5–11. Table 2 and Fig. 1 represent the final composite ranking of states by unexpected change in firearm homicide rates from 1990–1994 to 2015–2019. Western states typically showed reductions in firearm homicide over time (e.g., Oregon, California), while mid-western states (e.g., Missouri, Wisconsin, Indiana) showed increases during the same period. New Jersey and Delaware also showed increases over time, unlike neighbouring states in the North-East. New York, District of Columbia, and Hawaii were identified to be low state outliers, whereas Delaware, New Jersey, and Missouri emerged as high state outliers (Table 2, Table 3).

Table 2.

State ranks based on O/E ratios for unexpected change from 1990–1994 to 2015–2019.

State	Approach 1: Relative change	Approach 2: Firearm vs non-firearm	Approach 3: Unexpected county change	Composite ranks
				Mean	Final
NY	1	6	4	3.67	1
DC	2	11	2	5.00	2
HI	9	2	7	6.00	3
CA	3	21	1	8.33	4
RI	5	5	26	12.00	5
TX	6	27	3	12.00	6
OR	17	12	9	12.67	7
AZ	18	16	6	13.33	8
MT	20	3	18	13.67	9
UT	27	14	12	17.67	10
AR	12	22	19	17.67	11
FL	15	26	13	18.00	12
WA	16	19	20	18.33	13
WY	22	8	27	19.00	14
MI	8	33	17	19.33	15
LA	13	41	5	19.67	16
CT	4	18	39	20.33	17
NC	7	31	24	20.67	18
VA	10	42	11	21.00	19
ME	32	25	10	22.33	20
CO	37	10	23	23.33	21
MA	11	28	31	23.33	22
VT	19	15	40	24.67	23
SD	49	1	25	25.00	24
NV	21	13	42	25.33	25
ID	24	44	8	25.33	26
NH	25	17	35	25.67	27
GA	14	47	16	25.67	28
WV	23	20	37	26.67	29
ND	46	7	30	27.67	30
PA	39	32	15	28.67	31
NM	47	4	36	29.00	32
MD	31	34	22	29.00	33
MS	30	46	14	30.00	34
TN	29	43	21	31.00	35
OK	42	23	29	31.33	36
IA	48	9	38	31.67	37
MN	28	24	44	32.00	38
KS	35	30	33	32.67	39
SC	33	39	32	34.67	40
AL	34	49	28	37.00	41
NE	36	36	41	37.67	42
KY	41	35	43	39.67	43
OH	45	29	45	39.67	44
IL	26	45	49	40.00	45
IN	40	37	47	41.33	46
WI	38	40	48	42.00	47
MO	43	50	34	42.33	48
NJ	44	38	46	42.67	49
DE	50	48	50	49.33	50

O/E ratios, observed-to-expected ratios.

Fig. 1.

Final (composite) ranks of US states across estimation approaches.

Table 3.

Summarising low and high state outliers and changes in Firearm laws, 1990–2019.

State (rank)	Permissive change, 1990–2019				Restrictive change, 1990–2019
	Total changes	Firearm policy (type of change)	Specific law	Date	Total changes	Firearm policy (type of change)	Specific law	Date
Low state outliers (top 3)
NY (1/50)	0	n/a	n/a	n/a	13 implementations; 2 modifications	Purchasing licences (1 modification)	Modified PTP law to recertify permits issued before Jan 2013	01/01/2013
						Prohibited possessors (2 implementations; 1 modification)	Implemented DVRO laws	01/11/1996
							Modified (broadened) definition of “intimate relationship” in DVRO laws	21/07/2008
							Implemented ERPO law	24/08/2019
						Minimum age (1 implementation)	Implemented minimum age (21 years) of possession, as well as purchase and sale	01/11/2000
						Reporting lost and stolen firearms (2 implementations)	Implemented law to trace firearms used in crimes	08/08/2000
							Implemented reporting of lost and stolen firearms	01/11/2000
						Safe storage (1 implementation)	Implemented CAP law for intentional, reckless, or knowing provision to <16 years old	28/09/2019
						Background checks (4 implementations)	Implemented 2 background checks for licensed dealers: handgun & long gun	28/02/1994 & 30/11/1998
							Implemented background checks for private sales for handgun and long gun	16/03/2013
							Implemented extra time conditions for background checks for handgun and long gun	12/09/2019
						Firearm sales restrictions (1 implementation)	Implemented AWB for handgun and long guns	01/11/2000
						Firearm carry (1 implementation)	Implemented an open carry prohibition	01/11/2006
						Location restrictions (1 implementation)	Implemented a prohibition of possessing a weapon on school grounds	16/03/2013
DC (2/50)	1 implementation; 1 modification; 2 repeals	Firearm sales restriction (1 repeal) & purchasing licences (1 implementation)	Repealed the handgun ban, which was superseded with a PTP law	26/06/2008	9 implementations	Prohibited possessors (1 implementation)	Implemented a ERPO law	19/06/2019
		Firearm sales restrictions (1 repeal)	Repealed one gun per month	18/09/2015		Safety training (1 implementation)	Implemented safety training requirements for handguns and long guns	31/03/2009
		Firearm carry (1 modification)	Modified CCW law from may to shall issue	06/10/2017		Background checks (2 implementations)	Implemented 2 background checks for licensed dealers: handgun & long gun	28/02/1994 & 30/11/1998
						Waiting period (1 implementation)	Implemented waiting period (10 days) for handguns & long guns	22/10/2009
						Firearm sales restrictions (1 implementation)	Implemented ban on Saturday night special guns	27/12/2013
						Firearm carry (1 implementation)	Implemented (extended) an open carry prohibition to long guns	16/06/2015
						Location restrictions (2 implementations)	Implemented a prohibition of possessing a weapon on school grounds	01/08/1994
							Implemented a ban of CCW on school grounds	16/06/2015
HI (3/50)	0	n/a	n/a	n/a	10 implementations; 7 modifications	Purchasing licences (2 modifications)	Modified by adding a maximum waiting period (14 days) for first time applicants for handguns and long guns	29/06/1992
							Modified (expanded) the maximum waiting period (14 days) to all applicants	21/06/1994
						Prohibited possessors (4 implementations; 2 modifications)	Implemented prohibition for diagnosed mental health condition	19/06/1990
							Implemented DVRO laws	10/06/1993
							Implemented firearm seizure law for those who pose a domestic threat	01/07/1996
							Implemented prohibition of convicted criminal	26/05/2000
							Modified DVRO laws	07/06/2000
							Modified (expanded) firearm seizure law to reasonable grounds	03/07/2012
						Minimum age (1 modification)	Modified minimum age law to 21 years old or more for handguns and long guns	01/07/1994
						Registration (1 modification)	Modified registration of long guns within 5 days of acquisition	01/07/1994
						Reporting lost and stolen firearms (1 implementation)	Implemented reporting of lost and stolen firearms (handguns & long guns)	24/04/2019
						Safety training (1 implementation)	Implemented safety training requirements for handguns	01/07/1995
						Safe storage (1 implementation)	Implemented CAP law for negligent storage	29/06/1992
						Background checks (2 implementations)	Implemented 2 background checks for licensed dealers: handgun & long gun	28/02/1994 & 30/11/1998
						Waiting period (1 modification)	Modified waiting period law (14 days)	29/06/1992
						Firearm sales restrictions (1 implementation)	Implemented AWB for handguns	01/07/1992
High state outliers (bottom 3)
MO (48/50)	1 implementation; 4 modifications; 5 repeals	Purchasing licences (2 repeals)	Repealed PTP law & maximum waiting period (7 days)	28/08/2007	4 implementations	Prohibited possessors (1 implementation)	Implemented prohibition for adjudicated as mentally incompetent/incapacitated/disabled	28/08/2008
		Minimum age (1 repeal)	Repealed minimum age (21 years) for purchase and sale	28/08/2007		Background checks (2 implementations)	Implemented 2 background checks for licensed dealers: handgun & long gun	28/02/1994 & 30/11/1998
		Registration (1 repeal)	Repealed registration of firearms	28/08/2007		Location restrictions (1 implementation)	Implemented ban of CCW on school grounds	11/10/2003
		Background checks (1 repeal)	Repealed background checks for sales from dealers and private sales	28/08/2007
		Firearm carry (1 implementation; 2 modifications)	Modified CCW law to shall-issue (permit required)	26/02/2004
			Implemented open carry law which requires a license	08/10/2012
			Modified CCW law to shall-issue (permit not required)	01/01/2017
		Self-defense (2 modifications)	Modified castle doctrine, expanded to beyond he home to specific locations (e.g., vehicle, work place)	28/08/2007
			Modified castle doctrine, expanded anywhere a person has a right to be	14/10/2016
NJ (49/50)	0	n/a	n/a	n/a	16 implementations	Prohibited possessors (4 implementations	Implemented DVRO laws	14/01/2004
							Implemented firearm seizure laws for domestic violence	14/01/2004
							Implemented firearm removal of convicted domestic violent offender	01/08/2017
							Implemented ERPO law	01/09/2019
						Minimum age (1 implementation)	Implemented minimum age (18 years) of possession, as well as purchase and sale	01/01/2001
						Reporting lost and stolen firearms (3 implementations)	Implemented reporting of lost and stolen firearms	13/01/2008
							Implemented 2 laws to trace firearms used in crimes	18/09/2013 & 01/01/2014
						Safe storage (1 implementation)	Implemented CAP law for negligent storage	17/01/1992
						Background checks (2 implementations)	Implemented 2 background checks for licensed dealers: handgun & long gun	28/02/1994 & 30/11/1998
						Firearm sales restrictions (4 implementations)	Implemented AWB for handgun and long guns	30/05/1990
							Implemented a law restricting types of ammunition (e.g., nail gun ammunition ban)	01/04/2008
							Implemented one gun per month law	01/05/2009
							Implemented laws restricting various firearm types (e.g., 3D printed, firearm parts without a serial number)	08/11/2018
						Location restrictions (1 implementation)	Implemented a prohibition of possessing a weapon on school grounds	09/09/1992
DE (50/50)	0	n/a	n/a	n/a	8 implementations; 3 modifications	Prohibited possessors (2 implementations; 3 modifications)	Implemented prohibition for committed to mental health facility	21/07/1992
							Modified (expanded) 2 DVRO laws	24/07/1999 & 18/09/2007
							Modified (expansion of criteria) prohibition for committed to mental health facility	30/10/2018
							Implemented ERPO laws	27/12/2018
						Minimum age (1 implementation)	Implemented minimum age (18 years) of possession, as well as purchase and sale	08/07/1994
						Reporting lost and stolen firearms (2 implementations)	Implemented reporting of lost and stolen firearms	12/06/2013
							Implemented law to trace firearms used in crimes	14/09/2016
						Safe storage (1 implementation)	Implemented CAP law for negligent storage	12/07/1994
						Background checks (2 implementations)	Implemented 2 background checks for licensed dealers: handgun & long gun	28/02/1994 & 30/11/1998

AWB, assaults weapons ban; CAP, child access prevention, CCW, carrying a concealed weapon; DVRO, domestic violence restraining order; ERPO, extreme risk protection orders; PTP, permit to purchase; SYG, stand your ground.

3.1. Low state outliers: New York, District of Columbia & Hawaii

From 1990 to 2019, New York and Hawaii experienced no permissive changes in firearm laws (Table 3). District of Columbia underwent a handful of recent permissive changes, including the repeal of the handgun ban in 2008 (which was replaced by a permit to purchase (PTP) law) and of the one gun per month law in 2015. District of Columbia's firearm carry policy was also relaxed in 2017, moving from a may issue to a shall issue regime. However, all three low outliers implemented numerous restrictive changes to their firearm policies during the study period. New York, District of Columbia, and Hawaii expanded their laws on prohibited possessors, spanning from domestic violence restraining order (DVRO) to extreme risk protection orders (ERPO) to firearm seizure laws. New York and Hawaii enacted a series of restrictions to firearms sales (e.g., assaults weapons ban (AWB)), mandatory reporting of lost and stolen firearms, Child Access Prevention (CAP) laws for negligent storage, and introduced minimum age requirements for persons aged 21 years and older. District of Columbia already had these restrictive firearm policies in place by 1990. Additional changes included New York and District of Columbia prohibiting open carry and restricting firearms in schools/educational settings, while District of Columbia and Hawaii enacted safety training requirements and expanded their waiting periods for purchasing firearms.

3.2. High state outliers: Delaware, New Jersey & Missouri

Delaware and New Jersey did not make permissive changes to their firearm policies during the study period. On the other hand, Missouri repealed many of their previous restrictive firearm laws in the mid- to late-2000s, including a repeal of their permit-to-purchase (PTP) law, registration of firearms, background checks, and minimum age requirements. Missouri further expanded its open carry and self-defense laws (enacting a Stand Your Ground law) and moved to a shall issue policy for carrying a concealed weapon. By 2019, Missouri had one of the most permissive firearm policy contexts (Everytown, 2022).

Similarly to low state outliers, all three high state outliers expanded their firearm policies for prohibited possessors. For example, New Jersey enacted firearm seizure, DVRO and ERPO laws; Delaware enacted DVRO and ERPO laws, and Missouri enacted a mental health prohibition. New Jersey and Delaware further enacted mandatory reporting of lost and stolen firearms, CAP laws for negligent storage, and minimum age requirements for possession. However, unlike low outlier states, the minimum age requirement was for 18 years and older not for 21 years and older. New Jersey enacted a series of laws that imposed various firearm sales restrictions, including AWB, 3D printed guns, and restricting purchases to one gun per month. New Jersey and Missouri also restricted the carriage of guns in school/educational settings, with New Jersey banning all possession and Missouri prohibiting CCW in these locations.

3.3. Comparing low and high state outliers

Overall, low state outliers enacted fewer permissive changes than high state outliers from 1990 to 2019 – although this was primarily driven by Missouri. Low state outliers enacted a higher number and wider spectrum of restrictive firearm policies compared to high state outliers. Other than the enactment of the Brady Handgun Violence Prevention Act of 1994, which established a nationwide requirement that licensed firearms dealers must observe a waiting period and initiate a background check for firearm sales (Ludwig & Cook, 2000), restrictive firearm law changes among high outlier staters were primarily limited to prohibited possessors, safe storage, minimum age requirements, and location restrictions. Only New Jersey implemented restrictive changes to their firearm sales restrictions among the high outlier states. Meanwhile, the restrictive changes among low outlier states targeted a more diverse range of policy types, including more restrictive laws for purchasing licenses, firearm carriage laws (e.g., open carry prohibition), safety training requirements, and waiting periods.

4. Discussion

This study adopts a novel inductive approach to identify state outliers with large increases and decreases in firearm homicide and summarises relevant legislative changes over three decades. We identified states that consistently showed unexpected decreases or increases over time, irrespective of the modelling approach used. New York, District of Columbia, and Hawaii showed robust reductions in firearm homicide rates (i.e., low state outliers), whereas Delaware, New Jersey, and Missouri showed concerning increases over the same period (i.e., high state outliers) – despite the national decreasing trend. Overall, low state outliers made more diverse restrictive changes to their firearm policies and fewer permissive changes over time than high state outliers. Although we found potentially important differences between high and low outlier state firearm policy changes (e.g., minimum age 18 vs 21 years old), no law emerged as necessary or sufficient in shaping whether a state unexpectedly improved or deteriorated over time. Rather it was an accumulation of more restrictive and less permissive firearm laws that differentiated high from low outlier states.

This study extends previous work showing the importance of complementary approaches to advancing firearm research (Degli Esposti et al., 2022a, 2022b). Inductive research offers the opportunity to identify and learn from geographical areas (e.g., states, counties) which are “success stories”, while flagging at-risk areas as targets for future intervention (The Cost Conundrum | The, 2009). This approach has been applied informally and anecdotally in the media and podcasts (California's Answer to Gun Violence, 1977; The Daily, 2022), but had not yet been systematically applied using transparent, quantitative methods. In doing so, we identified clear examples of success states that have managed to tackle and reduce initially high rates of firearm homicides (e.g., New York, District of Columbia). We also identified states that are cause for concern (e.g., Delaware, Missouri), showing unexpected increases in firearm homicide rates over the last 30 years. Here, we document changes in firearm policies in these states, but future mixed methods research is needed to better understand potential drivers of change in firearm violence in these state outliers.

We found that it was not just one or two specific firearm policies that shaped changes in firearm homicide. Instead, it was an accumulation of the enactment of more restrictive firearm policies over time – which targeted a range of mechanisms, such as ownership, sales, carriage, and use – that was key for differentiating low from high state outliers. Although high state outliers also enacted numerous restrictive firearm policies during the study period, these policies typically targeted certain at-risk populations (e.g., prohibited possessors, CAP laws), rather than enacting policies that impacted the whole state population (e.g., purchasing licenses, firearm sales restrictions). Our findings indicate that a range of policy types, as well as laws changing firearm-related behaviours for the whole population, may be most effective in reducing firearm violence. While enacting policies that address certain risky populations remains important, a public health approach to tackling firearm homicide may be best served via policies which restrict firearm ownership and use for the entire population, reducing the whole population's distribution of risk rather than only reducing the risk for specific subgroups (Rose, 2001).

Our state-comparison highlighted some important differences between low and high state outliers for specific firearm policies. Missouri (a high state outlier) was the only state to enact SYG laws during this period. On the other hand, low state outliers enacted more waiting periods and CAP laws. These findings echo evidence generated from narrow examinations of specific firearm laws (Degli Esposti et al., 2022a, 2022b; Santaella-Tenorio et al., 2016; Schell et al., 2020; Zeoli et al., 2019). In addition, we identified differences in minimum age requirement laws for firearm possession and purchase/sale firearm between low and high state outliers. All three low state outliers had minimum age requirement laws which restricted possession and/or purchase to 18 years and older, whereas similar laws for high state outliers set an older age requirement of 21 years. This three year distinction might be critical for reducing firearm homicides, especially because persons aged 19–21 years account for a disproportionate burden of all firearm homicides. In order for minimum requirement laws to be effective for firearm homicide, a lower 18-year-old threshold may be needed. Prior research has underlined the importance of this age threshold for reducing avoidable fatalities from firearm suicide (Raifman et al., 2020), but additional robust evaluation evidence is needed to determine the role of the age threshold in minimum requirement laws for firearm homicides.

This study has a number of limitations. First, the state outliers we identified as showing unexpected changes in firearm homicide will be dependent on the period studied (1990–2019), particularly as we conceptualised change in two of the three estimation approaches as the difference between two time-points: 1990–1994 and 2015–2019. Non-linear trends may therefore be obscured. For example, while the District of Columbia showed overall large reductions in firearm homicide rates from 1990 to 2019, it experienced upticks in rates since 2010 (eFigs. 2–4), which might reflect more recent changes to firearm policies. Second, we only include data up until 2019 to avoid confounding from the COVID-19 pandemic. Third, defining and identifying state outliers is inevitably subject to arbitrary criteria. We aimed to guard against this limitation by triangulating across estimation approaches to identify states that consistently emerged as showing unexpected changes in firearm homicides. Fourth, our inductive approach required a detailed examination of only a handful of states (top and bottom three) to move away from the aggregate and shed finer-grained insights on the whole context of changing firearm policies in outlier states. Thus, we are unable to generalise findings to all US states.

4.1. Public health implications

New York and District of Columbia have effectively reduced firearm homicide rates in the last 30 years. On the other hand, Delaware and Missouri have seen a concerning rise and now account for a disproportionate burden of firearm homicides. State firearm policies have significantly changed during this period yet the relationship between these policy changes and firearm homicide is complex. We find no clear one-to-one relationship between enacting more restrictive firearm policies and reducing firearm homicides. Rather, it may be accumulation of more restrictive firearm policies which target a range of firearm-related behaviours, such as ownership, sales, carriage, and use. Policies that affect the whole population, such as purchasing licenses and firearm sales restrictions, may be more effective in tackling the overall burden of firearm homicides. This study identifies new lines for future research, which may be critical to understanding the role of gun control in preventing firearm deaths.

Role of the funding source

None.

Funding

This work was funded by the Laura and John Arnold Foundation and overseen by RAND Corporation via the National Collaborative on Gun Violence Research (NCGVR). The Arnold Foundation and RAND Corporation had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. The views expressed in this publication are those of the authors and not necessarily those of the Arnold Foundation or the RAND Corporation.

Author statement

Michelle Degli Esposti (MDE): Conceptualization, Methodology, Software, Validation, Formal analysis, Data Curation, Writing - Original Draft, Visualization, Project administration.

Jason Goldstick (JG): Methodology, Writing - Review & Editing, Supervision.

Jason Gravel (JG): Software, Resources, Data Curation, Writing - Review & Editing.

Elinore J Kaufman (EJK): Funding acquisition, Writing - Review & Editing.

M Kit Delgado (MKD): Funding acquisition, Writing - Review & Editing.

Therese S Richmond (TSR): Funding acquisition, Writing - Review & Editing.

Douglas J Wiebe (DJW): Conceptualization, Funding acquisition, Writing - Review & Editing Project administration, Supervision.

Access to data and data analysis

MDE and JG had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Declaration of competing interest

None.

Appendix A. Supplementary data

The following is the Supplementary data to this article.

Data availability

The authors do not have permission to share data.