to Moderate-Income Neighborhoods

Social Capital and Homeownership in Lowto Moderate-Income Neighborhoods Michal Grinstein-Weiss, Yeong Hun Yeo, Kim R. Manturuk, Mathieu R. Despard, Krista A. Holub, Johanna K. P. Greeson, and Roberto G. Quercia This study examined the relationship between homeownership and social capital among low- and moderate-income (LMI) households. Using data from the Community Advantage Panel Study, the authors used propensity score weighting and regression analyses to explore the relationship between LMI homeownership, neighborhood conditions, and social capital. After controlling for several important individual- and neighborhood-level characteristics, the authors found that homeownership is related to greater access to social resources in general but not to social resources within the neighborhood. Instead, resource generation within the neighborhood is largely predicted by neighborhood stability and perceived neighborhood size. Policy implications are discussed. KEY WORDS:

Community Advantage Home Loan Secondary Market Program; homeownership; low-income; resource generation; social capital

H

omeownership has long played an integral role in wealth and asset accumulation, community growth, and the promotion of positive social outcomes for individuals and families in the United States (Retsinas & Belsky, 2002; Rohe & Stewart, 1996; Shlay, 2006). Since the 1930s, government policies such as tax incentives, subsidy payments, and market regulation have promoted homeownership as a vehicle to stimulate economic growth (Carliner, 1988). Since the 1980s, they have also been used to spur community redevelopment and provide improved housing options for low-income families (Shlay, 2006). In light of economic and racial disparities in homeownership rates (Collins, 2002; Herbert, Haurin, Rosenthal, & Duda, 2005; Williams, Nesiba, & McConnell, 2005), recent economic and social policies have focused on promotion of homeownership as a way to provide long-term economic stability and other benefits to low- and moderate-income (LMI) families who may not have benefited from earlier policies. As a result, the rate of entry into homeownership for low-income and minority households has been increasing faster than that for other groups (Belsky & Duda, 2002). Research has supported the belief that the benefits of homeownership, such as life satisfaction and neighborhood stability, transcend wealth accumulation for both homeowners and communities (Rohe,

doi: 10.1093/swr/svs035

© 2013 National Association of Social Workers

Van Zandt, & McCarthy, 2002), although this body of research has primarily focused on middle- and high-income populations. It is important to examine whether LMI individuals and families also experience the positive benefits of homeownership found for middle- and higher-income households. Furthermore, the recent housing crisis, which has been blamed in part on the expansion of risky mortgages to LMI borrowers, underscores the need for a careful look at the costs and benefits of LMI homeownership. This study focused on one perceived benefit of homeownership: its potential impact on access to social capital. Specifically, using data from the 2007 Community Advantage Program (CAP) panel survey and the 2000 U.S. Census, this study examined the differences in the level of access to social resources between LMI homeowners and LMI renters and what role neighborhood economic and social conditions play in the relationship between resource generation (RG) and homeownership. LITERATURE REVIEW

Research has sought to identify externalities of homeownership and the potential social benefits that help determine the overall impacts of promotion of homeownership. Rohe et al., (2002) conducted a critical assessment of the research on the costs and benefits of homeownership and concluded there is evidence of positive benefits of homeownership, but they also cautioned that the

37

potential negative impacts should be investigated further. Numerous studies lend support to the belief that homeownership is related to increased involvement in local organizations, neighborhood stability, local problem solving, and satisfaction (DiPasquale & Glaeser, 1999; Rohe & Stegman, 1994b; Rohe & Stewart, 1996). Several studies on the relationship between homeownership and social capital have focused on the levels of participation in organizations as an indicator of investment in the neighborhood (DiPasquale & Glaeser, 1999; Rohe & Stegman, 1994b). In a longitudinal study on social capital, Rohe and Stegman (1994b) examined a sample of 143 low-income homebuyers participating in an affordable homeownership program in Baltimore, Maryland, sponsored by the Enterprise Foundation. A comparison group included 140 randomly selected, non-elderly, Section 8, low-income renters. They interviewed participants on a series of measures of community involvement while controlling for characteristics such as income, age, and education. They found that homeowners were more likely to participate in neighborhood and block associations but not in other community organizations, such as political or social organizations, compared with a matched set of renters. However, they also found that homeowners had lower levels of involvement with neighbors, suggesting that homeowners may be more likely to participate in formal social interactions at the neighborhood level but less likely to participate in informal social interactions. Rohe and Stegman (1994b) offered two explanations for these findings. First, the significant number of new homeownership units clustered together may have limited the existing informal networks that typically help integrate new residents into neighborhoods. Second, homebuyers are thought to have a higher economic interest in the conditions of the local area and thus may be more likely to participate in formal interactions such as block associations. Other studies on the effects of homeownership on both informal and formal local social interaction over time have supported the finding that homeownership is related to increased formal social interactions but not informal ones (Rohe & Basolo, 1997; Rossi & Weber, 1996). Rossi and Weber used three public use data sets—the General Social Survey, the National Survey of Families

38

and Households, and the American National Election Studies—to identify social correlates of homeownership. They found that renters are shown to be more sociable than owners in terms of spending time with neighbors, coworkers, or friends. Thus, homeowners are less neighborly than renters. However, the authors cautioned that the homeowners and renters in this sample differed in important ways, but only age and socioeconomic status were controlled for, so other variables could help explain the differences. For example, the authors suggested that homeowners may have more children in the household and thus may be more likely to interact socially with family rather than neighbors. Rossi and Weber concluded from their research that the overall social effects of homeownership are minimal and inconsistent. SOCIAL CAPITAL THEORY

The concept of social capital has been defined in myriad ways and can be conceptualized as either an individual resource (connections to others that yield material or social benefits; Bourdieu, 1985) or a collective resource (shared stocks of trust that contribute to collective problem solving; Putnam, 1995). For the purposes of this study, we focused exclusively on individual social capital as defined by Van der Gaag and Snijders (2004): Social capital is “the collection of resources owned by the members of an individual’s personal social network, which may become available to the individual as a result of the history of these relationships” ( p. 200). The present study draws on network theory (Lin, 1999), which characterizes social capital as assets in networks by highlighting investment in social relations with expected returns. Network theory, as defined by Lin, differentiates the access and the use of social capital. Access-based social capital is a collection of potentially usable social resources. Use-based social capital, on the other hand, refers to actions and consumption of resources to generate returns (Lin, 2001). We operationalize access-based social capital in this study as RG, which focused on the degree to which respondents had access to various social and economic resources through social networks. This study used a modified version of the Resource Generation instrument developed by Snijders (1999), a version based on Lin’s network theory, as a way to conceive of social capital as an individual pool of resources embedded in personal networks.

Social Work Research Volume 37, Number 1 March 2013

The theoretical relationship between social capital and homeownership has been explained in terms of financial motivation. The economic rational choice theory posits that homeowners have a higher financial stake in their neighborhoods and communities because house values are affected by neighborhood and community vitality. Thus, homeowners are more likely than renters to get involved in politics, community organizations, or other efforts to improve their neighborhoods in order to protect their investment (DiPasquale & Glaeser, 1999; Herbert & Belsky, 2006). For example, local land-use decisions affect property values, thus homeowners have a higher economic stake in these political decisions. Likewise, neighborhood safety is likely to affect home values, thus encouraging homeowner involvement in neighborhood watch groups. However, research that examined economic reasons for home buying and social capital found no evidence to support the economic rational choice theory, meaning that economically motivated homebuyers were no more likely than buyers with other motivations to have higher social capital (DiPasquale & Glaeser, 1999; Rohe & Stegman, 1994a). A second explanation, related to financial motivations that developed as part of the rational choice theory, is the cost of moving, which is typically higher for homeowners than renters. Due to the high transaction costs associated with moving, homeowners may be more likely to improve and maintain neighborhoods than to move away (DiPasquale & Glaeser, 1999; Herbert & Belsky, 2006). This increased involvement and residential stability mean that homeowners develop greater personal and emotional connections to neighbors, their homes, and the area. These connections motivate them to form ties with others (DiPasquale & Glaeser, 1999; Herbert & Belsky, 2006; Retsinas & Belsky, 2002). These ties to others may provide homeowners increased access to social capital, which is to say access to social resources held by others. NEIGHBORHOOD CHARACTERISTICS

LMI homeowners, who are more restricted than other homeowners to purchasing housing in affordable neighborhoods, are likely to differ in important ways from middle- and upper-income homeowners. It is therefore unclear whether the relationship between homeownership and social

capital will hold for this particular group. One way that LMI homeowners are likely to differ from homeowners in general is with respect to the characteristics of the neighborhoods in which they reside. LMI homeowners in impoverished neighborhoods may experience a different set of social processes that result in a different relationship between homeownership and social capital than that seen in existing literature for middle- and upper-income homeowners. In fact, some research has found that neighborhood poverty is negatively related to available social capital. For example, Caughy, O’Campo, and Muntaner (2003) found that in high-poverty neighborhoods, individuals’ general sense of community was lower. Sampson, Morenoff, and Gannon-Rowley (2002) provided insight into one potential reason for this relationship, in that they characterized highpoverty neighborhoods as centers of a variety of social problems, including homicide, suicide, infant mortality, low birth weight, teenage pregnancy, and sexual risk taking (Sampson, 2002). As a result, residents of such neighborhoods may feel less comfortable nurturing ties with one another, which would lead to low social capital. Low social capital is further theorized to be predictive of higher rates of crime (Sampson & Groves, 1989). This suggests that there may be a damaging negative feedback loop in highpoverty neighborhoods where crime and other social problems result in reduced social capital, creating a disorganized environment in which social problems increase. If such a characterization of highpoverty neighborhoods is accurate, we would not expect increased social capital for homeowners living in those neighborhoods. Limited mobility due to homeownership may impose financial costs that exceed any increased benefits of social capital (DiPasquale & Glaeser, 1999) by preventing these households from leaving a neighborhood that is experiencing increasing disinvestment and depreciation (Rohe, Van Zandt, & McCarthy, 2002). On the other hand, some literature has suggested that more economically disadvantaged neighborhoods have high levels of socially supportive networks and resources. One example is the ethnographic work of Stack (1974), who found that lower-income individuals tended to develop strong social support networks to cope with the uncertainties of economic hardship. In a quantitative study, Boisjoly, Duncan, and Hofferth (1995) similarly found that families in very poor neighborhoods

Grinstein-Weiss et al. / Social Capital and Homeownership in Low- to Moderate-Income Neighborhoods

39

reported access to increased social capital from relationships with friends and lower social isolation. However, Boisjoly et al. defined social capital as access to time and monetary help from friends and family as measured through the individual’s perceived access to social capital, not whether the individual actually received the help. Thus, the contradictory findings of high social capital in these studies and not in others could be due to the fact that social capital is defined and measured differently. If neighborhood disadvantage does contribute to higher levels of social capital, then we would expect LMI homeownership in disadvantaged neighborhoods to show a similar or stronger positive relationship between homeownership and social capital, as has been shown in higher-income homeowners. Of course, the relationship between neighborhood disadvantage and social capital is likely more complicated than simple positive or negative associations. Other variables, such as the overall stability of the neighborhood and the mobility of the individual in question, likely have an effect on one’s social capital. The influence of these variables is supported by findings that residential stability is related to local friendship networks and fewer social problems (Sampson & Graif, 2009; Sampson et al., 2002). Furthermore, Greenbaum (2008) found that low-income families forced to move to more advantaged neighborhoods as a result of public housing demolition actually had much lower social capital after the move due to the disruption of their existing social networks. Such research suggests that individuals with homes in disadvantaged yet relatively stable neighborhoods may have higher levels of social capital, whereas homeowners who have recently moved into a neighborhood may have lower levels of social capital, regardless of other neighborhood factors. Furthermore, along with the difficulty of disentangling the effects of neighborhood characteristics from homeownership, research has suggested that the effects of neighborhood characteristics may vary by group. Harkness and Newman (2003) used data from the Panel Study of Income Dynamics (waves 1968–1993) to compare results between homeowners and renters while allowing for the interaction between tenure and neighborhood characteristics. To follow up on research that suggested that growing up in a home-owning family confers benefits on children, their research

40

focused on children between the ages of 11 and 15 years. Their results showed that neighborhood characteristics mattered more for the children of homeowners than for children of renters. Thus, homeowner children were more negatively affected by neighborhood distress and more positively affected by the level of homeownership in the neighborhood than renter children. Three issues arise when examining the literature on the relationship between homeownership and social capital: (1) the need to differentiate between the effect due to the homeownership rate and that due to residential stability (DiPasquale & Glaeser, 1999; Harkness & Newman, 2003); (2) selfselection bias: people who are naturally more likely to be homeowners (those who are employed, married, and have higher income) may also have higher levels of social capital (Rohe et al., 2002); and (3) the neighborhood-level social and economic characteristics that may influence the degree to which residents participate in the civic life of their neighborhood (Rohe et al., 2002). Our study aimed to address these issues in the following ways. First, we controlled for neighborhood stability through two indicators, percentage of residents living in the same house at least five years and percentage of owners in the neighborhood. Second, to control for selection bias, this study used propensity score weighting (PSW) to match homeowners to renters. Third, we directly tested whether neighborhood conditions such as poverty, unemployment, and high mobility had an impact on the social resources held by those in the neighborhood. Social capital has been used to explain outcomes at both the individual (Hao, 1994) and community (Putnam, 1995) level. Although researchers have extensively documented what social capital does for people, we know less about how it is generated in neighborhoods and communities. This study aimed to examine the relationship between LMI homeownership and social capital and to explore neighborhood effects on social capital. Based on the theoretical framework outlined previously, we developed the two following research questions: (1) Do LMI homeowners report having access to greater social capital than a comparable group of renters in models controlling for individual and neighborhood characteristics? and (2) Are neighborhood characteristics related to the social capital of LMI homeowners and renters?


METHOD

Data and Sample

This study used data from the 2007 CAP panel survey for individual-level information and data from the 2000 U.S. Census for neighborhoodlevel information. CAP is a secondary-market mortgage pilot program for LMI households. It enables borrowers with lower credit scores to obtain prime financing for homeownership. To qualify for CAP, applicants must meet at least one of the following three criteria: (1) have an income less than 80% of the area median income (AMI); (2) have racial or ethnic minority status and income less than 120% of AMI; or (3) purchase a home in a high minority (greater than 30% concentration of minority populations) or lowincome (less than 80% AMI) census tract area and have an income less than 120% of AMI. Started in 1994 in North Carolina by the SelfHelp Ventures Fund, a community development financial institution, CAP expanded nationally in 1998 through a partnership with the Ford Foundation and Fannie Mae (Federal Reserve Bank of Philadelphia, 2004). The goals of CAP are to demonstrate the credit worthiness of LMI borrowers to secondary mortgage lenders and to provide evidence to lenders and policymakers that LMI borrowers are “bankable.” Through an annual panel survey, CAP also provides important information about the effects of LMI homeownership on a variety of economic and social outcomes. Since its inception, the program has funded more than 46,000 mortgages across the United States. The median loan amount is $78,800, and the median income of a CAP borrower is 60% of AMI. Thirty-nine percent of the borrowers are minority, and 44% are female-headed households. More than half (52%) of the loans have an original loan-to-value ratio of 97% or above, and 44% of the borrowers had a credit score below 660 at origination. The CAP panel survey, conducted by Research Triangle Institute International, consists of six annual in-depth telephone interviews of a sample of CAP homeowners and a matched comparison group of renters. The renter panel was chosen to match the homeowners in the owner panel based on neighborhood location and income criteria of the CAP program. The annual survey started in 2003 for the homeowner panel and in 2004 for

the matched renter panel. Because of the large number of study participants and the rigorous design, CAP provides an excellent opportunity to advance research on LMI homeownership. The 2007 CAP panel survey included 2,071 respondents from the owner panel and 893 from the renter panel. Among them, respondents who changed homeownership status since the first wave of the CAP survey were removed from the analysis for this study (n = 361). Thus, the analysis sample in this study included only respondents who were homeowners for at least four years or renters for at least four years. There were no significant differences in individual characteristics and RG between the final sample and the 361 excluded cases, except with regard to age. The analysis sample was older (M = 34.72, t = −3.54, p < .001) compared with the excluded cases (mean = 34.72). We also excluded 23 respondents over the age of 65 from the analysis because of skewed distribution of age and the small frequency. Further, we removed 564 respondents reporting household income greater than $60,280 (that is, 120% of the 2007 U.S. median household income). Trimming of household income ensured that our sample was composed of LMI households. Therefore, the final analysis sample included 1,918 LMI respondents (1,235 homeowners and 683 renters) from 1,452 census tract areas. Measures

In the analyses presented here, RG was used as a measure of social capital (Snijders, 1999) that focused on the degree to which respondents had access to various social resources through their social networks. There were two dependent variables for this study: (1) within-neighborhood RG (neighborhood RG) and (2) general resource generation (general RG). Scales measuring both general RG and neighborhood RG consisted of eight items, and both scales appeared to have good internal validity and consistency. Factor analyses indicated that the eight items for both general RG and neighborhood RG comprised a single factor for each concept, confirming that all eight items measure the same concept. Subsequent reliability tests also showed a good reliability for both general RG (α = .81) and neighborhood RG (α = .80). For neighborhood RG, the respondent was asked if she or he knew anyone outside the household,


41

but within the neighborhood, who could provide a given resource. Response options for each measure of neighborhood RG were dichotomous (0 = no; 1 = yes), and the variable neighborhood RG was a composite score of eight dichotomous items. For general RG, the respondent was asked the number of people they knew who could provide them with a given resource. The responses excluded household members but included people both within and outside a respondent’s neighborhood. To make general RG easily comparable with neighborhood RG, responses to general RG items were converted to dichotomized measures of whether or not respondents knew one or more people who could provide a given resource. Thus, like neighborhood RG, the outcome variable used for general RG in these analyses was a composite score of eight dichotomous measures. The independent variables for this study included various individual characteristics and neighborhood conditions (see Table 1 for descriptive statistics). Individual-level independent variables included the following 12 variables: (1) tenure (0 = renters; 1 = owners); (2) gender (0 = female; 1 = male); (3) age (in years); (4) a set of dummy variables indicating race/ethnicity: white (the reference group), black, Hispanic, and any other race/ethnicity; (5) a set of dummy variables for education level: no high school diploma, high school diploma or GED (the reference group), some college, and bachelor’s degree or more; (6) a dichotomous variable for the employment status of the respondent (0 = non-employed; 1 = employed); (7) a set of dummy variables for marital status: partnered, married (the reference group), separated/divorced/widowed, and never married; (8) the number of adults in the household; (9) the number of children in the household; (10) a set of dummy variables for total annual household income: less than $10,000 (the reference group), $10,000– $19,999, $20,000–$29,999, $30,000–$39,999, $40,000–$49,999, and $50,000 or more; (11) a set of dummy variables for perceived neighborhood size, representing how respondents defined their neighborhood: the block or street you live on (the reference group), several blocks or streets in each direction, the area within a 15-minute walk, and an area larger than a 15-minute walk; and (12) a dichotomous measure of moved to a new neighborhood since the last survey wave (0 = no; 1 = yes), to adjust for the length of time living in the neighborhood.

42

In addition to various individual characteristics, this study also included perceived neighborhood size and recent movers as control variables. It is reasonable to expect the respondents who conceived a broader neighborhood boundary to have more social resources within the neighborhood than the respondents who defined a narrower neighborhood boundary. Thus, it was important to control each respondent’s different definition of the neighborhood based on size. It was also important to control the length of time in which a respondent had resided in the neighborhood. The recent movers into a neighborhood would be expected to have a lower level of social resources within the neighborhood than their counterparts. In this study, neighborhood information came from the 2000 U.S. Census. Census tract information was assigned to each respondent on the basis of his or her address reported at the time of the survey. Two neighborhood variables were the concentrated economic disadvantage (CED) scale (Caughy, Hayslett-McCall, & O’Campo, 2007; Sampson, Raudenbush, & Earls, 1997) and the neighborhood stability scale (Morenoff, Sampson, & Raudenbush, 2001; Swaroop & Morenoff, 2006). Among various neighborhood characteristics, these two neighborhood variables were selected to represent neighborhood conditions in our study because neighborhood economic surroundings and neighborhood stability were expected to be important neighborhood factors related to social capital, as suggested by previous studies. The CED scale represents the relative economic condition of a neighborhood and was constructed using the following four census tract indicators: percentage of individuals below the poverty line, percentage of people receiving public assistance, percentage of people unemployed, and percentage of female-headed households with children. To develop the CED scale, each indicator was first standardized and then a composite score was divided by 4, the number of indicators (α = .91). The population stability scale was developed by combining two census tract indicators: percentage of homeowners and percentage of residents who lived in the same house for at least five years. These two measures were standardized and combined into a composite score, which was then divided by 2, the number of measures (α = .68).


Analytic Plan

Although the CAP panel survey was intended for LMI homeowners and renters to be matched in terms of income eligibility for CAP program and location, they appeared to differ greatly on almost all individual characteristics (for example, gender, race, marital status, income). The imbalance between the two groups raises questions about the potential effects of sample selection and endogeneity. Self-selection bias exists to the extent that LMI people in our sample with the ability to more easily access and form social capital may have also been more likely to have been homeowners at baseline. In other words, homeownership can be both a cause and a product of social resources. Tenure would be endogenous if the decision or ability to be a homeowner was correlated with observable or unobservable factors that affect the level of social capital. In the absence of random assignment of tenure status at baseline, the estimate of causal relationship from tenure to social resources would be questionable. Of course, housing tenure cannot be assigned randomly. Alternatively, this study applied propensity score analysis to remedy potential problems of sample selection and endogeneity in the data (Freedman & Berk, 2008; Leslie & Ghomrawi, 2008). Among various propensity score methods, this study used PSW for the following reasons. First, most other propensity matching methods (for example, kernel matching, full optimal matching, matching estimator) do not produce a coefficient of each covariate. Because we were interested in neighborhood effects and a tenure effect on resource generation, presenting estimations of neighborhood effects after propensity score analysis was critical. Second, PSW handles non-normal distributed outcomes unlike other matching methods. Third, one-to-one matching (for example, greedy matching) also provides an estimation of each covariate and works with nonnormal distributed outcomes. However, it substantially decreases a sample size, which can be criticized for sample representativeness. PSW in our data was found to be efficient at making the sample balanced (see Appendix A); thus, we used it for our propensity score scheme. Regression adjustment with PSW proceeded as follows. A logistic regression predicting group membership (homeowner versus renter) was

conducted to estimate a propensity score, which was the conditional probability of a participant to be a homeowner and based on observable characteristics. Logistic regression included the same set of independent variables as those later used in regression models to predict RG. Propensity score weights were calculated as the inverse of the propensity score for the treatment group (that is, homeowners) and as the inverse of one minus its propensity score for the nontreatment group (that is, renters). Because our dependent variables were count variables with overdispersed distribution, negative binomial regression was used to assess the relationship between tenure and RG. This study ran three models for each outcome. The first model, which controlled for individual and neighborhood characteristics, was conducted without applying PSW. This model did not account for the endogeneity and sample imbalance issue. The second model added PSW to remedy possible endogeneity and sample selection problems. To explore whether neighborhood stability and economic conditions alter the significance of the relationship between tenure and RG, a third model was run with PSW, but it did not include neighborhood characteristics. The strength of relationships between covariates and RG was evaluated using incidence rate ratios (IRRs), calculated by exponentiating the negative binomial regression coefficient. IRRs can be interpreted as the relative change in the incidence rate of outcome variable Y for a one-unit change in a given independent variable X. RESULTS

Differences of sample characteristics between LMI homeowners and LMI renters are presented in Table 1. In spite of efforts to match LMI homeowners with LMI renters in terms of income eligibility for CAP program and location, the differences between homeowners and renters in our sample were significant. For all individual indicators, homeowners had different socioeconomic profiles than renters. The majority of owners were male, white, and married. However, most renters were female, nonwhite, and not married. Most owners were located in higher income categories, but the majority of renters were in lower income categories ( p < .001). Regarding census-level indicators, homeowners lived in less economically


43

Table 1: Characteristics by LMI Homeowners and Renters in the 2007 CAP Survey (Analytic Sample) Owners (n = 1,235) Characteristic

Individual characteristics Male Age Race/ethnicity White Black Hispanic Others Education Less than high school graduate High school graduate Some college BA and more Employed Marital status Partnered Married Separated/divorced/widowed Single Household characteristics Number of children Number of adults Annual household income