The Effect of Gun-Free School Zones on Crimes Committed with a Firearm in Saint Louis, Missouri

Abstract

There have been no peer-reviewed, quantitative research studies on the effectiveness of gun-free school zones. The objective of this study was to use a cross-sectional, multi-group controlled ecological study design in St. Louis, MO city that compared the counts of crimes committed with a firearm occurring in gun-free school zones compared to a contiguous area immediately surrounding the gun-free school zone (i.e., gun-allowing zones) in 2019. Gun-free school zones were measured and analyzed in two ways. In the primary analysis, boundaries of the tax parcels were used for each school as the beginning of the gun-free school zone. Results from this analysis, after adjustment for pair-matching and confounding, were null. In the secondary analysis, gun-free school zones were measured as beginning at the geographic centroid of the school’s address. After adjusting for the pair-matching and confounding, this analysis showed 13.7% significantly fewer crimes committed with a firearm in gun-free school zones compared to gun-allowing zones. These results suggest that gun-free school zones are not being targeted for firearm crime in St. Louis, MO.

Introduction

Gun-free zones are areas where the use and carry of guns are prohibited by citizens. Gun-free zones exist in several forms across private businesses, public spaces, and government, but the most well-known example of the implementation of gun-free zones is the passing of the 1990 Gun-Free School Zones Act [1]. This federal law makes it illegal for an individual to knowingly possess a firearm in a place that they have a “reasonable cause to believe” is a school zone [1]. According to the National Criminal Justice Reference Service (NCJRS), federal, state, and local authorities are instructed to post signs around schools (public private, and parochial) indicating that the carrying of firearms within 1000 feet of school grounds is prohibited. Violating the Gun-Free School Zones Act may result in a fine, imprisonment up to 5 years, or both [2].

In theory, gun-free school zones can potentially change the risk of gun injury. For example, by removing, or even simply reducing, the number of guns in a gun-free zone, the risk of firearm-related injury due to criminal activity, otherwise small-scale arguments, or unintentional discharge of a gun may be decreased or eliminated [3]. Simply, where firearms are not present, firearm deaths cannot occur. However, this argument relies on the presumption that all individuals passing through the gun-free zone are following the law. An individual intending to do harm or threaten others with a gun could be emboldened in a gun-free zone because they may believe that there is less risk of encountering armed resistance. In this case, more crimes or threats with a firearm may occur in gun-free zones compared to areas that are not gun-free, henceforth called gun-allowing zones, since gun carriers are not able to defend themselves [3, 4] Relying on this latter hypothesis, gun rights activists are quick to blame gun-free zones as a cause of mass shootings, especially those that occur on school grounds. For this theory to be true, defensive gun use—where citizens armed with firearms stop gun violence from occurring—must happen at a rate that surpasses whatever preventative effect exists due to the prohibition of guns in gun-free zones. However, very little is known about the extent to which defensive gun use occurs in the USA, and rates of defensive gun use have varied greatly based on the method by which they were estimated [5]. As a result, both hypotheses, that gun-free zones are protective or harmful, may be true.

One challenge to determining the effectiveness of gun-free school zones is the selection of a location to study, as each jurisdiction may have varying laws, rates of gun carrying, and gun culture that could influence the association between gun-free school zones and firearm injury. Saint Louis was chosen as the location for this study for several reasons. Missouri is a relatively permissive state in terms of firearm laws [6], and approximately half of its residents own firearms [7]. If defensive gun use does indeed prevent firearm violence, then it could more likely show up in Missouri than in states with lower gun ownership rates, making Missouri an excellent test bed for a study of gun-free zones. Moreover, St. Louis, Missouri, has one of the highest firearm crime rates of crimes in the USA. In 2019, St. Louis had the highest rate of homicide of any major US city: there were 64.54 murders per 100,000 residents [8], most of which were committed with a firearm [9] and Forbes ranked it second in its list of most “dangerous” cities [10]. In addition, there has been extensive debate in St. Louis about whether or not to expand or eliminate gun-free zones. For example, in 2020, the St. Louis Board of Aldermen voted to declare public parks as “childcare facilities,” effectively making parks gun-free [11]. Nearly simultaneously, legislation (HB258) in Missouri was proposed to repeal certain gun-free zones, although this policy was ultimately opposed by the mayor of St. Louis [12]. Knowing the effect of gun-free school zones could inform policy in St. Louis in terms of the effectiveness of gun-free zones and gun-free school zones [3].

Furthermore, to the best of our knowledge, there have been no peer-reviewed, quantitative research studies on the effectiveness of gun-free school zones. This study will be the first to estimate the impact that gun-free school zones have on gun violence. The lack of research is due to the many methodological constraints of studying gun-free zones, including the challenge of finding appropriate control or comparison groups. This study innovates ways to contend with this challenge in order to evaluate if firearm violence is more or less likely to occur in gun-free school zones compared to comparator gun-allowing zones, in St. Louis, Missouri.

Methods

Data and Measures

The outcome for this study was any crime committed with a firearm in St. Louis, MO city, and county. These data were retrieved from the St. Louis, MO (metropolitan) Police Department [13] and the St. Louis County Police Department websites in April 2023 [14, 15].

The exposed units in the analysis were gun-free school zones, places where it is illegal for an individual to carry a firearm within 1000 feet of a school [2]. A list of the locations of all schools in St. Louis, MO in 2019, including pre-school, K-12 public, and K-12 private schools, was received from public [16] (and private [17]) school review, which provides “detailed profiles of USA public (and private) schools and their surrounding communities [16].” The, exposed unit, the gun-free school zones were conceptualized in two different ways. In the primary analysis, the tax parcels were used as the boundaries of the area of the school, and then, 1000 feet were added from the border of the tax parcel to account for the gun-free school zone. The school grounds themselves were excluded from the primary analysis due to possible issues of population flow and lack of a comparable control. The tax parcels were obtained from the St. Louis, MO government open data datasets and Saint Louis County GIS Service Center Scenarios [18]. The unexposed, gun-allowing zone extended 414.2 feet from the edge of the gun-free school zone. A detailed explanation and figure of this calculation are included in Appendix A. The exposed and unexposed units were then treated as a matched, paired unit.

In the secondary analysis, the geographic location of the address of the school was used as the center of the school (the centroid), and the gun-free zone was imagined as a circle with a 1000-foot radius surrounding this point. This is equivalent to a circle with an area of 3,141,592.7ft². The unexposed, gun-allowing units extended 1414.2 ft from the centroid. This distance was calculated so that the area was equal between the gun-free school zone and gun-allowing unexposed unit in the secondary analysis (or the equivalent to a tax parcel without any area in the primary analysis). A detailed explanation and figure of this calculation are included in Appendix A. The exposed and unexposed units were also then treated as a matched, paired unit.

Importantly, in both cases (but more so in the primary analysis), gun-free school zones and their corresponding gun-allowing zones overlapped (i.e., two schools next to one another). When this occurred, these schools were counted as a single unit, and attention was given in the analysis to account for differences in area via the use of an offset.

Possible confounders, including median age, median income, race, education, crime rate, population flow, and street length, were considered in the development of this study. The demographic variables and population flow were properly accounted for through the design of the study, as the unexposed units immediately surround the exposed units. Therefore, median age, median income, race, education, and crime rate should be approximately equal in both the gun-free school zone and the corresponding gun-allowing zone. Furthermore, the design of the study was purposely chosen to account for population flow, which is difficult to accurately measure. Because someone would need to travel through the gun-allowing zone to enter the gun-free zone, it was presumed that the population flow was approximately equal within the gun-free school zone and the gun-allowing area. However, differences in street length may still have introduced bias between the exposed and unexposed units because street networks may be denser immediately surrounding a school. Street length is related to population flow, as people usually travel on streets, and street length—via population flow—is known to be a cause of crime [19, 20]. Therefore, street length was added additionally as a confounder in the regression.

Mapping

The crimes committed with a firearm, gun-free school zones, and surrounding gun-allowing zones were then mapped. In Fig. 1, the primary analysis of the gun-free school zone, where the gun-free zone extends from the boundary of the tax parcel, is illustrated. The secondary analysis, where the gun-free zone extends from the centroid where the address is geolocated, is illustrated in Fig. 2.

Fig. 1.

Primary analysis: gun-free zone extends from the boundary of the tax parcel

Fig. 2.

Secondary analysis: the gun-free zone extends from the centroid where the address is geolocated

Analysis

Once mapped, the number of crimes committed with a firearm that occurred in gun-free school zones compared to gun-allowing zones were counted and described. Each gun-free school zone unit was pair matched with the surrounding gun-allowing zone, which accounted for issues of community level confounding, including median age, median income, race, education, crime rate and (partially) population flow. Street length was also included in the model to further adjust for population flow and the network density differences that may have arisen due to the mapping. Furthermore, due to the combining of units due to the overlapping borders, the area covered by gun-free school zones and gun-allowing zones was no longer equal. (Gun-free school zones predictably covered more total area than the gun-allowing zones, because the gun-allowing zones were more likely to intersect with each other and therefore lose area.) Therefore, a generalized linear-mixed model fit by restricted maximum likelihood, of the quasi-Poisson family (for the count response), with an offset that accounts for differences in area was utilized. This regression accounted for the correlations between units introduced via the pair-matching, and subsequently accounted for the spatial correlation between the exposed and unexposed units. The model statement and description are included in Appendix B. In a sensitivity analysis, we split the outcome, crimes committed with a firearm, into two categories: robberies and aggravated assaults. We did this to evaluate if there were differences in the result based upon crime type.

Results

Unadjusted Results

In 2019, there were 615 pre-kindergarten through 12th grade schools (including private) in St. Louis city and county. In the primary analysis, when the tax parcel boundaries were used as the beginning of the 1000 feet radius around the school ground (tax parcel), the number of units in the analysis dropped to 177 paired units due to gun-free school zones overlapping with their neighbors. There were 4691 total crimes committed with a firearm in St. Louis city and county. Of those, 3321 were cases of aggravated assault with a firearm and 1342 were cases of robberies where a firearm was used. The remaining 28 crimes were classified as state violence/defaced firearm. One thousand five hundred ninety-six crimes committed with a firearm occurred in gun-free school zones, which covered 91.33 mi², and 885 crimes committed with a firearm occurred in the surrounding gun-allowing zones, which covered 53.17 mi². Therefore, the rate of crimes committed with a firearm in gun-free school zone was 17.47 per square mile and 16.64 per square mile in the gun-allowing zones.

In the secondary analysis, once overlapping gun-free and gun-allowing zones were combined, the analysis consisted of 260 paired units. One thousand ninety crimes committed with a firearm occurred in the gun-free school zones, which covered 56.56 mi², or a rate of 19.27 crimes committed with a firearm per square mile in St. Louis in 2019. In the gun-allowing zones, 914 crimes committed with a firearm occurred, which covered 46.59 mi² of St. Louis, or a rate of 19.62 crimes committed with a firearm per square mile.

Adjusted Results

Using the boundaries of the tax parcels as the beginning of the 1000 feet radius, and after adjusting for the matching between the gun-free school zones and the gun-allowing zones, street length, and the differences in area due to the combination of units, the results were null (IRR: 1.00, 95%CI: 0.999, 1.002, p-value: 0.608). Results were similarly null when the outcome was divided to include only aggravated assaults with a firearm or only robberies involving a firearm.

In the secondary analysis, after adjusting for the paired matching between the gun-free school zones and the gun-allowing zones, street length, and the differences in area due to the combination of units, there were statistically significant 13.7% fewer crimes involving a firearm in gun-free school zones, with a confidence interval ranging from 22.6 fewer to 3.9% fewer (p-value: 0.008) (Table 1). When the outcome was divided into aggravated assaults and robberies, the results were nearly identical in magnitude, although statistically insignificant. For aggravated assaults, there were statistically insignificant 14.0% fewer aggravated assaults involving a firearm, with a confidence interval ranging from 25.5% fewer to 0.6% more aggravated assaults involving a firearm (p-value: 0.061). For robberies, there were statistically insignificant 13.9% fewer robberies involving a firearm, with a confidence interval ranging from 30.2 fewer to 6.1% more robberies involving a firearm (p-value: 0.061).

Table 1.

Quasi-Poisson model for the count of crimes committed with a firearm in gun-free versus gun-allowing zones in St. Louis city and county (2019)

	IRR	95% CI	p-value
Primary analysis	1.000	(0.999, 1.002)	0.608
Secondary analysis	0.863	(0.774, 0.961)	0.008

Discussion

The objective of this study was to use a cross-sectional, multi-group controlled ecological study design in St. Louis, MO city that compared the counts of crimes committed with a firearm occurring in gun-free school zones compared to a contiguous area immediately surrounding the gun-free school zone (i.e., gun-allowing zones) in 2019. In the primary analysis—when the boundaries of the tax parcels were used for each school as the beginning of the gun-free school zone and the school ground were excluded—resulted in null results. There was no difference in the crimes committed with a firearm between the gun-free school zone and the corresponding gun-allowing zone. In the secondary analysis—when gun-free school zones were conceptualized as beginning at the geographic point of the school’s address—after adjusting for the pair-matching and confounders, there were statistically significant 13.7% fewer crimes committed with a firearm in gun-free school zones compared to gun-allowing zones.

These results suggest that gun-free school zones are not being targeted for firearm crime due to the presumed lack of law-abiding individuals who would engage in defensive gun use. Furthermore, the secondary analysis suggests that gun-free school zones may even prevent crimes committed with a firearm committed very near the school or on school grounds; however, results of this analysis should be interpreted with caution, as there is not a good comparison group for the school grounds (given all school grounds are gun-free) and confounding due to population flow may still exist. The difference in magnitude and significance between the primary and secondary analysis may also be due to the way that gun-free school zones are conceptualized. While the law states that gun-free school zones extend 1000 feet from school grounds, the definition of “school grounds” is not well defined. Indeed, state [21] and city governments [22] and gun rights activists [23] have represented the borders of gun-free school zones as the secondary analysis of this study, despite the federal law indicating that they begin from school grounds. Both enforcement agencies and individuals may not know where school grounds—and subsequently 1000 feet from the border of the tax parcel—begins or ends, which may be a reason for the differing results. Nevertheless, the polygon centroid is not inherently meaningful (i.e., it is unlikely that many people could identify the centroid of a school polygon, let alone a distance buffer around that point), and in some instances, these radial buffers include land area that is within the school bounds.

Although no study has evaluated the effect of gun-free school zones on crime committed with a firearm, several studies have examined the relationship between gun ownership and crime. Generally, when gun ownership is higher, gun-related crime is also higher [24, 25]. Even at the household level, the presence of a firearm has been shown to be associated with a higher risk of homicide [26] and suicide [27]. These studies therefore suggest that the presence of a firearm increases the risk of a firearm-related injury or death. Unless defensive use with a gun is happening more often than actual crime committed with a gun, which is unlikely [28, 29], a reduction in firearm prevalence should theoretically lead to a reduction in firearm crimes, despite claims that gun-free school zones make schools less safe. In this study, we did not find higher rates of crime committed with a firearm in gun-free school zones.

The design of this study was novel and was created to deal with several unique epidemiologic challenges. Population flow presented as a probable confounder: places that have higher levels of foot traffic or are denser in population are more likely to have both higher rates of crime committed with a firearm [30] and more likely to be around a school, and thus in a gun-free school zone. Population flow was difficult to adjust directly in the analysis because there are currently no universal measures of this variable: at best, population (or a form of) can be used as a proxy, but residential population may be very low for a commercial area, despite population flow being high due to business and commerce. Other ways of adjusting for population flow, like cell phone records [31], were not practical for this study. Therefore, it was advantageous to account for population flow through the design of this study by having the unexposed units immediately surround the exposed units. However, this only partially accounted for population flow, as we would still expect that foot traffic would be highest in and around the school. For this reason, we expect that the results could be biased, making gun-free zones in this analysis look less protective than they may actually be.

Given the smallest available spatial unit where socioeconomic data are available is the census block group, and many of the exposed and unexposed units lie within a single census block, we believe that the matched design of the study has limited any meaningful neighborhood-level bias (including differences in age, education, median income, and race, as a proxy for racism). However, there may still be some unmeasured confounding on the street/block level. Street lengths, a measure closely related to population flow [19, 20], still varied between the exposed and unexposed units and was added as a confounder in the regression. Another confounder may be the presence of a police department. In a sensitivity analysis, a measure of the presence of a police station within the borders of the gun-free and gun-allowing zones was added as a covariate. The addition, however, did not meaningfully affect the results.

The unexposed and exposed units, as described previously, shared a similar distribution of confounders, were more likely to be alike than pair-matched units that were further away, and were therefore correlated. The issue of spatial autocorrelation was accounted for in this analysis in two ways: (1) any units that were near enough to one another to touch were collapsed into a single unit and (2) via pair-matching. While the combination of touching units accounted for a large component of the spatial correlation between paired units, it did not address the spatial autocorrelation that occurred within them. However, accounting for the pair-matching in the analysis phase via a linear-mixed model was analogous to directly accounting for spatial autocorrelation; both are accounting for the fact that pair-matched units are more likely to have similar characteristics than those that are not pair-matched [32]. In fact, the study was purposely designed so that this would be true to account for the difficult to measure confounders.

Another limitation is that the outcome of this study, by default, only includes a sample of all crime committed with a firearm. To be registered by the county and metro police departments, the crime needed to be reported and recorded, and it is documented that most shootings are not. ShotSpotter, a “gunshot detection technology that uses sophisticated acoustic sensors to detect, locate and alert law enforcement agencies and security personnel about illegal gunfire incidents in real-time [33],” was compared to actual police reported gunfire incidents in Washington DC and Oakland California. Only 12% of the gunfire incidents recorded by ShotSpotter were reported to the police, indicating most gun crime went undocumented [34]. In this study, if the distribution of underreporting of crimes committed with a firearm is equal between the exposed and unexposed units, then this would bias the true estimate toward the null. However, it is reasonable to believe that areas closest to a school might be more likely to report crime than areas further from the school, due to increased diligence by the teachers, staff, and security in reporting crime. If crime was more likely to be reported closer to a school than further away, then this would lead to bias and would make gun-free school zones look less protective against crimes committed with a firearm than they are in reality.

Furthermore, it was assumed that the gun-free school zones were homogenously enforced as gun-free, while that is not true in practice. Private residences that fall within the gun-free school zone, for example, would be exempt from the law. This would bias the estimate toward the null, making the effect of gun-free zones appear less protective, as the area covered by the gun-free school zone would be smaller. The aim of this study was also to only examine gun-free school zones and not all gun-free zones. Therefore, there were locations in the gun-allowing area that may be gun-free. While these establishments would bias the estimate away from the null, it is unlikely to be a meaningful change due to most firearm-related crimes incurring outside or in private homes [35].

Finally, this study was cross-sectional. The Gun-Free School Zones Act [1] was enacted in 1990, a period before extensive geospatial tracking of crime was common. We were therefore unable to test whether the addition of gun-free school zones resulted in an increase or decrease of firearm deaths overtime.

Conclusion

Gun-free school zones are a politically contentious intervention, and no prior studies have examined whether they are associated with harm or benefit due to gun violence incidence. This study suggests that gun-free school zones are not associated with higher rates of crimes committed with a firearm in Saint Louis, Missouri. This study is the first of its kind to evaluate the quantitative effect of gun-free school zones on crimes committed with a firearm and should be repeated in other cities or jurisdictions before it is widely generalized.

Appendix

Appendix A

In the secondary analysis, the geographic location of the address of the school was used as the center of the school (the centroid), and the gun-free zone was imagined as a circle with a 1000-foot radius surrounding this point. This is equivalent to a circle with an area of 3,141,592.7 ft². To create unexposed, gun-allowing units of the same area, the area of the circle was doubled (therefore 6,283,185.4 ft²), which is equivalent to a radius of 1414.2 ft. Therefore, when the inner circle is subtracted from the outer circle, the two areas—gun-free school zone and gun-allowing zone—are equal. For consistency, 414.2 feet (1414.2 minus 1000 feet) was used in the primary analysis as the distance extending from the gun-free school zone to the end of the gun-allowing zone.

Appendix B

$$O_{\mathit{ij}}\sim QuasiPoisson\left(exp,\left(n_{\mathit{ij}}\right)\right)$$

$$n_{\mathit{ij}}={\beta }_0+b_i+{\beta }_i{X}_{\mathit{ij}}+{\beta }_i{L}_{\mathit{ij}}+log\left(A_i\right)$$

$$b_i\sim N(0,{\sigma }_b^2)$$

Where:

i indexes on pair IDs

j = {1,2} indexes observations within IDs.

n = counts of interpersonal shootings for unit i

A = area for each unit i, acting as an offset

β ₀ = overall intercept

b _i = random intercepts

X = gun-free zone = 1; gun-allowing zone = 0

L = street length (km)