In Guns We Trust: A Reexamination of the Collective Security Hypothesis

Introduction

Scholars have long attempted to understand the factors that motivate individuals to purchase firearms (Cao, Cullen, & Link, 1997). Among contemporary surveys, self-defense and home defense remain the most commonly reported reasons (General Social Survey [GSS], 2015; Swift, 2016). One of the more heavily investigated explanations for defensive gun ownership, the collective security explanation, centers on citizens’ perceptions of government’s ability to ensure their safety. Specifically, this perspective suggests that individuals who perceive law enforcement as ineffective at controlling crime, or who have fear of experiencing victimization, are more likely to arm themselves to deter personal victimization (McDowall & Loftin, 1983).

Research exploring this hypothesis has provided inconsistent support—some studies find police effectiveness influences individual and aggregate levels of firearm ownership (Kleck & Kovandzic, 2009; Lizotte & Bordua, 1980; Smith & Uchida, 1988), while others do not (Whitehead & Langworthy, 1989). Even less of an empirical consensus exists over how victimization or perceived risk of victimization influences gun ownership (for review, see Kleck, Kovandzic, Saber, & Hauser, 2011). The inconsistent findings are likely due to methodological differences between studies, including the measurement of gun ownership (Kleck, 2015). Furthermore, the literature suffers from several other methodological shortcomings including the inability of researchers to correctly model causal order between gun ownership and fear of crime, the failure to include relevant covariates such as both objective and subjective measures of neighborhood crime, and narrow measures of perceptions of police that may not fully encompass individual attitudes toward agents of formal social control (Kleck, 2015; Kleck et al., 2011).

Although we see clear relevance for understanding the collective security hypothesis, few studies have explored how community characteristics (e.g., poverty, residential mobility) influence gun ownership. Economic disadvantage, transiency, and racial composition are at least marginally associated with neighborhood crime, while informal social control can mediate these relationships (e.g., Bursik & Grasmick, 1993; Sampson, Raudenbush, & Earls, 1997; Shaw & McKay, 1942). We suspect informal social control may also mediate the relationship between perceptions of police, as well as neighborhood conditions, and gun ownership. If individuals feel that the state or local police fail to ensure residents’ safety, but residents do so collectively on their own, they may refrain from gun ownership for defensive purposes.

Using data from a survey of Detroit residents, the current study seeks to determine whether gun ownership is more prevalent among citizens who report greater levels of dissatisfaction with police. Given that aggregate-level influences (e.g., crime rates) also impact gun ownership (Kleck & Kovandzic, 2009), we include a host of important neighborhood-level factors to better understand the relationship between perceptions of police effectiveness, neighborhood conditions, and gun ownership. This research improves upon previous studies by incorporating a direct measure of defensive gun ownership, a more comprehensive measure of satisfaction with police performance, and indicators of neighborhood conditions.

Gun Ownership

Gun ownership research is largely organized around motivating factors for the purchase of firearms and individual-level characteristics associated with ownership (Cao et al., 1997). To date, a number of empirical evaluations have been conducted to determine who owns guns in the U.S. and why (Cao et al., 1997; Dixon & Lizotte, 1987; Lizotte & Bordua, 1980; Smith & Uchida, 1988). Although scholars have examined issues related to firearms in earlier decades, a congressional moratorium on firearm research and data collection using funds from the Centers for Disease Control and Prevention and the National Institutes of Health has slowed the advancement of knowledge in this area more recently (Kellermann & Rivara, 2013). As a result, few recent studies have assessed whether motives for owning guns have changed over time.

A typology of gun ownership offered by Cao and colleagues (1997) suggests key factors likely to influence gun ownership. They argue wealthier individuals are more likely to own guns simply because they can afford them, and empirical evidence supports this relationship (Dixon & Lizotte, 1987; Smith & Uchida, 1988). Others have suggested education influences gun ownership, arguing less-educated individuals are more likely to own a firearm; less support is found for this hypothesis (Dixon & Lizotte, 1987). Other research has assessed whether a gun culture exists, or whether individuals are socialized toward pro-gun values through self-arming for protection or through employment that requires use of firearms (e.g., military, police, and security). There is some evidence to suggest gun cultures exist, with research demonstrating concentrations of gun owners in rural areas and the Southern states (Dixon & Lizotte, 1987; Marciniak & Loftin, 1991). Moreover, males are far more likely than females to report owning a gun, suggesting gender-specific socialization for gun ownership (Dixon & Lizotte, 1987).

Finally, some scholars suggest perceptions of failure of the criminal justice system to effectively control crime and disorder can increased gun ownership (McDowall & Loftin, 1983). According to the collective security hypothesis, individuals of this mind-set arm themselves to prevent crime and victimization because they have little faith in the police and courts. Moreover, victimization, including perceived risk, can influence decisions to obtain a firearm. Both of these mechanisms—perceptions of law enforcement and victimization—are hypothesized to influence the likelihood an individual will purchase a firearm for defensive purposes.

In a similar vein, individuals may rely on firearms for personal safety because they hold cynical views toward law enforcement or perceive the police as unjust. Specifically, research suggests members of disadvantaged social groups are more likely to express cynicism and perceive injustice in the application of laws (Gau & Brunson, 2010; Sampson & Bartusch, 1998). Urban minority communities are often characterized as holding cynical views toward law enforcement because they believe police do not care about them or because they, or someone they know, have experienced disrespect or hostility from police (Anderson, 2000; Carr, Napolitano, & Keating, 2007; Gau & Brunson, 2010; Kubrin & Weitzer, 2003). Scholars find police officers use excessive force more frequently in disadvantaged minority communities (Terrill & Reisig, 2003), and aggressive order maintenance tactics erode trust in the police in such communities (Brunson, 2007; Gau & Brunson, 2010). In particular, procedurally unjust police practices and structural conditions of disadvantaged neighborhoods are associated with the cultural adaptation of legal cynicism (Anderson, 2000; Kirk & Matsuda, 2011; Kirk & Papachristos, 2011; Sampson & Bartusch, 1998). Kirk and Papachristos (2011) argue that individuals are more likely to personally remedy threats and disputes when they feel police are unresponsive and legal cynicism is high. As such, gun ownership, a means of addressing potential threats, may be higher in areas where legal cynicism is greater and residents hold police in contempt. Retaliatory homicide and violent crime are often elevated in areas with more pervasive legal cynicism and perceptions that police are not procedurally just because residents feel that they cannot rely on police to solve problems (Kane, 2005; Kubrin & Weitzer, 2003).

The research assessing how perceptions of law enforcement influence gun ownership often shows that those who perceive police as ineffective in their ability to provide protection to the public are more likely to own guns (Kleck & Kovandzic, 2009; Lizotte & Bordua, 1980; Smith & Uchida, 1988). Similarly, both McDowall and Loftin (1983) and Kleck and Kovandzic (2009) show inverse relationships between police strength and rates of gun ownership.

No clear consensus has been reached about how actual or perceived risk of victimization influences the likelihood of purchasing a firearm for self-defense. Kleck and colleagues’ (2011) summary of the literature on fear, victimization, and perceived risk and their relationship with gun ownership demonstrates inconsistencies across studies in terms of measurement of constructs and inclusion of variables central to the theoretical framework of this perspective. Of the 17 studies discussed, only two included crime rates, only five measured perceived risk of victimization, and more than half were unable to determine whether the respondent owned a gun for defensive purposes. However, most of the more rigorous studies reported by Kleck and colleagues (2011) show a positive association between gun ownership and fear of crime or perceived risk of victimization (Cao et al., 1997; Kleck et al., 2011; Lizotte & Bordua, 1980; Smith & Uchida, 1988), with a few exceptions (Sheley, Brody, Wright, & Williams, 1994; Williams & McGrath, 1976). Similarly, most studies do not support the expected positive relationship between previous victimization experiences and gun ownership (Cao et al., 1997; Glaeser & Glendon, 1998; Kleck & Kovandzic, 2009; Williams & McGrath, 1976), although inconsistencies exist here too (Lizotte & Bordua, 1980; Whitehead & Langworthy, 1989).

As for crime rates (i.e., homicide, burglary, violent crime) and their influence on gun ownership, some studies show that crime rates indirectly influence individual gun ownership through fear of crime (Lizotte & Bordua, 1980). Macro-level research tends to find positive associations between rates of crime and gun ownership (Bice & Hemley, 2002; Kleck, 1979; McDowall & Loftin, 1983). Methodologically sound assessments of this relationship do not support the hypothesis that more guns lead to more crime, but rather the opposite (Kleck, 2015).

In all, literature testing the collective security explanation of gun ownership is somewhat limited. Many of the studies that have tested the propositions of the collective security explanation are ill-suited because of how gun ownership is measured (Kleck et al., 2011). Specifically, some studies are unable to differentiate between respondents who own guns for hunting or sport versus those who own them because of home and personal safety concerns. Second, some prohibit the determination of whether the respondent personally owns the gun instead of another household member. Assuming the respondent owns the gun can lead to inaccurate inferences about how fear, victimization, and perceptions of police influence gun ownership. Finally, many surveys are unable to determine the type of gun owned. Individuals primarily purchase handguns, not rifles, for self and home protection; thus, it is imperative studies examining the collective security explanation include measures of handgun ownership or ask about motive for acquiring a firearm. Unfortunately, most studies fail to satisfy these criteria.

In addition, variation across studies in the measurement of key constructs outlined in this perspective—including perceptions of police, gun ownership, and fear of crime—has produced inconsistent empirical findings. Studies of the collective security hypothesis have employed a variety of measures to capture the ability, or perceived ability, of agents of formal social control to ensure public safety and control crime. For instance, single-item measures asking respondents to rate the quality of police services in their neighborhood on Likert-type scales ranging from very poor to outstanding have been used (Smith & Uchida, 1988). Although this measure is clearly relevant for understanding how perceptions of police influence gun ownership, its narrow focus may not capture the dynamic nature of perceptions of police services or performance. For example, community relations with police may be rated as excellent by residents, while their ability to control crime on neighborhood streets may be viewed as less than satisfactory.

The Current Study

The current study seeks to expand upon previous research examining the collective security hypothesis in a number of important ways. First, we use a dependent variable that captures whether the individual respondent owns the gun in question and specifies whether the gun is used for home-defense. ¹ Second, the study includes a broader measure of satisfaction with police that captures not only respondent assessments of overall satisfaction with police but also their satisfaction with police services in their neighborhood, and the ability of police to control crime. Third, this study adds to the limited empirical literature examining the influence of macro-level neighborhood characteristics on gun ownership at the individual level (Cao et al., 1997).

Several hypotheses are tested in this study. First, because previous research on the collective security hypothesis finds police satisfaction is inversely related to gun ownership, we hypothesize respondents who have more negative views of police performance will be more likely to own a firearm for defensive purposes. Second, we hypothesize the relationship between satisfaction with police and gun ownership should be at least partially mediated by neighborhood informal social control. Scholars have shown concentrated disadvantage, racial/ethnic heterogeneity, and residential mobility are associated with elevated crime and disorder in neighborhoods (e.g., Sampson & Groves, 1989; Sampson et al., 1997; Shaw & McKay, 1942). The relationship between these measures of structural disadvantage and crime has often proven to be subject to at least partial mediation by informal social control, or the perceived ability of residents to collectively exercise social control over their neighborhood (Sampson et al., 1997). These same neighborhood factors associated with crime may also influence gun ownership. For example, the recognition that many residents in one’s neighborhood appear to be transient (i.e., greater residential mobility) may encourage the belief that neighbors have little stake in ensuring a crime-free neighborhood or intervening on behalf of other residents, thus prompting the purchase of a firearm for home defense. Greater levels of racial/ethnic heterogeneity within neighborhoods might foster mistrust between residents, leading fearful neighbors to purchase firearms for personal safety. If the relationship between neighborhood conditions and crime can be limited by informal social control, it may also insulate against motivations to purchase firearms. Specifically, residents who feel they can rely on their neighbors to intervene against minor forms of deviance may view gun ownership as unnecessary to ensure safety. Finally, as there is little consensus about whether victimization influences gun ownership, we take an exploratory approach to understanding this relationship in lieu of an additional hypothesis.

Method

Procedure

Data for the current study came from three sources: the Detroit Social Control Survey (DSCS), the U.S. Census Bureau, and official crime incident data reported to the Detroit Police Department (DPD). These data were merged into a single file and matched by census tract. There were 561 tracts in Detroit containing population and housing characteristics of residents in the area. On average, tracts contained 3,285.42 (SD = 1,308.77) people during the study period.

The DSCS served as the basis for the data set in this study, as it collected individual-level measures from residents. The purpose of the DSCS was to examine factors influencing citizens’ formal and informal responses to crime and disorder, perceptions of safety, and general satisfaction with police. Data were collected in October and November of 2009 using a three-stage cluster sampling procedure. First, all neighborhoods in Detroit were identified using the 2000 Census boundaries and census data were gathered on demographic and economic conditions of each neighborhood. Block group-level indicators of sociodemographic and structural factors used for stratification included the following proportions of the population: below the poverty line; Black residents; residents unemployed; female-headed households with children; not living in the same residence since 1995; and renter occupied households. The second stage of clustering involved selecting contiguous clusters of these stratified census block groups. The neighborhood clusters were classified into low, medium, and highly disadvantaged areas, and one area was selected from each (i.e., low, medium, and high). Finally, between 10 and 20 households within each block group in the selected neighborhood clusters were surveyed. Pairs of researchers conducted door-to-door, in-person interviews of every house that looked inhabited and allowed access to the front entrance. The rejection rate from residences was approximately 2:1. All participants were required to be at least 18 years old. The final sample size was 408 residents living in 10 census tracts (n = 21 census block groups).

Table 1 presents descriptive statistics for the sample, which are representative of the city of Detroit as a whole (see Papp, Smith, Wareham, & Wu, 2017). The sample was split nearly evenly with slightly more females (54.1%) than males (45.9%). The average respondent was 44.1 years old. The majority of the sample was non-White (88.7%), unmarried (64.9%), and did not have children living in the home (54.4%). Just more than one third of the sample (37.5%) owned guns for the purpose of home defense. This is substantively similar to past research that has examined the prevalence of gun ownership by household (GSS, 2015). ²

OPEN IN VIEWER

Table 1:

Descriptive Statistics (N = 380)

Variable	M	Minimum	Maximum	SD
Individual-level characteristics
Gun ownership	0.38	0	1	—
Male	0.46	0	1	—
Age	44.13	18	88	17.10
Non-White	0.89	0	1	—
Married	0.35	0	1	—
Children in house	0.46	0	1	—
Education	3.08	1	7	1.48
Income	5.70	1	12	4.26
Renter	0.40	0	1	—
Community safety factor	6.17	3	12	2.50
Police satisfaction factor	8.07	3	12	2.28
Victim	0.35	0	1	—
Neighborhood-level characteristics
Legal cynicism index	14.62	13.65	15.78	0.88
Informal social control index	22.97	20.83	25.19	1.47
Concentrated disadvantage factor	0.69	−0.63	2.24	1.04
Immigrant factor	−0.38	−0.81	0.44	0.31
Residential mobility factor	0.02	−1.11	2.16	1.00
Crime rate factor	0.93	0.00	1.59	0.60

Note. There were 10 Census tract clusters in the data with the number of participants in these neighborhoods in the following descending order: 85, 61, 55, 42, 41, 23, 23, 22, 21, and 19.

Additional measures from the U.S. Census and official crime incident data reported by DPD were linked to residents in the DSCS. Census data were compiled at the census tract level to measure aggregate levels of concentrated disadvantage, residential stability, and immigration. Finally, official records held by DPD were collected to measure census tract crime rates. These measures are discussed further below.

Missing Data

All individual items used in this study had low levels of missing data (between 2% and 8%). However, when these items were combined for composite measures of informal social control, legal cynicism, and police satisfaction, this led to a larger amount of missing data, approximately 20%. Consequently, multiple imputation was used to determine how the missing data may impact the results. Prior to imputation, all individuals missing more than half of the items used to create any scale were removed from the sample. Imputed values were not used for any of the demographic variables or the measure of victimization. Because imputation was not used to impute every missing value, there still remained a small amount (6%) of missing data after imputation, resulting in a final sample of 380 usable cases.

Multivariate linear regression imputation was conducted in Stata 14 to impute missing values. This method uses a Markov Chain Monte Carlo Method procedure to impute missing values. All variables found in Table 2 were used as predictors of missing values in the imputation. Fifty iterations of the imputation were conducted. The results presented in Table 2 reflect the pooled results of the 50 imputation iterations. It is important to note that the same analyses associated with our results (Table 2) were also conducted on the data before imputation to determine the sensitivity of the results to imputation. The results were substantively similar and so the results including the imputed data are reported.

OPEN IN VIEWER

Table 2:

Binary Logistic Regression Models Predicting Gun Ownership (N = 380)

	Reduced model			Full model
Variable	OR	β	SE	OR	β	SE
Individual-level characteristics
Male	2.31***	0.84	0.24	2.37***	0.86	0.24
Age	0.99	−0.01	0.01	0.99	−0.01	0.01
Non-White	6.43***	1.86	0.50	7.90***	2.07	0.51
Married	2.54**	0.93	0.27	2.62**	0.96	0.28
Children in house	0.62	−0.48	0.27	0.64	−0.45	0.27
Education	0.84	−0.18	0.10	0.82	−0.20	0.10
Income	1.04	0.04	0.04	1.03	0.03	0.04
Renter	0.45**	−0.80	0.28	0.42**	−0.86	0.29
Victim	1.21	0.19	0.25	1.19	0.18	0.25
Community safety	1.02	0.02	0.05	1.01	0.01	0.05
Police satisfaction	0.86*	−0.15	0.06	0.87**	−0.14	0.06
Neighborhood-level characteristics
Informal social control				0.98	−0.02	0.07
Legal cynicism				1.08	0.08	0.36
Concentrated disadvantage				1.19	0.17	0.34
Immigrant				1.75	0.56	0.52
Residential mobility				1.22	0.20	0.24
Crime rate				0.53	−0.63	0.36
Intercept		−2.86	0.89		−2.72	5.55

Note. OR = odds ratio.

*

p < .05. **p < .01. ***p < .001.

Results

Table 2 presents the results of the binary logistic model estimated with gun ownership as the dependent variable. ⁴ The model was conducted stepwise. In the reduced model, all of the individual-level measures were entered into the model for gun ownership. In the full model, the neighborhood-level measures were entered along with the individual-level measures. The neighborhood variables were entered into the model in this way to better understand how the inclusion of these variables influenced the individual-level variables. As can be seen from the results in the reduced model, gender, race, marital status, home ownership status, and satisfaction with the police were all predictive of gun ownership at a p < .05 level. Males had over twice the odds of gun ownership (odds ratio [OR] = 2.31), non-Whites over 6 times the odds (OR = 6.43), married individuals twice the odds (OR = 2.54), and renters half the odds (OR = 0.45), compared with their counterparts. In addition, for each unit increase on the police satisfaction scale, respondents had about 14% less odds of owning a gun. Put another way, those who were less satisfied with the police had higher odds of owning a gun. ⁵

The coefficients of the individual-level variables remained largely unchanged when neighborhood variables were entered into the model. Individual-level variables that remained significant in the full model included gender (OR = 2.37), race (OR = 7.90), marital status (OR = 2.62), home ownership status (OR = 0.42), and police satisfaction (OR = 0.87). This latter finding provides partial support for the collective security explanation of gun ownership, confirms our first hypothesis, and supports the results of previous research (Kleck & Kovandzic, 2009; McDowall & Loftin, 1983; Smith & Uchida, 1988). Given that this study is a more recent test of this perspective, it is worth noting that perceptions of police appear to remain a salient influence on self-armament across time. Alternatively, respondent victimization and perceptions of community safety are not significant predictors of gun ownership in either model. These findings are supportive of some previous research (Glaeser & Glendon, 1998; Kleck & Kovandzic, 2009; Sheley et al., 1994), but at odds with other studies (Cao et al., 1997; Kleck et al., 2011; Smith & Uchida, 1988). Moreover, this portion of the collective security explanation of gun ownership is unsupported. This particular finding, however, may be a product of the cross-sectional nature of this study as victimization could have occurred before or after the respondent purchased a gun and fear of crime can influence gun ownership and vice versa (Kleck et al., 2011).

The substantively large effect of race is also important to consider with caution due to limited variation in this variable. To be more specific, approximately 89% of the sample was non-White, leaving only 42 Whites in the comparison group. This creates concern about the stability and interpretability of this finding.

In addition, an equality of coefficients test was conducted to answer the second research question (Clogg, Petkova, & Haritou, 1995), of whether informal social control mediates the relationship between police satisfaction and gun ownership. When informal social control was entered separately into the model on its own (not shown in Table 2) after the inclusion of all the individual characteristics, or in combination with the other neighborhood characteristics, it did not significantly change the point estimate of police satisfaction. This suggests that it has no mediating effect on police satisfaction for this sample.

It is also important to note that no neighborhood variable showed a significant relationship with the dependent variable. However, the effect size for the aggregated measure of crime rate was substantively large (OR = 0.53), but nonsignificant, likely due to the small number of neighborhood clusters (n = 10 census tracts). The observed nonsignificance of each neighborhood-level characteristic suggests that respondents in this sample may not arm themselves because of disadvantaged conditions, transiency, racial composition, or beliefs about whether neighbors will intervene to prevent crime. It appears that gun ownership may be more closely tied to beliefs about the ability of agents of formal social control to prevent crime.

Finally, additional analyses were conducted to explore the potential interaction between police satisfaction and crime rate. That is, does the association between a resident’s perceived satisfaction with police and his or her willingness to own a gun for protection depend on crime rates in the neighborhood? An interaction was created between the composite crime rate factor and police satisfaction and entered into the model presented in Table 2. However, this interaction was not significant and so it is omitted for succinctness. ⁶

Discussion and Conclusion

In our reexamination of the collective security hypothesis, we find that satisfaction with police, a measure we believe takes into account the complexity of perceptions of police among the public, is a robust predictor of gun ownership. Consistent with our first hypothesis, those who maintain higher satisfaction with police, including their ability to prevent crime, are less likely to own a firearm for defensive purposes. This finding supports research that has observed similar results (e.g., Kleck & Kovandzic, 2009; Lizotte & Bordua, 1980; McDowall & Loftin, 1983; Smith & Uchida, 1988). Alternatively, the nonsignificance of the fear of crime and victimization variables across our models demonstrates no support for this part of the collective security explanation of gun ownership. Those who have previously experienced robbery, assault, or burglary are no more likely to own guns than those who have not experienced these forms of victimization. These findings are supportive of some prior research (e.g., Glaeser & Glendon, 1998; Kleck & Kovandzic, 2009; Sheley et al., 1994; Williams & McGrath, 1976), although they are at odds with the results of other studies (e.g., Cao et al., 1997; Kleck et al., 2011; Lizotte & Bordua, 1980; Smith & Uchida, 1988; Whitehead & Langworthy, 1989).

Regarding our second hypothesis, which stated informal social control or the ability of neighborhood residents to discourage deviance or intervene when criminal behaviors are observed, should partially mediate the relationship between police satisfaction and gun ownership, we find no support. In fact, none of the neighborhood characteristics measures influenced residents’ odds of gun ownership. The finding indicating that the collective ability of residents in a neighborhood to control or prevent crime and deviance had no effect on gun ownership was somewhat surprising. It appears that residents in our sample may own firearms for reasons unrelated to neighborhood structural conditions. Perhaps residents believe crime control to be primarily under the purview of law enforcement rather than neighborhood residents, which may explain the strength of the association between defensive gun ownership and formal social control. Indeed, studies have demonstrated that participation in coproduction of crime prevention among citizens is quite low and that individuals tend to prefer home security systems, guns, and dogs over collective responses such as a neighborhood watch (Rosenbaum, 1988). Research has also suggested that informal social control may be limited and age graded (Wilkinson, 2007), meaning that residents may intervene when children engage in deviant behavior, but not for teenagers or adults.

Although the findings from this study provide important and renewed insight into the predictors of gun ownership for protection, a few limitations warrant discussion. First, preliminary analysis of a random effect, multilevel model indicated no need to adjust for the pooled nature of the data. The random effect, multilevel model, however, only included 10 census tracts, and although Raudenbush and Bryk (2002) suggest 10 groups in a multilevel model can produce substantively meaningful results, the variation across the random effect is diminished. Although the survey data could be aggregated to the smaller census block group, which would have increased the number of groups available (21 instead of 10), the crime and census data were not available at this smaller aggregation level. Future research should replicate this study across a greater number of census tracts or smaller units of aggregation.

Second, the data from this study only come from the city of Detroit. Detroit is an urban, predominately African American, highly impoverished, and high-crime city. For this reason, the findings may not generalize to other cities that are more rural, affluent, or have less crime. Related to this point, the data were only collected from 10 census tracts in Detroit, and while the sampling procedures were designed to create a representative sample of the city, it cannot be said for certain that the findings are representative of the entire city. Moreover, surveys conducted in high-crime cities like Detroit may have increased odds of selecting respondents with criminal records, which can be problematic for two reasons. First, respondents who were legally barred from owning a firearm may have been untruthful about their gun ownership. Second, it is possible that some participants who were administered the DSCS engaged in criminal behavior themselves, which could impact their perceptions of police and the likelihood they own a firearm for defense. Nonetheless, minority communities often go overlooked when it comes to collecting survey data on gun ownership (Arthur, 1992), so this data limitation might also be considered an advantage. Future studies should explore the relationship between gun ownership, police satisfaction, and neighborhood conditions using data collected from other locations to determine the generalizability of our results.

Third, true casual ordering in our key variables cannot be established from the data in this study. The DSCS was a cross-sectional survey, and thus there is no way to determine whether the variables examined caused respondents to own a gun for protection or whether these relationships could be reversed. For instance, it cannot be determined whether dissatisfaction with police led an individual to purchase a firearm or if the firearm purchase preceded their opinions held about police when the study took place. This misspecification of causal ordering in cross-sectional studies of gun ownership is referred to as an endogeneity bias (Kleck et al., 2011). Our findings should be viewed with this limitation in mind, as respondents may have developed negative attitudes toward police after acquiring a firearm. However, there is less evidence to suggest that this is a concern for perceptions of police and gun ownership compared with the association between fear of crime and gun ownership (Kleck et al., 2011).

Finally, the nature of our measure of gun ownership is another especially important limitation to consider. Although our measure of gun ownership is unique from items used in previous research to assess this construct, we believe it still captures the essence of defensive gun ownership because the item requires respondents to indicate that they consider their gun a security mechanism, although it is limited in its ability to tell us the original motive for acquisition. If the gun in question served multiple purposes (e.g., used for hunting and home defense), this could have several ramifications for our findings. First, this measure may be inconsistent with our theoretical framing because assessing the collective security explanation of gun ownership requires a measure of defensive gun ownership. Without clearly differentiating defensive motives for the purchase of a firearm, it is difficult to say whether our findings offer the most valid test of this hypothesis. Second, we may have presented a biased estimate of the relationship between gun ownership and police satisfaction. Hauser and Kleck (2013) argue that when gun ownership of all types is included in studies of this nature, downwardly biased estimates of the relationship between defensive gun ownership and fear of crime may be produced, meaning that we may have underestimated the actual relationship between gun ownership and police satisfaction. More specifically, if we were able to pare our sample down to only defensive gun owners, we may have found a stronger negative association between these variables. Still, readers should interpret our results with this limitation in mind.

The policy implications of this research are clear, but somewhat precarious. Our results demonstrating that lower satisfaction with police may result in higher levels of gun ownership may lead some to argue that improving perceptions of police may reduce the number of guns owned by individuals in America. Given that gun violence is declining nationally but remains a serious issue for a number of large cities (Planty & Truman, 2013), a reduction in the perceived necessity for a gun may offer one potential solution to this issue. It is unclear, however, whether improving ratings of police among residents of areas with high gun violence would have any effect on the rate of gun ownership or crimes committed with guns. Nonetheless, this may be a potential avenue for exploration among those interested in reducing gun violence. Community policing initiatives, which often report positive changes in attitudes toward police as a result of officer engagement with communities (Gill, Weisburd, Telep, Vitter, & Bennett, 2014), would be well suited to this.

It is also important to consider the psychological implications of our findings. First, few psychological assessments for the motives of gun ownership have been pursued (Stroebe, Leander, & Kruglanski, 2017). For instance, the first psychological model of defensive gun ownership was only recently advanced by Stroebe and colleagues (2017). They found that defensive gun ownership is driven by both the specific threat of experiencing victimization in one’s immediate context as well as more general or diffuse perceptions about whether the world is a dangerous place. This suggests that research examining fear of crime and victimization and their relation to gun ownership should consider respondents perceptions of both aspects of fear. Future studies might also consider how individuals may have general or specific attitudes toward the police (Brandl, Frank, Worden, & Bynum, 1994) which could impact whether or not they purchase a firearm for defense. Perhaps those who have more positive experiences with police personally (specific), but have more negative global attitudes (or vice versa) may be differentially inclined to own a firearm for defense. Future studies should explore this possibility.