Housing Foreclosure as A Geographically Contingent Event: Columbus Ohio 2003–2007

Abstract

This article seeks to better understand geographic manifestations of housing foreclosure, moving beyond the usual portrayal that highlights, e.g., race/ethnicity and income. We depart from the usual analytical strategy which centers on factors that subsume high proportions of variance. Instead, this is the starting point for considering constellations and idiosyncratic but formative characteristics—contingencies—that further understanding of, e.g., why two households with identical attributes experience different outcomes. Empirical focus is on Columbus Ohio, 2003–2007. Regression analysis identifies central tendencies, followed by regression tree procedures that reveal variable combinations which alter correlational expectations. Unique areas are examined by neighborhood reconnaissance, exploratory data analysis, interviews, and archival research. Relevant factors include race/ethnicity and socio-economic characteristics. Beyond that, differing variable combinations lead to different outcomes, as do processes such as neighborhood life cycle, institutional actions/involvement, and year of home purchase/construction relative to housing de/inflation and mortgage market characteristics.

Key words:

• Columbus Ohio
• Contingency Foreclosure
• Housing
• Race/Ethnicity
• Subprime Lending
• Urban.

Acknowledgements

Comments from participants of the 2008 and 2009 meetings of the Association of American Geographers and European Regional Science Association, 2008 International Geographical Congress, 2009 International Sociological Association, 2009 Western Regional Science Association, 2010 International Conference of Population Geographies VI, and at Arizona State University, as well as Kevin Cox, Kevin Grove, Roy Heidelberg, John Paul Jones, Linda Lobao, Ed Malecki, Elvin Wyly as editor of Urban Geography, and anonymous reviewers are appreciated. We also value technical and data acquisition assistance provided by Ohio State's Center for Urban and Regional Analysis, Wenqin Chen in particular, and DataSource of Community Research Partners, Columbus Ohio. Finally, we acknowledge Alan Findlay who, as Chair of the IGU Commission on Population Geography, set out population vulnerability as a major theme, which inspired the approach taken in this article.

Notes

² Contingency per se has been given a good deal of attention in geography, largely as an abstract concept intertwined with realism, place context, and among specific topics, the meaning of race (Jones and Hanham, 1995; Holloway, 2000; Sayer, 2000). Our approach avoids such debates, simply taking the term at face value; i.e., that any given relationship is uncertain in so far as it can be altered by intervening circumstances or characteristics. In this vein, Holloway (2000, 199) says “social processes are modified and altered [in so far as] they are contingent upon the contexts in which they are embedded.” Albeit in a slightly different context, Holloway (198) also says “The multiple dimensions of identity do not just intersect as independent additive forces, they interact.” Our use of contingency also is similar to that of Holloway (1998) in “Exploring the Neighborhood Contingency of Race Discrimination in Mortgage Lending in Columbus, Ohio.”

³ As an example where two attributes acting conjointly have greater impact than the sum of each taken separately, Hogan and Marandola (2005, 463) note with regard to population vulnerability that instead of “focusing only on race, [class,] or gender … researchers should try to ‘unpack intersectionality’, looking at the multiple discrimination suffered, for example, by [Black women in poverty] … .”

⁴ These terms are not equivalent in statistics; e.g., explained variance would primarily be associated with multivariate tools and central tendency with distributional measures such as mean and variance. Conceptually, however, we see these as similar in so far as a set of statistically significant variables represent the central tendency of all variables (or forces) acting on the phenomenon being accounted for.

⁵ The approach taken here is applicable to a wide range of social science topics and to social science epistemology more generally, but elaborating that point is beyond the scope of this paper. It also has a long tradition, dating at least to General Systems Theory thinking of the 1950s through 1970s, as epitomized by Chorley's (1962) emphasis on “open system thinking” and Boulding (1956, 198) who notes “General Systems Theory … hopes to develop something like a ‘spectrum’ of theories … which may perform the function of a ‘gestalt’ in theoretical construction … directing research towards the gaps … [thus] reveal[ed].”

⁶ For many, a geographical perspective necessarily involves spatial statistics, cartographic analysis, etc., which attests to the success of programs such as the Center for Spatially Integrated Social Science (www.csiss.org). While agreeing with this thrust, we also invoke a broader view of what a geographer brings to social science inquiry.

⁷ Indeed, the mortgage-backed securities (MBSs) market—comprised of, e.g., investment houses, pension funds, money managers, etc.—exploded from $1.2 billion in 1985 to a $1,252 billion peak in 2006 (Securities Industry and Financial Markets Association, 2008). As indicated, buyers felt they were insured against risk through over-collateralization, and by bonds being divided into components with differing levels of exposure, known as tranches. The riskiest tranches, in turn, could be pooled, repackaged into new debt, and again, over-collateralized so as to receive a satisfactory credit rating (Salmon, 2007). In this context, “over-collateralized” refers to the situation where a bond selling for a given price is secured by mortgages which, taken together, exceed that price.

⁸ Without overtly advocating a correspondence with natural or human-spawned disasters, we nevertheless must note (as “food for thought”) that foreclosure came upon society as an external, overpowering, startling phenomenon that wreaked havoc among the population at large. Another connection is vulnerability research's focus on the cause, mitigation, control, and/or prevention of disasters—which is analogous to foreclosure-related actions such as bank bailouts, judicial proceedings, and the Dodd–Frank Wall Street Reform and Consumer Protection Act of 2010. Similarly, controlling the consequences for society at large, socio-economic systems, and/or individuals is represented in foreclosure by, e.g., the American Recovery and Reinvestment Act of 2009 (ARRA). The analogy also resonates in that both research enterprise have been concerned with delineating characteristics of susceptibility—e.g., as done for New Orleans under Katrina, by mapping a vulnerability landscape based on the physical and socio-economic properties of census tracts (Finch et al., 2010). More abstractly, Swyngedouw (1996, 2006) notes that cities and processes shaping them represent the outcomes of both political-economic/capital accumulation forces and those of nature—a “dialectic between nature and society” or an urban-based political ecology (1996, 69)—a perspective which, in our opinion, applies equally well to the traditional concerns of disaster research. An empirical example of linking nature-type metaphors and urban decay is Wallace and Wallace's (2000) study of urban health in New York City (e.g., increased tuberculosis, homicide, low birth weight) in which they “hypothesize that a massive socioeconomic catastrophe … which affected the poor neighborhoods of New York City in the 1970s, impose[d] an evolutionary shift in community structure and function,” using as an analogy “the impact of the giant meteorite … between the Cretaceous and Triassic Eras [which] shifted the direction of ‘time's arrow’ for global biota.” (1247)—and employing an “amplification factor” approach from ecosystem science to gauge resiliency and vulnerability, even though the latter might not be discernible (1248). More generally, we read Swyngedouw and the Wallaces as saying that nothing is either purely natural or socio-ecological—rather that both are present within urban areas and the sites of natural disaster.

⁹ Available data confined our analysis to Franklin County, rather than the entire seven-county Columbus MSA.

¹⁰ These statistics are approximate in that different sources provide noticeably different numbers. For example, parcel-level data from The Franklin County Sheriff Department indicate foreclosures in 2001 through 2008 to be 2,263, 2,962, 3,414, 4,657, 4,601, 5,918, 7,064, and 6,999; while data from the Ohio Supreme Court indicate 8,876 for 2006, 9,145 for 2007, and 9,307 for 2008 (Rothstein and Mehta, 2009). One element of difference is the distinction between cases filed and cases actually brought to term. Related to this, foreclosure is a judicial process in 21 states, including Ohio; elsewhere, it can be initiated by lenders outside of the court system. In general, the judicial process is more intractable, drawn out, and involves more bureaucratic hurdles (Streitfeld, 2010).

¹¹ As used here, “ecological” refers to variables that represent spatial aggregates such as census tracts or block groups; i.e., as noted by Openshaw (1984, 19) “in ecological correlation … the variables are descriptive properties of areas rather than of individuals.” In the current context, then, our block group analyses are undergirded by individual characteristics that are represented by that unit (e.g., income, race, housing age), an approach sometimes referred to as urban ecology. This term traces its origin to the Chicago School efforts of the early and middle twentieth century, but more currently is simply associated with analysis of spatial units using secondary data sources such as the United States Census (Berry and Kasarda, 1977, 6–7). As such, it refers to characteristics of a place (e.g., census tract) in terms of its SES, Demog, Built Env etc., composition—as in the Urban Ecology perspective of sociology and related disciplines, or what geographers might refer to as Urban Morphology.

¹² An LQ value of 1.0 indicates the BG and county rates are identical; <1.0 indicates the BG foreclosure rate is lower than the county; and >1.0 indicates a foreclosure rate that is higher.

¹³ Regression tree methods, a popular version of which is CHAID (Chi-Square Automatic Interaction Detection), have been most commonly used in marketing to set out target groups, and in policy settings to gauge the likelihood of different outcomes. The approach also has contributed to scholarly endeavors such as Mandel (2001) on the configuration of women's livelihood strategies in Benin. A summary of the methodology is provided by Fomunung et al. (1999, 342–343) who apply it to vehicle emissions. For broader coverage, see Wrigley (2002, ch 10, Statistical Models for Discrete Choice Analysis). Less technical descriptions are readily found through web references; e.g., StatSoft Electronic Textbook (www.statsoft.com/textbook/chaid-analysis/) and SmartDrill Data Mining (smartdrill.com/CHAID.html). These and many other references appeared by searching for the keyword CHAID.

¹⁴ Regression tree's intersection with, but distinction from, cluster analysis techniques also should be noted. Both provide groupings of areal units and for each grouping, their average for a critical variable, here the areal unit's foreclosure rate expressed as a location quotient (LQ). In RT analysis, however, group formation is based on independent variables, their efficacy in accounting for variance in the critical variable, and the average LQ describes the group but is not a basis for grouping. The typical cluster analysis (e.g., K-means clustering) also provides the average LQ for each group, but unlike RT, that variable is itself the basis for group formation (Norusis, 2012).

¹⁵ RT LQ indicates that the given location quotient values are those derived from regression tree analysis. Since these are akin to the average LQ, calculated over all block groups in the branch, they differ from the BG LQ itself.

¹⁶ The one exception is 197 BGs that have less than 20.7% African American Homeownership, M-PEmploy greater than 27.1%, and greater than 72.7% owner occupied houses built 1990-2000. As elaborated below in our discussion of Marion Franklin-Groveport, this appears related to the purchase of homes as the housing bubble was expanding and mortgages themselves were becoming increasingly more risky.

¹⁷ Neighborhood reconnaissance of each area revealed few on-the-ground differences in housing stock between adjacent BGs. There remains, however, the possibility of highly localized contagion effects, whereby one or two foreclosures increase that likelihood for neighbors. Assessing this hypothesis requires temporal data on foreclosure at the parcel-level, which is examined in Murray et al. (2013).

¹⁸ Double-wide refers to trailer, or modular home, settlements in un-zoned, usually non-urban areas that begin with a single trailer and subsequently upgrade to one that is twice the original width. Key Informant 1 (KINF-1) is a long standing real estate agent, community developer for Columbus and Ohio State University, and former member of Columbus’ University Area Commission, which includes Ohio State.

¹⁹ Foreclosure rates for West Point and Lake Darby—and for other areas highlighted in this section—were derived from two sources. Franklin County Auditor (www.co.franklin.oh.us/Auditor) provided parcel-level information such as location, subdivision name, and transaction history. Franklin County Sheriff Department (sheriff.franklincountyohio.gov) provided foreclosure information at the parcel-level for 2003–2007 and 2001–2009.

²⁰ That individual actions or relationships should not be inferred from aggregate statistics is well known, but it is nevertheless done by researchers and others. For example, suppose that Lake Darby subdivision, as described, also consisted of middle-income, but frugal minorities such that foreclosure was primarily among Whites. Under that scenario, an ecological fallacy error could lead to the (tempting) conclusion that foreclosure and minority ownership are related—which might be true elsewhere, but not in the present case!!

²¹ Symptomatic of the shift is Hague Avenue Methodist Church, a long-time neighborhood anchor for community organizations. This was renamed as the Cross Roads Methodist Church in recognition of its diminished community ties, its need to attract new members, and to revitalize more generally. Regarding American Home Mortgage, KINF-1 reports that its home office was in Melville New York; its main Ohio office in Cincinnati with branches in Dayton and Columbus; and that it went out of business in late 2007. Another active company was Red Brick Mortgage which aimed its messages to low income households, but avoided those with very low income because of heightened risk. For more on the TIN (or ITIN) segment of recent mortgage markets, see Jordan (2007).

²² Renting was more common than flipping because rents provided a good return relative to home purchase prices.

²³ Similar institutional actions, focused on the Near South portion of Figure 4-C, have been undertaken, but only in recent years. These are motivated by the presence there of Nationwide Children's Hospital which, together with the City and Ohio State University, is actively seeking to build upon and expand its national prominence (see, e.g., www.nationwidechildrens.org/community-relations).

²⁴ The names are pseudonyms used by Sharma, not those of the interviewees.

²⁵ This observation also applies to Hilltop-Bottoms and German Village-Near South communities. In those instances, however, the incomplete/incorrect finding probably would have highlighted income and/or SES more broadly, thus missing the neighborhood life cycle contingency.

²⁶ Again, at least one example here had an ethnic/racial component—KINF-2 who was “blinded by his mission” to save the neighborhood—which could lead to an incorrect finding under a more standard methodology.