Association of Involuntary Displacement of People Experiencing Homelessness and Crime in Denver, CO: A Spatiotemporal Analysis

Abstract

In 2022, approximately 580,000 people experienced homelessness in the United States. In response, many cities have implemented “camping ban” policies enforced by involuntary displacement of homeless encampments. Displacement has been cited as a strategy to protect public health and safety. However, there is mixed evidence that displacement is effective in reducing crime, while it is associated with other adverse health outcomes. To evaluate the neighborhood-level association between displacement and crime, we performed a retrospective (November 2019 to July 2023) pre-post spatiotemporal analysis using administrative data from Denver, CO. We used the Knox test statistic to detect excess clustering and change in total crime, as well as crime stratified by the National Incident-Based Reporting System (NIBRS) category, within spatiotemporal proximity to displacement events. We found that, on average, clustering of crime is high both before and after displacement. Within a 0.25-mile radius, displacement is associated with a statistically significant but modest decrease in crime, between − 9.3% within 7 days (p < 0.001) and − 3.9% within 21 days (p = 0.002). We found no consistent change in composite crime at a 0.5- or 0.75-mile radius. Hyperlocal decreases were driven by significant decreases in public disorder and auto theft, while crimes against persons increased and displayed high clustering post-displacement. There were no changes in any other offense type. Involuntary displacement is not consistently associated with changes in clustering of crime and may exacerbate violence in nearby areas.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11524-024-00924-1.

Introduction

More than 580,000 people in the United States (U.S.) experienced homelessness in 2022, a growth of 6% since 2017 [1]. Increasingly visible unsheltered homelessness in U.S. cities, accounting for 40.1% of the overall homeless population [2], has garnered contentious debate among residents and policymakers over homelessness governance policies. One strategy that has become common in U.S. cities is involuntary displacement of homeless encampments [3]. Also called “street evictions,” “clearances,” “cleanups,” and “sweeps,” involuntary displacement is a police-based response involving the dismantling of encampment communities and the dispersion of unsheltered people experiencing homelessness (PEH). Municipal authorities conduct displacements under the purview of myriad laws and policies, including “quality of life,” “nuisance,” and “vagrancy” ordinances and “sit-lie” and camping bans [4].

Policymakers have justified involuntary displacement as necessary to improve the health and safety of their communities [5, 6]. Of particular concern to urban residents is encampment-associated crime. Crime is a social determinant of health, as exposure to nearby crime is associated with stress, obesity, adverse birth outcomes, and poor mental health [7]. Areas around encampments may experience nearly three times greater clustering of crime compared to the city overall [8]. Furthermore, public perception of increased homelessness-associated crime (indicated by opinion polling and rapid rises in 311 complaints) has outpaced the growth of unsheltered homelessness, fueling demands for police-based responses including displacement [9, 10].

Camping bans have recently been subject to legal challenges on the basis that imposing criminal or civil penalties on PEH for sleeping outside constitutes an infringement on the fundamental right to survive protected by the second, eighth, and fourteenth amendments of the US Constitution [11]. The U.S. Supreme Court case City of Grants Pass v. Johnson, decided June 28, 2024, rejected these claims, giving states and cities broad authority to develop homelessness policy which may include involuntary displacement [12]. In consideration of this, there is a need for evidence-based evaluation of various policy options to inform effective and equitable public health and homelessness policy.

While cities pursue displacement intending to relieve neighborhoods of crime [5, 12], the spatial association between displacement and crime outcomes has not been thoroughly evaluated. The scant evidence available suggests that involuntary displacement is not consistently associated with changes in crime. An evaluation of Los Angeles’s Safer City Initiative, a carceral police intervention designed to reduce nuisance, property, and violent crime in Skid Row, found only modest reductions in crime and continued overall trends of high crime [13, 14]. A 2017 case study from The Bronx found no significant difference in nearby crime complaints following an encampment closure [15]. Additionally, involuntary displacement leaves PEH at greater risk of experiencing theft, physical assault, and sexual assault, suggesting that displacement may not have a positive neighborhood-level impact on crime or that potentially beneficial effects of the policy are not equitably distributed [16–18].

Involuntary displacement is otherwise associated with deleterious health effects on PEH, as experiencing displacement worsens sleep and mental health, causes disconnection from health and social service outreach, isolates and undermines the protective effects of community, increases exposure to hazardous environments, and often results in confiscation of medications and clothing [18–22]. A modeling study conducted by several authors of this paper demonstrated involuntary displacement could increase deaths by nearly 25% over a 10-year period [19]. In 2023, the American Public Health Association released a consensus statement that involuntary displacement harms the health and safety of PEH [23].

Given the paucity of evidence surrounding the impact of displacement on neighborhood crime and the potential adverse health outcomes associated with displacement, we aimed to evaluate the spatiotemporal relationship between involuntary displacement events throughout a nearly 4-year period and short-term changes in nearby crime in a large U.S. metropolitan area. We hypothesized that observed clustering of crime near former encampment sites would not change from pre- to post-displacement time periods.

Methods

Study Design

We performed a retrospective spatiotemporal analysis using data from Denver, CO, between November 2019 and July 2023. Although Denver’s camping ban was implemented in May 2012, displacement data were not available prior to the study period. We chose the aforementioned period due to comprehensive data availability for crimes and involuntary displacement events.

Data Sources

All homeless encampment displacements were recorded by the City and County of Denver City Council following public reporting requirements from the Denver Department of Transportation and Infrastructure (DOTI). All crimes were recorded by the Denver Police Department into the National Incident-Based Reporting System (NIBRS) [24], publicly available via the Denver Open Data Catalog. Criminal incidents are reported into NIBRS when they become known to law enforcement, and each distinct and mutually exclusive offense during an incident is reported separately.

Records of involuntary displacement (documented as “large-scale encumbrance cleanups”) include the date, time, and description of the encampment location. Every displacement in our dataset occurred between 5:00 and 8:00 AM, and all but 10 occurred on a Tuesday, Wednesday, or Thursday. Granularity of encampment location varied, provided as either an intersection (e.g., “7th Avenue and Lincoln Street”), a city block (e.g., “Emerson Street from 18th Avenue to 19th Avenue”), or an area bounded by multiple streets (e.g., “E. Colfax Avenue, Fillmore Street, 14th Avenue, and Clayton Street”). To create a uniform dataset of point variables, encampment locations that were provided as city blocks or bounded areas were first geocoded manually as line or polygon features, respectively; displacements along a city block were assigned the midpoint of that block as their point location and displacements whose bounds form a polygon were assigned the polygon’s centroid [25].

Each reported crime included a location (latitude and longitude) and date in which the crime began (“first occurrence date”). Crimes are categorized in NIBRS as crimes against persons, crimes against property, or crimes against society [24], and further divided into 13 offense types: aggravated assault, arson, auto theft, burglary, drug and alcohol crimes, larceny, murder, other crimes against persons, public disorder, robbery, theft from motor vehicle, white-collar crime, and all other crimes (Table 1). Sexual assault and domestic violence crimes were excluded because the location of these crimes is redacted to protect victim privacy. Encampment locations were mapped, geocoded, and visualized using QGIS version 3.30.0.

Table 1.

Count of crimes by NIBRS category and offense type

Crime category	Offense type	Count	Percent of total
Crimes against persons		28,465	10.74%
	Murder	291	0.11%
	Aggravated assault	12,229	4.61%
	Other crimes against persons	13,327	5.03%
	Sexual assault	2618	0.99%
Crimes against property		156,707	59.13%
	Robbery	4582	1.73%
	Burglary	19,482	7.35%
	Larceny	37,005	13.96%
	Auto theft	43,860	16.55%
	Theft from motor vehicle	47,102	17.77%
	Arson	602	0.23%
	White collar crimes	4074	1.54%
Crimes against society		51,034	19.26%
	Public disorder	40,187	15.16%
	Drug and alcohol crimes	10,847	4.09%
	All other crimes	28,805	10.87%
Total		265,011	100%

Statistical Analyses

We employed the Knox test statistic to quantify the magnitude of clustering of crimes which were spatiotemporally proximal to displacements, as well as the change in clustering from pre- to post-displacement periods [26–28]. The Knox test is a small-area spatiotemporal method previously used in research on crime, overdoses, cancer, infectious diseases, and other epidemiological outcomes to test for the presence of non-random “clustering” of one set of geographic points around another within a defined distance and time period [26–28]. Observed counts of crime incidents within certain spatial and temporal windows around displacements were compared to expected values which reflect baseline crime patterns in displaced areas during random time windows when displacement did not occur. The null hypothesis of the Knox test is that the outcome, crime, is randomly distributed and not associated with displacement [28].

The Knox test statistic—the observed value—is calculated by counting the number of crimes whose coordinates and occurrence date fall within certain spatial and temporal windows of a particular displacement and then averaging this count across all displacements. The equation for the test statistic κ is below, with ${\left(x,,,y\right)}_j^c$ and $t_j^c$ representing the coordinates and date of crime j, ${\left(x,,,y\right)}_i^s$ and $t_i^s$ representing the coordinates and date of displacement i, r representing the catchment radius (or spatial bounds) of interest, $t_0$ and $t_1$ representing the temporal bounds of interest, and n representing the total number of displacements.

$$\kappa =(\sum 1\{\Vert {\left(x,,,y\right)}_j^c-{\left(x,,,y\right)}_i^s\Vert <r,-t_0<t_j^c-t_i^s<t_1\})/n$$

We compared observed values to 95% confidence intervals for the expected pre-post difference in crime under a null distribution. We estimated expected values by generating null distributions of Δκ, κ_before, and κ_after through a Monte Carlo bootstrapping process in which random permutations of actual displacement dates were matched with actual displacement locations, while the dates and locations of crimes were kept fixed [26, 27]. In this method, crime occurring in locations where displaced encampments resided during times in which a displaced encampment was not present served as a control by which baseline expected values of crime and changes in crime were estimated. To preserve time series autocorrelation and limit confounding from covariates that could influence the likelihood of crime on a particular day (e.g., weather, holidays, major events), the pool of dates from which we resampled only contained dates in which displacement occurred. We used 200 permutations of displacement dates to quantify uncertainty. Test statistics outside the 95% confidence interval represent a meaningful deviation in clustering of observed crimes from expected values.

The primary outcome was the average change in the clustering of crimes per displacement from pre-displacement to post-displacement periods (“change in crime”), denoted by Δκ. Positive values of Δκ indicate more observed crimes after displacement compared to before—greater clustering of crime—while negative values indicate reduced clustering. Secondary outcomes were average counts per displacement of all crimes in pre-displacement periods (“crimes before”) and average counts per displacement of all crimes in post-displacement periods (“crimes after”), denoted by κ_before and κ_after, respectively. These outcomes were evaluated separately to isolate and contextualize local crime trends before and after displacement, independent of the effect of displacement [27]. Change in crime is equal to the difference between crimes after and crimes before.

To identify temporal patterns, we calculated outcomes at three time periods t: 7-, 14-, and 21-day periods before and after displacements, reflecting similar studies ²⁷. The pre-period included the 7, 14, or 21 days directly preceding the displacement date, and the post-period included the 7, 14, or 21 days following displacement. Because the time of day when displacements occur may vary, the primary analysis excluded the date the displacement occurred.

To identify spatial patterns, we calculated outcomes at three catchment areas, within radii (r) of 0.25, 0.50, and 0.75 miles of displacements. These radii—approximately two to six city blocks in Denver—represent a range of definitions of “neighborhood” or “area-level” in the epidemiologic literature on crime; exposure to crime within these distances has been linked to adverse health outcomes [7, 29, 30]. To calculate the bounds of spatial windows around each displacement, we used the R function “ml2d” in the okara package to transform Euclidean distances into decimal degrees, using the city center of Denver as the base latitude.

All analyses were conducted using R version 4.2.3.

Stratified and Sensitivity Analyses

In addition to our primary composite analysis, we performed two stratified analyses: (1) by NIBRS crime categories (crimes against persons, property, and society) and (2) by each of 13 NIBRS offense types. “Crimes against persons” include the offenses of murder, aggravated assault, and other crimes against persons; “crimes against property” include arson, robbery, burglary, larceny, auto theft, theft from motor vehicle, and white-collar crime; and “crimes against society” include public disorder and drug and alcohol crimes.

We conducted several sensitivity analyses to assess the robustness of our findings:

1. We included displacement date in the post-period rather than excluding it. We chose to include it in the post-period rather than the pre-period because displacement is typically conducted in the early mornings, thus capturing crimes occurring throughout that day.

2. We calculated outcomes within the two “donuts” between radii—areas between 0.25 and 0.5 miles and areas between 0.5 and 0.75 miles. This was done to isolate effects in outlying areas and test for crime displacement.

3. We excluded displacements that reoccurred in the same location less than 7 days after an earlier removal.

4. We excluded displacements that temporally overlapped, removing displacements occurring within 21 days of another from the exposure dataset.

5. We excluded displacements that spatially overlapped, removing displacements whose 0.25-mile buffer overlapped with another from the exposure dataset.

6. We drew less-granular displacement locations from random within lines or polygons, rather than assigning the midpoint or centroid.

7. We excluded displacements occurring during the COVID-19 lockdown (March–May 2020).

8. We performed a citywide analysis considering the boundaries of Denver as the catchment area.

Results

Three hundred and three unique encampment displacements were recorded during the study period (Fig. 1). The frequency of displacements peaked in 2021 to 10.0/month and then decreased to 8.2/month in 2022 and 5.1/month in 2023. The reported size of displaced encampment areas increased over time. While most displacements before 2022 occurred in the city core, more recent displacements increasingly occurred outside Denver’s Central Business District.

Fig. 1.

Map of all documented encampment displacements in Denver, CO, from November 1, 2019, to July 17, 2023, color-coded by year. Point, line, and polygon features were manually geocoded based on textual descriptions of boundaries of displaced areas

There were 265,011 crimes recorded during the study period. The majority were crimes against property, followed by crimes against society and then crimes against persons (Table 1). Total crimes increased from 4661/month in 2019 to 6360/month in 2023.

In general, we observed small decreases in overall crime following displacement within areas immediately adjacent to encampments, but no consistent changes in crime beyond a 0.25-mile radius (Table 2). In areas within 0.25 miles—approximately two blocks—of encampment locations, there was a significant but modest decrease in crime within 7, 14, and 21 days of displacement. On average, within 0.25 miles, displacement was associated with a decrease in crime between 9.3% at 7 days (p < 0.001), 5.6% at 14 days (p < 0.001), and 3.9% at 21 days (p = 0.002). Within 0.5 miles, we observed a decrease in crime of 3.0% within 7 days (p = 0.007). At all other distance-time combinations, observed changes in crime were within expected ranges.

Table 2.

Association of crimes and displacement, across all combinations of distance (0.25, 0.5, 0.75 miles) and time periods (7, 14, 21 days)

Outcome	Radius	Time	Expected value1	95% CI, expected value1	Observed value2	% change pre-post	p-value
Δκ (change in crime from pre- to post-displacement periods, per displacement)	0.25 mi	7 days	0.07	(− 0.54, 0.68)	− 1.31	− 9.31%	< 0.001***
		14 days	0.20	(− 0.75, 1.15)	− 1.54	− 5.57%	< 0.001***
		21 days	0.53	(− 0.81, 1.87)	− 1.56	− 3.85%	0.002**
	0.5 mi	7 days	0.09	(− 1.15, 1.33)	− 1.62	− 2.96%	0.007**
		14 days	1.07	(− 1.07, 3.21)	0.05	0.05%	0.35
		21 days	2.03	(− 0.92, 4.99)	0.53	0.33%	0.32
	0.75 mi	7 days	0.39	(− 1.28, 2.05)	− 0.13	− 0.12%	0.54
		14 days	2.34	(− 0.69, 5.38)	2.89	1.32%	0.72
		21 days	4.13	(0.27, 8.00)	2.11	0.65%	0.30
κbefore(average count of crimes occurring in pre-displacement periods, per displacement)	0.25 mi	7 days	12.93	(12.41, 13.44)	14.07	NA	< 0.001***
		14 days	25.71	(24.80, 26.62)	27.67	NA	< 0.001***
		21 days	37.99	(36.92, 39.06)	40.48	NA	< 0.001***
	0.5 mi	7 days	52.60	(51.50, 53.69)	54.79	NA	< 0.001***
		14 days	104.22	(102.43, 106.01)	107.93	NA	< 0.001***
		21 days	154.57	(152.06, 157.08)	159.54	NA	< 0.001***
	0.75 mi	7 days	109.18	(107.68, 110.67)	110.89	NA	0.02*
		14 days	216.59	(213.91, 219.26)	218.95	NA	0.08
		21 days	321.04	(317.71, 324.37)	325.87	NA	0.004**
κafter(average count of crimes occurring in post-displacement periods, per displacement)	0.25 mi	7 days	13.00	(12.48, 13.51)	12.76	NA	0.37
		14 days	25.91	(24.98, 26.85)	26.13	NA	0.65
		21 days	38.52	(37.28, 39.76)	38.92	NA	0.53
	0.5 mi	7 days	52.68	(51.63, 53.74)	53.17	NA	0.37
		14 days	105.29	(103.18, 107.40)	107.98	NA	0.01*
		21 days	156.61	(153.84, 159.37)	160.07	NA	0.01*
	0.75 mi	7 days	109.56	(108.00, 111.13)	110.76	NA	0.14
		14 days	218.93	(216.08, 221.78)	221.85	NA	0.04*
		21 days	325.18	(321.10, 329.25)	327.98	NA	0.18

Bolded values denote statistical significance at p<0.05; *** denotes p-values < 0.001; ** denotes p-values < 0.01; * denotes p-values < 0.05

¹Expected values were generated through Monte Carlo bootstrapping of past crime data

²Observed values represent Knox test statistics

In pre-displacement periods, we observed significantly high clustering of crime in 8 of the 9 scenarios (Table 2). In post-displacement periods (in areas where encampments formerly resided), we observed significantly high clustering in 3 of the 9 scenarios and within the expected range in all others (Table 2). There was no scenario in which time periods after displacement recorded significantly low clustering.

Stratified Analyses

Crime Category

Among crimes against persons, there was no significant change in crime from pre- to post-displacement periods aside from an increase of 7.4% at 14 days (p = 0.01) and 6.5% at 21 days (p = 0.006) within the “donut” between 0.25- and 0.5-mile radii (Supplemental Table S49). Pre-displacement clustering of crimes against persons was within expected ranges. In post-displacement periods, crimes against persons exhibited significantly high clustering within 0.5 miles (p = 0.06 at 7 days; p = 0.003 at 14 days; p = 0.003 at 21 days) and 0.75 miles (p = 0.004 at 7 days; p = 0.006 at 14 days; p = 0.02 at 21 days) (Supplemental Tables S1–S3).

Crimes against property decreased beyond expected ranges, between 1.5 and 10%, in areas within 0.25 miles (p = 0.003 at 7 days; p < 0.001 at 14 days; p < 0.001 at 21 days) and 0.5 miles (p = 0.001 at 7 days; p = 0.001 at 14 days; p = 0.02 at 21 days) (Supplemental Tables S4–S6). Crimes against property exhibited significantly high clustering in pre-displacement periods within 0.25 miles of encampments (p = 0.07 at 7 days; p < 0.001 at 14 days; p = 0.001 at 21 days), and significantly low clustering in 8 out of 9 combinations of post-displacement periods within 0.25, 0.5, and 0.75 miles (ranging from p = 0.07 at 0.5 miles and 21 days to p = 0.004 at 0.75 miles and 21 days).

Crimes against society did not exhibit significant changes in clustering and displayed significantly high clustering in both pre-displacement periods (all p < 0.05) and post-displacement periods (all p < 0.05 except within 0.25 miles at 7, 14, and 21 days) (Supplemental Tables S7–S9).

Offense type

Within 0.25 miles, decreases in crime were driven by two offense types: public disorder (which includes criminal mischief, loitering, prostitution, harassment, disturbing the peace, and similar crimes) and auto theft, which together accounted for 49–59% of the total decrease (Table 3). We found no consistent change in the other 11 offense types within this radius (Table 3; Supplemental Tables S10–S64).

Table 3.

Descriptive change in number of crimes from pre-sweep to post-sweep period (Δκ), stratified by crime category and offense type

Offense type	Change in crimes3
Total	Decrease at 0.25 miles
Crimes against persons	Increase between 0.25 and 0.5 miles
Murder	Increase at 0.75 miles
Aggravated assault	No change
Other crimes against persons	Increase between 0.25 and 0.5 miles
Crimes against property	Decrease at 0.25 and 0.5 miles
Robbery	No change
Burglary	No change
Larceny	No change
Auto theft	Decrease at 0.25, 0.5, and 0.75 miles
Theft from motor vehicle	No change
Arson	No change
White collar crimes	No change
Crimes against society	No change
Public disorder	Decrease at 0.25 miles
Drug and alcohol crimes	No change
All other crimes	No change

³Result reported if significant (p < 0.05) in at least 2/3 time periods within certain radius of displacements; 5 radii were tested, representing areas within 0.25, 0.5, and 0.75 miles of displacements and areas between 0.25 and 0.5 miles and 0.5 and 0.75 miles of displacements

Two offense types classified as crimes against persons—murder and “other crimes against persons,” which primarily include simple assaults—recorded increases beyond expected ranges following displacement within at least one catchment area outside of 0.25 miles (Table 3; Supplemental Tables S31–S36). Among offense types classified as crimes against property, only auto theft saw decreases beyond expected ranges at any radius (S19–S21).

Sensitivity Analyses

Sensitivity analyses did not qualitatively change results (Supplemental Tables S65–S88).

Discussion

We performed a spatiotemporal analysis over a 44-month period in a large city to understand the relationship between involuntary displacement of homeless encampments and area crime. Our thorough analysis included every reported displacement in the City of Denver and every reported crime with a location and date available. In general, we found that involuntary displacement was not consistently associated with changes in area crime across several time periods and distances. Our findings are novel in that they are the first to comprehensively and objectively assess this relationship, providing evidence that involuntarily displacing groups of PEH is not an effective crime prevention or reduction strategy.

While there was no consistent evidence of changes in area crime, our finding of a significant reduction in clustering of crime within a 0.25-mile radius deserves further context. First, the absolute change in composite crime was relatively modest—at most a decrease of 9%, or one fewer crime than would be expected at random. Additionally, in Denver, as in other large cities with sizeable populations of unsheltered PEH like Los Angeles, sites where encampments formerly resided are sometimes fenced off and rendered inaccessible following displacement. In doing so, the area within which crimes could possibly occur becomes smaller in post-displacement compared to pre-displacement periods [21]. Moreover, PEH are significantly more likely to experience crime victimization than the general population [31], so a decrease in crime near former encampment areas may reflect a decreased concentration of crime victims.

Independently assessing the time periods before and after displacement provides further context. In the pre-displacement period, we observed high clustering of composite crime. Within a framework of complaint-oriented policing, this suggests that displacement may be reactive to crime or perceptions of risk of crime [10]. As homelessness has become more visible, encampments are subject to more citizen and business demands for a police response, which culminates in displacement. Cities have encouraged citizen reporting of homeless-related complaints via 311 calls and forums like Nextdoor [10, 32]. Importantly, crime remained at expected or high levels in post-displacement periods, suggesting that crime was not ameliorated or prevented by displacement. Taken together, there does not appear to be a safety benefit to encampment displacement for the surrounding community.

Different categories of crimes exhibited contrasting patterns. In post-displacement periods, there was a high degree of clustering of arguably the most serious category of offenses: crimes against persons. Concentrated clustering of murders and assaults in areas adjacent to displaced encampments suggests that displacement may decrease neighborhood safety, possibly by destabilizing encampment communities, disrupting established hierarchies, and causing fights between encampment residents [17]. These results corroborate previous findings. As noted by Herring et al., “the constant churning of move-along orders provoke conflict among individuals trying to survive in limited public spaces”[10]. Additionally, Mayer et al. state that displacement causes the disruption of trustworthy encampment communities leading to increases in larceny and sexual assault experienced by PEH, especially women [17]. Crimes against property exhibited a significant decrease following displacement; this decrease was attributed solely to auto theft, with no other category recording an unexpected change. This is a novel finding and warrants further exploration. Finally, crimes against society exhibited high clustering both before and after displacements and did not change, although public disorder decreased. “Quality of life” crimes including public disorder are discretionary and disproportionately enforced against unhoused people [33]. These crimes may, therefore, be more reflective of policing priorities and perceived threats rather than an objective indicator of public safety.

In the wake of the Grants Pass decision, our findings inform a salient issue across many US cities. As policymakers attempt to balance the health and safety needs of all their constituents, including unhoused residents, we have provided evidence that displacement may not be an effective or equitable strategy for preventing or reducing crime.

Limitations

There are several limitations to this analysis. First, police-recorded crime data is an imperfect proxy for all crime; police-recorded crime is correlated with actually occurring crime, but is also correlated with police presence, especially so in heavily Black, Latinx, unhoused, and other overpoliced neighborhoods [7]. We addressed this by disaggregating results by offenses which are more or less likely to reflect police discretion. Second, over half of violent and property crimes do not get reported to police, and reporting rates vary between populations [34]. Although PEH are more likely to experience victimization than housed people, they are less likely to report crimes to the police due to widespread distrust and trauma associated with law enforcement [35]. However, the Knox test mitigates these issues, as underreporting within a neighborhood is likely to remain constant in the short term. Third, using an ecological study design, we are unable to draw causal conclusions. Fourth, displacement data did not include exact encampment locations or the size of the population displaced. We explored this in sensitivity analyses which did not qualitatively change our results. Finally, by measuring crime counts rather than rates, the Knox test assumes stable population counts within catchment areas from pre- to post-displacement periods. Future work could leverage methods like spatial regression to test for confounding and effect modification by factors including frequency of policing or demographic composition of neighborhoods.

Public Health Implications

Our study provides evidence that involuntary displacement of PEH does not reliably or sustainably reduce crime. In conjunction with a robust body of evidence suggesting that involuntary displacement is harmful to the health and safety of PEH, including increased morbidity, mortality, and crime victimization among this population, the implications of this study warrant consideration of shifting discourses, policies, and resources purporting to promote “public safety.” Policymakers should consider moving away from criminal-legal responses to homelessness towards those that address structural determinants of homelessness and crime.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contribution

PP, CJ, SKN, and JAB conceptualized the project and developed the study design. PP developed the coding algorithm and ran the analyses. All authors contributed to the development of variables. PP and JAB wrote the first draft of the manuscript, and all authors critically reviewed the manuscript. All authors offered significant feedback on drafts of the paper and approved the final manuscript.

Funding

This research was supported by the National Institute on Drug Abuse (DP2DA051864 and K01DA051684).

Data Availability

Data are publicly available at the request of the city agencies.