Lessons Learned from a Citywide Abandoned Housing Experiment

Abstract

Problem, research strategy, and findings:

The negative impact of vacant and abandoned housing in city neighborhoods is extreme, affecting health and quality of life, promoting violence, and leading to further abandonment. One approach to addressing abandoned housing is to intervene with low-cost interventions that provide a visual sense of ownership. We tested whether a low-cost remediation of abandoned and vacant houses or a trash cleanup intervention would make a noticeable difference in the levels of nearby disrepair, disorder, and public safety. The abandoned housing remediation and trash cleanup interventions were a test of compliance with municipal ordinances. We used an experimental design to test the causal effects of the ordinances, and because the scale of abandonment was too large to provide treatment to all abandoned houses in the city. We used systematic social observation methods to rate changes in disrepair, disorder, and litter at housing sites and on the city blocks they were located, and police reported data on gun violence and illegal substance uses. Our experimental design allowed us to see if observed disrepair, disorder, and public safety improved after working windows and doors were installed on abandoned houses compared with a trash cleanup around properties or a no-intervention control condition. Our results showed significant changes in observed disrepair, disorder, and gun violence and illustrate the benefits of experimental evaluations of place-based changes to the built environment.

Takeaway for practice:

Improving compliance with ordinances to remediate abandoned housing can make a noticeable difference in disrepair in neighborhoods and contribute improved public safety. We illustrate how planners can use field experiments in partnership with city agencies, nonprofit community groups, and local universities to discover novel approaches to advance place-based changes to the built environment that can help economically disadvantaged communities abate problems of physical disorder.

Introduction

Vacant and abandoned houses is a significant issue in many legacy cities in the United States that were once centers of manufacturing and industry (Mallach & Brachman, 2013). Property abandonment is the consequence of population loss following deindustrialization, racial segregation, suburbanization, and systemic racism (Alderson, 1999; Burchell & Listokin, 1980; Cohen, 2001; Dewar & Thomas, 2012; Rothstein, 2017). The problem of vacant and abandoned houses is especially acute in poor Black neighborhoods that have suffered the legacy of redlining, racial discrimination in real estate practices, and continued disinvestment (Massey, 1990; Pager & Shepherd, 2008; Skogan, 1986). Vacant and abandoned properties continue to represent a visible, and often widespread, sign of decades-long, structural disinvestment that undermines the perceived economic value and quality of life in many U.S. cities (Accordino & Johnson, 2000; Kim et al., 2020). For many former industrial cities, addressing the volume of vacant and abandoned houses has proven to be an intractable challenge.

In an effort to mitigate the problem of vacant and abandoned housing and buildings, cities have experimented with massive demolition programs, created land banks, and enacted ordinances that require vacant properties to be maintained by owners (General Accounting Office [GAO], 2011). Demolition programs often replace crumbling buildings with vacant lots, and create the problem of excessive vacant land that contributes to decreased tax bases for cities and substantial maintenance costs (Rigolon et al., 2021; Stern & Lester, 2021). Land banks have been criticized for being slow in selling vacant and abandoned properties for renovation or new development (Schmidt, 2022), though there are some more successful land bank models that work with community groups (Schilling & Hodgson, 2013; Schilling & Logan, 2008).

An alternative response to vacant and abandoned houses is a set of interim policies that seek to keep building structures intact until properties are renovated and reoccupied. Cities like Chicago (IL) and Philadelphia (PA) passed municipal ordinances that required owners of vacant and abandoned properties to maintain exterior windows and doors in good condition. Planners should be interested in whether compliance with vacant property ordinances reduces visible signs of disorder, disrepair, and serious crime in communities, as vacant and abandoned properties are often the centers of criminal activity (Porter et al., 2019; Spellman, 1993; St. Jean, 2007). Residents of predominately Black neighborhoods in Philadelphia have cited vacant and dilapidated houses among the key barriers to their sense of neighborhood wellbeing (Garvin et al., 2013). In legacy cities, Black neighborhoods are more likely to have experienced disinvestment than gentrification (Delmell, 2017), suggesting the urgent need to pursue and study in-place (or stationary) interventions that can quickly and inexpensively change the places in which economically marginalized residents live without requiring relocation.

We studied compliance with the Philadelphia doors and windows and litter property ordinances. In 2011 Philadelphia started enforcing an ordinance that required vacant properties that existed on blocks that were at least 80% occupied to have well-maintained exterior windows and doors to preserve the house (Philadelphia Property Maintenance Code Section PM-306.0; https://codelibrary.amlegal.com/codes/philadelphia/0e6b48f3-e095-484d-8733-9e5ef9fc1f65/philadelphia_pa_pm/0-0-0-7379). Prior to this municipal code, vacant and abandoned houses needed to have low-cost plywood or specialized metal grate coverings on open windows and doors. Owners of vacant and abandoned properties are also required to keep their lots litter free and without overgrown vegetation (Philadelphia Home Rule Charter Section 10–714; https://codelibrary.amlegal.com/codes/philadelphia/latest/philadelphia_pa/0-0-0-282563). To date, there is a lack of experimental research that has causally evaluated the effect of compliance with vacant and abandoned housing ordinances that are scalable at relatively low cost.

We rectify this gap in the scholarship by conducting a randomized field experiment to test the impact of compliance with the doors and windows and litter property ordinances on changes in physical disorder, gun violence, and illegal substance use activity. We found compliance with the doors and windows intervention was associated with an improvement in physical disorder, disrepair, and reductions in gun violence. The reductions in gun violence were associated with the improvement in the appearance of disrepair at housing sites. Cleaning up litter and weeding abandoned and vacant houses in compliance with the litter ordinance affected physical disorder at housing sites but had no association with gun violence.

There are two primary planning implications from this study. First, the study provides evidence that compliance with the doors and windows ordinance would improve the signs of disrepair in neighborhoods and could help mitigate the problem of gun violence in Philadelphia. Second, the study illustrates how to integrate systematic social observations into a field experiment to test the efficacy of place-based changes, and provides an example that planners can use to test doors and windows ordinances or other place-based planning interventions in other settings.

In the following section we discuss existing research on the consequences of vacant and abandoned houses on public safety, the theoretical framework that links vacant and abandoned buildings to public safety in communities, and the contribution our study makes to the literature. Next, we introduce our experimental design, data sources, and the analytical approach. Our presentation of results includes both descriptive and multivariate regressions for the experiment. We conclude with implications for planning, including the need to prioritize abandoned housing remediation efforts as a way to help reduce some of the negative implications of physical disrepair in communities, and how planners can advance the science of place-based changes to the built environment.

The Consequences of Vacant and Abandoned Houses on Safety

Physical Disorder and Disrepair

Well-maintained homes and yards are a key feature of the built environment that make neighborhoods attractive places to live and command higher property values. Vacant and abandoned homes with boarded up or broken windows and doors by contrast send a clear sign of physical disorder and disrepair in a neighborhood. The word disorder is problematic in that it suggests residents are behind the disorder, and may reflect biases that connect perceptions of disorder to the racial composition of neighborhoods (Sampson & Raudenbush, 2004). Although the concept of disorder conveys many meanings (Skogan, 2015), we focused on the observed “deterioration of the urban landscape” to measure physical disorder (Sampson & Raudenbush, 1999, p. 603). The presence of abandoned buildings, vandalism, overgrown vegetation, and garbage-strewn streets are all visible signs of physical disorder (Ross & Jang, 2000; Ross & Mirowsky, 1999; Skogan, 2015). A large body of research has found that neighborhood-level variation in physical disorder is associated with higher drug use and risky sexual behavior (Cohen et al., 2000; Latkin et al., 2007), increased crime (Branas et al., 2012; Chen & Rafail, 2019; Porter et al., 2018), heightened fear of crime (McGarrell et al., 1997; Ross & Jang, 2000; Skogan, 1990), poorer health (Hill et al., 2005; Ross & Mirowsky, 2001, South et al., 2018), and higher mortality (Cohen et al., 2003; Hannon & Cuddy, 2006). Research has found that residents are more likely to indicate their neighborhood lacks social order when they live in an area with greater perceived levels of physical disorder (Garvin et al., 2013; Ross & Mirowsky, 1999; Skogan, 1990).

Demolishing or Remediating Vacant and Abandoned Properties

There is a growing body of evidence that demolishing or remediating vacant and abandoned properties has an immediate positive impact on neighborhoods. Several studies have found a significant reduction in crime after vacant and abandoned properties were demolished (Jay et al., 2019; Larson et al., 2019; Porter et al., 2018; Stacy, 2017; Wheeler et al., 2018). Kondo et al. (2015) found that vacant and abandoned houses that had working windows and doors installed on them in compliance with a Philadelphia ordinance had significant reductions in crime compared with houses that did not comply with the ordinance. Branas et al. (2016) estimated that the costs of installing working windows and doors on vacant and abandoned houses resulted in nearly five times the return in tax-related benefits from reductions in crime.

The evidence from prior research suggests that demolishing or renovating vacant and abandoned housing may reduce the level illegal activity and violence in neighborhoods. The exact mechanisms through which demolishing or rehabilitating abandoned housing promotes these reductions is unclear. All published studies to date have been retrospective and relied on administrative data to assess the impact of housing demolition or rehabilitation programs on neighborhoods. Importantly, a major limitation of prior research is the lack of visual assessments of changes in physical disorder before and after changes to abandoned houses.

Systematic Social Observation

Administrative data collected by municipal agencies do not routinely capture measures of trash, debris, vandalism, or the level of disrepair of buildings on city blocks. Systematic social observations (SSOs; Reiss, 1971), or measuring the observation of social phenomena following explicit procedures (Mastrofski et al., 2009), provide an alternative method for collecting information on physical disorder (Raudenbush & Sampson, 1999). Research has used SSOs to assess physical disorder (Ndjila et al., 2019), typically including measures of building deterioration, broken windows, litter, vacancy, graffiti, and improper lot maintenance. Studies have found strong agreement between observers’ ratings of visual signs of physical disorder. Residents’ perceptions of disorder are also associated with SSO ratings (Hoeben et al., 2016; Jones et al., 2011; Perkins et al., 1992; Wei et al., 2005). SSO ratings may, however, be less susceptible to survey bias, such as respondents indicating a neighborhood has more physical disorder than is visually present based on the race and economic characteristics of the residents (Sampson & Raudenbush, 2004).

Conducting SSO research is more accessible than ever with emerging technologies that allow virtual auditing of the built environment. Google Street View (GSV), for example, is a tool that can help assess physical disorder in public spaces. Studies have found that SSO takes a fraction of the time for on-site observations and provides reliable and valid measures of physical disorder (Marco et al., 2017; Mooney et al., 2014, 2017; Odgers et al., 2012; Rundle et al., 2011). Although there can be spatial and temporal instability of image dates collected by GSV (Curtis et al., 2013), this is easily addressed if researchers make sure image dates and locations are consistent over time. Rigolon et al. (2021), for example, recently used visual images captured in the field and by GSV to monitor the progress and stewardship of land maintenance for vacant lots sold by the City of Chicago to private owners.

Conceptual Framework, Theory, and Rationale

The potential impact that remediating vacant and abandoned buildings has on disorder, disrepair, gun violence, and illegal substance outcomes is based on a recognized conceptual framework and past theory. This framework is buoyed by the rationale that vacant abandoned buildings are in great abundance in U.S. cities, have been singled out by community members as detrimental (Garvin et al., 2013; Hausman et al., 2009), and are modifiable, structural factors that, if remediated, could yield benefits to communities. Research has found that urban residents see vacant and abandoned buildings every day on their way to work or school, and describe these structures as visible detractors to community health, reducing their willingness to spend time outside with neighbors, creating trash, rodents, fear, havens for illicit drug use, and leading to more violent crime on city blocks (Hembree et al., 2005; St. Jean, 2007; Taylor, 2012).

Multiple theories posit that visible, environmental factors, such as vacant and abandoned buildings, lead to community decline by signaling that a community is uncared-for, incivilities are tolerated, and the ability of residents to engage in shared expectations of social control over neighborhood problems is eroded (Sampson & Raudenbush, 1999; Skogan, 1990; Taylor, 2012). As a result, residents are under more stress (South et al., 2015) and prevented from engaging in positive health behaviors whereas unhealthy behaviors, such as substance abuse and violence, become sheltered and more prevalent.

The concept of collective efficacy, or the willingness of residents to intervene in neighborhoods for the common good and out of shared mutual trust, suggests that a disordered built environment in a neighborhood may undermine social ties that foster informal social controls that mitigate criminal and antisocial behavior. For example, neighborhood residents may feel less empowered to intervene in the common good when there is no obvious owner maintaining a vacant and abandoned house and the property is left in a state of disrepair. Vacant and abandoned houses also signal that a site is not cared for and there is a “hole in the social fabric of a block” (Taylor, 1988, p. 186), potentially diminishing the perception that residents will act as guardians and intervene when they witness criminal activity in an area. Vacant and abandoned houses in a state of disrepair are also the inverse of the cues to care (CTC) framework that suggests there is an immediate signal of human intention or presence when a place is being well maintained (Nassauer, 2012).

Direct remediation of vacant and abandoned housing could be a potent intervention to interrupt long-standing problems of gun violence and illegal substance abuse occurring on city blocks by reducing the signs that an area is physically uncared for, and potentially improving the unconscious and conscious sense of informal social control of places.

Our Contribution

Our study makes three unique contributions to the planning literature on the effects of abandoned housing on neighborhood disorder and public safety. First, we provide the first causal study that documents the visual change in physical disorder captured by photo images around vacant houses and their surrounding blocks after having working windows and doors installed or trash and litter cleaned up. Although prior research has documented the impact of vacant and abandoned housing on disorder and crime, there are no experiment studies that document whether keeping vacant properties free of trash and debris or installing working windows and doors influences physical disorder and serious criminal activity. Second, we evaluated whether compliance with two straightforward municipal ordinances had benefits to physical disorder and public safety. Understanding the effectiveness of compliance with the doors and windows ordinance on physical disorder is particularly important because many cities simply require sealing up abandoned buildings, and residents in poor neighborhoods often express the sight of abandoned and boarded up homes as a blight that undermines positive aspects of a neighborhood (Garvin et al., 2013). Third, our study showcases how researchers can work with community groups and government agencies to test the efficacy that modifications in the built environment of places have on physical disorder, gun violence, and other outcomes that influence marginalized communities that have suffered from decades of disinvestment.

Research Strategy

Field Experiment

We conducted a field experiment of installing working windows and entry doors on vacant and abandoned houses that were in violation of the Philadelphia doors and windows ordinance. The primary aims of the experiment were to test whether a housing remediation that complied with the doors and windows ordinance would affect illegal drug trafficking and consumption, illegal drunkenness and drinking in public, and the occurrence of shootings and firearm violence compared with a simple trash cleanup or no treatment. South et al. (2022) provided further details on the experiment and additional analyses of public safety outcomes. Vacant and abandoned houses were selected for the experiment based on a random sample of four geographic regions of the city (north, northwest, south, and west). In each of the four regions of the city we compiled a master list of abandoned houses that were in violation of city ordinance. To form clusters, we randomly sorted the list of eligible houses. The first abandoned house on the list was then selected as an index house and a 0.125-mile-radius buffer was generated around its geographic centroid (latitude-longitude). All other eligible abandoned houses within this radius were used to form a cluster of approximately three to five houses. After a cluster of houses were formed, they were removed from the list, and then a new index house was selected that was at least 0.25 mile away to form a new cluster. This process guaranteed that no clusters overlapped and reduced the chance that there would be spillover across study arms.

A total of 63 clusters of abandoned and vacant houses were formed across four regions of Philadelphia and then randomly assigned to have working doors and windows installed (house remediation) by the Philadelphia Housing and Redevelopment Authority (PRA; n=23 clusters), receive trash cleaning and weeding (trash cleanup) by Pennsylvania Horticultural Society (PHS) contractors (n=20 clusters), or receive no treatment (control condition; n=20 clusters) starting in August 2018 until March 2019. In total, 258 abandoned houses (64 in the north, 103 in the northwest, 12 in the south, and 79 in the west) were enrolled in the experiment across the 63 clusters, of which 58 abandoned properties were assigned the housing remediation treatment, 93 received trash cleanup treatment, and 107 received no treatment. The 58 PRA sites also received trash cleanup and weeding on seven occasions starting 4 to 5 months (January 2019) after the installation of windows and doors. The study team visited the properties in this experiment on multiple occasions to ensure that doors and windows were not removed at PRA sites and that the trash was being picked up by PHS contractors. Ten of the 58 houses had to have windows and doors replaced in the summer of 2019 due to theft or damage. Technical Appendix A1 shows the flow the experimental design. Of the residents living in the experiment neighborhoods, 75% to 88% were Black with low median household incomes and high rates of unemployment.

Protocol for the Intervention

There were multiple steps needed to implement this vacant and abandoned housing experiment. First, the research team had to secure the legal arrangements between multiple city agencies that provided access to vacant and abandoned houses and indemnified property owners (including city agencies) from any potential injuries arising from conducting the interventions. Second, the research team had to secure contracts and memoranda of understandings with contractors to perform the housing remediation and the trash cleanup interventions. After roughly 8 months of meetings and negotiations (August 2016 to March 2017), license agreements were agreed upon between the research team, city agencies, and contractors. Cluster selection and random assignment occurred in the spring of 2018. The abandoned housing remediation and trash cleanup interventions began at the end of the summer of 2018.

Data and Measures

We conducted SSOs of virtual (photo-based) images of 258 abandoned houses and 360-degree panoramic views of the block faces in which they were located. The virtual images of houses captured the visual changes at the property level, whereas the images of the blocks measured any changes at the block level. A total of 634 photo images were taken for the abandoned and vacant houses, including 200 images of the 58 houses that received the remediation treatment, 226 images of the 93 houses that received the trash cleanup treatment, and 208 images of 107 houses that received no treatment. The study field team took photo-images of properties before the interventions while scouting out properties to determine eligibility for the experiment during March and April 2017. The PRA carpenters and PHS landscape crews took picture images of properties on the day of the remediation or trash cleaning after the treatments. To extract photo-based images for control houses after the interventions were completed, we used the GSV public application-programming interface (API). We were able to extract GSV pictures taken between June and October 2019 of the façade of all but two control houses. Pre-treatment and post-treatment images were taken with insufficient quality for coding for 25 of the 258 houses selected for the experiment. The final analytic sample for SSO coding consisted of 233 houses, including 56 of 58 houses that received the remediation treatment, 76 of 96 houses that received the trash cleanup treatment, and 101 of 107 houses that received no treatment. Figure 1 provides a visual sample of three houses that received the doors and windows house remediation, trash cleanup, or no treatment condition.

Figure 1.

Visual example of three properties rated on disrepair and disorder.

We used GSV to extract panoramic photo images of the blocks surrounding of the houses before and after the intervention. First, we used the package streetview to extract the closest GSV panoramic images associated with each property address in the intervention (https://github.com/robolyst/streetview). We used code improvements on the GSV package to stitch together the tiles to create 2-dimensional panoramic photos of the block surrounding each abandoned house (https://github.com/cplusx/google-street-view-panorama-download). Images taken before December 2018 were classified as pre-treatment. All images taken in 2019 were classified as post-treatment. A total of 466 panoramic photo images were extracted, including at least one photo for 95% (n=246) of the blocks that contained one of the 258 abandoned houses. Of these 246 blocks, 81% (n=224) had a picture for the pre-treatment period and 98% (n=242) had a picture for the post-treatment period between June and October 2019.

The outcomes variables for the analysis from the SSOs include measures of disrepair and physical disorder around each abandoned house and the block-faces in which they were located. These measures were adapted from validated neighborhood audit scales (Clarke et al., 2010; Earls et al., 1999) that have been used in multiple studies (Forsyth et al., 2008; Marco et al., 2017; Mooney et al., 2014, 2017; Odgers et al., 2012; Plascak et al., 2020; Rundle et al., 2011). Photo images of vacant and abandoned houses were coded by a diverse team of three independent trained raters on disrepair and disorder based on a 10-point scale (1 being the worst and 10 the best; see Technical Appendix A5 for details on coding). Disrepair included items such as peeling paint or crumbling brick on the building façade, boarded-up windows and doors, and broken, cracked, or missing stairs. The three ratings were reverse coded so that higher values meant more disrepair, and combined into an average scale with an inter-rater (alpha) reliability of 0.92. Physical disorder included items such as trash and debris strewn about; plant and weed overgrowth on the property, including the connected sidewalk; and graffiti on the building façade. The three ratings were reverse coded so that higher values meant more disorder, and combined into an average scale with an inter-rater (alpha) reliability of 0.87.

For the block face images, three trained raters were asked to assess the conditions of buildings, surrounding land, and litter on each block. These measures were adapted from a previously validated GSV neighborhood audit tool (Rundle et al., 2011). Raters were asked to assess on a Likert scale the poor conditions of building (structures, façades, and windows) on each block (none, 0%; very few, 1%–33%; some, 34%–66%; many, 67%–99%; and all, 100%). Sagging foundations, broken shingles, and collapsing roofs were signs of poor building conditions, whereas cracked brick, peeling paint, graffiti, broken or boarded-up windows, or missing windows were visual signs of poor façade and window conditions on a block. The nine separate measures of building, façade, and windows conditions (3 measures × 3 raters) were reverse coded for ease of interpretation and combined into an average scale, with an inter-rater (alpha) reliability of 0.84. The overall conditions of surrounding land, sidewalks, and roads were rated on a scale of poor, moderate, or good. The nine separate measures were combined into an average scale with an inter-rater (alpha) reliability of 0.77. Litter was assessed using the Philadelphia Litter Index, with ratings of 1 to 4 (1 = little to none; 2 = can be collected by a single person; 3 = requires collection by a team; and 4 = requires collection by a large team or heavy machinery; CleanPHL, n.d.). The three separate ratings of litter were combined into an average scale with an inter-rater (alpha) reliability of 0.78.

To determine the impact of the doors and windows remediation and trash cleanup interventions on Philadelphia police-reported gun violence (weapons violations, gun assaults, shootings) and illegal substance use (illegal substance trafficking or use, public drunkenness) outcomes, 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 from the centroid of each vacant and abandoned house. In this way, every vacant and abandoned house site had a unique value based on the specific distances of gun violence and illegal substance use outcomes that occurred around that house. Weapons violations included people reporting having been threatened with a gun or in possession of illegal guns. Gun assaults included reported threats from a gun, being shot at, or actually shot. Shooting incidents involved people being shot with a firearm. Illegal substance trafficking and use included any police incident report for narcotics traffic, sales, or possession. Public drunkenness included arrests and misdemeanor citations for public intoxication. We aggregated these monthly values into an average for the 18 months before (pre) and during (post) the field experiment.

Statistical Analysis

We relied on a difference-in-differences mixed-effects regression model to estimate the causal effect of the house remediation and trash cleanup interventions had on visible signs of disorder and disrepair at housing sites and block faces. For each regression model, we analyzed the relationship between the treatments on measures of disorder and disrepair at the housing site level, and building conditions, surrounding conditions, and litter at the block face level. We relied on contamination-adjusted intention-to-treat (CA-ITT) analyses of gun violence and illegal substance use outcomes to test the causal effect of changes in disorder and disrepair induced by the experiment. Specifically, we estimated a two-stage least-squares mixed-effects regression model to assess the effects of changes in visible signs of disorder and disrepair at housing sites and block faces induced by the experiment on changes in gun violence and substance use outcomes. All regression models accounted for the spatial clustering of houses nearby each other through the inclusion of a cluster-level random effect and a fixed effect that controls for the four sections of the city and captures larger aerial level differences (see Technical Appendix A6).

Limitations

Our randomized field experiment of vacant and abandoned housing remediations has several limitations. First, our results were limited to the duration of the trial and the settings in Philadelphia. We cannot extrapolate these results to periods after the trial ended or to other cities, yet we believe the context of Philadelphia is similar to that of other legacy cities where poor neighborhoods have suffered from historical patterns of disinvestment. Figuring out whether the observed improvements in physical disrepair and reductions in gun violence would occur in other cities will require implementing similar ordinances in other locations and studying their efficacy.

Second, our experiment narrowly measured changes within 18 months during the intervention. The results we observed may fade over time as other changes occur on the blocks that received the housing remediation interventions. In addition, the results from this experiment were not a comparison of the enforcement of municipal ordinances relative to more systemic investments in disadvantaged communities that may produce longer term benefits. Our study cannot attest to whether other place-based changes, such as providing more systematic investments in housing development and inclusionary zoning, would also produce improvements the physical environment and public safety.

Third, it was not feasible to vary the level of vacant housing remediation dosages within each cluster. Ideally, we would have randomly varied the number of remediations within each cluster to inform whether installing working windows and doors on all vacant and abandoned houses was more beneficial than just a few locations that residents may deem as problematic.

Fourth, our outcome measures of gun violence and illegal substances came from official police records. The shooting data collected by the police has been validated by a comparison with hospital data on gunshot victims (Branas et al., 2008), but other police data under-report the actual occurrence of crime victimizations (Baumer & Lauritsen, 2010).

Finally, the study was limited to examining the efficacy of a relative low-cost aesthetic improvement to vacant and abandoned houses. Conducting place-based experiments of more comprehensive changes to improve the built environment on blocks within economically marginalized communities is an important avenue for future research.

Results

Descriptive Changes in Physical Disorder, Gun Violence, and Illegal Substances

Descriptive statistics for the variables used in the analyses are presented in Table 1. The results show that the average level of disorder and disrepair was similar across the three different experimental study arms before the interventions, ranging from 4.3 to 4.8 in rating of physical disorder and 5.4 to 6.7 in disrepair. A score of 5 on the disorder scale suggested the presence of moderate levels of trash, debris, or overgrown plants that had not seen attention in several weeks. A score of 5 on the disrepair scale suggested that property had moderate issues with façade, porch, or attached stairs, but appeared that it could be inhabited. These data suggest that raters on average noted multiple instances of disorder in the form of trash, debris, and overgrowth of plants and weeds around vacant and abandoned housing sites. Similarly, raters noted multiple forms of disrepair to buildings at baseline in term of boarded-up windows and doors and broken, cracked, or missing stairs on vacant and abandoned housing sites.

Table 1.

Average level of observed physical disorder, disrepair, and litter.

Outcome	House Remediation		Trash Cleanup		No Intervention Control
Housing sites	Pre	Post	Pre	Post	Pre	Post
Disorder	4.330 (1.486)	3.380 (1.367)	4.809 (1.465)	3.967 (1.445)	4.296 (1.450)	4.654 (1.834)
Disrepair	6.736 (0.793)	3.551 (1.264)	6.394 (1.403)	6.250 (1.504)	5.470 (1.271)	6.414 (2.025)
No. houses	56	56	76	76	101	101
Block faces	Pre	Post	Pre	Post	Pre	Post
Buildings	1.928 (0.439)	1.57 (0.408)	1.807 (0.500)	1.682 (0.468)	2.008 (0.623)	1.977 (0.592)
Surroundings	1.966 (0.371)	2.035 (0.356)	2.042 (0.435)	2.095 (0.409)	1.892 (0.354)	1.914 (0.328)
Litter	2.095 (0.637)	1.932 (0.723)	1.872 (0.668)	1.749 (0.518)	2.031 (0.626)	1.915 (0.697)
No. blocks	49	49	73	73	98	98
Housing sites	Pre	Post	Pre	Post	Pre	Post
Weapons	19.84 (6.686)	20.68 (7.850)	21.60 (8.230)	24.01 (11.05)	20.00 (5.912)	22.93 (8.386)
Gun assaults	4.449 (1.483)	5.020 (1.824)	4.689 (1.945)	5.607 (2.694)	4.448 (1.452)	5.862 (2.471)
Shootings	2.741 (1.104)	3.092 (1.157)	2.830 (1.288)	3.483 (1.726)	2.693 (1.081)	3.320 (1.369)
Illegal substance	14.47 (23.06)	15.72 (25.75)	29.12 (49.21)	35.36 (66.00)	14.38 (11.56)	14.78 (10.48)
Drunkenness	0.397 (0.436)	0.810 (0.795)	0.264 (0.315)	1.067 (1.548)	0.158 (0.118)	0.824 (0.877)
No. houses	56	56	76	76	101	101

Note: Mean displayed with standard deviation in parentheses. Weapons, gun assaults, illegal substance, and drunkenness calculated as mean of monthly (18 months pre and post) kernel density per square mile. Data sources: Systematic social observations, Google Street View, Philadelphia Police Department.

The descriptive measures show that on average physical disorder and disrepair measures improved at housing remediation sites after the intervention. The average rating of disorder and disrepair decreased from 4.3 to 3.4 and 6.7 to 3.4 at housing remediation sites, respectively, after the intervention. At trash cleanup sites disorder dropped from 4.8 to 3.9, but there was little movement in building disrepair. By contrast, at the no-intervention control sites disorder and disrepair rose marginally. These findings suggest that the housing remediation intervention delivered noticeable changes to the level of disrepair at housing sites so that it appeared that the houses could be occupied by a visual assessment. For trash cleanup, the results suggest only some slight improvement in disorder. These findings are consistent with the intent of the experimental interventions and suggest they were delivered as planned.

For the gun violence and illegal substances outcomes, the descriptive results suggest that the average monthly rate of weapons violations, gun assaults, and shootings had similar levels during the pre-intervention period, but these rates increased the least among the housing remediation sites. At housing remediation sites, for example, there were on average 19.84 weapons violations per month in the pre-intervention period compared with 20.68 in the post-intervention period. By comparison, weapons offenses rose more in the control sites from 20 per month in the pre-intervention period to 22.93 in the post-intervention period. The description the rates in the pre- and post- periods is only an illustration of trends and does not provide any strong statistical identification.

Did Disorder and Disrepair Change After the Housing Remediation and Trash Clean-up Interventions?

The results of the regression analyses that compared the relative changes in disrepair and disorder at housing sites before and after the intervention and the conditions of the surrounding blocks are presented in Table 2. Disrepair changed significantly (p<.001) around housing sites that received the installation of working windows and doors as part of the house remediation intervention. Specifically, disrepair declined by an estimated −1.67 around houses that received the doors and windows remediation compared with control houses that received no treatment. This equates to a 27% (−1.67/6.02) proportional improvement in the average rating of disrepair. The results also show that visual assessments of physical disorder significantly (p<.001) improved by an estimated −1.37 around houses that received the house remediation and −1.29 around houses that received trash cleanup interventions relative to the control houses that receive no treatment. In the house remediation and trash cleanup interventions this equates to a proportional reduction of −31% and −30% in disorder.

Table 2.

Changes in observed physical disorder, disrepair, building conditions, surroundings, and litter.

	Housing Sites		Block Faces
	Disorder	Disrepair	Building Conditions	Surroundings	Litter
Trash cleanup	−1.297*** [−1.880,−0.714]	0.0989 [−0.439,0.637]	−0.0942 [−0.285,0.0967]	0.0241 [−0.121,0.169]	−0.00764 [−0.262,0.246]
House remediation	−1.370*** [−2.009,−0.731]	−1.677*** [−2.268,−1.086]	−0.312** [−0.528,−0.0957]	0.0476 [−0.116,0.211]	−0.0476 [−0.335,0.240]
Average dep	4.329	6.023	1.856	1.981	1.928
N	460	466	440	440	440
Chi-square	42.73***	180.08***	27.02***	8.30	14.92*

Notes: 95% confidence intervals in brackets. Models also control for four sections of the city, post-period, and treatment arm. Data source: Systematic social observations, Google Street View.

^*p < 0.05,

^**p < 0.01,

^***p < 0.001.

Did Building Conditions and Litter on Blocks Change After Housing Remediation and Trash Cleanup Interventions?

Table 2 also shows that results from the regressions models of changes in building conditions in both house remediation and trash cleanup interventions relative to the control condition. The results show that building conditions only improved by a statistically significant amount on blocks where houses received the installation of working windows and doors (p<.001). For those block faces where houses were remediated there was −.32 improvement in the average rating of building conditions. This equates to a proportion improvement of approximately −17% in average building conditions on a block (−.31/1.85). A value of 1 meant that 1 to some (1%–33%) of the buildings on a block had poor conditions in terms of structure, façades, and windows. The observed level of surrounding conditions and litter did not improve in either the housing remediation or trash cleanup intervention blocks. These results suggest that the changes in disrepair and disorder observed were largely localized to the housing sites.

Did Housing Remediation and Trash Cleanup Affect Gun Violence and Illegal Substance Use Outcomes?

The regression analysis results that compared the relative changes in gun violence and illegal substance use outcomes before and after the intervention are presented in Technical Appendix A3. The results show that weapons violations and gun assaults all dropped by statistically significant levels (p<.01) in the housing remediation intervention relative to the no-treatment control condition. Shootings were reduced by a nearly significantly level (p=0.109). These reductions equate to approximately a −9.6% change in weapons violations, −16.7% change in gun assaults, and −9% change in shootings in the housing remediation locations relative to control sites. A visual depiction of these regression estimates for the gun violence outcomes is displayed in Figure 2. The reductions in weapons violations and gun assaults outcomes were also significantly larger in the housing remediation than the trash cleanup study sites (p<.001) according to Wald tests comparing the differences between groups. For the trash cleanup intervention there was a null effect on gun violence outcomes relative to the no-treatment control condition. The results suggest that illegal substances increased in the trash cleanup sites relative to the no-treatment control condition, but the increase was not statistically significant (p<.05).

Figure 2.

Change in Gun Violence Outcomes

Source: Philadelphia Police Department

Did Changes in Disorder and Disrepair Affect Gun Violence and Illegal Substances?

The results from the CA-ITT analyses of gun violence and illegal substances outcomes that used the random assignment to the housing remediation or trash cleanup interventions as an instrumental variable to adjust for the level of improvement in disorder and disrepair at housing sites are presented in Table 3. The F-test from the first stage regression provides the key indicator of whether the random assignment was a suitable cause for a change in the level of disorder and disrepair in the trash cleanup and housing remediation arms. An F-test above 10 is generally considered evidence of a sufficiently strong predictor of a change (Angrist & Pischke, 2009). Table 3 shows that the trash cleanup arm was not sufficiently predictive of changes in disorder or disrepair to be a valid instrument, as all the F-tests were below a value of 5.0. For the housing remediation arm the intervention produced a sufficiently large change in disrepair (F-tests all above 10) to indicate it was a valid instrument. The results from the CA-ITT show the improvement in disrepair at housing sites caused by the installation of working windows and doors predicted a significant reduction in weapons violations, gun assaults, shootings, and drunkenness. An improvement of 1 point in disrepair predicted by random assignment to the housing remediation intervention predicted a −.712 (monthly) reduction in weapons violations, a −.362 reduction in gun assaults, and a −.22 reduction in shootings. The results imply that reductions in gun violence outcomes attributed to the housing remediation intervention occurred through its impact on improving physical disrepair.

Table 3.

Changes in crime from housing site-level changes in disorder and disrepair.

Disorder	Weapons	Gun assaults	Shootings	Illegal substance	Drunkenness
Trash cleanup	−0.323 [−1.057,0.410]	0.0616 [−0.228,0.352]	−0.186* [−0.349,−0.022]	−5.227** [−8.398,−2.056]	−0.300* [−0.544,−0.055]
Ave. dep	22.09	5.167	3.078	22.31	0.567
N	348	348	348	348	348
F-test	3.5	3.5	3.5	3.5	3.5
Disrepair	Weapons	Gun assaults	Shootings	Illegal substance	Drunkenness
Trash cleanup	−12.65 [−29.00,3.696]	−4.378 [−9.948,1.193]	−2.713 [−6.135,0.709]	−10.58 [−26.00,4.844]	−3.104 [−6.961,0.753]
Ave. dep	22.04	5.152	3.071	22.16	0.566
N	354	354	354	354	354
F-test	2.0	2.83	2.83	0.43	2.83
Disorder	Weapons	Gun assaults	Shootings	Illegal substance	Drunkenness
Housing remediation	1.287* [0.265,2.309]	0.508* [0.120,0.896]	0.145 [−0.0227,0.313]	−0.769 [−2.101,0.563]	0.0585 [−0.108,0.225]
Ave. dep	21.07	5.019	2.980	14.78	0.531
N	308	308	308	308	308
F-test	4.0	4.2	4.0	3.8	4.6
Disrepair	Weapons	Gun assaults	Shootings	Illegal substance	Drunkenness
Housing remediation	−0.712** [−1.158,−0.267]	−0.362*** [−0.534,−0.191]	−0.226*** [−0.317,−0.135]	−0.681 [−1.422,0.0593]	−0.186*** [−0.265,−0.107]
Ave. dep	21.04	5.005	2.974	14.76	0.531
N	314	314	314	314	314
F-test	15.6	17.4	16.4	25.0	27.8

Note: 95% confidence intervals in brackets. F-test of first stage of experiment (post*study arm) on disorder and disrepair. Models also controls for four sections of the city and post-period. Data sources: Systematic social observations, Google Street View, Philadelphia Police Department.

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

The results from the estimates of changes in gun violence and illegal substances outcomes predicted by changes in block-level building conditions are presented in Technical Appendix A4. All CA-ITT results had first-stage F-statistics less than 10, indicating there was not sufficient change at the block level for the trash cleanup or housing remediation treatment arms to be adequate instruments. These results were consistent with our SSO analysis in suggesting the effects on building conditions were localized to the housing sites.

Policy Significance

Our main findings from this experimental evaluation of compliance with the Philadelphia doors and windows ordinance have important implications for planners and policymakers confronting the legacy of vacant and abandoned houses in cities.

• Planning agencies should work with other city agencies to ensure that occupied residential blocks with vacant and abandoned houses have a remediation plan in place that includes low-cost treatments, such as the replacement of windows and doors in need of repair, façade cleaning, and regular trash removal.

• Remediating abandoned houses may help reduce the negative implications of physical disrepair on gun violence, an important indicator of public safety.

• Planners should pilot test place-based changes to the built environment that can help economically disadvantaged communities abate problems of physical disorder that arise from vacant and blighted spaces.

• A research partnership between planners, community residents, local nonprofit service providers, and city agencies could help advance the science of place-based changes in a way that is useful for communities that have suffered from years of disinvestment.

Conclusion

Vacant and abandoned buildings pose significant challenges to the health, safety, and economic viability of neighborhoods. With our experiment we sought to test the relative effectiveness of two abandoned building remediation interventions: the installation of working doors and windows, and property maintenance through trash removal. By randomly assigning abandoned housing to receive functional doors and windows in place of missing or boarded-up ones, we tested the causal effect of the doors and windows ordinance relative to cleaning up trash in compliance with a property maintenance ordinance on litter and illegal dumping or a control condition that received no treatment. By relying on photo documentation, GSV, and SSOs, we visually assessed changes in disrepair, disorder, and litter around the houses that were part of this experiment as well as their surrounding block faces, and the effect that visual changes to the built environment had on changes in gun violence and illegal substance use outcomes. The installation of working windows and doors led to a significant improvement in the appearance of disrepair of buildings, but did not change ratings of surroundings or litter on block faces, suggesting that the intervention produced only a noticeable difference around the affected properties. The changes to the conditions of disrepair at housing sites appeared to contribute to significant relative reductions in gun violence outcomes, suggesting that remediating vacant and abandoned properties to a level that they appear occupied helped improve public safety.

This is the first citywide experiment to randomly assign the remediation of vacant abandoned houses in compliance with a city ordinance and prospectively document nearby changes in physical deterioration of buildings and surrounding disorder and public safety outcomes. The average cost for installing working windows and doors on houses was $5,900, suggesting that a modest investment in remediating abandoned houses helped improve disrepair on a block and gun violence. Although experimental research is growing in planning (Honey-Rosés, 2019; Honey-Rosés & Zapata, 2021; Monkkonen & Manville, 2019), more experiments should be conducted to test how modifications of the built environment that “change behavior, city dynamics or a sense of place” (Honey-Rosés, 2019, p. 567) prior to large-scale implementation.

Biography

John MacDonald (johnmm@upenn.edu) is a professor in the Departments of Criminology and Sociology at the University of Pennsylvania. Ahuva Jacobowitz (aj2476@cumc.columbia.edu) is a PhD candidate in epidemiology in the Mailman School of Public Health at Columbia University. Jason Gravel (jgravel@temple.edu) is an assistant professor in the Department of Criminal Justice at Temple University. Mitchell Smith (mrsmit75@asu.edu) is a PhD candidate in the School of Criminology and Criminal Justice at Arizona State University. Robert Stokes (robert.stokes@csusb.edu) is a professor in the Department of Public Administration at California State University, San Bernardino. Vicky Tam (tamvw@chop.edu) is a senior GIS analyst in the Department of Biomedical and Health Informatics at the Children’s Hospital of Philadelphia. Eugenia South (Eugenia.South@pennmedicine.upenn.edu) is an assistant professor in the Perelman School of Medicine at the University of Pennsylvania. Charles Branas (c.branas@columbia.edu) is a professor of Epidemiology in the Mailman School of Public Health at Columbia University.

Data Availability

The data that support the findings of this study are available on request from the corresponding author (J.M.). The data are not publicly available due to their containing information that could compromise the privacy of research participants.