«Urban Problems and sPatial methods VolUme 17, nUmber 1 • 2015 U.S. Department of Housing and Urban Development | Office of Policy Development and ...»
A final comment addresses the theoretical framework and statistical techniques described here. As explained previously, these techniques explicitly consider the spatial distribution of homicides in which the goal was to show the existence of a geographic diffusion to areas immediately surrounding the direct focus of the policy efforts described. Nonetheless, the inference made from the empirical analysis does not imply a formal causality test between army intervention and rising homicides in absolute terms; other factors, such as clashes among drug cartels or groups within them, could be influential factors.
Note: States with army intervention as part of the “Operativo Conjunto” are in blue.
Authors Miguel Flores is a professor and researcher in the School of Government and Public Transformation at Tecnológico de Monterrey.
Amado Villarreal is a professor and researcher in the School of Government and Public Transformation at Tecnológico de Monterrey.
References Anselin, Luc. 1995. “Local Indicators of Spatial Association-LISA,” Geographical Analysis 27: 93–115.
———. 1993. “The Moran Scatterplot As an ESDA Tool To Assess Local Instability in Spatial Association.” Paper presented at the GISDATA Specialist Meeting on GIS and Spatial Analysis, Amsterdam, The Netherlands, December 1–5. West Virginia University, Regional Research Institute, Research Paper 9330.
Chainey, Spencer, and Jerry Ratcliffe. 2005. GIS and Crime Mapping. London, United Kingdom:
John Wiley & Sons.
Cohen, Jacqueline, and George Tita. 1999. “Diffusion in Homicide: Exploring a General Method for Detecting Spatial Diffusion Processes,” Journal of Quantitative Criminology 15: 451–493.
Fearon, James, and David Laitin. 2003. “Ethnicity, Insurgency, and Civil War,” American Political Science Review 97: 75–90.
Fotheringham, Stewart. 2009. “The Problem of Spatial Autocorrelation and Local Spatial Statistics,” Geographical Analysis 41 (4): 398–403.
Merino, Jose, and Victor Gomez. 2012 (December). “Cuerpos sin Nombre,” NEXOS 420. http:// www.nexos.com.mx/?p=15084.
Payan, Tony. 2006. The Three U.S.-Mexico Border Wars: Drugs, Immigration, and Homeland Security.
Santa Barbara, CA: Praeger Security International.
Ratcliffe, Jerry. 2010. “The Spatial Dependency of Crime Increase Dispersion,” Security Journal 23:
Rodriguez-Oreggia, Eduardo, and Miguel Flores. 2012 (January). Structural Factors and the “War on Drugs” Effects on the Upsurge in Homicides in Mexico. CID Working Paper No. 229. Cambridge, MA: Center for International Development at Harvard University.
Zimring, Franklin, and Gordon Hawkings. 1997. Crime Is Not the Problem: Lethal Violence in America. Oxford, United Kingdom: Oxford University Press.
Abstract This article details the use of a spatial difference-in-differences approach for measuring the effect of a vacant land greening program in Philadelphia, Pennsylvania, on nearby property values. Vacant land is a ubiquitous problem in U.S. cities, and many have recently begun to explore greening programs as an interim management strategy for vacant lots, in the hopes they will reduce the negative influence of vacancy and help to spur neighborhood development. The methods used here draw on previous approaches to modeling effects of greening on property values but expand on them to explore means of incorporating spatial relationships and allowing for spatial nonstationarity, in which the process being modeled changes across space. Spatial methods were used not only to derive data and choose appropriate observations but also to compare global and local versions of the analysis to assess spatial patterns and differences in outcomes across the study area, ultimately showing that, although greening vacant land increases surrounding property values, it does not do so uniformly across urban neighborhoods.
Introduction The urban revitalization literature is chock full of ideas for how to improve distressed neighborhoods, but actually determining the effects of interventions has proven to be more challenging.
One of the most commonly studied effects is the change in property values; these effects are generally studied because the housing market is seen as a good indicator of the desirability or perceived quality of a neighborhood (Galster, Tatian, and Accordino, 2006). Most methods for making these assessments rely at least in part on hedonic regression models, in which the value of an individual property is seen as reflecting a bundle of values of individual amenities, which would include Cityscape 51 Cityscape: A Journal of Policy Development and Research • Volume 17, Number 1 • 2015 U.S. Department of Housing and Urban Development • Office of Policy Development and Research Heckert both characteristics of the property itself and characteristics of the neighborhood, and a regression equation is used to estimate the value of each individual amenity or property characteristic (Rosen, 1974). Thus the effect of proximity to an intervention on property values might be assessed by comparing property values at varying distances to the intervention and checking the coefficient of the distance variable to see if lower distances correspond to higher values.
The difference-in-differences approach is an econometric case-control test that investigates whether an intervention influences an outcome over time by comparing observed differences in a case sample that receives the intervention to observed differences in a control sample that does not.
This approach enables isolation of the treatment effect above and beyond any difference that would have been expected regardless of the treatment (Meyer, 1995). The difference-in-differences model specification has been used with hedonic modeling of property values to assess the effects of a range of neighborhood interventions, including new housing development (Ellen et al., 2001;
Ellen and Voicu, 2005), establishment of community gardens (Voicu and Been, 2008), and tree planting programs (Wachter and Wong, 2008).
Incorporating Space When it comes to measuring the effects of interventions such as new housing development or tree planting on neighborhoods, space cannot be ignored. A neighborhood is a fixed location in which residential and commercial buildings, amenities such as parks, and disamenities such as vacant lots all coexist with their distinct spatial relationships informing the influences they have on each other and the neighborhood as a whole. The value of any component of that neighborhood cannot be divorced from the values ascribed to the other components of the neighborhood, because it is reliant in part on those values. This reliance means that space and spatial relations must inform any attempt to measure those effects. That being said, the attempts to measure the effects of intervention cited previously take space into account in only one way—by incorporating the distance to an intervention into the equation to estimate its effect. These attempts often take the form of creating distance bands from the intervention within which a property is or is not located (Ellen et al., 2001;
Wachter and Wong, 2008). For example, Ellen et al. (2001) studied the effects of new housing development on existing property values by specifying houses as affected if they were within three different distance bands of the new development—500 feet, 1,000 feet, or 2,000 feet—and designating control lots as outside the distance in question but within the same ZIP Code.
These models are helpful, but they often assume a single model that describes relationships and values that are constant across the entire study area. They are unable to account for potential differences in effects across locations, known as spatial nonstationarity, unless the drivers of those differences are known in advance and can be incorporated into the models as interaction terms.
More explicitly, spatial methods are required to find differences that are not known from the beginning or are unable to be incorporated because of a lack of appropriate data.
This article details the methods used to measure the effects of a greening intervention to manage vacant land on surrounding property values in Philadelphia, with a focus on the means by which spatial patterns were assessed. The program, Philadelphia LandCare (PLC), uses a simple greening approach to treat vacant lots by removing any existing trash or debris, bringing in topsoil, planting
new grass and a few trees, and erecting a split-rail fence to prevent dumping and give the lot a more managed look. During the first decade of the program, more than 5,000 individual parcels received treatment through the PLC program. For more details of the program, see Jost (2010) and Heckert and Mennis (2012).
For this research, which is described in more detail in Heckert and Mennis (2012), I adapted the difference-in-differences approach for spatiotemporal analysis of changes in property values surrounding treatment and control lots through use of a sampling strategy that ensured control lots mimicked the spatial distribution and economic characteristics of treated lots, while also remaining spatially distant enough to prevent diffusion of treatment, whereby any effect from the treatment might also be demonstrated by the control because of proximity. I modified the approach by creating a geographically weighted variant, using geographically weighted regression to explore geographic variation in the effects of the greening program.
Data and Methods This analysis relied on four primary spatial datasets and on several supplemental datasets. The primary datasets were (1) data on lots that were treated as part of the PLC program represented as points at the center of each of 747 contiguous groups of lots that were greened together between 1999 and 2006; (2) data on vacant lots in Philadelphia in 2010 also represented as points at the center of each group of contiguous vacant lots; (3) a set of points representing all Philadelphia residential real estate sales valued at more than $1,000 between the years 1999 and 2007, with sales prices adjusted for inflation to 2007 dollars; and (4) boundaries for neighborhood planning districts, breaking the city into seven regions. Additional datasets included shapefiles with locations of commercial corridors and schools and a real estate market typology created by a local community development financial institution, which ranked the 2008 real estate markets in each census block group on a 9.0-point scale from distressed (1.0) to regional choice (9.0). The purpose of each dataset is described in more detail in the following section.
The difference-in-differences specification uses a case-control methodology where each case—in this instance, each lot that was ultimately treated through the PLC program—is matched with appropriate controls—in this instance, lots that could have been treated through the PLC program but were not. Although the data on the PLC program and the vacant land data for Philadelphia both started as data on individual parcels, adjacent parcels were merged together for analysis. This merging was necessary because the PLC program was implemented on groups of adjacent lots that look and feel like a single entity, even if they are technically separate properties, and the effects of two adjacent lots cannot be reasonably separated from each other. Following similar logic, the vacant lots used as controls were combined based on adjacency.
It is very important to note that site selection for the PLC program was by no means random and, thus, controls could not simply be assigned randomly from the universe of all vacant lots. The Pennsylvania Horticultural Society (PHS), which developed and manages the PLC program, describes several criteria that are used to determine lots for inclusion in the program. First and foremost, PHS targets communities with large concentrations of blighted vacant lots—PLC is not a program designed for neighborhoods with strong real estate markets and low populations of vacant
lots to choose from. Within target neighborhoods, lots are chosen based on loose criteria intended to identify lots with most potential for effect, so that large collections of lots near schools or commercial corridors are prioritized (Jost, 2010). In an effort to select control lots that were closest in characteristics to the treated lots, I restricted the set of all vacant lots to those located within 500 feet of a school or commercial corridor before selection of control lots.
One challenge for selecting control lots was attempting to prevent diffusion of treatment effects, whereby the effect of a greened lot would also happen and be felt in the area of nearby nongreened lots. The concern here was that a large number of untreated vacant lots are also located in close proximity to treated lots, which is not surprising, given that the program targeted areas with large numbers of vacant lots. Although it was desirable to keep control lots as far as possible from the PLC lots to avoid the possibility that property values surrounding them also increased because of proximity to PLC lots, it was also necessary to ensure that controls were located in similar neighborhoods and thus represented appropriate counterfactuals. To mitigate diffusion of treatment effects, all vacant lots within 250 feet of a treated lot were excluded from the pool of potential control lots. To ensure that controls were in similar neighborhoods to treated lots, the final selection of controls matched each treated lot to three randomly selected controls from the pool, with the controls matching the treated lot in both the section of the city and the real estate typology score for the block group of the lot. Thus the matches did not guarantee that the control lots were in exactly the same neighborhood but required that the control lots face similar real estate market conditions and be in the same relative portion of the city.
For the specification of the difference-in-differences model, the unit of analysis was taken to be the lot, with values assigned to represent the value of residential properties at the location of the lot.