Structural Advantages of Good Health in Old Age

Abstract

This article tests the hypothesis that health influences older adults’ position within a defined social structure. Building on a recent synthesis of social gerontology and network analysis, good health was expected to be associated with more network constraint and less network integration—two indicators of autonomy and access to social resources and information within a network. The study was conducted within a continuing care retirement community, a unique site offering several advantages for a novel test of the health-begets-position hypothesis. Consistent with this hypothesis, residents with the best health had positional advantage in the network. Results also highlight the particular importance of good health during the initial period of community tenancy.

Keywords

health retirement community social networks

A basic insight from social network scholarship is that personal relationships hold the potential to dispense resources, information, and other varied benefits (Burt 2000; Granovetter 1973). Hence, how one is positioned among a set of others goes a long way in determining how much of the “good stuff” one derives from social life.

Of course, the realities of human development imply that the attainment and consequences of network position are complicated by age-related variables. Each stage of the life course may present distinct challenges, but old age is a stretch of time in which many people first encounter serious barriers to social participation and functioning. Scholars in the field of social gerontology have demonstrated, using a wide range of theoretical devices and methodological tools, that later-life health declines can disrupt social well-being in the form of strain on family relationships (Kaufman and Uhlenberg 1998), difficulty attending voluntary associations (Herzog et al. 1989), and erosion of former friendships (Shippee 2008).

A recent batch of empirical research has married these two perspectives—the basic orientation of social network analysis and the substantive concerns of social gerontology—with exceptional clarity and theoretical acuity. In particular, Benjamin Cornwell (2009a, 2009b) has advanced the proposition that health is related to social network position among older adults. With high-quality data from the recent National Social Life Health and Aging Survey (NSHAP; Cornwell et al. 2009), Cornwell finds that higher levels of functional and cognitive health enable people to occupy positions of power and independence in a network. For instance, healthier people have greater “bridging potential” within their group of core discussion partners. That is, they are more likely to report that their close relations do not know one another, suggesting that they can serve as intermediaries among otherwise unconnected people in their network and can remain relatively independent of others’ influence (Burt 1992).

For the remainder of this article, I refer to this concept—that health influences where someone is located in a network—as the health-begets-position (HBP) hypothesis. The idea is rich with possible implications, but there have been relatively few opportunities to adequately test it. Other network correlates of health besides position—network composition, size, and interaction (all recounted below)—lend themselves more easily to empirical analysis because they require less knowledge about the network as a system, focusing instead on immediate aspects of a focal individual’s reported relationships. Network position, on the other hand, would ideally account for what happens beyond first-order social ties; a thorough picture of how health influences one’s location in a social structure should account for ties of ties, ties of ties of ties, and beyond.¹

The chief purpose of this article is to further examine the HBP hypothesis (a) with a novel data set from a complete retirement community population and (b) with a set of outcomes related to network position requiring information from a full network and heretofore not examined in the gerontological literature. Specifically, I investigate how health is associated with network integration, a variant of closeness centrality that measures how well “an individual is connected to many diverse others in a network” (Valente and Foreman 1998: 90), and network constraint, an indicator of limited bridging potential that represents “the extent to which a person’s network is concentrated in redundant contacts” (Burt 2000: 373). Both constructs provide a fresh opportunity to test the HBP hypothesis.

Background: Social Networks and Health

Broadly speaking, the HBP hypothesis is one piece of a puzzle that asks how social networks and health are interrelated. Beyond the converse possibility that social networks influence health (Berkman et al. 2000), there are at least three other hypotheses in the extant literature concerning how health affects some aspect of social networks.²

First, health may influence how intensively people interact with their social ties. A health-begets-interaction (HBI) hypothesis may suppose that good health is an enabling resource for frequent contact. For example, a study of independently living Dutch seniors found that better health was associated with providing more frequent support to their social contacts (van Tilburg and van Groenou 2002). On the other hand, poor health may induce more interaction with a tie because it signifies the need for social support and caregiving (Cornwell, Schumm, and Laumann 2008). This pattern of receiving more contact in periods of health duress may be especially salient when the social tie is with a kin member.

Second, a health-begets-size (HBS) hypothesis anticipates that health shapes the number of contacts that a person has in his or her personal network. Similar to the logic of the HBI hypothesis, good health may serve as a resource that allows people to fulfill the demands of multiple relationships and maintain larger networks. It remains somewhat unclear, however, whether this general process is generalizable across age groups. Cornwell et al. (2008), for instance, report no association between self-rated health and size of core discussion partner network in a national sample of older American adults. In a recent study of adolescent friendship networks, on the other hand, Haas, Schaefer, and Kornienko (2010) show that youth with poor health ratings have fewer friends and are more likely to be isolates in their schools.

Much of the ambiguity in the HBI and HBS hypotheses is resolved when network composition is considered. A health-begets-composition (HBC) hypothesis asserts that health influences who composes a person’s network, which has implications for its size and the volume of interaction it contains. For instance, people in poor health may never see a host of casual contact that used to compose their friend network, yet have small, time-intensive networks composed of people in their immediate family.

Within the HBC tradition, network composition often means simply distinguishing kin from nonkin network ties (Cornwell 2009a). Other scholars provide a broader typology, differentiating networks as diverse, friend concentrated, church focused, family centered, or restricted (Litwin and Shiovitz-Ezra 2011a, 2011b). Consistent with an HBC assumption, there is some evidence that health is associated with network composition. For example, older adults who report “very good” health are 45% more likely to have a friend-concentrated network than are seniors in fair or poor health (Litwin and Shiovitz-Ezra 2011a).

In summary, three aspects of social networks—how much interaction, how many alters, and who composes the network—have been the primary focus of several studies in the social gerontological literature. An HBP perspective likewise expects that health is a crucial factor in enabling or constraining social functioning among seniors but marks somewhat of a departure from the other three hypotheses.

The HBP Hypothesis

The HBP hypothesis assumes that there is a relevant web of connections beyond an individual ego that are nevertheless influenced by a person’s health. Although the three above approaches are concerned with the properties of first-order social ties (e.g., who i, or ego, is connected to), the HBP hypothesis involves second-order ties and beyond (e.g., given i’s connection to j and k, who j and k are connected to). Figure 1 illustrates this conceptual extension; Panel A shows the basic information relevant for the HBI, HBS, and HBC hypotheses, whereas Panel B shows the information that is necessary for HBP inferences.

Figure 1.

Hypothetical scenarios for assessing hypotheses about health and social networks.

There are multiple strategies for operationalizing network position; this article follows and expands on the approach already taken in studies assessing the HBP hypothesis (Cornwell 2009a, 2009b). As a starting point, Granovetter (1973) famously noted that weak ties extending beyond one’s tight circle of close relations are crucial for opening access to diverse sources of information. Furthermore, being the bridge between these nonoverlapping sets of people provides its own set of interpersonal advantages, such as the ability to “broker” or strategically manipulate relationships between otherwise disconnected people, the role of being a “go-to” person for needed resources and information, and the capacity to remain autonomous from any given person’s control (Burt 2000; Emerson 1962).³ In general, the assumption is that being well integrated among a diverse host of people in the network—that is, having alters who are otherwise not likely to be connected with one another—opens up channels of information that would be unavailable in more circumscribed friendship circles and provides interpersonal power and independence. Indeed, Cornwell (2009b: 133) notes the particular importance of these various resources among “older adults who [tend to] fear loss of control and autonomy in their later years.”

Older adults, however, are more likely than younger people to experience health problems, which could limit their accessibility to the resources embedded in a network. The underlying processes are an extension of how health shapes interaction, size, and composition of personal networks. For instance, if health difficulties prevent an older person from participating in activities that give him or her opportunities to interact with new people, then he or she will likely recede into a more constrained position in the network (see Feld 1981 for a theoretical account of this general process). Likewise, if health is required for the energy and effort of maintaining personal network size, then it should also serve as a helpful asset for network bridging behavior; past studies on this topic emphasize that bridging structural holes in a network is taxing relational work because it requires juggling the demands of multiple relationships without the supportive scaffolding of a common group (Cornwell 2009a, 2009b). Finally, network composition may play a role in the HBP process. Specifically, health problems are associated with a higher proportion of frequently seen kin ties, a factor that partially explains why less healthy seniors are less likely to name people as core discussion partners who are themselves not connected (Cornwell 2009a).⁴

This final point—the close affinity between composition and position constructs—marks one of the limitations of existing tests of the HBP hypothesis. In the NSHAP study design, any participant who cites only family members as a tie (i.e., people with whom he or she discusses important matters) will by necessity manifest low bridging potential and appear to occupy a highly constrained position in the network.⁵ A main reason, then, that network position and composition are especially difficult to disentangle is that prior studies infer network position from the individual self-reports of (presumably) independent study participants. That is, “position” is allowed to extend only as far as the handful of people a subject in the survey chooses to name. Moreover, because survey participants are culled from a national random sample, the subject’s alters are outside of the sample and not connected in any way to any other subject in the study. Under this strategy, there is no way to know who ego’s alters are connected to—or if they even see themselves as tied to ego.

There is an alternative to beginning with individually sampled people and extrapolating upward to approximate personal network structure. A full network strategy obtains information from all people in a self-contained population and examines how those people fit within the overall structure of relationships. This setup gives primacy to the entire context of actors—any of which could, in principle, be linked to another—and it therefore enables a more direct test of the HBP hypothesis. In short, for the essence of a network “position” to make sense, there should be a system in which the position is located. This scenario, in turn, is made most concrete with a full network design. Figure 2 demonstrates how the conception of ego’s structural position changes according to his or her position in relation to all other actors in the network.

Figure 2.

Hypothetical scenarios for inferring network position.

Context and Considerations for the Current Study

In the present study, the population used to define a full network comes from all independent living residents in an American continuing care retirement community (CCRC).⁶ This type of population is ideal to assess the HBP hypothesis for several reasons. First, the CCRC under consideration represents a naturally bounded population; residents of the community live under the same roof, eat their meals in the same dining room, and participate in the same set of social, physical, educational, and religious activities. Using a meaningful and unambiguous community averts the boundary-specification problem that occurs when an analyst must create arbitrary demarcations to define a network (Wasserman and Faust 1994: 30-34). Intentional communities are typically distinctive in their demographic makeup—and CCRCs are certainly no exception—but they provide the opportunity to observe basic social processes on a far more tractable scale than could be achieved in more general populations (Vaisey 2007).

Second, though the population in this study was almost entirely White (> 98%), educated, and well-off, the demographic uniqueness of a CCRC population can actually be an asset. Racial, ethnic, and social class homogeneity effectively operates as a “control” to help rule out potential sources of spuriousness.

Third, by defining the relevant network a priori as the set of relations among people within a given community, we are on firmer ground to assess whether health influences network position without the complicating influence of network composition. When people vary in how they define their own network, second-order inferences about network structure become problematic.⁷ This study focuses on nonkin social relations with retirement community peers; hence, there is no ambiguity across study participants about who “belongs” in the network.

Finally, the full network population used in this study allows a distinction to be made between ties sent by ego and those received by ego (i.e., they are directed ties). Prior research demonstrates that self-reports of social interaction are considerably biased (Bernard et al. 1984), and the bias is nonrandom such that certain respondents will predictably overreport their relations and others will do the opposite (Feld and Carter 2002). Rather than taking ego at his or her word, Feld and Carter (2002) recommend relying on incoming directed ties (i.e., information from each alter concerning their tie with ego rather than vice versa). That way, potential reporting biases will average out rather than cluster within particular subjects.

Research Question

The following research question guides the analysis; it is intended to build on and extend initial tests of the HBP hypothesis: Is health associated with greater levels of network integration and lower levels of network constraint?

As a multidimensional construct, health is assessed by both an overall index of physical and mental well-being and two indicators of sensory impairment that are important for social contact—sight and hearing. Network integration refers to the “distance” a given person is to all others in the network, and network constraint denotes the level of redundancy in a person’s social ties (i.e., whether alters tend to be tied to one another). Though both are related to bridging potential and independence, the concepts are inversely related to one another; consistent with the HBP hypothesis, good health is expected to be associated with more integration and less constraint. Both position constructs take advantage of a full network design and are described in more detail in the method section below.

Method

Population and Study Protocol

At the beginning of the 6-month data collection period, the population of CCRC residents included 158 independently living persons. The CCRC had a separate facility that included assisted living and nursing care residents, but these persons were not included in the study population.

Over the course of data collection, seven persons died, one moved away, nine transitioned from an independent living apartment to another setting within the facility, and six lived independently with a spouse but were cognitively unfit for an interview. This left 135 people for a valid study population, and interviews were conducted with 123 of these residents (91% response rate).

Interviewees were recruited by face-to-face contact at social activities (e.g., meals, copresence at activities), by visiting residents at their apartments, and by telephone calls. As an incentive, participants were put in a drawing to win a $25.00 gift certificate to a popular local restaurant. Interviews were conducted in a small office in the facility or privately in interviewee’s own apartments. All interviews were conducted privately, so separate appointments were made with each person in the case of partnered residences (38% of respondents were partnered). All components of the study met approval by the university institutional review board.

Measures of Network Position

This study uses two measures of network position. The basic building blocks of each measure come from reports about social interaction gathered during the study interviews. Participants were shown a succession of figures, depicting each of the residences floors in the retirement community. Names of the residents were included next to their apartment in the floor plan, and interviewees were asked whether they “spend time interacting or socializing with [NAME] in a given week, beyond just passing by or saying hello.”⁸ If the answer was an affirmative, participants were then asked to approximate how much time they spent socializing “in a typical, or average, week.” Relations were designated by reports of half an hour or more of interaction. Time spent with married or nonmarried cohabiting partners was the only type of interaction not measured in this protocol.

Next, a 123 × 123 matrix was created, each column containing ego’s reported relations from the other 122 alters indexed in the matrix rows. A relation from alter j to ego i is defined as x _ji. The matrix serves as the basis for calculating constraint and integration.

The logic of constraint focuses on whether a person’s relationships are concentrated with others who are interconnected with an identical set of alters—“the more constrained the actor, the fewer opportunities for action” and autonomy (Borgatti, Jones, and Everett 1998: 3). Therefore, if actor i has ties with j, k, and l, i’s constraint will be higher if j and k are tied than if they are not. More specifically, constraint is affected by what proportion i and j make up of all i’s ties and the proportion by which i’s ties with k also involve k and j. As given by Burt (1992: 55), the basic representation of i’s constraint via each j is,

C_{i j} = {(p_{i j} + \sum_{k} p_{i k} p_{k j})}^{2}

where C _ij = constraint of i for each tie j, p _ij = proportion i’s ties that are invested in j, p _ik = proportion of i’s ties that are invested in k, and p _kj = proportion of k’s ties that are invested in j. High values of the product of p _ik and p _kj mean that i’s tie to k repeats connections to j, thereby indicating few structural holes for i to bridge among his or her personal network connections (Burt 1992). In this way, constraint denotes position, serving as “an (inverse) indicator for the utility [i.e., information or resource advantage] that actors can extract from a network” (Buskens and van de Rijt 2008: 378). Prior to its use in the final analysis, constraint was transformed to its natural logarithm because of its skewed distribution.

The second measure of network position is integration. Valente and Foreman (1998: 92) express the concept as,

I (i) = \frac{\sum_{j \neq k} R D_{j k}}{N - 1}

where I(i) is the integration score for actor i, RD_ji is the reverse of the geodesic path length spanning j and i, and N is the number actors in the network. Geodesic path length refers to the shortest path between two actors, such that if l reports a tie to k, k to j, and j to i, then the geodesic path length from l to i is equal to three, barring any alternate and more direct routes in the network. The average distance for each actor from every other actor in the network was reversed and averaged so that higher values mean more sociometric “closeness.” This integration value was then standardized so that it ranges from zero to one.

It is worth emphasizing that both indicators of network position take into account information from all other actors in the population. This is in contrast to more basic network measures such as ego-reported centrality, which is essentially a count of how many ties a respondent reports, or measures of network density calculated from a sample-based survey.

Health Measures

Overall health measures were drawn from the Rand 36-Item Health Survey (hereafter RIS) and from previous health surveys conducted on older samples. Individual RIS items were scored according to the recommendations of the study team (Ware et al. 1993); each respondent was then assigned an overall “health” value averaged across the series of health subscales (physical functioning, disturbance of normal roles because of physical reasons, disturbance of normal roles because of emotional problems, energy or fatigue levels, emotional well-being, social functioning and disruption of social activities, bodily pain, and overall health evaluations). To avoid conflating aspects of the independent and dependent variables, the social activity dimension was removed from the overall index. As a single measure, the 34-item index has high reliability (α = .92). All analyses are conducted after z transforming the measure.

In addition to the RIS health subscales, two questions about sensory health were incorporated from the National Social Health and Ageing Project (Williams, Pham-Kanter, and Leitsch 2009). Specifically, respondents were asked to evaluate their hearing and their vision as excellent, very good, good, or poor. Responses were coded so that better sensory health had higher scores (1–5).

Control Variables

Regression analyses control for several variables that may be jointly related to health and to network position. Age is a continuous variable subtracting year of birth from year of interview; gender is a binary variable (1 = female, 0 = male); residents who were married or were cohabiting were considered “partnered.” Tenure in the facility and whether the respondent lived in the local community are two factors that likely influence people’s network position in the CCRC. Tenure refers to how many years a respondent has lived in the retirement community and is operationalized with several categorical variables—recent residents are those with a tenure of less than 3 years, and long-term residents are those with a tenure of 18 years or more (3–16 years in the facility is the reference category in regression analyses).⁹ Local is a binary variable denoting whether the resident lived in the county prior to his or her move to the CCRC.

Results

Descriptive statistics are presented in Table 1. The values reveal that respondents living in this CCRC were relatively old (mean of 86 years), a majority were female (72%), and slightly over a third were partnered. Though some residents had been living in the community for less than a year, the average resident had been there over 5 years. Slightly over 40% of the residents had been in the facility for less than 3 years. Nearly three out of four residents had lived in the local area prior to moving into the CCRC.

Table 1.

Descriptive Statistics of Study Population.

	Range	M	SD
Overall health (SF-36)^a	16.35–95.59	58.52	17.45
Vision	1–5	3.27	1.13
Hearing	1–5	3.00	1.12
Network position
Constraint (ln)	−2.35–0.21	−1.57	0.50
Integration	2.80–4.26	3.93	0.22
Covariates
Age	74–96	85.95	4.35
Female	0–1	0.69
Partnered	0–1	0.38
Local	0–1	0.72
Recent resident	0–1	0.43
Long-term resident	0–1	0.05

N = 123.

Two items that referred to social interaction were removed from the SF-36, leaving 34 total items. Measure was z-transformed for use in multivariate analysis.

Ordinary least squares regression was used to assess the association between indicators of network position and health while controlling for the covariates shown in Table 1. Initial analysis determined that the indicators of residential tenure—both recent and long-term residency relative to the midrange category—were the strongest predictors of both dependent variables. Though it was uncorrelated with health,¹⁰ tenure in the community is clearly an important process for people’s position in the network. It takes some time for newcomers to be known by their peers and to learn the nuances of community life in a CCRC; long-term residents represent an older cohort of residents who have largely died out and been replaced by a more recent flow of community members. Those in the middle category are best positioned for low constraint and high integration.

In light of these considerations, I present the regression estimates in a sequence that highlights overall health’s effects (a) without controlling for tenure, (b) adjusting for tenure, and (c) interacted with tenure using multiplicative terms. F tests for difference in R² were used to assess improvement in model fit from the more basic to the more complete equation. For the interaction of health and tenure, the F tests revealed that including health × recent tenure was preferred to the noninteractive model, but health × long-term tenure did not improve the model for either dependent variable. Hence, the latter equations are not presented but are available on request.

Table 2 presents the regression results. Consistent with the HBP hypothesis, a standard deviation increase in overall health is associated with a .22 standard deviation decrease in network constraint (p < .05), net of basic covariates (Model A). Self-reported vision and hearing, however, were not associated with network constraint. Gender was the only other significant predictor in this model; on average, women had better bridging potential in the network than did men, and the effect size was identical to that of health.

Table 2.

Results of Regressing Network Position on Health.

	Constraint (ln)						Integration
	Model A		Model B		Model C		Model A		Model B		Model C
	b (SE)	β	b (SE)	β	b (SE)	β	b (SE)	β	b (SE)	β	b (SE)	β
Overall health	−0.108*	−.22	−0.089*	−.18	−0.027	−.05	0.051*	.24	0.043*	.20	0.012	.05
	(0.047)		(0.042)		(0.052)		(0.020)		(0.019)		(0.023)
Vision	0.029	.07	0.012	.03	0.009	.01	−0.021	−.11	−0.014	−.07	−0.012	−.06
	(0.044)		(0.040)		(0.039)		(0.019)		(0.018)		(0.017)
Hearing	−0.081	.17	−0.037	−.08	−0.049	−.05	0.037	.19	0.020	.11	0.026	.13
	(0.047)		(0.043)		(0.043)		(0.020)		(0.019)		(0.019)
Age	−0.007	−.06	−0.006	−.05	−0.006	−.06	0.001	.03	0.000	.01	0.001	.02
	(0.011)		(0.010)		(0.010)		(0.005)		(0.004)		(0.004)
Female	−0.244*	−.22	−0.258**	−.23	−0.257**	−.23	0.088	.18	0.091*	.19	0.091*	.19
	(0.105)		(0.094)		(0.093)		(0.045)		(0.042)		(0.041)
Partnered	−0.053	−.05	−0.008	−.01	0.001	.00	0.050	.11	0.034	.08	0.029	.07
	(0.097)		(0.087)		(0.086)		(0.042)		(0.039)		(0.038)
Local	−0.163	−.15	−0.089	−.08	−0.100	−.09	0.063	.13	0.033	.07	0.038	.08
	(0.098)		(0.089)		(0.088)		(0.043)		(0.040)		(0.039)
Recent resident			0.449**	.44	0.443**	.44			−0.178**	−.41	−0.175**	−.40
			(0.083)		(0.082)				(0.037)		(0.036)
Long-term resident			0.517*	.20	0.530*	.21			−0.169	−.16	−0.175	−.16
			(0.215)		(0.212)				(0.096)		(0.094)
Health × recent resident					−0.163*	.20					0.083*	.23
					(0.082)						(0.036)
Constant	−0.530		−0.981		−0.866		3.645		3.854		3.795
	(0.997)		(0.914)		(0.904)		(0.432)		(0.406)		(0.399)
R ²	.13		.32		.35		.12		.28		.31

N = 123.

p < .05. **p < .01.

Keeping with the topic of network constraint, Model B adjusts for tenure in the retirement community. Both recent and long-term residency are associated with higher levels of constraint, the former category predicting a nearly half-standard deviation increase in the dependent variable (β = .44, p < .001). Consistent with the strong influence of tenure, Model B’s R² value (.32) is significantly improved from the initial equation, F(2, 123) = 15.96, p < .01. Nevertheless, the standardized overall health coefficient decreases slightly in size from Model A but remains negative and statistically significant (β = –.18, p < .05).

The final step in the modeling sequence for network constraint was to investigate whether health interacted with tenure. Though not a crucial test for the HBP hypothesis, the interaction equation acknowledges the unique dynamics of a contained population in which people join at staggered times—it may be an instructive interaction to examine. Model C includes the term health × recent resident, as the addition of this interaction improves R² from .32 (Model B) to .35, F(1, 123) = 3.98, p < .05. As mentioned above, interactions using the long-term resident variable failed to significantly increase R². The health × recent resident term is negative and statistically significant, similar in effect size to the overall health coefficient in the prior to models (β = –.20). The overall health main effect term becomes nonsignificant in Model C.

A parallel modeling strategy was employed for the right half of Table 2, which focuses on network integration as the dependent variable. The overall health coefficient in Model A is positive and statistically significant (β = .24, p < .05). The effect of self-reported hearing does not reach the conventional .05 level of significance (β = .19, p = .075), and vision likewise fails to predict integration.

As before, model fit improves significantly with the inclusion of tenure (R² increase of .16 from Model A to Model B), though health remains a statistically significant predictor with a standardized coefficient of 0.2. Self-reported hearing is nonsignificant. Recent residency is again the most powerful predictor in the model and indicates less integration, whereas long-term residents do not appear to have lower integration (β = –.16, p = .08). Finally, Model C adds the interaction between health and recent residency, a positive and significant coefficient (β = .23, p < .05) whose inclusion significantly improves model fit from Model B, F(1, 123) = 5.22, p < .05. As with constraint, the association between overall health and integration was observed for relative newcomers to the community (close to half the sample) but was not found among other residents.

Discussion

This study offers a test of the hypothesis that health influences position in a social network. Specifically, it examined whether health was related to two network constructs that denote autonomy and access to diverse resources and information: network constraint and network integration.

The results support the essence of the HBP hypothesis, that good health may allow people to fulfill the demands and reap the rewards of connecting disparate groups in a local social structure (Cornwell 2009a, 2009b). The association between health and network position was documented by a broad index of overall health, though sensory function failed to manifest the same pattern. Retirement community residents who reported better overall health had (a) a lower proportion of their social ties concentrated among people who were also tied to one another and (b) easier and more direct access to any other given person in the network. Less constraint suggests that people can seize brokerage opportunities between various groups and have more freedom to move selectively between social partners (Burt 1992). More integration implies better access to “ideas, influences, or information” that flow through indirect social contacts (Granovetter 1973: 1370; Valente and Foreman 1998). Assuming that both constructs leverage a diffuse set of benefits, healthier seniors were advantaged in both respects. Though this is supportive of the HBP idea developed in earlier studies, caution should nevertheless be exercised when interpreting these findings; older adults in good overall health may have higher levels of integration and less constraint, but this does not demonstrate anything about whether (and how) they actively use this bridging potential.

One of the additional issues emerging from the analysis was the importance of tenure in the retirement community. In one sense this specific issue is orthogonal to the HBP hypothesis per se; on the other hand, the acknowledgment of temporality helps put the assumed processes into a more realistic context. The particularities of retirement community life dictate a fair degree of change over time as residents come and go—a reality manifested in these data by the total range of tenure (< 1 year to 21 years) as well as the heavy concentration of recent residents (43% with < 3 years residency). Tenure captures an important temporal dimension of network position, as it takes time for people to get to know others and find their place in the network’s topology. Indeed, both recent residents and (to a lesser degree) long-term residents were more constrained and less integrated than those with midrange tenancy; moreover, this factor exceeded even health in importance for predicting network position.

Making the tenure issue a complex one is that it draws attention to the steady turnover of residents in the bounded communities that are CCRCs; older adults eventually die or require more intensive levels of care, vacating their living space for new residents. These newcomers are then themselves adjusting to a community in flux, forming ties and finding their position within a Heraclitean structure. Moreover, as people’s alters exit the social structure, they may find themselves separated from parts of the network that they were formerly connected to through their former friends, or they may be thrust into an unexpected bridging position because their co-intermediaries have left. Both scenarios would operate outside of the direct control of ego and may not match his or her capabilities or inclinations. For instance, bridging unconnected alters can be a very demanding task that requires physical stamina from ego (Cornwell 2009a); some people may prefer a more embedded position that puts a lower onus on their own capacities.

The present study examined resident tenure to confront this complex temporal dynamic of the study context. Consistent with the HBP perspective, overall health was consequential for two aspects of network position when residential tenure was accounted for as a control variable in multivariate models. Nevertheless, the evidence suggests that health is far more consequential at the early stages of CCRC life. Specifically, the interaction terms between recent residency and health were statistically significant for both integration and constraint, and the inclusion of this term removed the main effects of health. This finding implies that health has an association with bridging potential in the early stages of adjustment to a CCRC, but it tapers in relevance over time. The principle of network endogeneity—the self-reinforcing processes characterizing a system of co-interacting actors—implies that one’s initial location in a network is especially consequential for their future position; for example, being highly constrained during the first several years of tenancy would limit people’s exposure to future arrivals. Nevertheless, specific inferences about the dynamics of this population remain speculative.

A desirable next step would be to observe longitudinal change in both health and network position and to identify whether physical decline is associated with increasing levels of constraint and decreasing integration. Thus far, other tests of the HBP hypothesis have similarly relied on cross-sectional data, and this limitation has been well noted in past studies (Cornwell 2009a, 2009b). Nevertheless, the tenure-related findings underscore that “more [longitudinal] research on the structural implications of health is badly needed” (Cornwell 2009a: 100). This study, like its predecessors, is limited by the current lack of follow-up data.

An aspect of the current study that distinguishes it from prior tests of the HBP hypothesis, however, is the use of directed data from a full network. Prior research on this topic has made good use of a national sample of older Americans, the NSHAP survey (Cornwell et al. 2009), but has been unable to go beyond ego-centered networks in the general population. An advantage of the current strategy is that we do not have to rely on people concocting their own sense of “the network” and inferring structural position from individuals’ idiosyncratic response patterns. Here, the network is a given—it is the set of relations among residents of a defined community. Networks based on self-reported social ties are by nature cognitivist constructions, as people have the ability exercise different criteria for defining social interaction with their peers (Krackhardt 1987). Prior scholars have warned about the inaccuracies inherent in such reports but demonstrate that bias is minimized when analysts utilize alters’ reports (Feld and Carter 2002). The fact that ego’s heath was related to network position based solely on alters’ accounts lends crucial support to the HBP hypothesis by subjecting it to a unique and rigorous test.

Of course, the uniqueness of the particular retirement community requires careful consideration for evaluating this study’s findings and its contribution to the growing literature on social networks, aging, and health. The fact that the observed network in this study comes from a single population—predominantly White, highly educated, and wealthy enough to afford CCRC living—should perhaps raise questions about applicability to more diverse communities, or at least those with a different demographic portfolio. Furthermore, social relations in this community are, by design, between older nonkin adults; therefore, the network-bridging potential or access to distal alters examined here is of a local variety and makes no assumptions about the broader world.¹¹ These considerations in mind, the particularity of the study setting can be seen in some ways as a strongpoint alongside its obvious weaknesses. The bounded nature of this community not only makes it a tractable site for observing basic social processes (Vaisey 2007) but also helps offers a tightly controlled test for whether the patterns inferred from the general population (i.e., the NSHAP) also occur in a micro context. Both approaches are potentially valuable, and consistency between extant research and the present findings attests to the viability of the HBP hypothesis.

Conclusion

This study’s results support the idea that health is important for network position. Given the recentness of interest in this hypothesis, it was important to extend initial ego-centric research with an alternate population and with full-network, directed measures. As with any other nascent and intriguing idea in social science, the HBP hypothesis merits scrutiny concerning its general applicability and scope conditions. Several of such issues were confronted in this study. For example, do the expected processes generalize to specific contexts, such as a self-contained community? Does health influence network position even when the social system, by necessity, restricts the possible composition of people’s networks? The fact that the HBP thesis holds under these circumstances helps support the generalizability of its claims. Of course, many questions cannot be answered in the current study and remain fruitful areas for future research—chief among them is the following: What are the actual mechanisms that explain the health–position association? Further investigation in this area should examine whether network position changes are primarily a function of ego’s own health-related choice and behavior, or whether the decisive agency resides with alters who approach, withdraw, or react otherwise in accordance with ego’s health. Likewise, future research should carefully consider multiple aspects of health in older age; the current study examined only an overall indicator of health status and two aspects of sensory function. Although vision and hearing failed to emerge as significant predictors in this study, past research suggests that cognitive impairment may be an important factor for bridging potential (Cornwell 2009a, 2009b). In light of these considerations, an important area of future research will be to clarify how these closely interconnected aspects of brain health and aging—sensory and cognitive functioning (Baltes and Lindenberger 1997)—are related to functioning in social networks.

In a broader context, the current study contributes to a burgeoning literature that acknowledges the importance of health for social relations. A long-standing tradition in social gerontology emphasizes the crucial nature of social interaction for prolonging life and enhancing health (Berkman et al. 2000). In reality, the association is likely recursive and the causality bidirectional; not recognizing health as an independent variable in this relationship risks overestimating how much it is influenced by social networks (Haas et al. 2010). Following this logic, recent studies suggest that better health enables people to maintain larger networks and interact more voluntarily with friends (Litwin and Shiovitz-Ezra 2011a; van Tilburg and van Groenou 2002). Network position, however, encompasses a more complex set of processes because it involves indirect ties and invokes a more concrete conception of social structure. Health, the evidence increasingly suggests, is quite consequential for this social dynamic of later life. As longitudinal data become available, both ego-centered studies from national samples and narrower community-focused studies can further clarify the causal mechanisms for health and network position.

Footnotes

Acknowledgements

I thank Scott Feld, Ken Ferraro, Karen Fingerman, and Jill Suitor for their guidance during an earlier stage of this research, and I appreciate the helpful critiques of the Research on Aging reviewers. Kirk Fatool, Ann Howell, and Sarah Poorman provided exceptional assistance with data collection and preparation.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Science Foundation (Dissertation Improvement Grant 1003772), the Purdue Research Foundation, and the Center on Aging and the Life Course at Purdue University.

Notes

Author Biography

Markus Schafer is an assistant professor of sociology at the University of Toronto. His research focuses on obesity, adult health problems associated with childhood misfortune, subjective aspects of aging, and the connection between health and social networks among older adults.

References

AARP. 2007. Continuing Care Retirement Communities. Retrieved from http://www.aarp.org/families/housing_choices/other_options/a2004-02-26retirementcommunity.html

Association for Homes and Services for the Aged. 2007. Retrieved from http://www.aahsa.org/aging_services/default.asp#1

Baltes

P. B.

Lindenberger

1997. “Emergence of a Powerful Connection Between Sensory and Cognitive Functions Across the Adult Life Span: A New Window to the Study of Cognitive Aging?” Psychology and Aging 12:12-21.

Bearman

P. S.

Parigi

2004. “Cloning Headless Frogs and Other Important Matters: Conversation Topics and Network Structure.” Social Forces 83:535-57.

Berkman

L. F.

Glass

Brissette

Seeman

T. E.

2000. “From Social Integration to Health: Durkheim in the New Millennium.” Social Science & Medicine 51:843-57.

Bernard

H. R.

Killworth

Kronenfeld

Sailer

1984. “The Problem of Informant Accuracy: The Validity of Retrospective Data.” Annual Review of Anthropology 13:495-517.

Borgatti

S. P.

Jones

Everett

M. G.

1998. “Network Measures of Social Capital.” Connections 21 (2): 27-36.

Burt

R. S.

1992. Structural Holes: The Social Structure of Competition. Cambridge, MA: Harvard University Press.

Burt

R. S.

2000. “The Network Structure of Social Capital.” Pp. 345-423 in Research in Organizational Behavior, edited by Sutton

R. I

Staw

B. M.

, vol. 22. Greenwich, CT: JAI.

10.

Buskens

van de Rijt

2008. “Dynamics of Networks If Everyone Strives for Structural Holes.” American Journal of Sociology 114:371-407.

11.

Cornwell

2009a. “Good Health and the Bridging of Structural Holes.” Social Networks 31:92-103.

12.

Cornwell

2009b. “Network Bridging Potential in Later Life: Life-Course Experiences and Social Network Position.” Journal of Aging and Health 21:129-54.

13.

Cornwell

Schumm

L. P.

Laumann

E. O.

2008. “The Social Connectedness of Older Adults: A National Profile.” American Sociological Review 73:185-203.

14.

Cornwell

Schumm

L. P.

Laumann

E. O.

Graber

2009. “Social Networks in the NSHAP Study: Rationale, Measurement, and Preliminary Findings.” Journals of Gerontology Series B: Psychological Sciences and Social Sciences 64B (Suppl. 1): i47-i55.

15.

Emerson

R. M.

1962. “Power-Dependence Relations.” American Sociological Review 27:31-41.

16.

Feld

S. L.

1981. “The Focused Organization of Social Ties.” American Journal of Sociology 86:1015-35.

17.

Feld

S. L.

Carter

W. C.

2002. “Detecting Measurement Bias in Respondent Reports of Personal Networks.” Social Networks 24:365-83.

18.

Granovetter

1973. “The Strength of Weak Ties.” American Journal of Sociology 78:1360-80.

19.

Haas

S. A.

Schaefer

D. R.

Kornienko

2010. “Health and the Structure of Adolescent Social Networks.” Journal of Health and Social Behavior 51:424-39.

20.

Herzog

A. R.

Kahn

R. L.

Morgan

J. N.

Jackson

J. S.

Antonucci

T. C.

1989. “Age Differences in Productive Activities.” Journal of Gerontology 44:S129-38.

21.

Kaufman

Uhlenberg

1998. “Effects of Life Course Transitions on the Quality of Relationships Between Adult Children and Their Parents.” Journal of Marriage and Family 60:924-38.

22.

Krackhardt

1987. “Cognitive Social Structures.” Social Networks 9:109-34.

23.

Litwin

Shiovitz-Ezra

2011a. “The Association of Background and Network Type Among Older Americans: Is ‘Who You Are’ Related to ‘Who You Are With’?” Research on Aging 33:735-59.

24.

Litwin

Shiovitz-Ezra

2011b. “Social Network Type and Subjective Well-Being in a National Sample of Older Americans.” The Gerontologist 51:379-88.

25.

Scanlon

Layton

B. D.

1997. Report to Congressional Requesters: How Continuous Care Retirement Communities Manage Services for the Elderly. Washington, DC: U.S. General Accounting Office.

26.

Shippee

T. P.

2008. “Social Cohesion in Communities With High Residential Mobility: The Paradox of Aging in Place.” PhD diss., Purdue University, West Lafayette, IN.

27.

Vaisey

2007. “Structure, Culture, and Community: The Search for Belonging in 50 Urban Communities.” American Sociological Review 72:851-73.

28.

Valente

T. W.

Foreman

R. K.

1998. “Integration and Radiality: Measuring the Extent of an Individual’s Connectedness and Reachability in a Network.” Social Networks 20:89-105.

29.

van Tilburg

van Groenou

M. B.

2002. “Network and Health Changes Among Older Dutch Adults.” Journal of Social Issues 58:697-713.

30.

Ware

J. E.

Snow

K. K.

Kosinski

Gandek

1993. SF-36^– Health Survey Manual and Interpretation Guide. Boston, MA: New England Medical Center, Health Institute.

31.

Wasserman

Faust

1994. Social Network Analysis: Methods and Applications. Cambridge, UK: Cambridge University Press.

32.

Williams

S. R.

Pham-Kanter

Leitsch

S. A.

2009. “Measures of Chronic Conditions and Diseases Associated With Aging in the National Social Life, Health, and Aging Project.” Journals of Gerontology Series B: Psychological Sciences and Social Sciences 64B (Suppl. 1): i67-i75.