Predicting Housing Abandonment with the Philadelphia Neighborhood Information System

Abstract

Several large US cities, including Chicago, Los Angeles, New York, and Philadelphia, have developed information systems to distribute property-level housing data to community organizations and municipal agencies. These early warning systems are also intended to predict which properties are at greatest risk of abandonment, but they have rarely used statistical modeling to support such forecasts. This study used logistic regression to analyze data from the Philadelphia Neighborhood Information System in order to determine which properties were most likely to become imminently dangerous. Several different characteristics of the property, including whether it was vacant, had outstanding housing code violations, and tax arrearages as well as characteristics of nearby properties were identified as significant predictors. Challenges common to the development of early warning systems -including integrating administrative data, defining abandonment, and modeling temporal and spatial data -are discussed along with policy implications for cities like Philadelphia that have thousands of vacant and abandoned properties. ABSTRACT: Several large US cities, including Chicago, Los Angeles, New York, and Philadelphia, have developed information systems to distribute property-level housing data to community organizations and municipal agencies. These early warning systems are also intended to predict which properties are at greatest risk of abandonment, but they have rarely used statistical modeling to support such forecasts. This study used logistic regression to analyze data from the Philadelphia Neighborhood Information System in order to determine which properties were most likely to become imminently dangerous. Several different characteristics of the property, including whether it was vacant, had outstanding housing code violations, and tax arrearages as well as characteristics of nearby properties were identified as significant predictors. Challenges common to the development of early warning systemsÐincluding integrating administrative data, defining abandonment, and modeling temporal and spatial dataÐare discussed along with policy implications for cities like Philadelphia that have thousands of vacant and abandoned properties.Over the past 30 years, community groups, municipal governments, and researchers

Housing Abandonment and Demolition: Exploring the Use of Micro-Level and Multi-Year Models

“…While many past studies of housing abandonment in rust belt cities used data at the census tract or block group level [11,17,35], few have used property level spatial data, especially for multi-years, to understand the dynamic patterns and the impacts of abandonment and demolition on neighborhoods [9,36]. There is no universal definition of land abandonment or housing abandonment, and vacancy-related data often exist in different formats with different spatial extents and are compiled by various agencies or organizations for different purposes and years [6,11,37].…”

“…This modeling was used to build on previous studies of vacancy and Buffalo's local housing market [3,9,11,17,36,[39][40][41]. There are three hedonic price models on housing sales prices, three logistic regression models on vacancy and abandonment, and three logistic regression models on demolition.…”

Employing community data to investigate social and structural dimensions of urban neighborhoods: An early childhood education example

“…Differential radii are defined around the child, placing the child at the center of their own neighborhood. Employing raster measures in examinations of context supports a child-centered approach to understanding environmental influences and has the potential to more fully exploit the richness of individual household and property-level administrative data (Hillier, Culhane, Smith, & Tomlin, 2003).…”

“…Brooks-Gunn and associates (1993) suggest that demographic and censusbased neighborhood variables are at best crude markers for the full range of contextual conditions that could buffer or exacerbate neighborhood effects on child development. With so many potentially influencing factors at the neighborhood level, disagreement exists regarding: (a) the appropriateness of census tract-level aggregation (which generally encompasses 1,500 -8,000 individuals in arbitrarily defined geographic polygons) as an adequate representation of neighborhood; (b) the sufficiency of census data to capture important variation in the social and physical environments of neighborhoods; and (c) the infrequent availability of data provided by the US Census 1 Hillier, Culhane, Smith, & Tomlin, 2003;Leventhal & Brooks-Gunn, 2000).…”

“…Such data can be aggregated at user-defined intervals and coincident with the enrollment of participants in research projects (Hillier & Culhane, 2005). In addition, administrative data allow for the investigation of neighborhood at lower levels of aggregation (in contrast to the census tract) that might more closely approximate actual neighborhood phenomena Hillier et al, 2003). Sampson and colleagues posit, ''it may be that adults and children are located in distinct ecological niches within larger neighborhoods, suggesting the need to disaggregate analyses and study smaller ecological units such as block groups'' (Sampson et al, 1999, p. 657).…”