Personal networks and mortality in later life: racial and ethnic differences

ABSTRACT

Background

Social network characteristics are associated with health outcomes in later life, including mortality. Moreover, there are well-established mortality disparities across race and ethnicity. Although previous studies have documented these associations separately, limited research considers the two in tandem. The present study addressed how the associations between social network characteristics and mortality differ across race and ethnicity in later life.

Methods

Data were from the National Social Life, Health, and Aging Project. At baseline, 3005 respondents were interviewed with regards to their health and social networks. Five years later, 430 respondents had died. Logistic regression models were used to estimate the odds of all-cause mortality over the study period.

Results

Network size and kin composition were negatively associated with mortality, whereas density was positively associated with mortality. There was a stronger negative association between the kin composition and mortality for Hispanic respondents compared with white and Black respondents.

Conclusion

The present study contributes to the large literature documenting the link between social networks and health by highlighting the importance of analyzing networks through a sociocultural lens.

mortality, older people, race

Introduction

Older adults often benefit from close social relationships through which they exchange support and derive a sense of belonging.¹ This recognition, coupled with the aging population of the USA, has resulted in a large body of research dedicating to exploring the intricacies of social networks in later life.^2–5 Personal social networks (i.e. webs of social relationships surrounding focal individuals) are a fundamental aspect of health, and have been associated with numerous primary risk factors such as cerebrovascular disease, cardiovascular disease and dementia.^2,6–10 Consequently, the types of networks that older adults maintain significantly influence their mortality risk.^7,11–13

Personal networks are hypothesized to influence mortality through multiple pathways. Network size plays an important role as individuals with larger networks have access to more social resources (e.g. emotional support, information sharing) compared with those with smaller networks.^12,13 Beyond network size, the structure and composition of personal networks are also linked to health and illness.^14,15 Network density (i.e. the degree of interconnectivity between individuals) highlights the extent to which network members can coordinate efforts to support a focal individual during a time of need.^16,17 Although tightly knit personal networks can promote a sense of belonging and buffer against undesirable stressors,^18,19 an alternative perspective highlights the advantages of being embedded within a personal network in which not all network members know each other.²⁰ Maintaining social ties to a loosely connected group of people offers health benefits in its own right, including access to novel information (e.g. alternative health care options)²¹ and providing a sense of independence.²² Frequent contact with family is also shown to influence health in later life.^23,24 Whereas friendships are likely to dissipate over time, familial relationships endure across the life course and serve as primary sources of emotional and instrumental support toward the end of life.^25,26 In addition to receiving support from family, older adults often provide support to various relatives, which has been shown to offer positive health benefits.²⁷ At the same time, family networks may accelerate health declines if there are ongoing family tensions that cause either stress or excessive demands.^28,29

Although empirical studies extensively document the link between social networks and mortality,^7,12,13,30 the majority of these studies ignore the fact that social networks look and operate differently depending on race/ethnicity.^31–34 Past research often assumes that social networks are equally protective for all older adults despite the fact that unique sociocultural values structure people’s ability to draw upon social resources.^33,35,36 In general, whites adopt an individualistic culture in which they view themselves as independent actors whose behaviors are not necessarily governed by group norms. Therefore, whites tend to maintain relatively loosely connected personal networks that incorporate a mix of family and friends such that each individual occupies their own social world.^33,37 Although family plays a significant role in their lives, older whites are more likely to be viewed as burdens to younger generations rather than upheld as pillars of the family.^38,39 Given their individualistic culture, older whites are expected to derive greater health benefits from loosely connected networks and fewer benefits from kin-centered networks compared with those from more collectivistic cultures. Hispanics, meanwhile, emphasize a strong commitment to family, especially older generations.^40,41 The concept of familismo, which refers to intergenerational solidarity and provision of care to aging family members, has been posited as a potential cause of the epidemiologic paradox (i.e. mortality rates are lower for Hispanics than are expected given their socioeconomic standing).^42–44 Maintaining dense networks that consist primarily of kin is therefore expected to offer older Hispanics with significant protection against mortality. Blacks tend to adhere to a similar collectivistic culture in which self-identity is interwoven with a larger social group.^45,46 Although blacks often report having relatively small personal networks, their cultural obligation toward family causes their networks to be dense and kin-centered.^34,37,47,48 Yet unlike Hispanics and whites, blacks often report higher incidences of ‘network stress’ (i.e. stress related to the life events of others) as they are more likely to be connected to socially disadvantaged network members.^49,50 Consequently, the potential network stressors that accompany family ties may offset the protective advantages of kin-centered networks for older blacks.

Collectively, understanding the multidimensionality of social networks is important because the health benefits and detriments of networks are dependent on who is in them and how they are interconnected.^11,14 Despite widespread emphasis on the value of social networks, there remains a paucity of research examining whether network exert similar health effects across different sociocultural groups. This study addresses this gap by assessing whether race/ethnicity moderates the association between social network characteristics and all-cause mortality in later life.

Methods

The National Social Life, Health, and Aging Project (NSHAP) is a nationally representative panel study of older Americans. Starting in 2005–06, the NSHAP used a multistage probability sampling design to survey 3005 non-institutionalized older adults between the ages of 57 and 85.⁵¹ Although the NSHAP did not explicitly incorporate race/ethnicity into their sampling design, they managed to achieve an oversample of black and Hispanic respondents. Five years after the initial wave, respondents were recontacted to be participate in a second wave. In total, 2261 of the original 3005 respondents were interviewed in 2010–11. Of the 744 nonparticipating respondents, 430 were confirmed dead at the time of contact, 304 were alive and 10 had an unconfirmed status. Because mortality status is the only variable to be used from the follow-up wave, all original respondents are retained in the analyses excluding those missing key variables (i.e. mortality status, social network data, race/ethnicity).

Measures

The primary outcome variable was ‘mortality status’ at the second wave (0 = alive, 1 = deceased). Mortality status was confirmed through either a proxy interview with a family member/friend or a public record search. The cause of death was not asked due to the sensitivity of the topic. Therefore, the analyses use all-cause mortality as the sole outcome.

Social networks were assessed using the core discussion network name generator.⁵² Respondents were asked to provide the names of up to five people with whom they discussed important matters during the past 6 months. Hence, network size ranged from 0 (no names provided) to 5. After delineating the networks, NSHAP field interviewers asked further questions about each network member, including how often each member spoke with the other network members (‘every day’, ‘several times a week’, ‘once a week’, ‘once every 2 weeks’, ‘once a month’, ‘a couple times a year’, ‘once a year’, ‘less than once a year’, ‘have never spoken’). By dichotomizing ties between network members (tie = spoke at least once a year), ‘network density’ was calculated as the proportion of total possible ties between network members. Values ranged between 0.0 (no network members were in contact with each other) to 1.0 (all members were in contact with each other). Respondents were also asked about their relationship with each network member (e.g. spouse, partner, child, sibling, in-law, friend, neighbor, minister, doctor, etc.). These items were used to calculate the ‘kin composition’ of each respondent’s network. Values ranged between 0.0 (no kin named in the network) to 1.0 (only kin named in the network). One shortcoming of these latter two measures is that they excluded respondents who failed to name at least two network members. To address this limitation, respondents with zero network members were assigned density and kin composition values of 0. Respondents with one network member were assigned a density value of 0 and a kin composition of 0 or 1 depending on their relationship with the network member. Figure 1 illustrates an example social network.

Fig. 1

Example of a social network. The white node represents the respondent and black nodes represent network members. Density = 0.6 (6 ties between network members divided by 10 possible ties).

Demographic variables for respondents included race/ethnicity (white, black, Hispanic), age (in years), education (less than high school [HS], HS, some college, college), marital status (married, divorced/separated, widowed, never married), activities of daily living (ADL)⁵³ and depressive symptoms (via a modified version of the Center for Epidemiologic Studies Depression Scale [CESD] Scale).⁵⁴ ADL, which represent a respondent’s functional health, were measured by averaging seven questions assessing the degree of difficulty that respondents have performing daily tasks (α = 0.85). Depressive symptoms were measured by averaging 10 questions that assess respondent’s affect (α = 0.80).

Analytic approach

The mean and proportional differences between racial and ethnic groups for demographic characteristics, social networks and mortality were evaluated using F-tests for continuous variables and χ² tests for categorical variables. Logistic regression models were used to assess the odds of all-cause mortality after the 5-year follow-up. Model 1 assesses the odds of mortality for network size, kin composition and network density as well as the odds of mortality for blacks and Hispanics in comparison to whites. This model controls for age, sex, education, marital status and health. The next two models assess whether the associations between social network characteristics and mortality are moderated by race/ethnicity. Model 2 tests for an interaction between network density and race/ethnicity. Model 3 tests for an interaction between kin composition and race/ethnicity. To help visualize the statistically significant relationships, predicted probabilities are provided. The delta method is used to provide accurate group comparison using these probabilities.⁵⁵ Associations were considered statistically significant at α = 0.05.

In terms of exclusion criteria, 30 respondents were excluded due to missing data on W2 mortality status (n = 10), social networks (n = 17) or race/ethnicity (n = 3). Because this study focuses on whites, blacks and Hispanics, 73 respondents who reported their race/ethnicity as ‘other’ were also excluded. After excluding these cases, the analytic sample size was 2902.

In order to account for potential selection bias, author employed inverse probability treatment weighting.⁵⁶ A logistic regression model was used to predict whether a respondent was included in the analytic sample using sociodemographic and health variables as predictors. The inverse probability from this model was multiplied by the NSHAP-supplied survey weights to adjust for the selection bias and complex multistage sampling design. All statistical analyses were performed using STATA 15.⁵⁷

Results

Summary statistics are presented in Table 1. The sample was majority white (72%) and female (52%). The mean age was 69.3 years (standard deviation [SD] = 7.9). White respondents reported the largest networks (⁠|$\overline{\mathrm{x}}$| = 3.42 members) of all three groups. Not only did Hispanic respondents report ties to the few number of network members (⁠|$\overline{\mathrm{x}}$| = 2.86) but also their networks were the most interconnected (density = 0.70) and were composed of the highest proportion of kin (0.74).

Table 1

Sample descriptive statistics

	All (n = 2902)		White (n = 2101)		Black (n = 503)		Hispanic (n = 298)		F/χ2
Deceased	0.14		0.14		0.16		0.11		3.74
Network characteristics
Size (0–5)	3.42	(1.47)	3.57	(1.42)	3.12	(1.56)	2.86	(1.47)	11.80***
Density (0–1)	0.67	(0.38)	0.67	(0.36)	0.62	(0.41)	0.70	(0.41)	3.14
Kin composition (0–1)	0.66	(0.34)	0.66	(0.34)	0.64	(0.37)	0.74	(0.36)	9.66***
Race/ethnicity
White	0.72		1.00
Black	0.17				1.00
Hispanic	0.10						1.00
Educational attainment									414.88***
Less than HS	0.23		0.14		0.41		0.60
HS	0.27		0.29		0.23		0.13
Some college	0.28		0.31		0.23		0.17
College	0.22		0.26		0.13		0.10
Marital status									102.65***
Married	0.62		0.66		0.44		0.66
Divorced/separated	0.12		0.10		0.22		0.14
Widowed	0.22		0.21		0.28		0.17
Never married	0.04		0.03		0.06		0.03
Age	69.34	(7.86)	69.67	(8.02)	68.91	(7.31)	67.80	(7.42)	6.69***
Female	0.52		0.51		0.56		0.49		4.35
Health
CESD	1.91	(0.37)	1.90	(0.35)	1.97	(0.40)	1.94	(0.44)	1.98
ADL	0.20	(0.40)	0.18	(0.36)	0.26	(0.46)	0.24	(0.47)	2.67

Mean/proportions are shown (SD in parentheses).

*P < 0.05, **P < 0.01, ***P < 0.001.

Data: NSHAP.

Table 2 presents the odds ratios (ORs) from the logistic regression models. Model 1 shows the ORs for the network characteristics and race/ethnicity, controlling for all other demographic covariates. Network size (OR = 0.89, confidence interval [CI] = 0.81–0.98) and kin composition (OR = 0.60, CI = 0.39–0.92) were each negatively associated with mortality such that respondents with more network members as well as those with higher proportions of kin had lower odds of death. Density was positively associated with mortality (OR = 1.35, CI = 1.01–1.82) indicating that respondents who had tightly knit networks had higher odds of mortality than respondents with loosely connected networks. Hispanics had lower odds of mortality compared with whites, controlling for covariates (OR = 0.39, CI = 0.20–0.77).

Table 2

Logistic regression estimates of associations between network characteristics and all-cause mortality

	Model 1		Model 2		Model 3
Network characteristics
Size	0.89*	(0.81–0.98)	0.89*	(0.81–0.98)	0.90*	(0.81–0.99)
Density	1.35*	(1.01–1.82)	1.30	(0.93–1.84)	1.36*	(1.01–1.84)
Kin composition	0.60*	(0.39–0.92)	0.60*	(0.39–0.92)	0.69	(0.41–1.15)
Race/ethnicity (ref: white)
Black	0.78	(0.54–1.13)	0.62	(0.38–1.00)	0.51	(0.25–1.04)
Hispanic	0.39**	(0.20–0.77)	0.47	(0.16–1.37)	1.49	(0.71–3.11)
Age (by decade)	1.85***	(1.52–2.24)	1.85***	(1.52–2.24)	1.86***	(1.53–2.25)
Female	0.61**	(0.43–0.86)	0.61**	(0.43–0.86)	0.59**	(0.42–0.84)
Education (ref: <HS)
HS	0.91	(0.62–1.36)	0.92	(0.61–1.36)	0.89	(0.61–1.34)
Some college	0.66	(0.42–1.04)	0.67	(0.42–1.05)	0.68	(0.42–1.09)
College	0.48***	(0.32–0.73)	0.48***	(0.32–0.73)	0.49**	(0.32–0.75)
Marital status (ref: married)
Divorced/separated	1.57*	(1.05–2.36)	1.57*	(1.04–2.37)	1.49	(0.97–2.29)
Widowed	1.44	(0.95–2.20)	1.44	(0.95–2.18)	1.49	(0.99–2.23)
Never married	1.85	(0.81–4.24)	1.81	(0.80–4.12)	1.76	(0.79–3.94)
ADL	2.58***	(1.84–3.60)	2.60***	(1.84–3.68)	2.77***	(1.96–3.95)
CESD	1.33	(0.97–1.82)	1.33	(0.97–1.82)	1.35*	(1.01–1.84)
Interactions
Black × density			1.43	(0.73–2.82)
Hispanic × density			0.80	(0.33–1.89)
Black × kin composition					1.93	(0.65–4.50)
Hispanic × kin composition					0.14***	(0.04–0.26)
F statistic	14.95***		12.76***		11.54***
N	2902		2902		2902

ORs are displayed (95% CIs in parentheses).

*P < 0.05, **P < 0.01, ***P < 0.001.

Data: NSHAP.

Models 2 and 3 introduce interactions terms between race/ethnicity and network characteristics. As seen in Model 2, the association between network density and mortality was not moderated by race/ethnicity. In Model 3, the kin interaction terms were significant for Hispanic (OR = 0.14, CI = 0.04–0.26). To ease interpretation of these interactions, Figure 2 plots the predicted probabilities of mortality from Model 3. As seen in the figure, Hispanic respondents’ probability of death significantly reduced as their networks consisted of greater proportions of kin. Hispanic respondents had a 0.04 probability of death when their networks were composed solely of kin. These respondents were significantly less likely to die compared with White respondents (Δ probability = 0.10, P < 0.01) and Black respondents (Δ probability = 0.11, P < 0.01) who also reported ties only to kin. Yet Hispanic respondents who reported zero ties to kin had a 0.25 probability of death. These kinless Hispanic respondents had a greater probability of death compared with kinless Black respondents (Δ probability = 0.13, P < 0.05) but not compared with kinless White respondents (Δ probability = 0.06, P = 0.29).

Fig. 2

Probability of all-cause mortality. Note: Probabilities are derived using average marginal effects from Table 2, Model 3.

Black respondents did not significantly differ from White respondents in their probability of death when their networks consisted entirely of kin (Δ probability = 0.002, P = 0.93) nor when their networks were devoid of kin (Δ probability = 0.07, P = 0.06). It is important to recognize that although there appears to be a positive trend line for blacks in Figure 2, there is no statistically significant difference between the probability for Black respondents whose networks consisted purely of kin and Black respondents whose networks were absence of kin (Δ probability = 0.10, P = 0.07). The seemingly negative trend line for whites in Figure 2 is also not significant as indicated by the main association for kin composition in Model 3 (OR = 0.69, CI = 0.41–1.15).

Discussion

Main finding of this study

Among a nationally representative sample of older Americans, race/ethnicity moderated the association between social networks and mortality. Network size and kin composition were negatively associated with mortality, whereas density was positively associated with mortality. Having a large proportion of kin in one’s core discussion network was more protective for Hispanics than it was for whites or blacks.

What is already known on this topic

A long line of empirical research has found that social interaction protects against negative health outcomes, including mortality. While early studies suggested that broad measures of social isolation were associated with elevated risk of death,¹³ recent research highlights the nuanced nature of social networks.¹⁴ Beyond the size of one’s social network, other network properties have proved important, including density and composition.^16,58 A separate line of research explores how individuals of different races and ethnicities form and maintain social networks.^34,59,60 Yet, minimal research has considered how social networks may influence health differently across race and ethnicity.

What this study adds

This study leverages social network data from a large nationally representative sample that allows for comparisons across racial and ethnicity groups. Similar to previous findings,¹² there was a negative association between network size and mortality. Network density was separately associated with mortality, albeit in the opposite direction. Respondents with loosely-connected core discussion networks exhibited lower odds of mortality across the five-year study period compared to respondents with tightly-knit networks. This finding suggests that bridging separate social spheres may offer health benefits via non-redundant information sharing and social independence.^21,22 Although older adults from collectivistic cultures (i.e. Hispanics, blacks) may theoretically benefit from occupying dense networks as opposed to loosely connected networks, there was no detectable interaction between race/ethnicity and network density.

The main finding to emerge was the importance of family within respondents’ core discussion networks. Family has been theorized to play a central role in the lives of older adults. Family members such as spouses, adult children and grandchildren are among the primary sources of emotional and instrumental support in later life.^1,23,27 Moreover, previous research finds that older adults are equally as likely to provide support to family members as they are to receive it.^27,61 Yet the influence of family on mortality appears to differ by race/ethnicity. White respondents, who tend to have more socially diverse networks than minority races,^33,34 exhibited no significant association between kin composition and mortality. Hispanic respondents, however, demonstrated a strong link between kinship and mortality. The latter finding reflects the collectivistic nature of familismo in Hispanic culture. Not only are older Hispanics likely to receive family support in the event of a health decline but also their high status within the family hierarchy may instill them with a sense of esteem that older adults from individualistic cultures lack.^41,43,44

Similar to their white peers, black older adults fared no better against the threat of death when their networks consisted solely of kin. Although blacks tend to adhere to a more collectivistic culture than whites, they are more likely to be exposed to multiple network stressors.⁴⁹ Due to their relative socioeconomic disadvantage, blacks tend to experience more negative life events (e.g. racial discrimination, incarceration, death of a loved one) compared with other racial and ethnic groups in America.^50,62–64 The stress caused by these life events has been shown not only to be detrimental to the health of those experiencing the event but also to their family members.^49,50 It is important to realize, however, that there was neither a positive nor negative association between kin composition and mortality for Black respondents in the present study.

Limitations of this study

Although the NSHAP offers impressive coverage of the personal social networks of older adults, time constraints prevented the collection of data on the race/ethnicity of their network members. Given the primary focus on race and ethnicity, it would have been useful to know if respondents were discussing important matters mainly with intragroup network members or if they were interacting with a racially and ethnically diverse network. Additionally, the name generating prompt that was used to elicit the social network truncated the number of network members that a respondent could name. This was done in the interest of saving time, although it could have artificially lowered certain respondents’ network size. Finally, data were not explicitly collected on negative network ties. Although the health literature predominantly emphasizes the positive implications of social networks, the presence of stressful or burdensome network members has been shown to have detrimental repercussions.^49,65

Conclusion

This study highlights the importance of social network characteristics, particularly kin composition, associated with mortality among older adults of different races and ethnicities. Although the specific mechanisms for these findings remain unclear, the divergent associations between kinship and mortality have important policy implications. For instance, Hispanics may benefit from welfare policies that presume that families will care for aging family members, whereas such policies may be less suitable for older whites and blacks who do not appear to derive equal health benefits from their families. Future studies aimed at assessing the nuanced associations between social networks and health outcomes across racial/ethnic groups are warranted to identify and improve existing population health disparities.

Adam R. Roth, Postdoctoral Fellow

Acknowledgements

The author would like to thank the two anonymous reviewers for their valuable feedback and the Inter-University Consortium for Political and Social Research for use of the data.

Funding

None.

Conflict of interest

None.