Changes in Neighborhood-Level Concentrated Disadvantage and Social Networks Among Older Americans

This study uses social disorganization theory as an orienting framework. According to this theory, unfavorable economic circumstances cause declines in the neighborhood institutions and informal networks that control community members’ behavior while enhancing their ability to act in their common interests (Ansari, 2013). The resulting criminal and delinquent behavior lead residents to remain within their homes, socialize only with their strongest ties, and avoid community engagement while creating generalized distrust of both neighborhood residents and nonresidents (Aneshensel, 2009). This disorganization reduces neighborhoods’ attractiveness to potential in-migrants and residents’ relatives and friends, who might be reluctant to visit their threatening neighborhoods (York Cornwell & Behler, 2015).

These declines lead to stress and depression (Kim, 2010), which compromise residents’ ability and motivation to develop and strengthen relationships, both within and outside of their neighborhoods (York Cornwell & Behler, 2015). In addition, depression leads to unattractive behaviors that reduce one’s social desirability (Schaefer et al., 2011).

As local gathering places, including recreational venues, churches, and senior centers, weaken within declining neighborhoods (Sampson, 2012), residents lose opportunities to develop relationships (Glass & Balfour, 2003). As these institutions also attract nonresidents, both local and nonlocal ties are compromised (Van Eijk, 2010). As older persons are likely to have numerous ties with relatives and friends living outside their neighborhoods, both are important.

Whether due to retirement, reduced functionality, or other life course developments, older persons show heightened spatial confinement (Glass & Balfour, 2003), which exacerbates these neighborhood effects. Furthermore, for older persons who have lost important roles, including through retirement or widowhood, neighborhood involvements are substitutes (Cornwell et al., 2008), increasing dependence upon the immediate environment (Lawton & Nahemow, 1973).

Stable and changing neighborhood circumstances should be distinguished. Under stable neighborhood disadvantage, already adapted older residents might show limited close network turnover. However, exposure to worsened neighborhood circumstances to which one is unadjusted might compromise older residents’ development of local and nonlocal relationships.

Scholarship suggests that disadvantaged communities’ residents tend to have larger networks of strong ties and smaller networks of weaker connections (e.g., Curley, 2005; Gilchrist & Kyprianou, 2011; MacDonald et al., 2005). Stack’s (1974) ethnography showed how residents of a disadvantaged urban American community profited from large networks of strong ties characterized by mutual rights and obligations. These networks were key to survival and happiness under circumstances of deprivation and discrimination (Stack, 1974).

Smaller networks of weaker ties within disadvantaged communities are attributable to processes of social disorganization that impede forming new acquaintanceships. As new close ties often emerge through strengthening earlier weaker relationships (Elfring & Hulsink, 2007; Ihm & Castillo, 2017; Lorenzen, 2007), worsening communities can prevent development of new strong ties, even among residents with sizable close networks.

This study’s context is the Great Recession of 2007–2009, which was the most extensive global economic shock since the 1930s Great Depression (Meltzer et al., 2013). This economic shock involved the collapse of the property market, causing severe losses within the financial sector and an acute recession during which the median American family’s net worth decreased by 40% (Meltzer et al., 2013).

Through this recession, the average American neighborhood saw a 4 percentage point rise in unemployment rate, a 1 percentage point increase in rate of abandoned homes, and a 2 percentage point rise in poverty rate (Owens & Sampson, 2013). Neighborhoods that were disadvantaged or included high concentrations of racial minorities or immigrants prerecession were harder hit (Owens & Sampson, 2013), contributing to substantial variability in extent of decline (Kneebone & Holmes, 2016).

Important neighborhood processes underlay the community-level consequences. Higher home foreclosure rates and decreasing home values caused neighborhood-level externalities that hampered social cohesion, decreased homes’ upkeep, and increased criminal activity (Lerman & Zhang, 2012). Furthermore, many prosperous neighborhoods’ affluence decreased (Solari, 2012).

This study asks how changing neighborhood conditions through this recession affected the sizes of and turnover within older residents’ close networks. Based on social disorganization theory, it hypothesizes that rising neighborhood-level concentrated disadvantage caused smaller close networks, less gain of close ties, and more loss of close ties.

This study employed individual-level variables from the National Social Life, Health, and Aging Project (NSHAP) survey. Developed through a complex multistage area probability sampling design, this longitudinal panel study of a representative sample of older Americans focuses on health, well-being, and relationships. Wave 1 (2005–2006) involved 3,005 respondents from 57 to 85 years of age (75.5% response rate), 75.2% (n = 2,261) of whom participated in Wave 2 (2010–2011). The first two waves’ timing permits analyses of the recession’s short-term effects.

To study the effects of neighborhood changes, only respondents not changing census tracts between the two waves were investigated (n = 1,788). In addition, this helped rule out the prospect that some respondents experienced pronounced network turnover only because of residential relocation. In accordance with other scholars (e.g., Estabrooks et al., 2003), neighborhoods were approximated through census tracts, which are “small, relatively permanent statistical subdivisions of a county or equivalent entity . . . Census tracts generally have a population size between 1,200 and 8,000 people, with an optimum size of 4,000 people” (United States Census Bureau, 2010, p. A-12). Respondents were linked with their census tracts through protected geodata obtained from the National Opinion Research Center via special contractual arrangements. This study included 387 census tracts. On average, there were 4.59 respondents per census tract (minimum = 1, maximum = 24).

All Wave 1 census tract–level variables were obtained from the 2000 Decennial Census. The 2006–2010 American Community Survey (ACS) provided all Wave 2 census tract–level variables; 2006 to 2010 data were averaged as each individual year included too few respondents per census tract (each census tract was assessed at all years from 2006 to 2010). Because of data limitations, the 2000 Decennial Census is the most effective option for Wave 1, whereas the 2010 Decennial Census could not be used for Wave 2. Regarding the former, there are no alternative census tract–level ACS data between 2000 and 2005. Concerning the latter, the 2010 Decennial Census does not include all six items of this study’s concentrated disadvantage index (see below). The 2006 to 2010 ACS is the most effective data set for Wave 2 as there are no alternative census tract–level ACS data that do not extend beyond 2010 (an upper limit to avoid associating an outcome [close network dynamics] with a cause [concentrated disadvantage] from later in time).

Respondents retained over time had more close ties, were younger, were more likely married, had longer neighborhood residence, were more educated, were more likely employed, had more household assets, were healthier, and more frequently attended religious services than those who moved or could not be reinterviewed. There were no other significant differences in this study’s variables.

This study’s outcomes denote size of and turnover within close networks. In response to “Looking back over the last 12 months, who are the people with whom you most often discussed things that were important to you?” respondents listed up to five close ties. If left off this list, one’s partner could serve as a sixth tie. Respondents were then asked whether they had any other close contacts, providing a seventh possible tie. Close networks spanned from zero to seven members. Respondents reported every Wave 2 tie included in Wave 1 and vice versa, allowing counts of new and lost close ties (network turnover).

This study employed the index of census tract–level concentrated disadvantage developed by Sampson et al. (1997), based on percentage of the population below the poverty line, households on public assistance, female-headed households, individuals unemployed, individuals below 18 years of age, and non-White individuals. This index had high internal reliability (Cronbach’s αs: Wave 1 = .91; Wave 2 = .84). Exploratory factor analyses showed that in both waves, these six items formed one factor (Kaiser criterion) and further provided factor loadings for each item at each wave. This index was created by standardizing all items, multiplying each by its Wave 1 factor loading, and averaging these six scores, separately by wave. This study’s focus was change in concentrated disadvantage, computed by subtracting Wave 2 scores by Wave 1 scores. A confirmatory factor analysis revealed that the factor structure of these six items significantly differed between the two waves; a factor structure presents the correlational relationships between a group of variables used to develop a latent variable. Because results did not substantively differ when applying the first- or second-wave factor loadings to both waves, the Wave 1 factor loadings were employed.

All contextual- and individual-level control variables were from Wave 1. This study postulates that the Great Recession may have been the underlying cause of both personal and neighborhood misfortunes that affected close networks. The full causal effect, through both individual and contextual mechanisms, is of interest; no Wave 2 variables were controlled to avoid blocking potential pathways.

At the census tract level, population density (persons per square mile; logged to reduce right skew), location within a metropolitan statistical area (MSA; reference category = not within an MSA), and residential instability (proportions of residents living in different homes in the years 2000 and 1995) were controlled. Because the recession more heavily affected already disadvantaged neighborhoods, Wave 1 concentrated disadvantage was controlled.

Numerous respondent-level variables were controlled as neighborhood-level variables might simply be aggregates of individual-level characteristics. These included sociodemographics: gender (ref. = men), age (divided by 10 for more substantial incidence-rate ratios [IRRs]), race/ethnicity (White [ref.], African American, Latino, other), education (no high school diploma [ref.], high school diploma, university degree), and total household assets (logged to reduce right skew). Years of residence in one’s local area were controlled as they affect ties with neighbors (Kasarda & Janowitz, 1974). Employment (working for pay [ref.], retired, not working for reasons other than retirement) and marital circumstances (married [ref.], living with a partner, separated, divorced, widowed, never married) were controlled as they affect networks (Börsch-Supan & Schuth, 2013; Cornwell, 2009; Cornwell et al., 2008; Shin et al., 2019).

Health variables were controlled as they impact networks (Sluzki, 2010). Self-rated physical health was on a continuous scale (1 = poor to 5 = excellent). Beyond being powerful predictors of mortality (Idler & Benyamini, 1997), self-rated overall health variables have high reliability and validity and have been recommended for research by the World Health Organization, the United States Centers for Disease Control, and the European Commission (Salomon et al., 2009).

A continuous index of functional health problems was developed from average scores on seven activities of daily living (walking across a room, walking one block, bathing, dressing, getting in/out of bed, eating, using toilet; 0 = no difficulty to 3 = unable to do). This index held high internal reliability (Cronbach’s α > .80). This measure is an earlier validated scale and physical disability is a strong predictor of future health and mortality (Smith et al., 1986).

A continuous index of depressive symptoms was created through 10 items from the 11-item short form version of the Center for Epidemiologic Studies Depression Scale (CES-D; a measure of loneliness was excluded as it implicates interactions within close networks). These 10 items included the extents to which respondents felt depressed, had restless sleep, did not feel like eating, could not get going, and so on, over the last week (1 = rarely or none of the time to 4 = most of the time). Two items denoting less depression were reverse coded. This index was developed through averaging standardized scores on all 10 items (high internal reliability: Cronbach’s α > .70). Radloff (1977) emphasized the CES-D’s high reliability, validity, and internal consistency.

Numerous social connectedness variables likely to affect close network dynamics were controlled. For each close tie, respondents reported frequency of contact (1 = less than once a year to 8 = every day) and emotional closeness (1 = not very close to 4 = extremely close), both of which were averaged across all ties. The proportion of one’s close network that was relatives was controlled. Further controlled was attendance at religious services (0 = once or more per week [ref.], 1 = about once per month, 2 = less than once per year to several times per year, and 3 = never).

This study examined change in neighborhood-level concentrated disadvantage as a focal influence upon close network dynamics. The first model regressed Wave 2 close network size upon Wave 1 close network size (dummy variable categories), both Wave 1 and change in concentrated disadvantage, and all control variables. Controlling for the dependent variable at Wave 1 (a “conditional change panel model”) allows estimation of how predictors are associated with a second wave outcome for fixed initial values of this outcome (Finkel, 1995). This permits analyses of changes in an outcome and reduces concerns with selection effects. The second and third models repeated the first analysis with new close ties and lost close ties as the dependent variables, respectively. Because Wave 1 close network size limited the ties potentially lost, it was included as an exposure term (thereby controlling effects based on close network size). Because all outcomes were count measures with little indication of overdispersion, Poisson regressions were employed.

The inverse probability weighting technique (see Hawkley et al., 2014) minimized bias caused by selective attrition. Causes for attrition included death, institutionalization, unsuccessfulness locating an interviewee, and relocation to a new census tract. An array of Wave 1 health and demographic variables, and initial concentrated disadvantage, predicted inclusion at Wave 2. Inverse predicted probability scores obtained through this logistic regression were multiplied by the NSHAP’s standard weights before application in the analyses. Those least likely to have remained within the sample were thus weighted more heavily. Nonetheless, if those most vulnerable were less likely to have remained within the sample, attrition might have caused underestimation of how neighborhood-level declines affected close networks.

Standard errors were adjusted for census tract–level clustering. Missing data were addressed through multiple imputation using chained equations (10 imputed data sets). Only household assets and length of community residence had 10% or more missing data (13.81% and 16.99%, respectively). Over two thirds of the sample (69.87%) had no missing data. While the multiple imputation process included the dependent variables, the final analyses excluded their originally missing cases, removing 12 respondents. The analyses employed the Stata 15 statistical software package.

Results were substantively identical when initial concentrated disadvantage was not controlled, implying that they hold regardless of initial conditions. Findings were also substantively identical with inclusion of those who relocated to a new census tract, suggesting their broad generalizability to older populations.

This study suggests that investing resources in neighborhood institutions that maintain community social capital might help older persons and communities remain strong during economic shocks. Feld (1981) discussed the significance of such social foci, including senior centers, exercise facilities, lawn bowling clubs, hobby groups, and coffee shops, for the development of new strong social ties. This might be especially important for older residents of declining neighborhoods.

A minority (17.84%) of lost close ties were due to network members’ deaths. The neighborhood-level processes that affect mortality might differ from those here analyzed and are beyond this study’s scope. Future research should address this topic.

This study’s focus is the short-term effects of the Great Recession. As neighborhoods change gradually, a longer timespan might reveal the full effects upon close networks, which might include significantly more ties lost.

A further limitation is that this study considers only the presence of close ties. Future research should study how qualities of ties, including frequency of contact and emotional closeness, change through neighborhood-level declines.

Future research should study connections specifically among neighbors confronting the same neighborhood changes. This research might reveal whether declining communities erode neighborly ties or lead neighbors to band together.