Skip to main content
JAMA Network logoLink to JAMA Network
. 2022 Dec 5:e225460. Online ahead of print. doi: 10.1001/jamainternmed.2022.5460

Effect of Abandoned Housing Interventions on Gun Violence, Perceptions of Safety, and Substance Use in Black Neighborhoods

A Citywide Cluster Randomized Trial

Eugenia C South 1,, John M MacDonald 2, Vicky W Tam 3, Greg Ridgeway 4,5, Charles C Branas 6
PMCID: PMC9857286  PMID: 36469329

This cluster randomized controlled trial assesses whether abandoned house remediation reduces gun violence and substance-related outcomes and increases perceptions of safety and use of outdoor space in low-income, Black neighborhoods.

Key Points

Question

Do structural interventions to abandoned houses lead to improvements in health and safety in low-income, Black neighborhoods?

Findings

In this citywide cluster randomized controlled trial of 63 clusters containing 258 abandoned houses and 172 participants, abandoned houses that were remediated showed substantial drops in nearby weapons violations (−8.43%), gun assaults (−13.12%), and to a lesser extent shootings (−6.96%). Substance-related outcomes were not reliably affected by the interventions, and no effect of either intervention was found for perceptions of safety or time outside for nearby residents.

Meaning

Abandoned house remediation was directly linked to reduced gun violence and may be considered in efforts to create safe and healthy communities.

Abstract

Importance

Structural racism has resulted in long-standing disinvestment and dilapidated environmental conditions in Black neighborhoods. Abandoned houses signal neglect and foster stress and fear for residents, weakening social ties and potentially contributing to poor health and safety.

Objective

To determine whether abandoned house remediation reduces gun violence and substance-related outcomes and increases perceptions of safety and use of outdoor space.

Design, Setting, and Participants

This cluster randomized trial was conducted from January 2017 to August 2020, with interventions occurring between August 2018 and March 2019. The study included abandoned houses across Philadelphia, Pennsylvania, and surveys completed by participants living nearby preintervention and postintervention. Data analysis was performed from March 2021 to September 2022.

Interventions

The study consisted of 3 arms: (1) full remediation (installing working windows and doors, cleaning trash, weeding); (2) trash cleanup and weeding only; and (3) a no-intervention control.

Main Outcomes and Measures

Difference-in-differences mixed-effects regression models were used to estimate the effect of the interventions on multiple primary outcomes: gun violence (weapons violations, gun assaults, and shootings), illegal substance trafficking and use, public drunkenness, and perceptions of safety and time outside for nearby residents.

Results

A master list of 3265 abandoned houses was randomly sorted. From the top of this randomly sorted list, a total of 63 clusters containing 258 abandoned houses were formed and then randomly allocated to 3 study arms. Of the 301 participants interviewed during the preintervention period, 172 (57.1%) were interviewed during the postintervention period and were included in this analysis; participants were predominantly Black, and most were employed. Study neighborhoods were predominantly Black with high percentages of low-income households. Gun violence outcomes increased in all study arms, but increased the least in the full remediation arm. The full housing remediation arm, compared with the control condition, showed reduced weapons violations by −8.43% (95% CI, −14.68% to −1.19%), reduced gun assaults by −13.12% (95% CI, −21.32% to −3.01%), and reduced shootings by a nonsignificant −6.96% (95% CI, −15.32% to 3.03%). The trash cleanup arm was not associated with a significant differential change in any gun violence outcome. Instances of illegal substance trafficking and use and public drunkenness outcomes were not significantly affected by the housing remediation or trash cleanup treatment. Perceptions of neighborhood safety and time spent outside were unaffected by the intervention. The study arms did differ in a baseline characteristic and some preintervention trends, which raises questions regarding other potential nonmeasured differences between study arms that could have influenced estimates. No evidence of displacement of gun violence outcomes was found.

Conclusions and Relevance

In this cluster randomized controlled trial among low-income, predominantly Black neighborhoods, inexpensive, straightforward abandoned housing remediation was directly linked to significant relative reductions in weapons violations and gun assaults, and suggestive reductions in shootings.

Trial Registration

isrctn.org Identifier: ISRCTN14973997

Introduction

Racial segregation is a defining feature across neighborhoods in the US. Segregation, the result of structural racism embedded in historical and ongoing government and private-sector policies, has resulted in decades of disinvestment in Black, urban communities. Lack of investment has led to widespread deterioration of neighborhood environmental conditions, including a proliferation of abandoned houses marked by crumbling facades, dilapidated or absent doors and windows, and/or broken plywood boards and the accumulation of trash nearby. Abandoned houses potentially pose a major health and safety threat for residents of the predominantly Black neighborhoods in which they are concentrated.

There are several mechanisms through which abandoned houses may harm health, including from gun violence and substance use. Abandoned houses are often easy to enter and can be a location of illegal activity, including the storage of firearms. Residents point to these spaces as fracturing social ties between neighbors, engendering fear and feelings of neglect, and contributing to poor mental health, that in turn may contribute to a social environment with low collective efficacy in which unchecked violence can proliferate. Finally, these spaces contribute to a stressful neighborhood environment, which may be an antecedent to the escalation of disputes involving firearms, as well as substance trafficking and use.

Prior research indicates that health and safety can be improved in Black neighborhoods using place-based interventions. Vacant lot cleaning and greening, for example, results in significant decreases to gun violence and fear, as well as improvements in social cohesion and mental health for nearby residents. The introduction of greater street lighting has also been shown to significantly reduce violent crime. A prior quasiexperimental analysis of abandoned house remediation—installing working doors and windows in all structural openings—was associated with a drop in gun violence, but to our knowledge, no randomized controlled trial evidence exists. Given this, we conducted the first citywide cluster randomized trial to test an inexpensive, structural abandoned housing intervention and its effects on gun violence, substance use–related outcomes, and perceptions of safety and use of outdoor space for nearby residents.

Methods

Study Design

A cluster randomized controlled trial of abandoned house remediation was conducted citywide in Philadelphia, Pennsylvania. This trial was approved by the University of Pennsylvania Institutional Review Board and registered with ISRCTN registry (ISRCTN14973997) while follow-up data were being collected and prior to the start of data analysis (trial protocol and statistical analysis plan in Supplements 1 and 2). This study was reported using the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline for the reporting of cluster randomized trials.

Cluster Formation

A master list was compiled of all abandoned houses from the Philadelphia Department of Licenses and Inspections and Office of Property Assessment. Houses were eligible if they violated the city’s Doors and Windows Ordinance and were excluded if no owner of record was identified or it was determined to be a vacant lot.

The list of eligible abandoned houses (n = 3265) was randomly sorted to ensure that cluster formation was randomized. Next, to form clusters, which was the unit for random assignment, the first abandoned house on the randomly sorted list was chosen as an index house, and a 0.125-mile radius buffer was generated around its centroid. All other eligible abandoned houses within this radius were removed from consideration as future index houses. Of these eligible abandoned houses in the cluster radius, 3 to 5 were used to form a cluster. Some clusters had abandoned houses that were not included in the study due to the high concentration of abandoned houses in the City of Philadelphia. This process then cycled to the next abandoned house on the randomly sorted list that was at least 0.25 miles away from the edge of prior clusters, guaranteeing no clusters overlap, until all clusters (n = 63) were formed. This process was stratified by each of 4 geographically distinct city sections to ensure that clusters were evenly spread across the city. At the end of this process, 258 abandoned houses were included in the trial, and the remaining 3007 on the master list were not in the trial.

Cluster Random Assignment

Within each city section, clusters were then randomly assigned to 1 of 3 trial arms: (1) full housing remediation, (2) trash cleanup, or (3) a no-intervention control condition (Figure 1). Author V.T. generated the master list, assigned the random allocation sequence, and randomly assigned clusters to trial arms. See eMethods in Supplement 3 for information about the repeat randomization procedure.

Figure 1. Before and After Example Photographs of 3 Trial Arms: Abandoned Housing Remediation, Trash Cleanup, and No-Intervention Control.

Figure 1.

Photo credits: A, Philadelphia Redevelopment Authority; B, Pennsylvania Horticultural Society; C, Penn Staff.

Of the abandoned houses initially selected, 45% were publicly owned and 55% were privately owned. For publicly owned properties, 100% of those randomly selected were enrolled in the trial. For privately owned properties, the research team used available city records to contact the house owner(s) to obtain consent for remediation. Ultimately, 9 privately owned homes assigned to full remediation were enrolled. Privately owned properties for which we were unable to obtain consent due to the owner being deceased, not responding, or not giving consent were replaced with publicly owned houses within the same cluster. Privately owned houses in the trash cleanup and control arms were fully retained in the trial since no permissions were needed and the physical structure and treatment of privately owned vs publicly owned houses were functionally the same.

Abandoned Housing Interventions and Control Group

The full housing remediation intervention involved (1) installing working windows and doors on the front and sides of each house, (2) removal or repair of deteriorated structures on the front facades, (3) cleaning trash and weeds in front of the house, and (4) maintaining a clean facade, trash cleanup, and weeding over the postintervention period. Maintenance was performed every 2.5 months on average during the postintervention period. Ten houses required reinstallation of windows and doors that were stolen. A single house required the repair of a deteriorated structure (porch).

The trash cleanup intervention involved removal of debris and weeds in front of the house, including the sidewalk, and graffiti, which was reported to the city for removal every 1.8 months on average during the postintervention period. No intervention took place at control sites. The trash cleanup and house remediation interventions occurred between August 2018 and March 2019. We examined data for 18 months before (preintervention) and 18 months after (postintervention) intervention implementation.

Random Sampling of Participants

Two preintervention home-based interview surveys in study clusters were conducted from June 10 to December 5, 2017, and from December 6, 2017, to June 26, 2018, and 2 postintervention home-based surveys were conducted from August 24, 2018, to April 17, 2019, and from June 1, 2019, to February 1, 2020. All participants completed at least 1 preintervention survey and 1 postintervention survey. A field team identified, obtained written consent from, and interviewed 5 participants per cluster. Participants were 18 years of age or older and were compensated $25 per interview. Both the field team and participants were blinded to the intervention.

Outcomes

Publicly available Philadelphia Police Department databases provided data for recorded weapons violations, gun assaults, shooting incidents, and incidents of illegal substance trafficking and use (illegal substance) and public drunkenness that occurred between January 2017 and August 2020 (see eMethods in Supplement 3). To determine the effect of the interventions on each outcome, we used the latitude–longitude locations of each incident to calculate a kernel density estimate of the smoothed monthly rate of each outcome per square mile at the centroid of each abandoned house. Kernel densities are a well-established approach that assigned more weight to incidents that occurred closer to a house and less weight to incidents that occurred farther from a house.

To evaluate perceptions of safety, participants were asked to strongly agree, agree, disagree, or strongly disagree with the following statement bounded within the past 30 days: “My neighborhood is safe.” To evaluate time outside, participants were asked if they always, often, sometimes, rarely, or never “spent time hanging out, relaxing, or socializing on porches, stoops, or front yards in [their] neighborhood” over the past 30 days.

Statistical Analyses

We used difference-in-differences mixed-effects regression models to estimate the effect of the housing remediation and trash cleanup interventions on each outcome. The unit of analysis was individual abandoned houses (i). Our models analyzed the relationship between the interventions and the kernel density weighted estimates of each outcome (Y) around each house. Difference-in-differences is a common approach with group randomized trials to account for the limited opportunity to randomly distribute all potential sources of bias evenly since the intervention operates at the group level.

Difference-in-differences estimates assume that the difference in average outcomes between the treatment and control arms is constant across periods, or that they have parallel trends. We first tested this assumption by first graphing the linear trends in each trial arm in the preperiod for each outcome. We then evaluated interaction terms for the following regression model:

Yit = β0 + β1Housing Remediationi + β2Trash Cleanupi + β3t + β4Trash Cleanupi × t + β5Housing Remediationi × t + δs(i) + ξc(i)

If β4 and β5 differ significantly from 0, that would constitute evidence of a violation of the parallel trends assumption.

Following the check for parallel trends, we estimated all outcomes using the following regression model:

Yit = β0 + β1Housing Remediationi + β2Trash Cleanupi + β3Postt + β4Housing Remediationi × Postt + β5Trash Cleanupi × Postt6t + δs(i) + ξc(i)

In each regression model, we included terms for each intervention arm (β1 and β2) to control for preintervention differences relative to the no-intervention control study arm; a term Postt, which is a 0/1 indicator that time (t) is in the postintervention period; and interaction terms whose coefficients provided the difference-in-differences estimate of the effect of each intervention during the postintervention period (with coefficients β4 and β5, respectively) relative to the preintervention period. Assuming the difference in outcomes between treatment arms would remain constant in the absence of treatment, the difference-in-differences estimates of the interventions during the postintervention period are the causal estimates of the treatment. The regression models also included a fixed effect for the linear time trend, β6, where t denotes the month (−18, …, 18); differences by region of the city using indicator terms, δs(i), where s(i) denotes abandoned house i’s city section. The regression model controls for dependence caused by cluster randomization by including a group-level random effect term, ξc(i), where c(i) denotes the cluster index for abandoned house i (1, …, 63). We compute and report cluster-robust standard errors for all analyses. For point-based estimates, we used a linear regression model because the kernel densities are weighted counts and not integers. For the perceptions of safety and time spent outside, the regression model did not include a linear time trend.

We performed additional robustness checks on the parallel trends assumption by estimating a trend-adjusted difference-in-differences model that allows for separate linear trends for each study arm (see eMethods in Supplement 3). Please see eMethods in Supplement 3 for sample size calculations, additional robustness checks, and sensitivity analyses.

For ease of interpretation, we converted coefficients from the linear models into the percentage change in outcomes. To do so, we took the difference-in-difference estimate and divided it by the expected postperiod outcome in the absence of a treatment effect. For the Poisson models, we converted the incident rate ratios into percentage changes. All P values were 2 tailed. In addition to P values, we report adjusted P values (q values) based on probability of false discovery given multiple outcome tests, a decision that was made prior to the start of data analysis. Cluster creation and data analyses were conducted using ArcGIS, version 10.1 (ESRI) and Stata, version 15.1 (StataCorp LLC). All results are relative changes in outcomes unless otherwise specified as absolute changes.

Results

Abandoned Housing Clusters, Neighborhood, and Participant Characteristics

An initial master list included 3630 abandoned houses, 3265 (89.9%) of which were deemed eligible for study inclusion. A total of 63 clusters (containing 258 abandoned houses) were enrolled into the trial and randomly allocated to 1 of 3 study arms: full housing remediation (23 clusters with 58 houses), trash cleanup (20 clusters with 93 houses), and control (20 clusters with 107 houses) (Figure 2). The study neighborhoods were predominantly Black with a high percentage of low-income households and high unemployment rates (Table 1). All covariates had a small standardized mean difference across arms and small eta-squared, except for race and ethnicity, which had between small and medium standardized mean difference (eTable 1 in Supplement 3).

Figure 2. Abandoned Housing Selection and Random Allocation.

Figure 2.

aOPA indicates Office of Property Assessment. An invalid ID meant we were either unable to locate the owner of record or identify a parcel polygon in the spatial data.

bBecause of original clusters dropping out due to difficulty obtaining permission from private owners, 6 clusters did not have participants enrolled.

Table 1. Baseline Census Block Group Characteristics for Study Clusters Demonstrating Balance Across the 3 Trial Arms.

Characteristic Full house remediation Trash cleanup intervention No-intervention control DHa DTa
Abandoned house clusters
No. 23 20 20 NA NA
Total study houses, No. 58 93 107 NA NA
Study houses per cluster, mean (SD) 2.52 (1.20) 4.65 (1.73) 5.35 (2.25) NA NA
Resident population, mean (SD), people 1132.87 (540.26) 881.45 (357.22) 1023.80 (481.88) .11 .14
Properties, No. (SD) 181.74 (50.84) 198.10 (54.49) 190.15 (52.71) .06 .14
Median household income, mean (SD) 22 029.86 (9731.31) 24 321.00 (10 221.43) 23 055.47 (7453.23) .05 .06
Unemployment rate, % (SD) 18.56 (10.30) 17.38 (11.63) 20.85 (11.82) .09 .14
Serious crimes,b mean (SD), crimes 1118.15 (509.52) 1251.58 (559.75) 1082.01 (383.30) .03 .16
Market value of parcels, mean (SD), $ 55 953.31 (66 227.77) 49 341.47 (61 073.99) 53 633.95 (56 043.63) .01 .03
Race and ethnicity, % (SD)
Black, non-Hispanic 76.79 (25.21) 75.43 (21.46) 87.61 (9.63) .25 .27
Hispanic 11.63 (21.39) 13.05 (19.76) 5.13 (8.15) .17 .20
White, non-Hispanic 9.12 (13.65) 7.47 (8.85) 3.69 (5.15) .25 .17
Participants
No. 48 67 57 NA NA
Age, mean (SD), y 53.3 (13.9) 49.4 (16.9) 47.8 (15.3) .019 .152
Race and ethnicity, No. (%)
Black 40 (83.3) 60 (89.5) 56 (98.2) .017 .141
Hispanic 6 (12.5) 5 (7.5) 1 (1.8)
White 2 (4.2) 1 (1.5) 0
Other 0 1 (1.5) 0
Education, No. (%)
<High school 12 (25.0) 15 (23.0) 10 (17.5) .009 .010
High school 23 (48.0) 34 (52.3) 32 (56.1)
Any college 13 (27.0) 16 (23.8) 15 (26.3)
Employment status, No. (%)
Employed 40 (83.3) 54 (80.6) 48 (84.2) .001 .010
Unemployed 8 (16.7) 13 (19.4) 9 (15.8)

Abbreviation: NA, not applicable.

a

D = standardized difference of means for H (housing remediation) and T (trash cleanup) to control are either small (0.2) or between small and medium (0.5).

b

Serious crimes include part I violent and property crimes. Violent crimes include criminal homicide, rape, robbery, and aggravated assault. Property crimes including breaking and entering, larceny-theft, and motor vehicle theft.

A total of 301 participants were interviewed during the preintervention period, and 172 (57.1%) of these original participants were interviewed during the postintervention period and are included in this analysis (Figure 2). This amounted to a 42.9% loss to follow-up. The demographics of participants included in analysis compared to those lost to follow-up did not differ by more than 0.20 SDs. Of the 301 participants, 92 (30.6%) were in the full remediation arm, 107 (35.5%) in the trash cleanup arm, and 102 (33.9%) in the no-intervention control. Across the 3 trial arms, participants were majority Black and employed (Table 1).

Gun Violence Outcomes

Preperiod trends in the gun violence outcomes did not significantly differ between study arms (eTable 2 and eFigure 1 in Supplement 3). Table 2 shows the unadjusted kernel density estimates of weapons violations, gun assaults, and shootings for each trial arm in the preintervention and postintervention periods.

Table 2. Unadjusted Average Monthly Count of Outcomes During the Preintervention and Postintervention Time Periods for Each Trial Arm and Absolute Change Between Time Periods.

Outcome Full house remediation (cluster n = 23, houses n = 58) Trash cleanup intervention (cluster n = 20, houses n = 93) No-intervention control (cluster n = 20, houses n = 107)
Pre (N = 1044)a Post (N = 1036)a Change Pre (N = 1674)a Post (N = 1674)a Change Pre (N = 1926)a Post (N = 1926)a Change
Primary kernel density analyses, meanb
Weapons violations 19.97 20.78 0.81 22.07 24.49 2.42 19.91 22.62 2.71
Gun assaults 4.47 5.02 0.55 4.81 5.68 0.87 4.43 5.75 1.32
Shootings 2.75 3.09 0.34 2.89 3.56 0.67 2.69 3.27 0.58
Illegal substance 14.42 15.76 1.34 32.66 39.82 7.16 14.54 14.94 0.4
Public drunkenness 0.39 0.82 0.43 0.29 1.14 0.85 0.15 0.78 0.63
Cluster n = 17, participants n = 48 Cluster n = 20, participants n = 67 Cluster n = 20, participants n = 57
Pre Post Change Pre Post Change Pre Post Change
Perceptions of safety and time outside, mean
Neighborhood safe 2.53 2.64 0.11 2.40 2.63 0.23 2.53 2.59 0.06
Time outside 3.45 3.97 0.52 3.27 3.75 0.48 3.14 3.72 0.58
a

N = count of houses per month × 34 to 36 months included in the data set.

b

The kernel density can be interpreted as the average monthly rate of the outcome per square mile in the designated trial arm.

For the housing remediation arm compared with control, the difference-in-differences estimates showed that the monthly weapons violations during the postintervention period compared with the preintervention period was reduced by −8.43% (95% CI, −14.68% to −1.19%; P = .02), gun assaults were reduced by −13.12% (95% CI, −21.32% to −3.01%; P = .01), and shootings were reduced by a nonsignificant −6.96% (95% CI, −15.32% to 3.03%; P = .17) (Table 3; eFigure 2 in Supplement 3). For the trash cleanup arm compared with control, the difference-in-differences estimates showed a nonsignificant decrease in weapons violations by −1.20% (95% CI, −9.47% to 7.44%; P = .79), a nonsignificant reduction in gun assaults by −7.13% (95% CI, −17.04% to 4.93%; P = .24), and a nonsignificant increase in shootings by 2.26% (95% CI, −8.57% to 12.27%; P = .69).

Table 3. Difference-in-Differences (Primary and Trend Adjusted) and Displacement Results for Abandoned House Remediation and Trash Cleanup Clusters Compared With No-Intervention Control Clusters on Gun Violence Outcomes.

Outcome Full remediation (cluster n = 23, house n = 58, N = 2080) vs no-intervention control (cluster n = 20, house n = 107, N = 3852)a Trash cleanup (cluster n = 20, house n = 93, N = 3348) vs no-intervention control (cluster n = 20, house n = 107, N = 3852)a
Coefficient (95% CI)b Percent change, % (95% CI) P value q Value Coefficient (95% CI)b Percent change, % (95% CI) P value q Value
Primary kernel density analyses, gun violence outcomes
Weapons violations −1.91 (−3.57 to −0.25) −8.43 (−14.68 to −1.19) .02 .35 −0.30 (−2.56 to 1.97) −1.20 (−9.47 to 7.44) .79 >.99
Gun assaults −0.76 (−1.11 to −0.41) −13.12 (−21.32 to −3.01) .01 .35 −0.44 (−1.17 to 0.30) −7.13 (−17.04 to 4.93) .24 >.99
Shootings −0.23 (−0.56 to 0.10) −6.96 (−15.32 to 3.03) .17 >.99 0.08 (−0.33 to 0.50) 2.26 (−8.57 to 12.27) .69 >.99
Primary kernel density analyses, substance-related outcomes
Illegal substance 0.87 (−2.09 to 3.84) 5.26 (−11.69 to 19.58) .56 >.99 6.76 (−5.12 to 18.65) 14.51 (−11.41 to 31.90) .27 >.99
Public drunkenness −0.20 (−0.71 to 0.31) −19.44 (−46.31 to 27.52) .44 >.99 0.23 (−0.64 to 1.10) 16.58 (−36.07 to 49.02) .61 >.99
Survey outcomes (participants n = 172)
Perception of safety 0.06 (−0.18 to 0.30) 2.1 (−6.5 to 10.2) .64 >.99 0.15 (−0.06 to 0.36) 5.3 (−2.5 to 12.1) .18 >.99
Time outside −0.15 (−0.65 to 0.34) −3.6 (−14.1 to 8.2) .57 >.99 −0.11 (−0.56 to 0.34) −2.9 (−13.0 to 8.2) .62 >.99
Trend adjusted model, gun violence outcomes
Weapons violations −1.42 (−4.54 to 1.70) −6.39 (−17.91 to 7.56) .37 >.99 0.65 (−2.74 to 4.05) 2.59 (−10.09 to 14.20) .70 >.99
Gun assaults −0.74 (−1.63 to 0.17) −12.78 (−24.59 to 3.20) .11 >.99 −0.01 (−1.07 to 1.06) −0.91 (−15.84 to 15.71) .99 >.99
Shootings 0.12 (−0.59 to 0.84) 3.96 (−16.05 to 21.49) .73 >.99 0.09 (−0.66 to 0.85) 2.58 (−15.68 to 19.28) .81 >.99
Trend adjusted model, substance-related outcomes
Illegal substance −3.15 (−8.40 to 2.11) −16.65 (−34.77 to 11.79) .24 >.99 −1.80 (−6.61 to 3.00) −4.33 (−14.23 to 7.01) .46 >.99
Public drunkenness −0.36 (−1.01 to 0.29) −30.69 (−55.27 to 25.92) .27 >.99 0.38 (−0.74 to 1.52) 25.10 (−39.45 to 56.91) .51 >.99
IRR (95% CI) Percent change, % (95% CI) P value q Value IRR (95% CI) Percent change, % (95% CI) P value q Value
Secondary displacement analyses, proximal zone (0-330 ft)
Weapons violations 0.71 (0.55 to 0.93) −28.56 (−44.82 to −7.52) .01 .16 0.85 (0.60 to 1.22) −14.80 (−40.31 to 21.62) .38 >.99
Gun assaults 0.77 (0.44 to 1.34) −23.37 (−56.21 to 34.09) .35 >.99 0.89 (0.52 to 1.51) −11.45 (−48.02 to 50.87) .66 >.99
Shootings 0.95 (0.51 to 1.78) −4.80 (−49.02 to 77.84) .88 >.99 0.94 (0.46 to 1.92) −6.24 (−54.13 to 91.66) .86 >.99
Secondary displacement analyses, displacement zone (330-660 ft)
Weapons violations 0.96 (0.81 to 1.14) −4.12 (−19.09 to 13.62) .63 >.99 0.97 (0.75 to 1.18) −3.13 (−20.54 to 18.10) .75 >.99
Gun assaults 0.96 (0.68 to 1.35) −4.54 (−20.76 to −32.49) .79 >.99 0.78 (0.57 to 1.08) −21.79 (−43.40 to 8.08) .14 >.99
Shootings 0.80 (0.48 to 1.34) −19.70 (−51.99 to 34.30) .40 >.99 0.86 (0.49 to 1.53) −13.76 (−51.42 to 53.42) .14 >.99
a

N = count of houses per month × months included in the analysis.

b

Coefficients for the primary analyses correspond to β4 and β5 shown in the model equation in Methods. The trend adjusted analyses report coefficients as the average of the postintervention αts and γts.

When comparing the trend-adjusted difference-in-differences estimates with the primary analysis (Table 3), the point estimates for weapons and gun assault outcomes for the housing remediation arm are in the same direction, but the effects are partially attenuated; they are also less precise given the additional uncertainty that trend extrapolation introduces into the difference-in-differences estimates.

Substance-Related Outcomes

Significant differences in preperiod trends were found among the substance-related outcomes across study arms (eTable 2 and eFigure 1 in Supplement 3). The kernel density estimates for substance related outcomes went up from the preintervention to postintervention time periods in all arms of the trial. For the housing remediation arm compared with control, the difference-in-differences estimates showed that the monthly illegal substance incidents during the postintervention period compared with the preintervention period was increased nonsignificantly by 5.26% (95% CI, −11.69% to 19.58%; P = .56). Monthly illegal substance incidents around abandoned houses in the trash clean arm compared with control houses increased by a nonsignificant 14.51% (95% CI, −11.41% to 31.90%; P = .27). Public drunkenness showed a nonsignificant reduction for the house remediation arm compared with control and a nonsignificant increase for the trash cleanup arm compared with control. When comparing the trend-adjusted difference-in-differences estimates to the primary analysis (Table 3), the illegal substance estimates changed direction, suggesting that the difference in preperiod trends in this outcome may have been responsible for the nonsignificant increases associated with the housing remediation and trash cleanup arms that were observed in the main analysis.

Safety Perceptions and Time Outside Outcomes

For the housing remediation arm compared with control, the difference-in-differences estimates showed a nonsignificant increase of 2.1% (95% CI, −6.5% to 10.2%; P = .64) in perceptions of the neighborhood being unsafe and a nonsignificant 3.6% decrease in time spent outside (95% CI, −14.1% to 8.2%; P = .57). For the trash cleanup compared with control, participants reported a nonsignificant increase of 5.3% (95% CI, −2.5% to 12.1%; P = .18) in perceptions of the neighborhood being unsafe and a nonsignificant 2.9% decrease in time spent outside (95% CI, −13.0% to 8.2%; P = .62).

Additional Robustness Checks, Displacement Analysis, and Sensitivity Analyses

Overall, robustness checks provided confidence in the gun violence outcomes and suggest that the nonsignificant changes in substance-related outcomes are likely driven by difference in preperiod trends and regression to the mean (eTables 2 and 3 in Supplement 3). Sensitivity analysis suggested minimal effect of restricting analysis to publicly owned houses or staggered rollout (eTables 4 and 5 in Supplement 3). No evidence of displacement of gun violence outcomes was found (Table 3; eTable 6 in Supplement 3).

Discussion

In this citywide cluster randomized trial of abandoned house interventions, full house remediation was linked to reductions in gun violence outcomes. Gun violence increased in all arms of the trial, reflecting citywide gun violence increases over the study period, but increased less around houses that received the housing remediation. These findings were not explained by preexisting differences in trends between study arms or displacement of gun violence to adjacent areas. The study arms did differ in some baseline levels and trends, which raises questions regarding other potential nonmeasured differences between study arms that could have affected the gun violence outcomes. Because illegal substance and public drunkenness outcomes were inconsistent across robustness checks, most notably violating the assumption of parallel preperiod trends, in these instances nonsignificant changes should not be interpreted as causal estimates. Nevertheless, the results for gun violence outcomes suggest that remediation of abandoned houses is a novel strategy that may be useful for improving community safety and health and is worthy of further exploration.

The abandoned houses that were randomly sampled citywide in this trial were in predominantly Black neighborhoods marked by low median household incomes and high rates of unemployment. Racial and economic segregation is the result of structural racism embedded in historical and ongoing government and private-sector policies that underpin the root causes of poor health and gun violence in US cities. Such conditions produce persistently challenged socioecological contexts, including widespread deterioration of physical environments in which gun violence is able to proliferate and community health is difficult to achieve. While abandoned housing remediation does not directly address the policies leading to segregation, this intervention does influence an upstream, structural contributor to gun violence.

Limitations

This study has limitations. First, the abandoned houses in the remediation arm were predominantly publicly owned. However, there was little functional difference between publicly and privately owned abandoned houses in the study as they all were in violation of city code and often co-located on the same city streets. The sensitivity analysis showed there would be little change in the results if we removed the privately owned properties from the analysis. A second limitation was that although all neighborhoods in the study had high percentages of Black residents, the control arm had the highest. All neighborhoods, however, were still highly segregated. Economic indicators—unemployment and median household income, which are markers of concentrated disadvantage and the legacy of structural racism—were similar across all arms. Finally, while we found no displacement of outcomes in the immediate areas surrounding study abandoned houses, we are unable to determine if these outcomes were shifted across larger geographies.

Conclusions

This cluster randomized trial adds experimental evidence to the idea that structural, scalable, and sustainable changes to neighborhood environments can potentially promote community safety. Abandoned housing remediation may help prevent gun violence in low-income Black neighborhoods that have been deeply challenged by long-standing disinvestment. Additional replication studies are warranted in follow-up to this first citywide randomized controlled trial of abandoned housing remediation.

Supplement 1.

Trial Protocol

Supplement 2.

Trial Protocol and Statistical Analysis Plan

Supplement 3.

eMethods.

eResults.

eFigure 1. Visual inspection of pre-period parallel trends

eFigure 2. Percent reduction in three gun-violence outcomes for Housing Remediation and Trash Cleanup interventions compared to control

eTable 1. Baseline census block group characteristics for study clusters demonstrating balance across the three trial arms including eta-squared

eTable 2. Test of parallel pre-intervention trends

eTable 3. Robustness check to guard against mean reversion

eTable 4. Sensitivity analysis including calendar month as a fixed effect to account for staggered rollout of intervention timeline for abandoned house remediation and trash cleanup clusters compared to no-intervention control clusters on gun violence and substance-related outcomes

eTable 5. Sensitivity analysis demonstrating difference-in-differences results for abandoned house remediation and trash cleanup clusters compared to no-intervention control clusters on gun violence and substance-related outcomes only for publicly owned properties

eTable 6. Unadjusted total counts of outcomes during the pre- and post-intervention time periods for each trial arm in the proximal and displacement zones, and absolute change from pre- to post-intervention time periods

eReferences.

Supplement 4.

Data Sharing Statement

References

  • 1.Massey D. American apartheid: segregation and the making of the underclass. Am J Sociol. 1990;96(2):329-357. doi: 10.1086/229532 [DOI] [Google Scholar]
  • 2.Bailey ZD, Krieger N, Agénor M, Graves J, Linos N, Bassett MT. Structural racism and health inequities in the USA: evidence and interventions. Lancet. 2017;389(10077):1453-1463. doi: 10.1016/S0140-6736(17)30569-X [DOI] [PubMed] [Google Scholar]
  • 3.Sampson RJ, Wilson WJ, Katz H. Reassessing “Toward A Theory of Race, Crime, and Urban Inequality”: enduring and new challenges in 21st century America. Du Bois Rev. 2018;15(1):13-34. doi: 10.1017/S1742058X18000140 [DOI] [Google Scholar]
  • 4.Krivo LJ, Peterson RD, Kuhl DC. Segregation, racial structure, and neighborhood violent crime. AJS. 2009;114(6):1765-1802. doi: 10.1086/597285 [DOI] [PubMed] [Google Scholar]
  • 5.Econsult Corporation . Vacant land management in Philadelphia: the costs of the current system and the benefits of reform. Published 2010. Accessed October 25, 2022. https://penniur.upenn.edu/publications/vacant-land-management-in-philadelphia-the-costs-of-the-current-system-and
  • 6.Branas CC, Rubin D, Guo W. Vacant properties and violence in neighborhoods. ISRN Public Health. 2013;2012:246142. doi: 10.5402/2012/246142 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Hembree C, Galea S, Ahern J, et al. The urban built environment and overdose mortality in New York City neighborhoods. Health Place. 2005;11(2):147-156. doi: 10.1016/j.healthplace.2004.02.005 [DOI] [PubMed] [Google Scholar]
  • 8.Cohen DA, Mason K, Bedimo A, Scribner R, Basolo V, Farley TA. Neighborhood physical conditions and health. Am J Public Health. 2003;93(3):467-471. doi: 10.2105/AJPH.93.3.467 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Accordino J, Johnson GT. Addressing the vacant and abandoned property problem. J Urban Aff. 2002;22(3):301-315. doi: 10.1111/0735-2166.00058 [DOI] [Google Scholar]
  • 10.MacDonald J. Community design and crime: the impact of housing and the built environment. Crime Justice. 2015;44(1):333-383. doi: 10.1086/681558 [DOI] [Google Scholar]
  • 11.Garvin E, Branas C, Keddem S, Sellman J, Cannuscio C. More than just an eyesore: local insights and solutions on vacant land and urban health. J Urban Health. 2013;90(3):412-426. doi: 10.1007/s11524-012-9782-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Sampson RJ, Raudenbush SW, Earls F. Neighborhoods and violent crime: a multilevel study of collective efficacy. Science. 1997;277(5328):918-924. doi: 10.1126/science.277.5328.918 [DOI] [PubMed] [Google Scholar]
  • 13.Bourgois P, Hart LK. Commentary on Genberg et al. (2011): the structural vulnerability imposed by hypersegregated US inner-city neighborhoods—a theoretical and practical challenge for substance abuse research. Addiction. 2011;106(11):1975-1977. doi: 10.1111/j.1360-0443.2011.03615.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.South EC, Kondo MC, Cheney RA, Branas CC. Neighborhood blight, stress, and health: a walking trial of urban greening and ambulatory heart rate. Am J Public Health. 2015;105(5):909-913. doi: 10.2105/AJPH.2014.302526 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Kondo MC, Andreyeva E, South EC, MacDonald JM, Branas CC. Neighborhood interventions to reduce violence. Annu Rev Public Health. 2018;39:253-271. doi: 10.1146/annurev-publhealth-040617-014600 [DOI] [PubMed] [Google Scholar]
  • 16.Branas CC, South E, Kondo MC, et al. Citywide cluster randomized trial to restore blighted vacant land and its effects on violence, crime, and fear. Proc Natl Acad Sci U S A. 2018;115(12):2946-2951. doi: 10.1073/pnas.1718503115 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.South EC, Hohl BC, Kondo MC, MacDonald JM, Branas CC. Effect of greening vacant land on mental health of community-dwelling adults: a cluster randomized trial. JAMA Netw Open. 2018;1(3):e180298. doi: 10.1001/jamanetworkopen.2018.0298 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Garvin EC, Cannuscio CC, Branas CC. Greening vacant lots to reduce violent crime: a randomised controlled trial. Inj Prev. 2013;19(3):198-203. doi: 10.1136/injuryprev-2012-040439 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Chalfin A, Hansen B, Lerner J, Parker L. Reducing crime through environmental design: evidence from a randomized experiment of street lighting in New York City. J Quant Criminol. 2022;38(1):127-157. doi: 10.1007/s10940-020-09490-6 [DOI] [Google Scholar]
  • 20.Kondo MC, Keene D, Hohl BC, MacDonald JM, Branas CC. A difference-in-differences study of the effects of a new abandoned building remediation strategy on safety. PLoS One. 2015;10(7):e0129582. doi: 10.1371/journal.pone.0129582 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Boruch R, May H, Turner H, et al. Estimating the effects of interventions that are deployed in many places: place-randomized trials. Am Behav Sci. 2004;47(5):608-633. doi: 10.1177/0002764203259291 [DOI] [Google Scholar]
  • 22.Philadelphia Police Department, Shooting Victims Data . Accessed March 11, 2021. https://www.opendataphilly.org/dataset/shooting-victims
  • 23.Branas CC, Cheney RA, MacDonald JM, Tam VW, Jackson TD, Ten Have TR. A difference-in-differences analysis of health, safety, and greening vacant urban space. Am J Epidemiol. 2011;174(11):1296-1306. doi: 10.1093/aje/kwr273 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Rosenblatt M. Remarks on some nonparametric estimates of a density function. Ann Math Stat. 1956;27(3):832-837. doi: 10.1214/aoms/1177728190 [DOI] [Google Scholar]
  • 25.Murray D. Design and analysis of group-randomized trials. Presented at: Director, NIH Office of Disease Prevention, OBSSR Summer Institute; July 20, 2015. Accessed October 25, 2022. https://prevention.nih.gov/sites/default/files/2018-06/2017-07-17%20OBSSR%20Summer%20Institute.pdf
  • 26.Murray DM, Taljaard M, Turner EL, George SM. Essential ingredients and innovations in the design and analysis of group-randomized trials. Annu Rev Public Health. 2020;41(1):1-19. doi: 10.1146/annurev-publhealth-040119-094027 [DOI] [PubMed] [Google Scholar]
  • 27.Benjamini Y, Drai D, Elmer G, Kafkafi N, Golani I. Controlling the false discovery rate in behavior genetics research. Behav Brain Res. 2001;125(1-2):279-284. doi: 10.1016/S0166-4328(01)00297-2 [DOI] [PubMed] [Google Scholar]
  • 28.Jacoby SF, Dong B, Beard JH, Wiebe DJ, Morrison CN. The enduring impact of historical and structural racism on urban violence in Philadelphia. Soc Sci Med. 2018;199:87-95. doi: 10.1016/j.socscimed.2017.05.038 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Fagan J, Davies G. The natural history of neighborhood violence. J Contemp Crim Justice. 2004;20(2):127-147. doi: 10.1177/1043986204263769 [DOI] [Google Scholar]
  • 30.Nardone A, Rudolph KE, Morello-Frosch R, Casey JA. Redlines and greenspace: the relationship between historical redlining and 2010 greenspace across the United States. Environ Health Perspect. 2021;129(1):17006. doi: 10.1289/EHP7495 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Nardone AL, Casey JA, Rudolph KE, Karasek D, Mujahid M, Morello-Frosch R. Associations between historical redlining and birth outcomes from 2006 through 2015 in California. PLoS One. 2020;15(8):e0237241. doi: 10.1371/journal.pone.0237241 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Branas CC, Macdonald JM. A simple strategy to transform health, all over the place. J Public Health Manag Pract. 2014;20(2):157-159. doi: 10.1097/PHH.0000000000000051 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.South EC, MacDonald J, Reina V. Association between structural housing repairs for low-income homeowners and neighborhood crime. JAMA Netw Open. 2021;4(7):e2117067. doi: 10.1001/jamanetworkopen.2021.17067 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Keizer K, Lindenberg S, Steg L. The spreading of disorder. Science. 2008;322(5908):1681-1685. doi: 10.1126/science.1161405 [DOI] [PubMed] [Google Scholar]
  • 35.Burt CJ, Kondo MC, Hohl BC, et al. Community greening, fear of crime, and mental health outcomes. Am J Community Psychol. 2022;69(1-2):46-58. doi: 10.1002/ajcp.12544 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Hohl BC, Kondo MC, Kajeepeta S, et al. Creating safe and healthy neighborhoods with place-based violence interventions. Health Aff (Millwood). 2019;38(10):1687-1694. doi: 10.1377/hlthaff.2019.00707 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement 1.

Trial Protocol

Supplement 2.

Trial Protocol and Statistical Analysis Plan

Supplement 3.

eMethods.

eResults.

eFigure 1. Visual inspection of pre-period parallel trends

eFigure 2. Percent reduction in three gun-violence outcomes for Housing Remediation and Trash Cleanup interventions compared to control

eTable 1. Baseline census block group characteristics for study clusters demonstrating balance across the three trial arms including eta-squared

eTable 2. Test of parallel pre-intervention trends

eTable 3. Robustness check to guard against mean reversion

eTable 4. Sensitivity analysis including calendar month as a fixed effect to account for staggered rollout of intervention timeline for abandoned house remediation and trash cleanup clusters compared to no-intervention control clusters on gun violence and substance-related outcomes

eTable 5. Sensitivity analysis demonstrating difference-in-differences results for abandoned house remediation and trash cleanup clusters compared to no-intervention control clusters on gun violence and substance-related outcomes only for publicly owned properties

eTable 6. Unadjusted total counts of outcomes during the pre- and post-intervention time periods for each trial arm in the proximal and displacement zones, and absolute change from pre- to post-intervention time periods

eReferences.

Supplement 4.

Data Sharing Statement


Articles from JAMA Internal Medicine are provided here courtesy of American Medical Association

RESOURCES