Neighborhood Conditions and Social Network Turnover among Older Adults

Abstract

Increasing research highlights heterogeneity in patterns of social network change, with growing evidence that these patterns are shaped in part by social structure. The role of social and structural neighborhood conditions in the addition and loss of kin and non-kin network members, however, has not been fully considered. In this paper, we argue that the residential neighborhood context can either facilitate or prevent the turnover of core network relationships in later life – a period of the life course characterized by heightened reliance on network ties and vulnerability to neighborhood conditions. Using longitudinal data from the National Social Life, Health, and Aging Project linked with data from the American Community Survey, we find that higher levels of neighborhood concentrated disadvantage are associated with the loss of older adults’ kin and non-kin network members over time. Higher levels of perceived neighborhood social interaction, however, are associated with higher rates of adding non-kin network members and lower rates of adding kin network members over time. We suggest that neighborhood conditions, including older adults’ perceptions of neighborhood social life, represent an underexplored influence on kin and non-kin social network dynamics, which could have implications for access to social resources later in the life course.

Extensive evidence links point-in-time properties of personal network relationships to individual well-being, access to social resources, and life chances more broadly (Berkman et al. 2000; Wellman and Wortley 1990). However, close personal relationships vary considerably in their continuity (Fischer and Offer 2020; Mollenhorst, Völker, and Flap 2014; Shaw et al. 2007; Suitor, Wellman, and Morgan 1997; Wellman et al. 1997). Individuals can experience the addition and loss, or shedding, of personal network members over time; or these two processes can occur in parallel, known as network turnover. These dynamics, distinct from cross-sectional network characteristics, can be consequential for how network resources are accessed and how individual outcomes are ultimately achieved across the life course (Bookwala 2016; Schwartz and Litwin 2017, 2019). For one thing, access to the resources that inhere in network ties may depend on their consistency. Alternatively, opportunities to shed straining or stressful relationships can reduce burdens – and exposure to new, diverse social settings can allow individuals to cultivate new ties and reshape the network in ways that enhance access to social resources.

Early research on social networks viewed tie formation as a product of individuals’ strategic or entrepreneurial action (Burt et al. 1998; Kadushin 2002), or other individual-level characteristics such as personality (Kalish and Robins 2006) and cultural worldviews or preferences (Lizardo 2006). Aging-related changes in daily experiences, capacities, and support needs can also shape network maintenance and shifts in patterns of exchange (Charles and Carstensen 2010; Wellman and Wortley 1990). However, more recent work implies that opportunities to form and maintain ties are socially structured. For example, economic scarcity and social-environmental instability are associated with greater contingency and instability among family ties (Amorim 2021; Goldman and Cornwell 2018).

A key possibility is that neighborhood contexts, through their structural organization, spatial configuration, and social composition could provide more or less fertile ground for patterns of tie formation or turnover. Mario Small’s work has been central in drawing attention to the role of neighborhood and organizational contexts in providing opportunities for the formation of social ties (e.g., Small & Adler 2019; Small 2007; Small 2009). Yet, empirical studies of neighborhood and social network dynamics, such as network turnover, have existed largely separate from one another, even though neighborhoods and social networks are both intrinsic to studies of social disadvantage, disparities, and well-being (Desmond and An 2015). Indeed, a large literature on “neighborhood effects” demonstrates that neighborhood disadvantage adversely shapes social and economic well-being, including educational attainment, cognitive development, and health (Hill, Ross, and Angel 2005; Sharkey and Faber 2014; Wodtke, Harding, and Elwert 2011), in ways that accumulate and persist across the life course (Sharkey & Faber, 2014; Wodtke et al., 2011). And, there has been growing interest in understanding how neighborhood conditions shape social network properties that represent social integration and provide access to social resources (Small 2007; York Cornwell and Behler 2015).

In this paper, we draw from urban sociological literature on neighborhoods as social contexts (e.g., Sampson 2012), and from work that has considered the residential neighborhood as a context for point-in-time social interaction (Small, 2007; Tigges et al., 1998), to consider how neighborhood conditions contribute to the dynamics of close network relationships. We suggest that the structural and social conditions of the neighborhoods provide residents with sets of resources and opportunities (or a lack thereof) which have implications for social relationships. These conditions may support relationship continuity, increase the risk of network losses, or promote network turnover.

We explore this topic using data from Rounds 2 and 3 of the National Social Life, Health, and Aging Project (NSHAP), a population-based panel study of older Americans collected at 5-year intervals between 2005/6 and 2015/16 that is linked with data on respondents’ residential census tracts from the American Community Survey (ACS). Older adults’ well-being is especially vulnerable to both changes in social connections (e.g., Alwin et al. 2018) and the neighborhood environment (Yen, Michael, and Perdue 2009), making this a particularly important population for understanding the social-environmental processes that underlie network change. Following the analysis, we discuss the finding that residential contexts matter for personal network change in nuanced ways, particularly regarding the turnover of kin versus non-kin ties. We conclude that social network turnover represents a relatively overlooked consequence of neighborhood-level conditions, and suggest that network turnover may be a key mechanism that contributes to the robust associations between neighborhood conditions and later life outcomes.

LITERATURE REVIEW

The Importance of Personal Network Change

Personal social networks are immensely valuable social resources that shape a wide range of individual outcomes (Berkman et al. 2000; Ellwardt et al. 2015; Offer 2020; Ruppel et al. 2022; Alwin et al., 2018; Wellman and Wortley 1990). There is reason to believe that the degree to which individuals can consistently access these ties has implications for how consistently they can draw on these resources. Turnover in one’s personal network can lead to uncertainty or lack of reliability in one’s access to advice, information, and instrumental and emotional support. Social network change may be especially consequential for older adults who are more likely than younger individuals to face certain life-course transitions (e.g., retirement, widowhood) and age-related declines in health, which could result in a reshuffling in their network ties (e.g., Ha, 2008). As older adults rely on support from their network members to navigate the challenges that accompany these transitions, their well-being may be especially vulnerable to personal network changes.

In addition to providing social support, the typically tight-knit structure of personal networks allows members to monitor one another’s well-being and coordinate support (Hurlbert, Haines, and Beggs 2000; Umberson 1992). This coordination also helps networks to enforce certain normative behaviors and decision-making (Coleman 1990). Thus, high levels of social network turnover can also compromise individuals’ sense of belonging and ultimately contribute to poorer integration and greater social isolation.

It is important to acknowledge that network turnover is not ubiquitously beneficial or detrimental; rather, the addition and removal of network members can have distinct implications for individual outcomes depending on individual and social-contextual circumstances (e.g., Feld, Suitor, and Hoegh 2007; Offer 2021). Personal network change could be advantageous when newly added ties reflect access to a new set of social resources, particularly the addition of non-kin (e.g., friends, neighbors) with whom other network members may not be familiar. Indeed, a key insight from social network theory is the idea that certain network structures (e.g., bridging) provide individuals with access to novel, non-redundant information and advice (Burt, 1992; Burt, 2005; Granovetter, 1973). The formation of new network confidants could offer unique perspectives and advice.

Likewise, network members who are dropped or otherwise withdrawn from an individual’s personal network may not always reflect a detrimental or undesirable change. Recent research suggests that approximately 15% of network members are considered to be at times “demanding or difficult” (Offer and Fischer 2018). Thus, network turnover could reflect individuals’ ability to shed ties that are not reciprocating support, or who are otherwise problematic, as in the case of network ties who are abusive (Schafer and Koltai 2015). In this context, network losses can be advantageous for an individual’s health and level of stress (Offer 2020), and even allow them more time and resources to invest in more rewarding network ties.

Why Do Personal Networks Change?

Several theoretical orientations have guided empirical studies of social network change. Some accounts emphasize that individuals are deliberate in forming and maintaining ties that provide the most returns in exchange for support and other resources (e.g., Homans 1950; Burt 1992). Other research considers “tie decay” – that is, the tendency for social ties to weaken or disappear over time – as a function of tie characteristics, including the strength and age of the relationship, the context of its development, and embeddedness through shared contacts (Burt 2000). The life-course perspective has also been used to consider how close social ties change or persist in light of events and transitions that characterize life stages (Alwin, Felmlee, and Kreager 2018), including the transition to adulthood (Bidart and Lavenu 2005), workforce changes (McDonald & Mair, 2010; Settels and Schafer 2018), widowhood (Zettel and Rook 2004), and institutional transitions (Small et al., 2015).

More recently, researchers have considered how social network changes are structured by social position. African American older adults and individuals with less education are more likely to experience network loss than White older adults and those with more education (Cornwell 2015). In midlife, too, college-educated individuals are more likely to experience an increase in network advantages (i.e., frequency of contact, social support) (Fischer and Beresford 2015), while less-educated older adults exhibit declines in support from non-kin as they age (Shaw et al., 2007). Higher levels of income are also associated with the stability of resourceful ties over time (Schafer and Vargas 2016). Some of this work alludes to the relevance of contextual factors such as housing instability and access to institutional resources, though little research has explicitly examined the implications of neighborhood properties for personal network turnover (Settels 2020). This perspective shifts the focus from individual or familial circumstances, recognizing that the conditions that characterize where people live can structure the needs, alternatives, and opportunities for contact with network confidants (Kalmijn, 2012).

Neighborhood Conditions and Network Turnover

Our main argument is that neighborhood structural and social conditions influence social network turnover in meaningful ways, with distinct implications for kin and non-kin network ties. We build from classic research on urbanism, which argued that the urban context fundamentally reshaped the networks of residents by leading to the weakening or loss of close social bonds, particularly kin ties, and their replacement with secondary or non-kin ties (e.g., Wirth 1938; Fischer 1982). More recently, this research focus has narrowed from the urban context writ large to the consideration of structural and social dimensions of the residential neighborhood that have implications for access to social and economic resources that profoundly shape life chances (Aneshensel, 2009; Sampson et al., 2002), and may also be relevant for processes of network tie formation (Small 2009; York Cornwell and Behler 2015). Broadly speaking, structural dimensions refer to neighborhood socioeconomic conditions, including the socioeconomic well-being of its residents and the overall residential stability (Ross, Reynolds, and Geis 2000; Schieman 2005; Wheaton and Clarke 2003). Social dimensions refer to the social environment within a neighborhood, such as local social ties and interactions, as well as the sense of trust and cohesion that serve as the building blocks of neighborhood collective efficacy and social organization (Morenoff et al., 2001; Sampson et al., 1997, 1999). Below, we expand more specifically on how these dimensions of residential neighborhoods could be consequential for kin and non-kin network turnover.

Concentrated disadvantage.

Neighborhoods characterized by concentrated disadvantage are those in which a large share of residents lack socioeconomic resources. These neighborhoods are characterized by high rates of poverty and/or low median incomes, high rates of unemployment, and relatively low levels of education. The concentration of individuals who lack these socioeconomic resources can contribute to a neighborhood environment that becomes characterized by declining infrastructure and the disappearance of resources and opportunities (Sampson, 2012; Wilson, 1987). Living in a neighborhood characterized by concentrated disadvantage can compromise one’s capacity to maintain stable social relationships by way of adversely influencing individual well-being (Settels 2020). Noise, inaccessible or poorly maintained public spaces, lack of transportation, and walkability, for example, are aspects of the physical environment that are associated with concentrated disadvantage and that can interfere with in-person visits with social ties, and also lead to stress (Diez Roux and Mair 2010). Coping with daily life in a resource-poor environment may also sap individuals’ time and energy, thereby reducing their abilities to form and sustain social ties through social interactions or exchanges of support (York Cornwell and Behler 2015).

Neighborhoods also house social institutions and organizations (e.g., religious organizations, community centers, nonprofit agencies) that provide residents with access to resources and information (Small, 2006; Wilson, 1987), as well as opportunities to form and maintain network ties (Small 2006). Neighborhoods characterized by concentrated disadvantage are less likely to contain these institutions and organizations and therefore may offer residents a relatively impoverished, less consistent resource network and fewer shared contexts that could otherwise enable tie formation (Sampson, 2012; Small & Adler, 2019). Indeed, establishments such as senior centers, barbershops, social service organizations, churches, and schools are important sites for developing and maintaining stable network ties with both residents and non-residents who visit these locations (Sampson, 2012; van Eijk, 2010), especially older adults (Torres 2018). More disadvantaged neighborhoods may provide only intermittent access to organizations that could connect residents with more resource-rich institutions and individuals outside of their neighborhoods (Small 2007). In this way, neighborhoods characterized by greater concentrated disadvantage may contribute to residents’ higher levels of social network turnover by offering fewer and less reliable locations for forming and sustaining network ties, particularly with non-kin. Relationships with non-kin may depend more on common meeting areas and shared spaces, whereas the stability of kin ties may depend less on neighborhood characteristics, and more on shared contacts and obligations within the family system (e.g., Martin & Yeung, 2006; Mollenhorst et al., 2011).

Finally, neighborhoods with high levels of concentrated disadvantage imply the concentration of individuals experiencing certain aspects of flux or turbulence in their own lives. These experiences could include high levels of job turnover, periods of unemployment and non-standard work hours, family complexity, as well as housing insecurity, and frequent residential changes (Wilson 1987 p. 61-63), which could directly implicate the stability of individuals’ personal network relationships, especially those with kin (Goldman and Cornwell 2018). As Wilson (1987, p. 60) notes, disadvantaged neighborhoods can alienate their residents by compromising sustained contact with social ties in more advantaged, “mainstream,” and stable places (Tigges et al. 1998; Wilson 1987). This may lead to the consolidation of ties within the local neighborhood while creating instability in ties with family and friends who live outside of the neighborhood.

Residential instability.

Residential (in)stability refers to the (in)consistency in the composition of households in a given neighborhood or census tract (Sampson et al., 1997), otherwise defined as “the flux of residents into and out of neighborhoods over time” (Ross, Reynolds, and Geis 2000, p. 581). Residential instability is thought to be generally disruptive to local social connectedness, leading to the loss of existing social network ties with other residents and preventing the development of new network ties within the neighborhood (Coleman, 1990; Sampson et al., 1999). Frequent shifts in residents can directly undermine individuals’ ability to form stable ties with their neighbors, simply by having an unstable neighborhood population from which to form social bonds.

At the neighborhood level, high levels of residential instability can weaken social capital and social organization, as the strength of community norms and sanctions that support social control are weakened by a more transient, less consistent population (Coleman, 1990; Shaw & McKay, 1942). Residential instability is associated with lower levels of reciprocated exchange and less intergenerational closure than are more residentially stable neighborhoods (Sampson et al., 1999). In this sense, residential instability may also indirectly lead to more unstable network ties, particularly with non-kin neighbors with whom there is no prior, long-standing relationship, due in part to lower levels of neighborhood social capital and social organization that could make residents more inclined to develop ties with one another.

Whereas prior literature supports our expectation that residential instability contributes to higher levels of network turnover, particularly among non-kin, an alternative possibility is that individuals respond to residential instability by maintaining an especially close network of trusted others. Akin to the concept of “adaptive cohesion” (Schieman 2005), individuals living in more residentially unstable neighborhoods may develop more stable personal network ties, particularly with kin, as a means of more efficiently and reliably accessing social resources amidst a neighborhood population that is in flux (Schieman 2005). In other words, instability at the neighborhood level may prompt lower levels of personal network turnover as individuals seek to maximize their consistent access to social support and minimize the degree to which neighborhood conditions undermine this access (Schieman 2005).

Neighborhood social ties and collective efficacy.

Interpersonal connections among residents are a fundamental social dimension of neighborhoods that shape communities’ abilities to act on collective goals, maintain informal social control, and exchange social capital that can benefit residents’ well-being (Coleman, 1988, 1990; Sampson et al., 1999). Collective efficacy is a neighborhood-level property; individuals may benefit from higher levels of collective efficacy even if they themselves do not frequently interact with other residents, because greater cohesion and informal control within the neighborhood helps to lower levels of delinquency and crime (Sampson & Groves, 1989), and contributes to the improvement of neighborhood infrastructure and resources including the built environment, commercial activity, and local organizations and institutions (Sampson, 1991). High levels of neighborhood collective efficacy also reduce individuals’ perceptions of neighborhood fear and mistrust (Ross and Jang 2000), which could promote visitation and other interactions with kin and non-kin ties.

Physical propinquity is a fundamental basis of tie formation (Blau, 1977; Small & Adler, 2019), and the connection between neighborly interaction and tie formation is well-established (Small & Adler, 2019). If social interaction and exchanges of support and resources among neighbors are normative within a local area, residents may find it easier to form new ties, particularly with local non-kin. Conditions of concentrated disadvantage and lack of local institutional resources can lead residents to cultivate local ties with neighbors for resource pooling and exchanges of support (Stack 1974; Pattillo 1998; Schieman 2005). From a practical perspective, ongoing relationships and exchanges among neighbors can make it easier for individuals to form new ties or replace lost ties with local non-kin.

However, another important consideration is that the perception of social connectedness or collective efficacy in one’s neighborhood may also have implications for network dynamics. Indeed, perceived or subjective neighborhood social characteristics can be strongly associated with individual outcomes including well-being, independent of objective neighborhood measures (e.g., Weden, Carpiano, and Robert 2008). Individuals who perceive a higher level of social connectedness among neighbors may perceive that they have a greater number of alternative sources of social integration and social support (Fischer 1982). Older adults who live in an environment that they perceive to be particularly cohesive or interconnected – regardless of its actual level of social interaction - could be more likely to become involved in social organizations, participate in local neighborhood events, and more generally be more inclined to form new network ties with individuals who they meet in their local area, potentially even substituting new network ties for older network relationships that are no longer as beneficial or pertinent. This may be especially the case for older adults who are aging in place, and those who have few kin ties to draw on (Mair 2019; Yen et al. 2009), as well as socially disadvantaged individuals seeking non-kin sources of social support (Desmond 2012). Thus, individuals’ assessments of their local social context can prompt or hinder social participation, which may contribute to changes in non-kin network ties.

THE PRESENT STUDY

This study aims to examine how residential neighborhood characteristics shape personal network turnover. We pay particular attention to how subjective neighborhood social conditions and objective structural factors carry different implications for patterns of turnover among kin and non-kin network ties, bringing together prior literature on neighborhood effects and social-structural factors that influence personal network change later in the life course. This paper is guided by the following overarching research question: How are neighborhood conditions associated with the turnover of kin and non-kin personal network ties among the older adult population? This study departs from prior work in this area, which has focused on patterns of network turnover more generally and has not considered the role of perceived neighborhood social ties (e.g., Settels 2020).

Indeed, non-kin tie turnover may depend on the availability of neighborhood resources that provide opportunities to develop and maintain ties with neighbors and friends (e.g., Mollenhorst et al. 2014; Torres 2018), as well as perceptions of how social neighborhood residents are more generally. Kin ties are more likely to be maintained out of normative obligations (Bloem, van Tilburg, and Thomése 2008; Fischer and Offer 2020; Thomése et al. 2005), but may also be especially vulnerable to socioeconomic strains that are relevant to the neighborhood context (e.g., housing insecurity, unemployment) and that could also limit availability for support exchange (Amorim 2021; Goldman and Cornwell 2018). This work motivates the separate considerations of kin and non-kin network changes as a function of neighborhood conditions, given the possibility of different social processes underlying any associations. Importantly, we emphasize that our theoretical framework is grounded in the notion that residential neighborhood characteristics can have implications for personal network turnover in general, regardless of whether personal network ties share the same residential neighborhood as the respondent. Put differently, we consider that residential neighborhood concentrated disadvantage, residential (in)stability, and perceived neighborhood social ties are consequential for the kin and non-kin network turnover regardless of whether kin and non-kin ties live, as these factors can compromise or facilitate the maintenance and development of older adults’ social connections broadly considered. Future research may investigate how neighborhood characteristics have distinct implications for local versus non-local network ties (York Cornwell and Goldman 2020).

DATA AND METHODS

This analysis relies primarily on data from Rounds 2 and 3 of the NSHAP, a population-based study of community-residing older adults in the United States (Suzman 2009). The overall goal of the NSHAP is to better understand how health and social context intersect to influence older adults’ well-being as they age. The original cohort (Round 1) includes 3,005 non-institutionalized older adults ages 57-85 at baseline (2005-2006), with a weighted response rate of 75.5%. Round 2 (2010-2011) includes 3,377 returning respondents and their co-resident partners, if applicable, yielding a conditional response rate of 89%. Round 3 (2015-2016) includes returning respondents and their partners, if applicable (N = 2,409), as well as a new cohort of respondents born between 1948 and 1965 and their co-resident partners (N = 2,368). The conditional response rate for returning respondents at Round 3 is 89.2%. At each round, data collection consisted of in-home interviews conducted by the National Opinion Research Center (NORC), which included the collection of personal social network information (described below). Following the in-home interview, respondents were also asked to complete a leave-behind questionnaire (LBQ) to be returned to NORC by mail.

Measures of Personal Network Turnover

At each round of the NSHAP, the in-home interviews began by asking respondents to provide information about their personal social networks. Respondents were prompted with an “important matters” name generator:

From time to time, most people discuss things that are important to them with others. For example, these may include good or bad things that happen to you, problems you are having, or important concerns you may have. Looking back over the last 12 months, who are the people with whom you most often discussed things that were important to you?

This instrument is often used in survey research to gather information about respondents’ closest, most supportive social ties (Marsden 1987; Paik and Sanchagrin 2013). Respondents could name up to five individuals (i.e., network alters) who comprised Roster A of their personal network.ⁱ Following the enumeration of network members, respondents were asked to categorize their relationship with each alter (e.g., spouse, partner, child, friend), report how often they spoke with each alter, and how often each alter spoke with every other alter.

At Rounds 2 and 3, after the administration of the “important matters” name generator, respondents were presented with a list of the network members who they had named at prior rounds and were asked to confirm the computer-identified matches between network members who were listed at multiple rounds. Figure 1 illustrates this process, whereby respondents identify matches in alters named in their Round 3 roster with those who were named as part of their Round 2 roster. From these across-round comparisons, we code alters named at Round 2 but not at Round 3 as lost. Alters named at Round 3 but not at Round 2 are coded as additions, and alters named at Rounds 2 and 3 are coded as stable (Cornwell et al. 2014). By combining the categorization of alters as lost, added, or stable with the information about how each network member is related to the respondent, we further classify each network change in terms of whether it pertains to a kin or non-kin network member (e.g., kin network loss, kin network addition).

Figure 1.

Sample screenshot of the network roster matching exercise completed by a hypothetical NSHAP respondent.

Scholars have used a range of approaches to measure changes in personal social networks over time. In some cases, network changes are modeled as differences in a feature of social network structure between two time points (e.g., differences in network density, proportion kin, overall number of losses and/or additions) (see Wellman et al. 1997). Other research focuses on social network consistency, as opposed to patterns of change (Faris and Felmlee 2019). Given that our main concern is with how neighborhood conditions can prompt the formation or loss of network members, we focus on personal network turnover. Additionally, because different processes can prompt additions and losses, and because these two types of changes are also linked with distinct individual outcomes, our analysis is designed to assess how neighborhood conditions predict the addition and loss of network members separately (Cornwell & Laumann, 2015; Feld et al., 2007).

Whereas overall (or net) differences in network size and proportion kin can indicate growth or decline in overall social integration, this type of summary measure can mask the actual degree of turnover in the network. A respondent could, for example, experience the loss of 5 network members and the addition of 5 new network members (i.e., complete network turnover), but would have an overall net change of 0 in their network size. Therefore, analyses that track the entrance or departure of specific network alters to and from older adults’ networks are conceptually different measures of overall network change.

For the main analyses, we focus on modeling six distinct types of network change: (1) the number of total network additions between Rounds 2 and 3, (2) the number of total network losses between Rounds 2 and 3, (3) the number of kin network additions between Rounds 2 and 3, (4) the number of kin network losses between Rounds 2 and 3, (5) the number of non-kin network additions between Rounds 2 and 3, and (6) the number of non-kin network losses between Rounds 2 and 3. The separate examinations of kin and non-kin changes allow us to consider whether these types of ties may be constrained by different norms, opportunities, and alternatives for contact (Kalmijn 2012).

Neighborhood Conditions

A key advantage of using the NSHAP to address this research question is that respondents’ subjective assessments of their neighborhood social characteristics can be linked with census tract measures from the U.S. Census and the American Community Survey (ACS), making it possible to obtain objective measures of neighborhood disadvantage. Subjective and objective measures of the social environment can differ in important ways (Bailey et al. 2014). The ability to consider both dimensions allows for a richer, more nuanced portrait of the social environment and its intersection with personal network turnover.

We create two measures that serve as our primary indicators of neighborhood structure. The first is a measure of neighborhood concentrated disadvantage. We create a scale using five measures from the 2000 U.S. Census and the 2005-2009 ACS, including the percentage of the tract population with income levels below poverty, the percentage of residents ages 25 and older who have less than a high school education, the percentage of residents over age 18 who are unemployed, the percentage of female-headed households, and the median household income in respondents’ census tracts at Round 2, consistent with prior research (Sampson et al. 1997; Wodtke et al. 2011). The scale demonstrates good reliability (alpha = .87). Higher scores on the scale reflect higher levels of concentrated disadvantage in respondents’ residential tracts.

The second measure of neighborhood structure is an estimate of residential instability. Following prior research (Sampson et al., 1999), we create a scale using the proportion of renter-occupied housing units and the proportion of residents who have moved to a different residence in the past year in respondents’ Round 2 census tracts. This scale demonstrates good reliability (alpha = .76), with higher scores representing greater residential instability.

The third neighborhood measure is a subjective indicator of neighborhood social ties. This measure is based on respondents’ Round 2 reports of how often they and others in their neighborhood: 1) visit with each other, 2) do favors for each other, and 3) ask each other for advice, where 0 = “never,” 1= “rarely,” 2 = “sometimes,” and 3 = “often” (York Cornwell and Cagney 2014). Each of these three items measures specific ways that neighbors interact with one another, arguably a more concrete measure of social exchange at the neighborhood level than more abstract constructs such as social cohesion (e.g., trust, shared values). We create a scale that averages respondents’ ratings of these three items, with higher scores representing higher levels of perceived neighborhood social ties. The scale demonstrates good reliability (alpha = .76).

Covariates

Several sociodemographic, life-course, and other contextual factors are likely to be associated with both changes in personal social networks and the relationship between neighborhood characteristics and social network changes. While concentrated disadvantage, residential instability, and social ties are the primary neighborhood characteristics that are of interest, other measures of neighborhood structure are also relevant to social network turnover. We account for whether respondents reside in an urban, suburban, or rural area (census tract), categorized according to the U.S. Census definition of a Metropolitan Statistical Area (MSA).ⁱⁱ Relative to rural and suburban areas, urban tracts exhibit greater population and institutional density. Higher density affords residents more opportunities to replace lost ties or form new social network relationships, particularly with non-kin (Fischer 1982). A larger pool of alternatives could also make it more difficult to sustain the same network ties over time.

Residential tenure may contribute to social network turnover, as changing residences can lead to a restructuring of one’s personal confidants (Bloem et al. 2008). We therefore account for the total time that respondents report that they have lived in their neighborhood at Round 2 (1=less than 1 year; 2=1-5 years; 3= 6-10 years; 4=11-15 years; 5=16-20 years; 6=21-25 years; 7=26-50 years; 8= more than 50 years). Finally, to account for respondents’ own residential mobility and potential exposure to different residential and social contexts, we include a control for whether respondents moved to a different tract during the five years.

A key focus of this analysis is whether any associations between neighborhood conditions and network change are explained by individual measures of life-course transitions, sociodemographic characteristics, and attainment that prior research shows to shape social network turnover. Individual-level covariates include age at Round 2 (divided by 10), gender, whether the respondent is Black, whether the respondent is Hispanic, and household income in the prior year (divided by 10,000 and then log-transformed). Life-course covariates include the respondent’s highest level of education attained (less than high school, high school or equivalent, some college, Bachelor’s or more), self-rated physical health (1 = poor; 5 = excellent), whether the respondent was married or living with a partner at Round 2, whether the respondent was widowed at Round 2, whether the respondent was retired at Round 2, and the number of children they had.ⁱⁱⁱ We also control for whether respondents became widowed between Rounds 2 and 3 and whether respondents retired between Rounds 2 and 3, as these are key life-course transitions that may contribute to shifts in social network structure (e.g., Settels and Schafer 2018; Zettel and Rook 2004). Given respondents’ close physical proximity to their household members, a larger number of household members may be associated with greater network continuity and more kin-based personal networks. This factor may also be associated with neighborhood conditions. We therefore control for the number of household members.

Analytic Strategy

We begin by comparing the distribution of network additions and losses between Rounds 2 and 3 across levels of concentrated disadvantage. We then proceed to a series of Poisson models that predict the number of kin and (separately) non-kin personal network additions and losses between Rounds 2 and 3 as a function of neighborhood and individual characteristics measured at Round 2. Poisson is an appropriate modeling strategy given that the set of outcome variables are counts of network members (added or lost). Likelihood ratio tests indicate no evidence of overdispersion.

The first set of multivariable models the number of kin network members added and lost between Rounds 2 and 3 as a function of neighborhood characteristics, followed by models that predict the number of non-kin network members added and lost during this same timeframe. For all models predicting social network additions, we use network size at Round 3 as the exposure variable. For models predicting overall social network losses, we use the number of kin and non-kin in the network, respectively, at Round 2 as the exposure variable, given that the number of overall kin and non-kin losses that one could experience between Rounds 2 and 3 depends on the number of these network members named at Round 2. All models predicting kin and non-kin changes include the number of kin and non-kin at Round 2, respectively, as a covariate.

A respondent’s residential mobility during the study period can also have significant implications for network turnover, potentially challenging the maintenance of formerly proximal ties, or offering new potential network ties in the destination local environment. At the same time, respondent residential mobility is likely a function of many of the same socioeconomic factors that shape both neighborhood concentrated disadvantage and personal network turnover. For example, proximity to and support exchange with nearby family and friends may explain lower rates of neighborhood residential mobility, particularly among lower-income households (Dawkins 2006; Mulder and Cooke 2009). Indeed, neighborhood attainment trajectories are patterned by race/ethnicity and socioeconomic status, with residential mobility tending to reproduce broader patterns of inequality (Sharkey 2012). Older adults who move out of a neighborhood may also be doing so to be closer to family and friends, which has direct implications for personal network turnover. We therefore include a second set of models that are stratified by whether the respondent moved between Rounds 2 and 3 to test the robustness of our findings when limited to respondents who do and do not remain in the same tract during the observation period.

Missing data.

Missing data is generally not problematic in the NSHAP. However, approximately 28% of Round 2 respondents do not provide information on household income. As income can be an important indicator of individual socioeconomic position, we use multiple imputation with chained equations (20 iterations using “mi impute” in Stata 14) to preserve cases in the analysis that have missing data on income and other covariates used in the models. We include the dependent variables in the imputation equation, but then exclude cases with originally missing values of a particular dependent variable from the respective analyses (von Hippel 2007).

The 2,354 respondents aged 50 and older who provided network data at Rounds 2 and 3 serve as the basis of the analytic sample. Models predicting network additions use network size at Round 3 as the exposure variable. Models predicting the loss of kin exclude 189 respondents who did not include any kin in their Round 2 network. Models predicting the loss of non-kin exclude 865 respondents who did not name any non-kin as part of their Round 2 networks. Whereas these exclusions are necessitated by the fact that we cannot model the loss of network alters who were not present at baseline, we recognize that respondents who have more or less kin-based personal networks may differ from one another on characteristics such as health/well-being and residential neighborhood. Tests of mean differences indicate that those who are excluded from analyses based on having no non-kin at Round 2 are more likely to live in neighborhoods with higher levels of concentrated disadvantage and lower levels of perceived neighborhood social ties at Round 2, while those excluded on the basis on no kin network members at Round 2 are more likely to live in neighborhoods with higher levels of social ties (p < .001 for all comparisons). Any significant effect of these neighborhood factors on the loss of network ties between Rounds 2 and 3 may therefore be conservative estimates, given that neighborhood factors could have influenced personal network composition before the start of the observation period.

Selection.

The main analytic sample is limited to respondents who provide network data at Rounds 2 and 3. This restriction calls for attention to selection bias. Sample attrition is more likely among those who are in worse health at Round 2 and who are socially disadvantaged, both of which are pertinent to studies of social network turnover and neighborhood disadvantage. To help address these differences, we derive the inverse Mills ratio (i.e., non-selection hazard) using a probit model to predict whether each Round 2 respondent was interviewed again at Round 3 (N = 2,371). Inclusion in Round 3 is modeled as a function of individual sociodemographic characteristics, life-course, and health measures from Round 2, as these factors are likely to predict attrition between survey rounds. The inverse Mills ratio is then derived as the probability density function of the linear prediction divided by the cumulative distribution function of the linear prediction. We include the inverse Mills ratio as a covariate in all multivariable models, helping to account for selective attrition (Heckman 1979; Mills 1926).^iv All analyses are weighted using Round 2 respondent-level weights provided by the NSHAP, and standard errors are adjusted to account for the stratified and clustered nature of the sample (O’Muircheartaigh et al. 2014).

RESULTS

Descriptive Analyses

Descriptive statistics in Table 1 indicate that most respondents experience changes in their networks between Rounds 2 and 3. Over 80% of respondents report the addition and, separately, the loss of at least one network member. The addition and loss of kin network members are less common but are still experienced to some extent by over half of respondents (57% and 67%, respectively). Likewise, 54% and 81% of respondents report the addition and loss of non-kin, respectively.

Table 1.

Proportion of Respondents Reporting Each Number of Social Network Change Outcomes Used in the Main Analyses.^a

# Network members	Addedb W2 ➔ W3	Lostc W2 ➔ W3	Kin Addedb W2 ➔ W3	Kin Lostd W2 ➔ W3	Non-Kin Addedb W2 ➔ W3	Non-Kin Loste W2 ➔ W3
0	.16	.15	.43	.33	.46	.19
1	.24	.26	.30	.36	.27	.41
2	.26	.26	.17	.18	.17	.24
3	.19	.19	.07	.08	.07	.10
4	.11	.11	.02	.03	.03	.04
5	.03	.03	.004	.01	.01	.01

^aProportions are unweighted and are calculated using one of the 20 imputed datasets.

^bCalculated among respondents in the analytic sample who named at least one network member at Round 3.

^cCalculated among respondents in the analytic sample who named at least one network member at Round 2.

^dCalculated among respondents in the analytic sample who named at least one kin network member at Round 2.

^eCalculated among respondents in the analytic sample who named at least one non-kin network member at Round 2.

Table 2 presents descriptive statistics for the main variables included in our analyses. Within our analytic sample, respondents reside in tracts that include, on average, 14% of households in poverty, a median household income of $55,895.77, 5% unemployed, and just under 45% with less than a high school degree. Just over 28% of housing units are renter-occupied, on average, with just under 15% movers. Respondents report moderate levels of perceived neighborhood social ties, with average scaled reports falling between “rarely” and “sometimes” when all three composite items are averaged. Over half (57%) of our analytic sample is female, while 14% of respondents are Black and 11% are Hispanic. The majority of respondents (75%) were married or living with a partner and were retired (70%) at the time of the Round 2 survey, with 8% experiencing widowhood between Rounds 2 and 3, and 15% becoming retired during this same period. Approximately 59% of the sample has more than a high school degree. On average, respondents are 70 years old at Round 2 and have lived in their neighborhood for 16-20 years. Just under 20% of respondents moved tracts between rounds.

Table 2.

Descriptive Statistics of Key Covariates in the Main Analysis (N = 2,354).^a

Variable	Proportion or Weighted Mean	Standard Deviation
Residential Neighborhood Context
Tract-level concentrated disadvantage scale (α = .87)b	−.07	.74
(Average of standardized items; range: −1.99 – 2.43)
Percentage in poverty	13.62	11.10
Percentage of female-headed households	12.65	7.86
Median household income	56,895.77	26,305.22
Percentage unemployed	5.07	2.93
Percentage with less than a high school degree	44.29	16.56
Tract-level residential instability scale (α = .76)b	−.01	.89
(Average of standardized items; range: −1.57 – 4.50)
Percentage of renter-occupied housing units	28.12	18.22
Percentage of movers	14.28	8.12
Perceived neighborhood social ties scale W2 (α = .76)b	1.55	.70
(Average of items; range: 0 = never; 3 = often)
How often do you and people in this area visit in each other’s homes or when you meet on the street?	1.77	.90
How often do you and people in this area do favors for each other?	1.93	.79
How often do you and other people in this area ask each other for advice about personal things?	.96	.88
Urbanicity of residential tract
Urban	.36
Suburban	.45
Rural	.20
Respondent Covariates
Age (in decades)	6.99	.71
Female (=1)	.57
Educational attainment	.57
Less than high school	.17
High school or equivalent	.24
Some college	.32
Bachelor’s or more	.27
Black (=1)	.14
Hispanic (=1)	.11
Income in the prior year (divided by 10,000 and logged)	1.46	.92
Married or living with a partner (=1)	.75
Widowed (=1)	.15
Became widowed R2➔R3 (=1)	.08
Retired (=1)	.70
Became retired R2 ➔ R3 (=1)	.15
Self-rated physical health (1 = poor; 5 = excellent)	3.41	1.01
Network size R2	3.88	1.30
Respondent moved to a different tract between R2 and R3 (=1)	.19
Residential tenure (years)	5.21	2.12
Number of household members	1.08	.99
Number of children	2.72	1.75

^aAll covariates are measured using the Round 2 survey of the NSHAP unless otherwise noted. Means and proportions are calculated using one of the 20 imputed datasets. Means are weighted using Round 2 respondent-level weights provided by the NSHAP that adjust for age and urbanicity.

^bMeans and standard deviations for scale items are based on non-imputed data at Round 2 from respondents interviewed at all three rounds and are weighted using Round 2 respondent-level weights provided by the NSHAP that adjust for age and urbanicity.

Bivariate analyses reveal a strong positive correlation between neighborhood-level residential instability and concentrated disadvantage (r = .53; p < .001). There are considerably weaker and non-significant negative correlations between perceived neighborhood social ties and concentrated disadvantage (r = −.025; p = .51) and between perceived neighborhood social ties and residential instability (r = −.007; p = .91)

Figure 2 illustrates the number of respondents experiencing each of the different possible combinations of overall network additions and losses between Rounds 2 and 3, among those residing in the lowest and highest quarters of the distribution of the concentrated disadvantage scale. Whereas the main analyses focus on kin and non-kin turnover as a function of neighborhood conditions, this bivariate exercise provides useful preliminary insight into whether network change in general may be patterned by aspects of the social environment. Compared to older adults living in neighborhoods characterized by higher levels of concentrated disadvantage, older adults in neighborhoods of lower concentrated disadvantage more frequently experience no network change between Rounds 2 and 3 (yellow bar). In addition, respondents in greater concentrated disadvantage neighborhoods more frequently experience only network losses (no additions; dark red bars) and net losses in network members between rounds (orange bars).

Figure 2.

Distribution of network members (overall) added and lost between Rounds 2 and 3, by bottom and top quarter of neighborhood concentrated disadvantage at Round 2.

Indeed, 37% of older adults living in the top quarter of neighborhood concentrated disadvantage experience a higher number of network losses than network gains, compared to 24.4% of those in the bottom quarter of concentrated disadvantage. Similar patterns emerge when looking at the addition and loss of kin network members. Between Rounds 2 and 3, 48.6% of those living in the top quarter of concentrated disadvantage at Round 2 experienced more losses of kin ties than they did gains, compared to 33.4% of those living in more advantaged neighborhoods at Round 2. Differences are smaller for non-kin, as 48.1% of those in the top quarter of concentrated disadvantage at Round 2 experience more non-kin losses than gains between Rounds 2 and 3, which is slightly more than the 45.0% of older adults living in the bottom quarter of neighborhood disadvantage at Round 2.

Multivariable Analyses

We begin by reviewing the Poisson models presented in Table 3 that predict the number of network kin and non-kin additions and losses between Rounds 2 and 3, adjusting for all covariates. Model 1 of Table 3 examines the role of neighborhood conditions in shaping the addition of network kin between rounds. Neither neighborhood concentrated disadvantage nor residential instability is significantly associated with higher rates of adding kin network members between Rounds 2 and 3. At the same time, higher levels of perceived neighborhood social ties are associated with a 13% lower rate of adding kin network members (IRR = .87, p < .001).

Table 3.

Incidence Rate Ratios from Poisson Models Predicting the Number of Kin and Non-Kin Additions and Losses Between Rounds 2 and 3.

	# Kin Network Members		# Non-Kin Network Members
	(1) Additions	(2) Losses	(3) Additions	(4) Losses
Concentrated disadvantage	1.03	1.11*	1.07	1.07*
	(.04)	(.06)	(.05)	(.03)
Residential instability	.97	1.00	.99	.98
	(.03)	(.04)	(.03)	(.03)
Perceived neighborhood social ties	.87***	1.03	1.22***	1.03
	(.03)	(.03)	(.05)	(.03)
Urbanicity (Ref = Urban)
Suburban	1.02	1.00	.92	.99
	(.07)	(.06)	(.06)	(.04)
Rural	1.06	1.07	.84*	.98
	(.07)	(.08)	(.07)	(.06)
Age	.98	1.16**	.95	1.00
	(.06)	(.06)	(.07)	(.05)
Female	1.05	.87**	.86*	.99
	(.06)	(.04)	(.05)	(.03)
Educational attainment (ref = less than high school)
High school or equivalent	.98	.88†	.84	.93
	(.09)	(.06)	(.10)	(.06)
Some college	.85	.88*	1.00	.92
	(.08)	(.05)	(.10)	(.06)
Bachelor’s or more	.78*	.77**	.99	.93
	(.08)	(.06)	(.11)	(.06)
Black	1.26*	1.14†	.79*	.97
	(.13)	(.08)	(.09)	(.05)
Hispanic	1.15	.97	.74**	.94
	(.08)	(.08)	(.08)	(.06)
Income	1.05	.95	.92*	1.00
	(.04)	(.03)	(.03)	(.03)
Retired	.97	1.02	.98	.91†
	(.08)	(.07)	(.09)	(.05)
Widowed	1.09	.95	.91	.91†
	(.11)	(.08)	(.09)	(.05)
Respondent moved to a different tract R2➔R3	.96	1.06	1.20**	1.07*
	(.06)	(.08)	(.07)	(.03)
Residential tenure	1.01	1.00	1.00	1.00
	(.01)	(.01)	(.02)	(.01)
Number of household members	1.03	1.02	.97	.98
	(.02)	(.02)	(.03)	(.02)
Number of children	1.10***	.99	.92**	1.02*
	(.02)	(.01)	(.02)	(.01)
F(df)	29.17*** (25, 47.8)	9.34*** (25, 47.9)	14.65*** (25, 47.9)	3.43***(25, 47.9)
N	2,325	2,165	2,325	1,489

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001 (two-sided tests). 95% confidence intervals in parentheses.

^aAll models control for the inverse Mills ratio, self-rated health, marital/partner status at Round 2, retirement between Rounds 2 and 3, widowhood between Rounds 2 and 3, and number of kin (models 1 and 2) and non-kin (models 3 and 4) network members at Round 2.

Black older adults are also significantly more likely to add kin network members (IRR = 1.26, p < .05), whereas older adults with a Bachelor’s degree or more are less likely than older adults with less than a high school degree to add kin (IRR = .78, p < 05). In the full model, number of children is associated with a 10% higher rate of kin network addition between Rounds 2 and 3 (IRR = 1.10, p < .001). Given that children comprise nearly 30% of alters in the full Round 2 sample, older adults with more children may have more alternatives to add to their network.

Moving to Model 2, we find that there is no statistically significant association between perceived neighborhood social ties or residential instability and the loss of kin between Rounds 2 and 3, while higher levels of neighborhood concentrated disadvantage are associated with higher rates of kin network loss (IRR = 1.11, p <.05). Loss of kin is more likely at older ages (IRR = 1.16, p < .01), but significantly less likely among older adults with some college education (IRR = .88, p <.05) and those with a Bachelor’s degree or more (IRR = .77, p <.01).

Turning to non-kin network turnover, Model 3 indicates that older adults who live in neighborhoods with higher levels of social ties add non-kin at a 22% higher rate (IRR = 1.22, p <.001) compared to those in neighborhoods with lower social ties. Neighborhood concentrated disadvantage and residential instability are not significantly associated with the addition of non-kin network ties. Relative to those who live in urban areas, older adults living in more rural or smaller towns added non-kin at a 16% lower rate. Black and Hispanic older adults added non-kin at significantly lower rates (21% and 26%, respectively) compared to other racial groups and non-Hispanic older adults, as did women and those with higher incomes (IRR = .86 and .92, p <.05, respectively). Notably, older adults who moved to a different residential tract between survey rounds were significantly more likely to add non-kin compared to those who do not move (IRR = 1.20, p <.01).

Turning to non-kin losses (Model 4), we observe a significant positive association between neighborhood concentrated disadvantage and the loss of non-kin network members (IRR = 1.07, p < .05). Other neighborhood characteristics are not significantly associated with non-kin network loss, however moving to a different residential tract is also associated with a 7% higher rate of non-kin loss (IRR = 1.07, p < .05) compared to older adults who did not change tracts between Rounds 2 and 3. Whereas number of children was associated with fewer non-kin additions (IRR = .92, p <.01), having more children was associated with a higher rate of non-kin network loss (IRR = 1.02, p <.05).

The relationships between neighborhood conditions and social network change can also be interpreted in terms of predicted values to help understand the meaning of these associations. Figure 3 shows the predicted number of kin and non-kin network members lost (average adjusted predictions) by the level of neighborhood concentrated disadvantage at Round 2. These predictions are derived from a version of Models 2 and 4 of Table 3 that relies on categorical measures (quarters) of concentrated disadvantage rather than continuous for ease of interpretation. On average, older adults residing in neighborhoods that fall in the top quarter of concentrated disadvantage experience approximately .28 more kin network losses (p <.05) and .15 more non-kin network losses than those residing in the bottom quarter of concentrated disadvantage. Put differently, higher levels of concentrated disadvantage are associated with a 75% higher probability of losing a kin network member compared to a non-kin network member relative to living in the lowest quartile of concentrated disadvantage.

Figure 3.

Predicted number of network losses between Rounds 2 and 3, by quarters of neighborhood concentrated disadvantage at Round 2. Predicted values represent average adjusted predictions that are derived from models that include the full set of covariates.

*p < .05; **p < .01; ***p < .001 (Two-sided tests).

Figure 4 illustrates the degree of network turnover predicted by different levels of perceived neighborhood social ties. Older adults who report the highest levels of perceived neighborhood social ties add, on average, .83 kin ties between Rounds 2 and 3, compared to 1.05 kin ties added, on average, among those who report the lowest levels of perceived neighborhood social ties (p < .01). To put these results in further perspective, the difference in the predicted addition of kin ties between the highest and lowest levels of perceived neighborhood social ties (.22) is comparable to the difference in the predicted number of kin additions between respondents with less than a high school degree and those with a Bachelor’s degree or more (.24). For non-kin ties, we observe the opposite pattern. Those who report the highest levels of perceived neighborhood social ties report adding, on average, 1.19 non-kin network ties between Rounds 2 and 3, which is approximately .32 more than the number of non-kin additions (.87) reported by those with the lowest levels of perceived neighborhood social ties (p < .001).

Figure 4.

Predicted number of network additions between Rounds 2 and 3, by level of perceived neighborhood social ties at Round 2. Predicted values represent average adjusted predictions that are derived from models that include the full set of covariates.

*p < .05; **p < .01; ***p < .001 (Two-sided tests).

Findings from our main models in Table 3 indicate that moving plays a significant role in personal network turnover, particularly for non-kin ties. Table 4 presents coefficients for the focal neighborhood variables in models that stratify respondents by moving status between Rounds 2 and 3. These models allow us to consider whether the findings are robust to the exclusion of movers, who may have greater exposure to new non-kin ties between rounds, and who may otherwise be different from individuals who cannot or choose not to move (more broadly, mover status may introduce selection issues). Among non-movers, higher levels of concentrated disadvantage were associated with significantly higher rates of kin loss (IRR = 1.13, p < .05), whereas greater perceived neighborhood social ties were associated with a 14% lower rate of kin addition (IRR = .86, p <.001) and a 27% higher rate of non-kin network addition (IRR = 1.27, p <.001). Among movers, we do not observe statistically significant associations between neighborhood measures and network turnover.

Table 4.

Incidence Rate Ratios from Poisson Models Predicting the Number of Kin and Non-Kin Additions and Losses Between Rounds 2 and 3, by Mover Status.

	Non-Movers (R2 to R3)
	# Kin Network Members		# Non-Kin Network Members
	(1a) Additions	(2a) Losses	(3a) Additions	(4a) Losses
Concentrated disadvantage	1.03	1.13*	1.06	1.05
	(.04)	(.05)	(.06)	(.03)
Residential instability	.96	1.02	1.02	.99
	(.03)	(.03)	(.04)	(.03)
Perceived neighborhood social ties	.86***	1.02	1.27***	1.05
	(.03)	(.03)	(.06)	(.03)
F(df)	19.50*** (24, 47.7)	8.56*** (24, 47.9)	8.45***(24, 47.8)	2.39**(24, 47.8)
N	1,913	1,783	1,913	1,195
	Movers (R2 to R3)
	# Kin Network Members		# Non-Kin Network Members
	(1b) Additions	(2b) Losses	(3b) Additions	(4b) Losses
Concentrated disadvantage	1.05	1.06	1.10	1.14
	(.12)	(.10)	(.08)	(.10)
Residential instability	1.01	.95	.88†	.99
	(.08)	(.06)	(.06)	(.06)
Perceived neighborhood social ties	.93	1.05	1.11	.97
	(.08)	(.06)	(.07)	(.04)
F(df)	9.97***(24, 46.7)	3.83***(24, 46.7)	8.20***(24, 46.7)	6.31***(22, 45.9)
N	412	382	412	294

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001 (two-sided tests). 95% confidence intervals in parentheses.

^aAll models control for the inverse Mills ratio, self-rated health, marital/partner status at Round 2, retirement between Rounds 2 and 3, widowhood between Rounds 2 and 3, and number of kin (models 1a, 1b, 2a, and 2b) and non-kin (models 3a, 3b, 4a, and 4b) network members at Round 2.

Supplementary analyses.

The Poisson models suggest that neighborhood conditions are relevant for explaining patterns of social network change. We conducted two sets of supplementary models to address potential limitations and extensions of this analytic design. First, research shows that personal social networks may exhibit, to some extent, homeostatic properties, meaning that losses or additions experienced at time 1 tend to be offset by compensating changes at time 2 (Cornwell et al., 2020; Fischer & Offer, 2020). These findings beg the question of whether the apparent linkage between neighborhood conditions and the social network changes between Rounds 2 and 3 exists independent of changes between Rounds 1 and 2. Changes between Rounds 1 and 2 are likely to shape subsequent changes and are also likely to be shaped by the same neighborhood conditions.

Appendix Table 1 presents the results from models based on respondents from all three rounds that include control variables that reflect changes in the corresponding network outcomes between Rounds 1 and 2 (i.e., losses and additions between Rounds 1 and 2). In most cases, prior losses and additions (between Rounds 1 and 2) significantly predict subsequent changes between Rounds 2 and 3. Even with strong evidence that personal network changes are a function of prior network changes, however, the role of neighborhood conditions observed in Appendix Table 1 is still generally consistent with the findings in the main analyses. Higher levels of concentrated disadvantage are associated with significantly higher rates of kin network losses (IRR = 1.16, p < .01, respectively). Perceived neighborhood social ties are associated with lower rates of adding kin ties (IRR = .86, p < .01), higher rates of adding non-kin ties (IRR = 1.20, p < .001), and higher rates of losing non-kin (IRR = 1.05, p < .05). Therefore, it seems that neighborhood conditions influence personal network turnover independent of prior changes in the network.

Although the findings are robust to alternative specifications of the lagged DV, the Poisson models also exclude respondents who have zero network members at Round 2 in models predicting network losses, and zero network members at Round 3 in models predicting network additions. This restriction leaves open the question of whether the findings can be generalized to people who have smaller networks, or whose networks are particularly kin-centric or non-kin-centric, given that network size is also a function of neighborhood conditions (Tigges et al. 1998; York Cornwell and Behler 2015). To address these concerns, we used a multilevel modeling strategy that nests repeated measures of individual and neighborhood covariates from each of the three rounds (Level 1) within individual respondents (Level 2). This type of model accounts for overtime within-person differences in time-varying individual and neighborhood measures, as well as between-person differences in time-stable factors. In addition to fixed effect parameters, we estimate random effects (individual deviations) for the intercept.

Whereas the Poisson models in the main analyses rely on tracking the entrance or departure of specific network alters, the multilevel models are predicting overall growth or decline in network size and kin composition (proportion kin). In this sense, the multilevel models are not capturing turnover based on who comprises the network, but rather more general measures of changes in social integration – overall and by alter type. Nevertheless, they provide a useful alternate view and robustness check using all three rounds of the NSHAP, and in a framework that accounts for both within and between respondent variation (Singer and Willett 2003).

As shown in Appendix Table 2, the associations between neighborhood conditions and trajectories of personal network change are generally consistent with some of the findings from the main analyses.^v Concentrated disadvantage is associated with declines in overall network size (IRR = .97, p < .05) and with marginally significant declines in proportion kin (b = −.02, p < .10). These results lend additional support to the main findings, specifically the notion that overtime trajectories of personal network size and composition are in part a function of neighborhood concentrated disadvantage.

DISCUSSION

This study was driven by growing empirical research that demonstrates the dynamic properties of older adults’ personal networks as a reflection of social stratification (Fischer and Beresford 2015; Schafer and Vargas 2016), along with evidence that neighborhood conditions shape some of the most consequential features of older adults’ personal networks (Bloem et al. 2008; Schieman 2005; Thomése et al. 2005; Settels 2020). This study aimed to contribute to a broader structural framework for examining social network change relative to prior traditions of examining more individual-level and interpersonal (network-level) predictors (e.g., Mollenhorst et al. 2011). These analyses collectively indicate that social and economic aspects of residential neighborhoods – in particular, neighborhood concentrated disadvantage and perceptions of neighborhood social ties – influence older adults’ personal network turnover in ways that differ across kin and non-kin ties. Higher levels of perceived neighborhood social ties are associated with higher rates of adding non-kin and lower rates of adding kin network ties. Higher levels of tract-level concentrated disadvantage are associated with higher rates of losing both kin and non-kin ties over time. Associations are generally robust to the exclusion of respondents who move between survey rounds.

Perhaps the most striking implication of this study is that older adults living in neighborhoods with higher levels of concentrated disadvantage are at higher risk of losing non-kin network members. These findings emerge when accounting for life course transitions, health, and social position, which prior research links with kin relationship dynamics (e.g., Fischer and Beresford 2015; Zettel and Rook 2004). Older adults who live in more socioeconomically disadvantaged neighborhoods may be more likely to rely on local kin and non-kin for various forms of support, in part to compensate for individual and neighborhood disadvantages (Kana’Iaupuni et al. 2005; Stack 1974). Over time, support demands or the inability to fulfill support needs may weaken close network ties. Local kin and non-kin ties may also move out of the neighborhood, making it difficult for older adults who remain living in more disadvantaged neighborhoods to maintain more geographically distal kin relationships. Further, the finding that no other neighborhood conditions examined in this study appear to shape the addition of kin ties suggests that older adults’ close family relationships may be especially vulnerable to structurally disadvantaged neighborhoods, without any apparent contextual force to counteract or compensate for this loss.

Important to recognize, however, is that kin network turnover may be influenced by life course events in the lives of kin that are not measured in this study, and that are not necessarily directly shaped by neighborhood factors (e.g., whether a child divorces or marries, death or declining health of a kin member). Whereas respondents’ self-rated physical health did not reach statistical significance in our models, neighborhood characteristics may compromise the health of network members who also live in more disadvantaged areas in ways that interfere with the maintenance of these ties (e.g., death or disability). Indeed, neighborhood disadvantage is associated with all-cause mortality (Denney, Saint Onge, and Dennis 2018) and poor health (Diez Roux and Mair 2010), while weak infrastructure and lack of transportation can make it difficult to care for network members who cannot travel (Denney et al. 2018; Meijer et al. 2012), ultimately straining these ties over time. Neighborhoods can also function as potential “ecological stressors” or stress buffers through their institutions and other resources, levels of physical disorder, social disorganization (crime, distrust), and available social support and social capital (Karb et al. 2012). High levels of individual stress or stress within one’s personal network could lead to a demanding personal network (Offer and Fischer 2018) that ultimately compromises network relationship stability.^vi

Older adults’ perceptions of neighborhood social ties emerged as especially relevant for the addition of both kin and non-kin network ties, albeit in opposite directions. Older adults who perceived higher levels of interaction among neighbors, in the form of favors, advice, and other types of social exchange, may be more likely to participate in such exchanges or to become involved in organizations, thereby having more opportunities to form ties with neighbors and other non-kin. As older adults’ needs for certain kinds of support and information change over time, a higher degree of non-kin turnover may signify that older adults who live in neighborhoods with greater perceived social interaction have a wider range of non-kin resources to recruit as needed. That this finding emerges among non-movers suggests that this is not merely due to changing neighborhoods; rather, neighborhoods with greater social ties likely offer older adults more opportunities for developing relationships with non-kin.

We speculate that different mechanisms could lead to distinct associations between residential neighborhood characteristics and patterns of turnover among network ties who do and do not reside in the same neighborhood as respondents (local versus non-local network ties). One possibility is those older adults who perceive higher levels of social interaction in their neighborhoods are more likely to become involved in local organizations and social venues. To the extent that these social experiences cultivate and strengthen local friendships, we may observe higher rates of adding local non-kin ties and the potential decay or dropping of non-local non-kin ties. Higher levels of neighborhood concentrated disadvantage may lead to lower levels of dropping local kin and non-kin alters to the extent that fewer resources in the local environment may promote longer-term relationships with co-residing kin and other local network members (kin and non-kin) as key sources of practical and emotional supports (York Cornwell and Goldman 2021), similar to the notion of adaptive cohesion (Schieman 2005). A dense network of local ties may also reinforce local relationships over time, as more embedded network ties may slow or prevent tie decay (Burt 2000). At the same time, if neighborhood disadvantage is associated with less social support availability at the local level, residents of more disadvantaged areas may seek out non-local ties to compensate for or substitute a lack of social resources within the residential neighborhood. Future rounds of the NSHAP that collect information about whether each network alter resides in respondents’ local neighborhood (currently only available from Round 3) will allow for empirical tests of these possibilities.

As perceived neighborhood social ties are also associated with a lower rate of adding kin network ties, this may reflect that older adults in more social neighborhoods are fulfilling social needs through the addition of non-kin in ways that render the recruitment of additional kin ties less necessary. Interestingly, perceived neighborhood social ties were only weakly correlated with concentrated disadvantage and residential instability. This finding departs from predictions made within social disorganization theory, which suggests that higher levels of socioeconomic disadvantage preclude the development of social ties (e.g., Sampson and Groves 1989). One possibility is that stressful neighborhood conditions facilitate the formation of ties within the local area, serving as buffers against fear or mistrust that stem from neighborhood disadvantage (Oh and Kim 2009; Ross and Jang 2000), analogous to the idea of “adaptive cohesion” (Schieman 2005).

Given recent evidence that suggests that a substantial portion of non-kin ties are not truly “lost” over time (Fischer and Offer 2020), the non-kin ties who are “lost” in this study may actually be “dormant,” or otherwise not listed as a result of the 5 alter limit in the network roster. In this sense, higher perceived levels of neighborhood social ties may be associated with the expansion of older adults’ more peripheral social networks.^vii The finding that higher perceived levels of neighborhood social ties are also negatively associated with the addition of kin ties supports the interpretation that areas with more interaction among residents provide more non-kin alternatives whose inclusion in the network may offset the need to recruit kin as key sources of social support. Social capital at the neighborhood level may structure opportunities for accessing different kinds of social capital at the individual network level.

One takeaway from this and other recent studies (e.g., Goldman and Cornwell 2018) is the notion that kin ties are particularly vulnerable to individual and neighborhood indicators of disadvantage. This pattern is somewhat surprising given prior work on network-based predictors of turnover that suggests that network members who know one another are less likely to leave the network (Feld 1981; Feld et al. 2007; Mollenhorst et al. 2011). Kin members are more likely to be embedded in older adults’ networks, but the lower rates of kin tie decay that one might expect from this embeddedness (Burt 2000) do not follow the findings that emerge here. These results call for greater attention to the social contexts surrounding kin relationships, and how the constraints of those contexts may undermine relationship stability.

Limitations and Future Directions

The NSHAP offers one of the first opportunities to examine how neighborhood-level measures influence changes in older adults’ personal social networks using a nationally representative sample. Nevertheless, this study is not without limitations. For one, the neighborhood measures that we use in this analysis rely on data from respondents’ residential census tracts. Although informative of individuals’ primary social contexts, older adults spend a considerable portion of their time outside of their residential neighborhood (York Cornwell and Cagney 2017). The spaces that older adults visit to conduct their day-to-day activities may vary in structural and social characteristics, and in ways that may contribute to different levels of personal network change. Future research using activity space data could examine whether extra-residential neighborhood characteristics also play a role in personal network turnover.

Second, the NSHAP does not collect detailed information on why older adults add or lose network confidants, although information on the mortality of “lost” ties could be leveraged in future analyses. Still, this leaves open the question of why exactly respondents add and lose network ties over time, and the specific ways that neighborhood conditions directly or indirectly influence these changes. A related limitation is that all network change is self-reported by the respondent. We are unable to discern whether a network loss, for example, was due primarily to actions or circumstances of the respondent, of the network member, or perhaps was mutual. Likewise, information on the social and economic circumstances of respondents’ network alters – including their neighborhood contexts – could speak to how neighborhood conditions experienced within the network as a whole might contribute to network turnover.

Finally, it is important to emphasize that these findings are only generalizable to the population of community-dwelling older adults in the United States. As the United States is unique from other comparably wealthy nations in its weaker, more fragmented social welfare and healthcare system, these factors may contribute to the patterns of network turnover that we observe across residential neighborhoods. Future research using other datasets is needed to determine whether the findings are generalizable to other countries.

CONCLUSION

Later life is a time of personal and social change, as older adults become both increasingly dependent on their social ties and increasingly vulnerable to their residential environments (e.g., Yen et al. 2009). This study offers compelling evidence that personal network turnover is constrained in nuanced ways by the structural and perceived social characteristics of residential environments, and in ways that may be consequential for older adults’ well-being. As social network turnover is increasingly studied as a stratified process (e.g., Fischer and Beresford 2015; Schafer and Vargas 2016; Zettel and Rook 2004), these findings call for greater attention to the intersection of personal network change and neighborhood social and economic conditions in contributing to inequality in individual and community-level outcomes.

Highlights.

• Neighborhood conditions influence the addition and loss of older adults’ personal network members over time.

• The role of neighborhood conditions emerges even when accounting for life-course transitions, health, and social position, which prior research links with personal network change.

• Older adults living in neighborhoods with higher levels of concentrated disadvantage lose kin and non-kin network members at higher rates than those living in more advantaged neighborhoods.

• Higher levels of perceived neighborhood social ties are associated with higher rates of non-kin network additions and lower rates of kin network additions.

Appendix Table 1.

Incidence Rate Ratios from Poisson Models Predicting the Number of Network Additions and Losses Between Rounds 2 and 3, Including Changes between Rounds 1 and 2 as Covariates.^a

	Number of Network Additions W2 to W3			Number of Network Losses W2 to W3
	Overall	Kin	Non-Kin	Overall	Kin	Non-Kin
Concentrated disadvantage	1.04	1.05	1.06	1.10*	1.16**	1.02
	(.03)	(.05)	(.06)	(.04)	(.06)	(.03)
Residential instability	.93**	.97	.86**	.92***	.89**	.98
	(.02)	(.04)	(.04)	(.02)	(.04)	(.02)
Perceived neighborhood social ties	1.05	.86**	1.20***	1.03	1.03	1.05*
	(.03)	(.04)	(.05)	(.02)	(.04)	(.03)
Overall Additions W1-W2	.93***			1.15***
	(.01)			(.02)
Overall Losses W1-W2	1.33***			1.04*
	(.02)			(.02)
Kin Additions W1-W2		.86***			1.11***
		(.03)			(.02)
Kin Losses W1-W2		1.46***			1.08**
		(.04)			(.03)
Non-Kin Additions W1-W2			1.04			1.08***
			(.03)			(.01)
Non-Kin Losses W1-W2			1.25***			1.13***
			(.08)			(.03)
N	1532	1532	1532	1547	1420	982

^†p < 10;

^*p < .05;

^**p < .01;

^***p < .001 (Two-sided tests). Standard errors in parentheses.

^aAll models include the full set of covariates that are included in the main analyses. Coefficients and statistical significance of neighborhood measures are consistent when additions and losses between Rounds 1 and 2 are modeled as factor variables.

Appendix Table 2.

Coefficients from Multilevel Models Predicting Growth/Decline in Network Size and Proportion Kin, Rounds 1 to 3.^a

Predictors	Network Size (Poisson)	Proportion Kin (OLS)
Fixed effect parameters
Concentrated disadvantage	.97*	−.02†
	(.01)	(.01)
Residential instability	1.02†	.001
	(.01)	(.01)
Age	1.05***	−.04***
	(.01)	(.01)
Female	1.10***	−.03*
	(.02)	(.02)
Black	.92*	−.01
	(.03)	(.03)
Hispanic	.86***	.07*
	(.03)	(.03)
Educational attainment (ref = Less than high school)
High school or equivalent	1.03	−.002
	(.03)	(.03)
Some college	1.11**	−.09**
	(.03)	(.03)
Bachelor’s or more	1.13***	−.07*
	(.04)	(.03)
Random effects parameters
Level 1 residual		.057**
Level 2 intercept	1.21e-33***	.033**
Number of observations	3577	3567
Number of respondents	1286	1286
Log presudolikelihood	−7192.88	−634.10

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001 (Two-sided tests). Standard errors in parentheses. Models are weighted using respondent-level Round 1 weights that adjust for probability of selection and non-response and are based on non-imputed data.

^aAll models include controls for marital status, retired status, urbanicity, residential tenure, number of household members, income (4 categories), number of children, whether the respondent moved to a different census tract between Rounds 2 and 3, and the inverse Mills ratio. Coefficients in the Poisson model (network size) are incidence rate ratios.