The Association Between Vacant Lot Redevelopment and Violent/Firearm Violent Crime: A Difference-in-Difference Analysis From 2007 to 2023

Abstract

Objectives. To investigate the association between vacant lot redevelopment and violent crime and firearm violent crime.

Methods. We used a quasi-experimental study with a difference-in-difference (DID) design. The study population was 254 vacant lots located in Philadelphia, Pennsylvania. The exposure was redevelopment, defined as repurposing the vacant lot into a permanent structure (e.g., housing) between 2007 and 2023. The outcome was violent crime subtypes and firearm violent crime subtypes reported to the police. Our primary analysis used a staggered DID estimator.

Results. We found a negative association between redevelopment and aggravated assault (DID = −56.55 crimes per square mile per year; 95% confidence interval [CI] = −97.86, −15.24), firearm aggravated assault (DID = −35.11 crimes per square mile per year; 95% CI = −55.41, −14.82), and overall firearm violent crime (DID = −27.26 crimes per square mile per year; 95% CI = −52.97, −1.54). We did not find significant associations for the other outcomes tested.

Conclusions. Our results support the hypothesis that vacant lot redevelopment may prevent or displace violent crime.

Public Health Implications. Built environment investments may prevent or displace violence. (Am J Public Health. 2026;116(1):95–102. https://doi.org/10.2105/AJPH.2025.308269)

Violence is a significant public health problem. The four most common violent crime categories are murder, rape, robbery, and aggravated assault.¹ In 2023, there were 1.2 million violent crimes reported in the United States.¹ Investment in the built environment by directly treating aspects of disinvested environments may prevent violence.^2,3

Lot vacancy is a treatable aspect of disinvestment.⁴ Many communities that experience high rates of gun violence also have a high volume of vacant lots and unmaintained land.⁵ Vacant lots are hypothesized to contribute to nearby violence through 2 complementary mechanisms: (1) vacant lots indicate some level of abandonment, leading to the perception that there are fewer eyes on the street and thus a safer place to engage in potentially violent activities, and (2) an unmaintained vacant lot is an open and relatively deserted space where drugs or weapons can be stored.^2,3,6,7 Vacant land restoration has therefore emerged as a potential strategy for violence prevention.

Most of the vacant land restoration strategies for crime prevention focus on vacant lot greening. Greening vacant lots by removing trash, mowing and landscaping, and adding community cues to care such as gardens, have been shown to reduce overall and violent crime and improve perceptions of neighborhood safety.^6,8–13 For example, a study in Philadelphia, Pennsylvania, found that the odds of homicide were significantly lower in neighborhoods with maintained and greened vacant lots relative to those that did not have maintained and greened lots.¹¹ Similarly, another study in Philadelphia using a randomized controlled trial design found that all crimes, gun assaults, and burglaries were significantly reduced after a vacant lot greening intervention.¹⁴

Investment in housing and neighborhoods has been shown to reduce neighborhood violence.^15–18 For example, a study found that in Philadelphia, a housing repair intervention was associated with a 21.9% reduction in total crime.¹⁵ It follows that redeveloping vacant lots into permanent structures such as housing may also reduce violence. However, perhaps because redevelopment is much more expensive on a per-lot basis than greening or repair, it has been less often studied as a violence prevention technique. Nonetheless, examining the effect of vacant lot redevelopment and violent crime and firearm violent crime could provide insight into potential mechanisms through which intervening on the built environment could help reduce crime in certain places.

We quantified the association between vacant lot redevelopment and violent crime and firearm violent crime reported to the police in Philadelphia from 2007 through 2023 using an analytic approach that accounted for staggered timing of vacant lot development. We hypothesized that there would be a reduction in violent crime and firearm violent crime as redevelopment removes the influence of abandonment and decreases the amount of open or deserted places.

METHODS

We conducted a quasi-experimental staggered difference-in-difference (DID) study to assess the relationship of vacant lot redevelopment with violent crime and firearm violent crime in Philadelphia from 2007 through 2023.

Study Population

We used the OpenDataPhilly Building Demolitions data set to identify vacant land parcels.¹⁹ We aimed to combine contiguous lots within the data set, as they are likely perceived as a single land parcel. To construct contiguous vacant lots, we used the lot coordinates to cluster the addresses based on 1 point of contact within 26.2 feet and verified a random subsample (n = 10) using Google Street View (Google, Mountain View, CA) to double check our parameters were accurate. This process identified 2117 unique vacant lots. We randomly selected 300 vacant lots and subsequently excluded vacant lots (n = 46) that (1) only had 0 or 1 Street View images such that we were unable to assess when the lot was redeveloped or (2) had sufficiently blurry images that we were unable to locate the lot or determine its status. Our final study population consisted of 254 vacant lots in the City of Philadelphia. In our study population, demolitions occurred between 2007 and 2013. Lots entered the study population when the demolition was complete.

Exposure

Our exposure was vacant lot redevelopment. Two researchers (NA and HYZ) visually audited each vacant lot in the study sample using available Street View imagery for all available time points to determine whether anything had been constructed on the lot of interest, with the goal of identifying the first image showing redevelopment. We double-coded 62 randomly selected vacant lots. We identified no coding differences in type of redevelopment or year of redevelopment. We identified 6 discrepancies related to visual concerns, where raters were not able to accurately view the lot because of blurry images or removal of images on Street View. We removed these 6 lots from our study population.

From these images, we identified 3 forms of redevelopment: housing or business structures, impermeable fences, and other permanent structures (which included parking lots, sheds, and garages). We excluded permeable fences (e.g., split rail fences typically installed as part of the Philadelphia Horticultural Society’s Land Care program¹⁴) as a form of redevelopment as they do not restrict access to the lot. We then created 2 exposure definitions of redevelopment for subsequent analyses. Our primary redevelopment definition tested the hypothesis that investment prevents or displaces violence. For this analysis, we defined redevelopment as new construction of housing or business structures, considering this to both visually indicate investment into the area and restrict public access to the no-longer-vacant lot. The secondary redevelopment definition tested the hypothesis that simply physically restricting access to the vacant lot (i.e., preventing drugs or weapons from being stored there) prevents or displaces violence. For this analysis, we defined redevelopment as fences (chain linked, wooden) and permanent structures (garages, sheds, or parking lots) in addition to housing or business structures, considering these to remove access to the lot, albeit with more variable signaling of investment. For both redevelopment definitions, we considered the lot to be redeveloped in the first year a Street View image showed redevelopment.

Outcome

Our outcome was violent crime and firearm violent crime reported to the police. We obtained georeferenced crime data from the Philadelphia Police Department from 2007 through 2023. Using police codes, we divided violent crime into the following four subtypes in line with the Uniform Crime Reporting Program: homicide, aggravated assault, robbery, and rape. We then separated firearm violent crime from 3 of the 4 subtypes to get a set of firearm violent crime subtypes: firearm homicide, firearm aggravated assault, and firearm robbery. There were no instances of firearm-related rape in our data set. Finally, we included an overall violent crime outcome, the sum of the 4 violent crime subtypes, and an overall firearm violent crime outcome, the sum of the 3 firearm violent crime subtypes.

We used kernel density estimation to generate a smoothed surface from georeferenced violent crime data for each crime type‒year combination. For each surface, we used a quartic (biweight) kernel and specified a bandwidth of 500 feet to reflect the hypothesized spatial influence of vacant lot redevelopment on nearby violent crime and to reflect the average size of a Philadelphia block.^20–22 We estimated surfaces at a 100-foot resolution and applied an edge correction to correct for edge cases.²³ The resulting surfaces estimate the density of events per square mile for a given year for each violent crime subtype and firearm violent crime subtype. The density surfaces for each crime type were spatially linked to each vacant lot. We generated kernel density surfaces in RStudio version 4.2.2 using the “spatstat” package (Posit, Boston, MA).²⁴

Statistical Analysis

We described Philadelphia census tract baseline characteristics using the 2010 American Community Survey data. Then, we estimated the association between redevelopment and violent crime and firearm violent crime densities using a DID design with a staggered estimator to account for redevelopment over different years. We estimated the average treatment effect after redevelopment using a binary variable with 1 for lots after redevelopment and 0 for lots before redevelopment or a lot that was never redeveloped. All regressions used ordinary least squares and included lot and year fixed effects. We implemented the heterogeneity-robust Gardner 2-stage estimator using the did2s package in RStudio.^25,26 We chose this approach as the Gardner 2-stage estimator uses all pretreatment data as a baseline, whereas other common estimator types use only the last pretreatment period as a baseline.^26,27 We felt incorporation of all pretreatment data was important given substantial year-over-year variation in violent crime and firearm violent crime. For ease of interpretation, we estimated relative change in addition to the absolute change in density.

A key identifying assumption for DID estimation is the parallel trends assumption—that the violence near lots in which redevelopment occurred would have continued along their same trajectories in the absence of redevelopment. Although the true parallel trends assumption is untestable, consistent with DID best practices, we can assess the parallel trends assumption by examining preintervention trends using event study analyses.^27,28 The event study evaluates the pretreatment dynamics between the treatment and control group by including an additional term that indicates the leads and lags of the treatment variable.²⁸ If violent crime and firearm violent crime rates for redeveloped and nonredeveloped lots were trending similarly before redevelopment, we expect estimated coefficients for event times before redevelopment to be small and statistically insignificant.

We conducted spillover analyses to measure immediate spatial displacement of crime to areas near the vacant lots included in this study. To do this, we used a “donut versus donut hole” buffer zone approach.¹⁴ First, we captured occurrence of crime as counts per year within a 152.4-meter bandwidth (∼500 feet) of vacant lots and then captured occurrence of crime as counts per year within 152.4 meters to 304.8 meters bandwidth (∼500 to 1000 feet) of vacant lots. We then used the same approach described for the primary analysis and estimated the association between redevelopment and police-reported violent crime and firearm violent crime outcomes using a DID design with a staggered estimator, ordinary least squares regression, and year and lot fixed effects (Table A, available as a supplement to the online version of this article at https://ajph.org).

We conducted all analyses in RStudio version 4.2.2. We conducted data analysis from March 2024 through December 2024. We determined statistical significance at the level of P < .05.

RESULTS

When we used the primary redevelopment definition (housing or business structures), 41 (16.1%) of lots were redeveloped. Lots that were redeveloped were clustered around Philadelphia’s Center City District (Figure 1). For the secondary redevelopment definition (housing or business structures, permanent structures, fences), 143 (59.3%) lots were redeveloped over the study period. Lots that were redeveloped were more likely to be areas with a higher average percentage of White respondents (20.5% vs 9.9%), higher average median income per year ($24 981 vs $21 156), and lower poverty percentage (38.6% vs 41.1%) compared with areas without vacant lot redevelopment (Table 1).

FIGURE 1—

Spatial Distribution of Redeveloped Lots and Vacant Lots in Philadelphia, PA: 2007‒2023

TABLE 1—

Census Tract Characteristics for Lots: Philadelphia, PA, 2010

Characteristica	Lot Statusb
	Vacant Lots	Redeveloped Lots
White, mean %	9.9	20.5
Poverty, mean %	41.1	38.6
Median yearly income, mean $	21 156	24 981

^a Characteristics pulled from the 2010 American Community Survey.

^b Each lot was attributed to the corresponding census tract, and the characteristics are averaged across census tract stratified by lot status. Vacant lots remained vacant throughout the entire study period, whereas redeveloped lots were redeveloped at some point during the study period.

Redevelopment and Violent Crime

For the primary redevelopment definition using housing or business structures, we found a statistically significant negative association between redevelopment and aggravated assault, firearm aggravated assault, and overall firearm violent crime, consistent with our hypothesis that redevelopment may prevent or displace violent crime and firearm violent crime (Table 2). When we compared the difference between postredevelopment and preredevelopment years among lots that were redeveloped with the difference during that time in lots that were not redeveloped, the average aggravated assault rate was 56.6 crimes per square mile per year lower (95% confidence interval [CI] = −97.9, −15.2) representing a roughly 30.1% decrease in aggravated assaults (95% CI = −29.0%, −31.2%). Similarly, the average firearm aggravated assault rate was 35.1 crimes per square miles per year lower (95% CI = −55.4, −14.8), or about a 30.8% decrease (95% CI = −29.0%, −32.6%), and the average overall firearm violent crime rate was 27.3 crimes per square mile per year lower (95% CI = −53.0, −1.5) or a 14.7% decrease (95% CI = −13.1%, −16.3%; Table 2). There were no significant associations between the primary redevelopment definition and other violent crime and firearm violent crime outcomes.

TABLE 2—

Difference-in-Difference (DID) Estimates of Violent Crime and Firearm Violent Crime per Square Mile per Year Comparing Redeveloped and Vacant Lots: Philadelphia, PA, 2007‒2023

Outcome	DID Estimatea (95% CI)
Primary redevelopment definition: housing or business structures b
Overall violent crime	−37.57 (−92.86, 17.73)
Aggravated assault	−56.55 (−97.86, −15.24)
Homicide	2.62 (−4.47, 9.71)
Rape	−0.05 (−6.71, 6.60)
Robbery	16.41 (−9.42, 42.25)
Overall firearm violent crime	−27.26 (−52.97, −1.54)
Firearm aggravated assault	−35.11 (−55.41, −14.82)
Firearm homicide	−3.30 (−7.11, 0.52)
Firearm robbery	11.15 (−4.59, 26.90)
Secondary redevelopment definition: housing or business structures, permanent structures, fences b
Overall violent crime	−8.48 (−52.00, 35.03)
Aggravated assault	−5.11 (−37.90, 27.69)
Homicide	−6.59 (−16.32, 3.15)
Rape	−5.75 (−13.42, 1.92)
Robbery	8.96 (−13.69, 31.61)
Overall firearm violent crime	11.42 (−16.98, 39.82)
Firearm aggravated assault	8.05 (−13.24, 29.34)
Firearm homicide	−6.99 (−13.95, −0.02)
Firearm robbery	10.36 (−3.46, 24.18)

^a DID regressions used Gardner’s 2-stage estimator, used ordinary least squares regression, and included year and lot fixed effects with standard errors clustered at the lot level. The unit of the DID estimates is crimes per square mile per year.

^b The primary redevelopment definition includes housing or business structures only. The secondary redevelopment definition includes housing or business structures, along with permanent structures and fences.

For the secondary redevelopment definition (housing or business structures, permanent structures, and fences), we found a statistically significant negative association between redevelopment and firearm homicide. The average firearm homicide rate was 7.0 crimes per square mile per year lower comparing postredevelopment and preredevelopment years among lots that were redeveloped contrasting with the change during that time in lots that were not redeveloped (95% CI = −14.0, −0.02) or a decrease of 29.8% (95% CI = −26.7%, −32.7%; Table 2). No other changes in violence were associated with our secondary redevelopment definition beyond what might be expected by chance.

Spillover analyses of the police-reported violent crime and firearm violent crime outcomes showed no indication of spatial displacement. In none of the outcomes did we see that there was a significant reduction in the 152.4-meter buffer surrounding vacant lots that was coupled with significant increases in the ring surrounding the vacant lots (Table A).

Event Study Findings

Event study specifications, which were used to evaluate the parallel trends assumption, are presented for the primary redevelopment definition in Figures A1 and A2 (available as a supplement to the online version of this article at https://ajph.org). Figures A1 and A2 show differences in the rates of violent crime and firearm violent crime before redevelopment, comparing vacant lot clusters that were redeveloped and those that remained vacant. The event study detected minor and variable differences in violent crime and firearm violent crime rates 4 years before redevelopment, though this estimate is small. However, the event study did not detect differences in violent crime and firearm violent crime in the other years before in lots that were redeveloped compared with lots that were not.

DISCUSSION

To our knowledge, this is the first study assessing the association between vacant lot redevelopment and violent crime and firearm violent crime. We found an association between vacant lot redevelopment and lower rates of aggravated assault, firearm aggravated assault, and overall firearm violent crime. However, our results did not hold when we included permanent structures and fences into the redevelopment definition (secondary redevelopment definition). This suggests that redevelopment of vacant lots into housing or business structures might hold a greater impact on aggravated assaults than more modest investments. More broadly, our results support the hypothesis that interventions that invest in neighborhoods (such as housing structures) may prevent or displace (that is, violent crime moving from one area to another) violent crime.

Our results are consistent with those from other studies exploring similar neighborhood violence-prevention mechanisms. One potential mechanism in which redevelopment may prevent or displace violence is through clear ownership of the lot. Replacing unsupervised lots with housing may force unwanted activities to relocate or be reduced,²⁹ suggesting more neighbors in place may increase collective efficacy, or the perception of mutual trust and willingness to help each other.^30,31 One study created a collective efficacy measure from neighborhoods in Chicago, Illinois, that reflected a combination of factors related to cohesion, social trust, shared experiences, and the willingness of neighborhood residents to act in support of these values, and found that areas with higher collective efficacy measures had lower rates of violence.³⁰

Another mechanism through which redevelopment may prevent or displace violence is through removing signs of abandonment. Findings from studies targeting this mechanism are mixed.^15,16,18,32 One study assessed abandoned housing and vacant land remediation through home repairs, full gut rehabilitation, and new builds in relation to all and violent crime in Columbus, Ohio, using synthetic control methods¹⁸ and found no significant effects on assaults or robberies across an 11-year time period.¹⁸ Another study assessing the 2011 Doors and Windows Ordinance in Philadelphia, which required property owners of abandoned buildings to install working doors and windows in all structural openings, found implementation of the ordinance was associated with a decrease in all assaults (−2% change) and firearm assaults (−4% change) over a 2-year time period.³²

Our secondary redevelopment definition, which focused on restricted access to the lot (i.e., by construction of housing, permanent structures, and fences rather than only housing), resulted in weaker findings than the primary redevelopment definition. We identified an association consistent with prevention of firearm homicides, which we were concerned may have arisen by chance as we had not specifically hypothesized an impact on homicide alone. It is possible that the differences between the primary and secondary redevelopment definitions represent housing or business redevelopment having a more consistent impact on violent crime compared with fences or permanent structures. One study assessing residents’ perceptions of the quality and benefits of green space from 2 neighborhoods in the Netherlands found that green areas that were accessible and usable mattered more than the mere amount of green space.³³ By analogy, the quality and utility of vacant lot redevelopment may impact violence differentially, which may explain the patterns we identified across redevelopment types.

Lastly, our study found no evidence of spatial spillover or displacement. This is in line with previous research that shows that given the methods we have available to capture displacement, displacement is often inconsequential.^34,35 Although we did not expect displacement to be present, our displacement analysis likely faces significant methodological problems in measuring displacement or diffusion.³⁴ Understanding and improving the methods used to quantify and assess spatial displacement is an important area for future research.

Limitations

First, our quasi-experimental design is subject to substantial assumptions that affect any causal interpretation. In particular, we observed that many of the redeveloped lots were in and around the Center City District of Philadelphia, whereas undeveloped lots were spread about the city. It is thus possible that redevelopment is occurring in a location in which violence was already dropping. While our parallel trends test did not indicate lack of parallelism and our staggered DID models account for time trends, an exchangeability violation could still lead to a biased estimate of the causal effect.

Second, we could not control for neighborhood block-level time-varying factors that could have changed over the study period differentially by lot—for example, if small-scale gentrification encourages redevelopment and displaces violent crime, residual confounding would still be present in our findings.

Third, as the range in which redevelopment might affect violence is not well defined, our analytic approach could not account for possible spatial spillover or displacement or relocation of violence to other areas. Although we did include a spillover analysis to assess displacement, these methods may not accurately capture displacement or diffusion.

Fourth, owing to lack of support for Poisson or negative binomial model specification in staggered DID software packages, we modeled violence density as a continuous variable—that is, our estimators assumed a Gaussian (normal) distribution, despite the fact that violence densities are constrained to be nonnegative. This likely did not affect our results for assaults, which are relatively common, but may have impacted our rarer outcomes, such as firearm homicides. For this reason, we have less confidence in our finding of the broader redevelopment definition preventing or displacing firearm homicide than we do in the narrower definition preventing or displacing assaults.

Fifth, our analysis only included violence reported to the police, and unreported violence may be common in areas with high vacancies. If redevelopment causes more violence to be reported, it would cause a bias toward the null.

Finally, although our event study was consistent with the hypothesis that violence near vacant lots before redevelopment was independent of whether a lot was developed, we cannot fully disentangle redevelopment processes from gentrification processes that affect neighborhood poverty, a root cause of violence.

Public Health Implications

Our work supports calls for sustainable interventions to prevent violence by modifying the built environment. The relationship between neighborhood physical environments and violent crime is of critical interest to public health and local governments seeking to address root causes of violence. These results may be useful to policymakers considering vacant land redevelopment as a component of neighborhood violence prevention.