Dawn of the Dead City: An Exploratory Analysis of Vacant Addresses in Buffalo, Ny 2008–2010

ABSTRACT:This article examines residential vacancy patterns in Buffalo, NY, using data from a unique data set. It includes variables from HUD Aggregate USPS Administrative Data on Address Vacancies, the American Community Survey (ACS) 5-year estimates for 2005–2009, housing choice voucher (HCV) records of local public housing agencies, and municipal in rem property records. Multiple regression is used to identify significant relationships between vacancy patterns, socioeconomic characteristics, and institutional factors. The findings from this analysis suggest that the percent of vacant residential properties increases in census tracts with elevated poverty rates, higher percentages of renters receiving rental assistance, and long-term vacancies. They also suggest that the percent of abandoned residential properties increases in census tracts with highly concentrated black populations, elevated poverty rates, long-term vacancies, and higher percentages of business addresses. We conclude that these relationships are unique to older core cities experiencing systemic population and job losses. These cities struggle with a distinct type of long-term vacant and abandoned structures, which we label zombie properties. They can be contrasted with vacant and abandoned properties in transitional or regenerating areas. We offer recommendations for further analysis of zombie properties in these urban settings. Notes1 At the time that this article was written, 2010 U.S. Census data were becoming available. For 2010, the U.S. Census reported that the total population in Buffalo was 261,301. This represented a 55.0% decline between 1950 and 2010. Correlations were calculated on the population counts by census tract comparing the 5-year estimates for 2005–2009 and full counts for 2010. The counts were highly correlated (p= 0.96). The 2010 full counts also fell within the margin of error reported for the five year 2005–2009 estimates.2 For 2010, the U.S. Census reported that the total population in the Buffalo-Niagara Falls metropolitan statistical area was 1,135,599. This represented a 15.8% decline in population between 1970 and 2010.3 For 2010, the U.S. Census reported that the total population in the suburbs of Buffalo was 874,199. This represented a 1.4% decline in population between 1970 and 2010.4 City-owned properties are largely comprised of in rem properties. In rem properties include residential and commercial tax foreclosure properties owned by municipalities. These properties are often transferred to municipalities after prolonged periods of abandonment and neglect.5 Estimates of vacant properties reported by the City of Buffalo have fluctuated in recent years. For instance, the City's Office of Strategic Planning identified over 23,000 vacant and uninhabitable structures in a prior report (United States Conference of Mayors, 2006). In contrast, the 2008 estimate was based on a survey submitted by the Mayor's Office. Because of inconsistent reporting on vacancies by the City of Buffalo, we consider U.S. Census data and HUD Aggregated USPS Administrative Data on Address Vacancies more reliable estimates of vacant property.6 Structural damage is often the result of extended exposure to extreme weather conditions, insect and rodent infestations, a lack of regular building maintenance, mold, and/or damage due to looters stripping fixtures and scrap metal from buildings.7 The term TOADS was introduced in reference to a variety of vacant and abandoned properties, including residential structures, commercial buildings, industrial sites, and vacant lots. Our focus is primarily on abandoned residential structures.8 The "other" category is used when the USPS is unable to categorize an address as either business or residential. For example, these addresses might be occupied by government, religious, or other nonprofit institutions. They might also be mixed-use properties, functioning as both businesses and residential units.9 The American Community Survey (ACS) is an annual survey of population and housing characteristics conducted by the U.S. Census Bureau. It is administered to 3 million households in the country each year. The ACS collects information previously collected in the long form of the decennial census. It is the largest survey, other than the decennial census, administered by the U.S. Census Bureau. The 2005–2009 ACS represents estimates based on a rolling average for five years of sampling. Because these data are based on a sample of the total population, the Census reports margins of error for individual variables in the ACS. Population counts reported in the 2010 Census were compared to the 2005–2009 ACS, and they fell within the margins of error reported.10 At the time this research was conducted, a limited amount of 2010 U.S. Census data was available. These data were part of the summary file 1 (SF1) data release, which includes full count data for general population and housing characteristics. More detailed data for population and housing characteristics were subsequently released. These data are part of the survey component of the U.S. Census which is included in the ACS estimates; 2010 ACS estimates and 2006–2010 estimates were released using the 2010 census tract boundaries. After adjusting for changes in tract boundaries, a Pearson's correlation coefficient was calculated for Buffalo's 2005–2009 ACS estimates and 2010 census population counts at the tract level (R= 0.0956); 2005–2009 ACS estimates were used in the analysis for this paper, in part, because of their compatibility with other data used in this analysis. This issue is elaborated upon in footnote 11.11 The necessity of using 2000 census tract boundaries represents a limitation of this study. HUD Aggregated USPS Administrative Data on Address Vacancies were only released at this level of analysis. Consequently, 5-year estimates for the 2005–2009 ACS were used in this analysis. However, there were some advantages to using these data. First, a number of census tracts in Buffalo were consolidated between 2000 and 2010. In addition, some census tracts were split in a manner that did not maintain the integrity of 2000 block groups. Using 2010 tract boundaries would have resulted in a loss of twelve census tracts, reducing the sample size from 89 to 77. The option of using block group data was also limited due to issues of suppression in the data released by the U.S. Census, and the inability to obtain block group or individual address data from the USPS.12 The number of census tracts and census tract boundaries remained constant in the city of Buffalo for the 1990 decennial census, 2000 decennial census, and all ACS years through 2009. In 2010 the U.S. Census adjusted tract boundaries in the city of Buffalo. The total number of census tracts was reduced from 89 to 77. The adjustments were made in response to population decline in the city.13 Linear regression was used in this analysis since the dependent variables were interval, continuous. Because of this characteristic, other methods, such as logistic regression and Poisson count models, were not applied to this analysis. Such analytic frameworks require dependent variables that are discrete and based on count data. Although this analysis was exploratory in nature, future studies should consider more sophisticated techniques that apply advanced models to such data. Ideally, such data would be reported at the address or parcel level and not aggregated at the census tract level. The availability of parcel level data would also enhance the use of methods focusing on spatial analysis and diagnostic tools designed to test for spatial autocorrelation. These techniques were not applied to this analysis and represent a limitation of the available data and research design.14 One method commonly used to measure segregation is the calculation of a white/black dissimilarity index. This index identifies the percent of blacks who would have to relocate in order to produce a completely integrated community. Hyper-segregation is suspected when the white/black dissimilarity index is above a value of 0.70 and minorities remain concentrated in a core city area that has experienced general population decline for several decades.15 Some may argue that segregation is, in part, driven by more benign forms of racial sorting and expressions of individual tastes and preferences. In this article, we emphasize the broader institutional factors that structure racial segregation in an inner-city context.16 The estimate for the percent of properties in rem was calculated using the total number of vacant addresses reported in the USPS data. This value was used to calculate the estimate since in rem parcels are composed of residential and nonresidential properties. It should be noted that the estimate for the percent of in rem properties is based on two point-in-time counts from different data sources.17 This analysis examined correlations between the independent variables and the dependent variables examined in the linear regression models. Establishing direct causal relationships is beyond the scope of this analysis.18 In addition to the distinction between housing units and addresses, each database approaches the identification of properties with different objectives. For instance, the U.S. Census has a tendency to overestimate the number of housing units since it is required to make an effort to collect decennial census data from all housing units regardless of their occupancy status. One consequence of the U.S. Census methodology is that its estimate of all vacant units includes "other" vacant units. Consequently, long-term abandoned properties are intermingled with census counts of vacancy. In contrast, the USPS only identifies addresses that received mail during the past five years in its count of properties. The USPS maintains a separate count of "no-stat" addresses which have no current postal status. In 2010 there were 3,593 residential addresses that fit into this category. However, it is not possible to differential among no-stat address. For example, no-stat addresses include those that are potentially abandoned, new construction that is not occupied, vacant lots, and non-vacant addresses where occupants receive mail at a post office box instead of through home delivery. Because there was no clear methodology to disentangle no-stat addresses, we followed HUD's general cautionary notes and excluded them from the analysis.19 The most notable distinction between these two dependent variables is that the percent of "other" vacant housing units accounts for all properties that are not being offered for rent or sale, being held for future occupancy, or limited to seasonal or occasional use. In contrast, in rem properties are comprised of a subset of vacant and abandoned properties that have amassed outstanding debt and for which the City of Buffalo has taken proactive steps to take possession. These properties may be viewed as worst-case abandoned properties or those that the City has prioritized for demolition, revitalization, or other actions.20 Population change was not significantly related to vacancy and abandonment in any of the models presented in Table 3, and multicolleniarity was not detected with this variable when regression diagnostics were run. In part, this is a reflection of population decline occurring citywide. We suspect this also reflects the degree to which other factors impacted neighborhoods differentially across the city, such as declining household sizes, gentrification, and demolitions. Further analysis at the micro level is necessary in the future to disentangle these effects.21 Separate models were run using residential and nonresidential in rem properties as the dependent variable. Significant independent variables were only identified in the latter case. The estimated percent of the population in poverty was correlated with nonresidential in rem property. However, examining all in rem property revealed relationships between abandonment and characteristics of the built environment that were pertinent to this analysis.22 Protocols of this nature exist for other data sets, like data compiled by the National Center for Education Statistics. HUD and the USPS should model protocols for restricted use of parcel level data after similar policies that are in place.