Article Text

Patterns of gun owner beliefs about firearm risk in relation to firearm storage: a latent class analysis using the 2019 National Firearms Survey
Free
1. Carmel Salhi1,
2. Deborah Azrael2,
3. Matthew Miller1
1. 1 Department of Health Sciences, Northeastern University, Boston, Massachusetts, USA
2. 2 Health Policy and Management, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
1. Correspondence to Dr Carmel Salhi, Northeastern University, Boston, Massachusetts, USA; c.salhi{at}northeastern.edu

## Abstract

Background Research on gun owners’ risk-related beliefs has focused on how gun owners answer discrete questions about firearm risk. The current study is the first to use latent class analysis (LCA) to: (A) identify groups of gun owners who share patterns of beliefs about firearm-related risk and (B) examine whether class membership predicts household firearm storage.

Methods We conducted LCA using the 2019 National Firearms Survey, a nationally representative survey of US adult gun-owners (n=2950). The LCA assigned gun owners to classes based on responses to four questions about firearm-related risk. Identified classes were included in logistic regression models predicting firearm storage, along with characteristics linked to storage in past research.

Results Three classes emerged: (1) owners who believe that guns unconditionally make the home safer and should generally be readily accessible (47%); (2) owners who believe that whether guns make homes safer or less safe depends on context (34%); (3) owners who believe that guns do not pose a risk if stored safely (19%). In adjusted models, compared with owners in class 1, those in classes 2 and 3 were less likely to store guns loaded and unlocked (class 2: OR 0.30, 95% CI 0.23 to 0.39; class 3: OR 0.23, 95% CI 0.16 to 0.32).

Conclusion Our LCA is a first step towards better understanding variation in patterns of beliefs among gun owners regarding the risks and benefits of firearms. Our results suggest that messaging aimed at promoting safer firearms storage might benefit from the empirically derived typologies we identify.

• firearm
• suicide/self?harm
• epidemiology

## Data availability statement

No data are available. Data for this study are from the 2019 National Firearm Survey.

## Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

## Introduction

Members of households that contain firearms are several fold more likely to die by suicide and unintentional firearm injury, compared with people who live in homes without firearms.1–8 For youth living in homes with firearms, suicide and unintentional firearm injury risk appear to depend on how firearms are stored (locked is safer than unlocked; unloaded safer than loaded; stored separately from ammunition safer than together with ammunition).9 A recent Monte Carlo simulation study estimated that if 20% of households storing at least one firearm unlocked moved to locking all guns (eg, from approximately 50% of households locking all firearms to 60%), approximately 100 youth firearm fatalities would be prevented in a single year (mostly suicides), as would approximately 300 non-fatal firearm injuries (mostly unintentional injuries).10

Despite the substantially elevated risk of suicide and unintentional firearm death that household firearms confer on all people living in the home, fewer than 10% of gun owners in a recent national survey acknowledged any such suicide risk, let alone the potential mitigating effect that safe storage might have on suicide.11 It is unclear to what extent gun owners’ beliefs about the causal relationship between easy access to firearms and suicide mortality informs their decisions about how to store firearms. Indeed, although several studies have examined factors that predict whether household firearms are stored unlocked, loaded (or both), these studies have generally focused on sociodemographic characteristics (eg, sex, age and presence of children in the home), reasons for owning firearms (protection vs hunting vs sport shooting), type and number of firearms owned and history of formal firearm training, rather than on the role of risk perception.12–18

To our knowledge, the only nationally representative study to examine whether perceptions about firearm risk are associated with how gun owners store their firearms was based on a single-item question in our prior survey, the 2015 National Firearms Survey, that asked whether a gun makes the home ‘safer’, ‘more dangerous’ or ‘it depends’.16 In that study, gun owners who asserted categorically that firearms in the home make homes safer were, as a group, far more likely to store guns loaded and unlocked.

The current study is the first of which we are aware to use latent class analytic methods to identify groups of gun owners who share common patterns of beliefs about the risks imposed and benefits conferred by household firearms. We assess to what extent these empirically derived classes predict how a nationally representative sample of gun owners store household firearms, above and beyond previously identified factors associated with storing firearms loaded and unlocked.

## Methods

### Study design and sample

We used data from a web-based survey designed by the authors and conducted by the survey research firm Ipsos to assess firearm ownership, risk beliefs and firearm storage among a nationally representative sample of US gun owners. Respondents were drawn from Ipsos’ KnowledgePanel (KP), a group of ∼55 000 US adults selected on an ongoing basis with an equal probability of selection. All panel members, except those on active duty in the US Armed Forces, were eligible to participate.

Panel members’ report of whether they live in a home with firearms is collected on enrolment in KP and then updated annually by Ipsos, allowing us to restrict invitations to participate to adults who had previously reported that they lived in a home with firearms. Email invitations to participate in the survey contained a link that sent them to the survey questionnaire. No description of the survey content accompanied the invitation. After 3 days, automatic email reminders were sent to all non-responding panel members in the sample. Of the 6721 panel members invited to complete the survey, 4379 started and 4030 completed the survey (response rate 65.2%, participation rate 92.0%), of whom 2950 personally owned firearms. For selection of general population samples from KP, Ipsos uses an equal probability selection method design by weighting the entire KP to benchmarks from the latest March supplement of the US Census Current Population Survey. Sampling weights supplied by Ipsos were applied such that estimates made from the survey are representative of US gun owners (≥18 years). Additional details are available in online supplementary appendix 1.

### Supplemental material

Participants were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

### Primary statistical analyses

Using latent class analysis (LCA), we identified mutually exclusive subgroups (‘classes’) of gun owners based on their risk beliefs. We considered parsimony, interpretability and statistical fit as indicated by the Bayesian Information Criterion (BIC) in choosing our preferred class solution.19 20Posterior probabilities of class membership assigned respondents to the class into which they had the highest probability of belonging. Once we arrived at a final class solution, we used logistic regression to examine whether membership in a particular class was associated with storing at least one household firearm loaded and unlocked, adjusting for demographic and ownership characteristics that have been linked to storage practices (eg, owning handguns and living in the South) in prior work.

All descriptive statistics, LCA and logistic regression models are weighted to be nationally representative of US gun owners. LCA was conducted in MPLUS V.7.21 Descriptive statistics and regression analyses were conducted in Stata V.14.2.

### LCA measures

Risk belief items: we assessed beliefs about risk–benefit tradeoffs related to firearms in the home using the four question sets related to risk perception included in the survey: (1) ‘Do you think having a gun in the house makes it a safer place to be or a more dangerous place to be?’ (safer/more dangerous). This question was followed, depending on the respondent’s answer, by ‘are there circumstances in which you could imagine having a firearm might make the house more dangerous/might make it safer?’. If respondents replied ‘yes’, their response was categorised as ‘it depends’ resulting in three response categories, yes/no/it depends (the variable used in our analyses); (2) ‘Having a gun in the home increases the risk of suicide’; (3) ‘If a gun owner has to take the time to unlock or load their gun, it’s no good for protection’; (4) ‘Guns should be stored locked and unloaded with the ammunition stored separately’. Questions 2–4 had five-point responses ranging from ‘strongly disagree’ to ‘neither agree or disagree’ to ‘strongly agree’. For parsimony and ease of interpretation, we combined responses of ‘strongly disagree’ with ‘disagree’ and ‘strongly agree’ with ‘agree’.

We conducted sensitivity analyses as a check on our decision to collapse the range of response categories. Specifically, we ran LCAs that: (1) fully disaggregated response categories (ie, separated ‘strongly agree’ from ‘somewhat agree’ and ‘strongly disagree’ from ‘somewhat disagree’); and (2) split the ‘it depends’ response into two categories based on the initial stem question response (ie, we distinguished between response patterns that started with ‘guns make the home safer’ but were followed by agreement that there may be situations where a gun makes the home more dangerous from (the few) response patterns that started with ‘guns make the home more dangerous’ but were followed by agreement that there may be situations in which a gun makes the home safer).

### Logistic regression measures

#### Personal gun ownership characteristics

Ownership of guns in a household was assessed by asking respondents whether they personally own a gun. Those identified as personal gun owners were then asked follow-up questions about the nature of that ownership, including: number of guns owned, reasons for ownership (eg, sport, hobby or personal protection), types of guns owned (ie, long gun and/or handgun) and having ever received firearm safety training.

#### Storage practices

After having ascertained how many handguns and how many long guns a respondent owned, respondents were asked separately about the number of handguns and then about the number of long guns stored: (1) loaded and unlocked, (2) unloaded and locked, (3) unloaded and unlocked and (4) loaded and locked. Locking was defined as the use of a ‘trigger lock, cable lock, in a lock box or in another locked container’. Presence of any loaded and unlocked firearm was used as the main outcome of interest in logistic regressions.

#### Other characteristics predictive of firearm storage

In logistic regression models with firearm storage (any gun loaded and unlocked) as the dependent variable and latent class membership as the primary independent variable, we adjusted for factors previously identified as predicting storage practices: living in a metropolitan area, gender, race and ethnicity, categorical age (ie, 18–24, 25–44, 45–64 and over 64 years of age), having a child <18 years in the home, veteran status, categorical income (ie, $0–$39 999, $40 000–$59 999, $60 000–$100 000 and over \$100 000), and region of USA (ie, West, Midwest, South and Northeast).12 13 22–24 We also included gun ownership variables that have been previously associated with storage practices: owning a gun for protection, owning a handgun and having prior firearm training.25 26

## Results

Table 1 presents descriptive univariate statistics for demographic characteristics and firearm-related risk belief items in our sample of 2950 gun owners. For individual risk belief items, most gun owners believed that whether guns made the home safer or not depended on situational factors (56%), that when guns were locked and unloaded they were not useful for self-protection (53%) and that guns do not increase the risk of suicide (62%). As many gun owners actively agreed (38%) as actively disagreed (37%) with the statement that firearms should be stored locked and separate from ammunition when not in use.

Table 1

Descriptive statistics of demographic and gun-related characteristics

Most gun owners were over 45 years of age (64%), white (78%), lived in metropolitan areas (80%) and had no children in the home under 18 years of age (69%).

Table 2 presents univariate statistics for the item–response probabilities and latent class membership probabilities of the final class model. The item–response probabilities express the probability that a particular response (eg, agree, disagree or neither) to a survey item occurs within each latent class. For gun owners’ beliefs about risk and safety of firearms in the home, we selected a three-class model balancing parsimony, interpretability and BIC (1-class BIC 22 080; 2-class BIC 21 320; 3-class BIC 21 196; 4-class BIC 21 147; 5-class BIC 21 164; 6-class BIC 21 198). Sensitivity analyses that accounted for the full range of intervals in the relevant risk belief items did not materially affect results. Specifically, analyses that: (A) distinguished ‘strongly agree’ from ‘agree’ and ‘strongly disagree’ from ‘disagree’ and (B) disaggregated the ‘it depends’ response into two subgroups based on the gun owners’ initial response to the dichotomous (‘safer’ vs ‘more dangerous’) stem question did not result in class structures or fit statistics that were materially different from those presented in primary analyses (not shown).

Table 2

Item–response probabilities and labels for latent class analysis

Conceptual labels for the classes are derived from the distribution of item–response probabilities. Higher item–response probabilities are considered defining characteristics of latent class membership. Latent class membership probabilities reflect how common each profile of firearm safety beliefs is among adult US gun owners. We labelled latent class 1 ‘Guns are Safe’, as gun owners largely agreed that guns make the home safer, do not increase risk of suicide, should not be locked or unloaded for self-protection and should not be stored locked and unloaded when not in use. We labelled latent class 2 ‘Guns can be Safe and Useful’, as members of this class tend to answer ‘It Depends’ or ‘Neither Agree nor Disagree’ to whether guns make the home safer, and their responses are generally equally distributed between agree, disagree and neither about whether guns increase the risk of suicide or that locking or unloading a firearm decreases its utility for self-protection. We labelled latent class 3 ‘Safe if Responsible’, as members of this class did not disagree that firearms should be stored locked and separate from ammunition when not in use and were unlikely to agree that locking or unloading affects the utility for self-protection.

Overall, each of the four risk belief items included in the LCA helped discriminate between at least two of the three classes. Members of the ‘Guns are Safe’ and ‘Safe if Responsible’ classes generally agreed that a gun makes the home safer and does not increase the risk of suicide. However, in contrast to the ‘Safe if Responsible’ class, ‘Guns are Safe’ members were more likely to believe that ready access to firearms was important. Members of the ‘Guns can be Safe and Useful’ class resembled the ‘Safe if Responsible’ class in questions about access to firearms. However, ‘Guns can be Safe and Useful’ members were far more likely to acknowledge suicide risk.

table 3 provides nationally representative class membership probabilities alongside descriptive statistics for demographic and gun-related characteristics by latent class. Overall, the majority of gun owners belong to the ‘Guns are Safe’ class (46.9%), while one-third belong to the ‘Guns can be Safe and Useful’ class (33.9%). The ‘Safe if Responsible’ class of gun safety beliefs represents the smallest proportion of gun owners (19.1%). Although the classes were broadly similar with regard to demographic characteristics, the ‘Guns are Safe’ class consisted of gun owners who were modestly more likely to be older, live in homes with no children, live in urban areas, and reside in the South or the West regions of the US.

Table 3

Class membership probabilities for gun owners and demographic characteristics by latent class

Table 4 describes firearm practices by latent class membership. The ‘Guns are Safe’ class was distinct in their ownership patterns, as most owned both handguns and long guns (60%), owned slightly over 2.5 more guns on average than the other two classes and were least likely to own only long guns. Furthermore, those in the ‘Guns are Safe’ latent class were by far the most likely to have a concealed carry permit (48%) and store at least one gun loaded and unlocked (50%). The ‘Guns can be Safe and Useful’ class was least likely to own a firearm for protection (56%) compared with ‘Guns are Safe’ (86%) and ‘Safe if Responsible’ (70%). Members of the ‘Guns are Safe’ class were almost three times as likely to store firearms loaded and unlocked (50%). However, similar and substantial numbers of each class stored at least one of their guns loaded and locked or unloaded and unlocked (‘Guns are Safe’: 39%; ‘Guns can be Safe and Useful’: 45%; ‘Safe if Responsible’: 44%).

Table 4

Firearm practices by latent class membership (weighted column %)

Table 5 shows the results of multivariate adjusted logistic regression predicting gun owners storing at least one household firearm loaded and unlocked. Latent class membership was as strongly associated with unsafe storage as was any other model characteristic (‘Guns can be Safe and Useful’: OR 0.30, 95% CI 0.23 to 0.39; ‘Safe if Responsible’: OR 0.23, 95% CI 0.16 to 0.32; ‘Guns are Safe’ referent). Owning any gun for protection (OR 3.35, 95% CI 2.34 to 4.79) and handgun ownership (OR 3.83, 95% CI 2.36 to 6.22) were strongly associated with loaded and unlocked storage and had similar magnitudes of association, in the opposite direction, to membership in our latent classes.

Table 5

The association between latent class membership and storage practices, adjusted odds ratios and 95% confidence intervals

Having a child under 18 years of age in the home was also associated with lower odds of the presence of an unlocked and loaded firearm (OR 0.45, 95% CI 0.31 t o0.67). Residing in a metro area was associated with lower odds of unlocked and loaded storage (OR 0.68, 95% CI 0.5 to 0.91), while residing in the South was associated with higher odds (OR 2.00, 95% CI 1.33 to 3.01). Gender (OR 1.22, 95% CI 0.95 to 1.58) and veteran status (OR 1.04, 95% CI 0.79 to 1.37) were neither strong nor significant predictors in this model.

## Discussion

Research on gun owners’ risk-related beliefs has focused on how gun owners answer discrete questions about firearm risk. The current study is the first to use LCA to identify groups of gun owners who share patterns of beliefs about firearm-related risk and to examine whether class membership predicts household firearm storage. Members of the largest of the three groups that emerged, the ‘Guns are Safe’ class, representing 47% of gun owners, tend to endorse statements that assert that firearms should be readily accessible in most circumstances. The next largest class, the ‘Guns can be Safe and Useful’ class, comprising approximately one-third of gun owners, is characterised by the importance of circumstance in determining perceptions of firearm safety. An additional 20% of gun owners fall into the ‘Safe if Responsible’ class, who, as a group, agree that when guns are not in use the safest form of storage is best, though like all gun owners, they largely disagree with the statement that guns increase the risk of suicide.

While the majority of owners in all classes owned for protection and more than two-thirds owned handguns, over 80% of owners in the ‘Guns can be Safe and Useful’ and ‘Safe if Responsible’ classes took some storage precaution by locking or unloading their firearms. Indeed, these classes were associated with safer storage at least as strongly as were well-established factors in the firearms literature (eg, handgun ownership, owning for protection and living in homes with children).12 13 Future studies should test whether safety messages that target the emergent classes separately prove more effective in promoting desired safety practices, such as locking and unloading firearms, than does a generic message delivered broadly to gun owners or to gun owner subgroups defined by demographics, characteristics of firearms possessed or single item–responses to questions about risk perception.

Our findings should be interpreted with several considerations in mind. First, the four risk belief items used in our LCA come from a survey that was designed primarily to elicit information about firearm acquisition, ownership and storage, and secondarily to explore risk perceptions related to suicide and self-defence. As such, the limited number and scope of questions related to firearm risk may have constrained the number of classes we found, as well as our ability to describe in more nuanced ways the classes identified. Second, as with all self-report surveys, respondents’ answers to some questions may have been subject to social desirability bias, which could have resulted in under-reporting of unsafe firearm storage practices. However, online panel surveys have been shown to reduce social desirability bias and yield more accurate estimates of respondent characteristics than telephone surveys.27 Third, differences between those who chose to complete the survey and those who did not may have affected the generalisability of our findings and the latent classes we identified.

Despite these limitations, the LCA we present is a first step towards better understanding variation in systems of beliefs about firearms held by US adult gun owners. Because the emergent classes are strongly associated with how household firearms are stored, our findings suggest the possibility that safety messages might be more effective if crafted and tested with our distinct typologies in mind.

### What is already known on this subject

• Members of households that contain firearms are several- fold more likely to die by suicide and unintentional firearm injury, compared with people who live in homes without firearms.

• Despite the substantially elevated risk of suicide and unintentional firearm death that household firearms confer on all people living in the home, few gun owners acknowledge any such suicide risk, let alone the potential mitigating effect that safe storage might have on suicide.

• Using latent class analysis, this study is the first to identify nationally representative groups of gun owners who share common patterns of beliefs about the risks imposed and benefits conferred by household firearms.

• We show how class membership predicts how gun owners store household firearms above and beyond previously identified factors associated with storing firearms loaded and unlocked.

## Data availability statement

No data are available. Data for this study are from the 2019 National Firearm Survey.

## Ethics statements

### Ethics approval

The Institutional Review Board of Northeastern University approved the study.

## Footnotes

• Funding This study was supported by the Joyce Foundation (grant #18–38517) and the National Collaborative on Gun Violence Research.

• Competing interests None declared.

• Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

• Provenance and peer review Not commissioned; externally peer reviewed.