A Bridge Too Far? Social Network Structure as a Determinant of Depression in Later Life

Abstract

Existing research has documented various determinants of mental health related to individuals’ social connections, but less is known about the role of the structural features of interpersonal networks. This is especially true in the case of bridging, which refers to ties to people who are otherwise disconnected from each other. By intersecting theories of social networks and gerontology, this study employs within- and between-person analysis with data from the National Social Life, Health, and Aging Project (NSHAP) to examine the association between social network bridging and depression in later life. The study finds that bridging, particularly between kin and non-kin members in the network, is associated with increased depressive symptoms in later life. This association is contingent on social support and strain respondents experienced, and it exhibits variations within individuals over time, especially among older adults in the youngest age cohort (57–64 years old included in NSHAP in 2005). In closing, the paper discusses the extent to which heterogeneous network structures may be one mechanism that shapes mental health trajectories in the context of later life-course experiences.

Introduction

Depressive symptomatology is an important indicator of older adults’ general health, potentially leading to mental disorders, increased morbidity, and even mortality (Pinquart and Duberstein 2010). While depressive symptoms are linked to social determinants, such as socioeconomic disadvantage and adverse life events, research also suggests that social relationships profoundly affect these symptoms (e.g., Nakagomi et al. 2023). This is especially true with respect to older adults’ relationships with their family and friends, which are intricately linked to social processes shaping mental well-being (for a review, see Umberson, Crosnoe, and Reczek 2010). Current research in this area often explores how the quality of older adults’ dyadic relationships influences their mental health, emphasizing the protective effect of positive kin and friend relationships, and the detrimental effect of ambivalent or negative interactions (Song et al. 2021).

Research also demonstrates that the structure of one’s network influences mental health, including network size, composition, contact frequency, and closeness with network members (Thoits 2011; Tsai and Papachristos 2015). While previous studies have identified certain network characteristics that promote mental well-being (e.g., diverse network, Fiori, Antonucci, and Cortina 2006), less is known about whether and how a lack of connections among one’s network members affects mental health status. Theoretical and empirical work suggests that this structural characteristic may become particularly crucial in later life because a network’s influence on mental health emerges from not only one’s direct ties to other network members but also the connections among network members across social domains (Coleman 2018; Ellwardt et al. 2020).

Bridging is one measure of social network structure that denotes a lack of connectivity among network members. A focal individual (i.e., ego) is said to occupy a bridging position or serve as a bridge when two or more network members (i.e., alters) have direct ties to the ego but are otherwise not connected to each other (Burt 1992, 2017). In this study, I pay special attention to whether bridging unconnected kin (e.g., by blood or marriage) and non-kin (e.g., friends) may be associated with higher depressive symptoms in later life, given that kin and non-kin ties represent distinct sources of resource, strain, and caretaking obligations. Existing research highlights social support, regulation, and strain as crucial mechanisms in linking interpersonal relationships to mental health (Nakagomi et al. 2023; Thoits 2011). Accordingly, bridging may also affect depression risks through these mechanisms. For example, it may obstruct access to coordinated social support that would otherwise be facilitated by a tight-knit network. Bridging separate social domains like kin and non-kin may also involve juggling competing norms and demands, requiring one’s time and efforts to maintain these separate ties (Santini et al. 2015). Particularly for older adults, increasing bridging prospects indicates a decline in solidarity and mutual proximity, deviating from the expectations of a well-integrated support structure (Agllias 2011). These processes over time may also increase social strain among older adults, which leads to depressive symptoms.

Another point to consider is the later-life contexts in which older adults’ social networks are structured according to their life stages or cohort to influence mental health. Although older adults’ social networks are generally seen as relatively steady, those in early later life stages or younger cohorts at the time of interview may shift into bridging roles later on due to turbulent events during later-life transitions, signaling an erosion of cohesion in the network that exerts mental health consequences (Cornwell 2009; Ertel, Glymour, and Berkman 2009). Thus, it is essential to distinguish between within-person changes and between-person comparisons (Hoffman and Stawski 2009), because while individuals occupying more bridging than others may tend more to feel depressed, their depressive symptoms might be exacerbated during times when they bridge more disconnected members or are less embedded in their networks than usual This study conducts within-between analysis to investigate the association between network bridging and depression in later life, using panel data from the National Social Life, Health, and Aging Project (NSHAP). This study also explores how this association varies by age or cohorts and whether it pertains specifically to bridging kin and non-kin members.

Background

Linking Social Network Structure to Mental Health in Later Life

The benefits and harms of social relationships for mental health in later life have been extensively documented with respect to the psychosocial and psychological pathways (Santini et al. 2015; Umberson, Crosnoe, and Reczek 2010). These pathways often involve processes of social support, social strain, and social regulation. Regarding social support, increasing evidence from the stress-buffering literature suggests that social networks protect against psychological distress by providing practical aid and emotional support, especially during stressful times (Cohen and Wills 1985; Thoits 2011). Conversely, social strain from demanding social ties, ambivalent interactions, and asymmetric exchange are associated with anxiety and stress that exacerbates depressive symptoms (Offer 2020; Rook 2015). This body of research underscores that individuals’ psychological well-being depends on the quality of their support system, including core family members (Offer 2020; Perry and Pescosolido 2012), and informal networks such as friendship (Ang 2018; Huxhold, Miche, and Schüz 2014). Still, less is known regarding how the broader configuration of kin and non-kin ties within the network may affect the determinants of depression. One step further, we know little about whether and how interconnections among individuals’ close members across social domains may change to yield implications for their mental health. This structural aspect is of particular importance to older adults, where ties that coordinate to provide care and monitor health can impact their health behaviors and management, ultimately determining their depression levels.

This study focuses on the concept of bridging as a critical aspect of social network structure, which has been discussed extensively in its social and economic benefits. Occupying the space between two otherwise unconnected contacts (structural holes) provides greater access to novel information and resources, which older adults often require to identify alternative care providers and make informed health decisions (Goldman and Cornwell 2015). Activities around bridging, such as controlling the flow of medical resources and strategically playing two disconnected parties against each other, in turn, enhance the bargaining power of older adults in social exchange (Burt 2017; Gould 1989). These activities additionally relax strict enforcement of norms and provide older adults with power, privacy, and autonomy by loosening their dependence on their close members in the network. The absence of such autonomy may result in emotional burdens, especially among older care recipients who may face increasing reliance on others (Krause 2007).

On the contrary, bridging may pose risks, especially regarding mental well-being. Some studies demonstrate that being embedded in a tight-knit network offers better access to some health benefits that depend on the presence of closure among one’s closest network members (Coleman 2018; Hurlbert, Haines, and Beggs 2000). For instance, older adults who bridge unconnected network members may be more susceptible to experiencing abuse and mistreatment than those with more closure in their network (Schafer and Koltai 2015). In this case, bridging restricts the alters’ capacity to share timely information and execute sanctions against harmful events. More generally, being embedded in a network with more connectivity benefits more from relationships based on trust and mutual obligations and perhaps more cohesive social environments, which are vital for exchanging social support, enforcing social regulation, and reducing social frictions. To this end, several reasons suggest that bridging sparse, non-overlapping relationships may be associated with higher depressive symptoms. This study aims to differ from the previous literature on the economic viability of bridging by highlighting deficits and costs in ties to disconnected network members, given its importance in many mechanisms linking social relationships to mental health.

Social Support

Older adults often prefer strong bonds with close kin, characterized by more emotional and material support exchanges over non-kin ties (Carstensen 1992). These sources of support are most readily accessible when family members are interconnected with each other. For instance, when children and relatives live in proximity, they often coordinate to share responsibilities, such as managing finances and ensuring the quality care of aging family members (Sicotte et al. 2008). On the other hand, increasing evidence suggests that older adults with family support alone experience more depressive symptoms than their peers who benefit from an integrated support structure including both family and non-kin ties, such as friends and neighbors (Fiori et al. 2006; Huxhold et al. 2014). These voluntary non-kin ties not only complement family support in care coordination but also alleviate pressure associated with obligatory kin support, providing them with a wealth of information and company that re-affirms their independence and self-worth (Krause 2007; Perry and Pescosolido 2012).

In contrast, bridging kin and non-kin may compromise the effectiveness of these resources. First, older adults who are not well enough or lack the incentives to bridge, often characterized by an “entrepreneurial spirit” trait (Burt, Reagans, and Volvovsky 2021), may feel incompetent and stressed when coordinating care and triangulating information across different domains, especially during pressing periods. Family members are often emotionally invested and obligated to the person’s recovery even when they are unfamiliar with the specific stressors, so they may provide well-intentioned but less effective advice compared to non-kin networks with similar experiences (Grace 2021; Thoits 2011). This potential incongruence in informational and emotional support between family and non-family members may leave older adults feeling unsupported and unaccepted within either domain (Myroniuk and Anglewicz 2015; Sapin, Widmer, and Iglesias 2016), potentially elevating depressive symptoms.

Social Regulation

Bridging kin and non-kin may also hinder access to available health benefits embedded within an interconnected network structure. Drawing on social control theory, the connectedness among kin and non-kin alters fosters their communication regarding ego’s health, thereby enhancing the collective ability to enforce social norms or sanction deviations from certain health practices (Coleman 2018). This mechanism is not limited to physical health but is just as consequential for mental well-being. Risk health behaviors, such as binge drinking, are often precursors to psychological distress (Ang 2021). Thus, the success of network alters’ joint efforts in monitoring the ego’s behaviors and mood swings could play a critical role in alleviating depressive symptoms. In particular, when both close family and professionals engage in health communication, individuals’ psychological symptoms are more likely to be interpreted as serious enough to necessitate professional treatment and press them to seek help (Thoits 2011). Individuals whose kin and non-kin members are detached from each other, in contrast, may experience ineffective norm regulation and competing sources of pressure, potentially leading to feelings of depression (Bearman and Moody 2004; Falci and McNeely 2009). Therefore, bridging disparate social relationships may undermine the collective influence of network members on individuals’ mental well-being (Schafer and Koltai 2015).

Social Strain

Lastly, bridging implies dissonant relationships or a lack of embeddedness in cohesive social groups, which may cause strain and stress. According to the social balance theory, the absence of ties among one’s closest network members can lead to psychological distress for the ego. People generally prefer their close contacts to be interconnected, as this closed or balanced network structure minimizes ambivalence and disturbing sentiments within the network (Chiang et al. 2020; Heider 1982). In the family context, bridging kin could be stress-inducing as it reflects the conflict or estrangement among family members, opposing the general desire for intergenerational solidarity and obligatory norms surrounding family connectedness (Connidis and McMullin 2002; Goldman 2016). Bridging poorly connected kin and non-kin is less normviolating than bridging kin but can still produce role strain that yields psychological problems. Individuals’ connections to kin and non-kin entail various role relationships (e.g., parent-child, friend-friend) attached to distinctive behavioral norms, scripts, and obligations (Thoits 2011), potentially pulling the focal individual in conflicting directions. Especially when the unconditional presence of family members becomes a source of conflict, it may compel individuals to seek connections in a separate domain, potentially causing stress and strain. Besides, assuming multiple roles across kin and non-kin domains could cause psychological distress due to stress that arises from incompatible role commitments or strain caused by accumulative role obligations (Adelmann 1994; Myroniuk and Anglewicz 2015). In this sense, the psychological distress resulted from contrast between kin and non-kin settings may be more salient than the differences observed in various non-kin relationships.

Figure 1 illustrates three possible types of bridging including individuals’ (ego’s) kin and non-kin alters. Panel A shows an individual who bridges kin. Panel B illustrates an individual bridging kin and non-kin pairs, and Panel C indicates a bridging position among a pair of unconnected non-kin. Older adults’ confidant networks may include any combination of these three bridging scenarios, from maximum bridging in a radial network to no bridging potential at all in a completely closed network. My primary hypothesis is that, first, bridging unconnected network members is linked to higher depressive symptoms through its associations with social support, regulation, and strain processes that link social relationships to mental health. Furthermore, this association may be specific to bridging kin and non-kin (Panel B), as the increased risk of depression is more likely to arise from the subjection to competing norms, expectations, and support systems in distinct kin and non-kin domains.

Figure 1.

Three Hypothetical Confidant Networks Illustrating Three Possible (Non)Kin-related Network Bridging

Network Structure in Later Life

Previous research illuminates how social networks are experienced across ages, later-life experiences, or generational factors, providing contexts to understand how network structures and mental health evolve in later life. From a life-course perspective, older individuals experience less emotional perturbation from solitude since they tend to have denser and emotionally closer social networks than their younger counterparts (Carstensen 1992). Such age-related differences over time are usually explained in terms of later life-course transitions (Waite et al. 2021). As younger adults assume increasing bridging roles around organizing family gatherings, coordinating healthcare, and committing to community services following major life events such as family members’ health decline, psychological distress may occur when a role domain is strained or conflicted (Cornwell 2009; Frazier and Brown 2023; Wrzus et al. 2013). In comparison, those who are older (and not working), reminiscent of Rosow’s “role-less role” hypothesis about the absence of institutionalized obligations and expectations in later life stages, may participate in fewer activities (Flood and Moen 2015). Therefore, declining strained experiences with age may be symptomatic of decreased engagement in diverse contexts.

Moreover, social strain may emerge from broader life-course experiences that are antecedents of a person’s kin-non-kin bridging positions. Mental problems may emanate from distress associated with an endogenous network process following one’s illness, where kin ties provide a safety net while non-kin connections disintegrate or reshuffle owing to reciprocity violations (Perry and Pescosolido 2012; Qu 2023). What’s more, depression is often linked to family reconfigurations such as bereavement, particularly widowhood, one of the most trying experiences in later life that enhances subsequent preferences for kin relations while creating opportunities for bridging one’s in-laws and friends for the surviving (Cornwell 2009; Roth 2020). Additionally, bridging kin and non-kin may result from challenging life events at early stages of later life, such as retirement, which involves the loss of certain institutional connections (e.g., colleagues in the workplace) that are largely separate from one’s close family domains. There is evidence that retirees adapt to this change by establishing new roles and increasing commitment to family and the community, such as volunteering or caring for others (Roth 2020; Wrzus et al. 2013), but they may become anchored in distress as a result of coordinating tasks from various domains, partly due to high demand of cognitive awareness when navigating varying social expectations, interaction styles, and behavioral scripts (Giddens 1984). For these reasons, bridging kin and non-kin may increase over life-course experiences and influence depression risks by adding interpersonal strain that is commonly associated with turbulent life transitions during early later-life stages.

Alternatively, these purported age or life-course variations may reflect generational differences in social network structure that relate to mental health. Existing research indicates that younger generations, such as the Baby Boomers (born 1946-1964), are less connected to the community through formal activities compared to older generations (e.g., the Silent Generation, born 1920-1947), which diverges their civicmindedness and emotional well-being in later life (Waite et al. 2021). Although there is less evidence for cohort differences within the older generations, a few studies have pointed out that the oldest people included in NSHAP in 2005 (75+ years old) had higher marital stability and stronger family solidarity compared to the youngest in 2005 (57-65 years old) who came of age at the beginning of marriage revolution and experienced the dramatic increase in divorce and re-partnering. This cohort also had increasingly heterogeneous marriage and family structures and intimate unions that did older cohorts (Laumann et al. 2000; Raley 2000; Sassler 2010; Waite et al. 2021). These generational differences in relationship structures might yield differential bridging prospects that partly account for the worse mental well-being among the younger cohorts over time.

For these reasons, I hypothesize that older adults in the youngest cohorts who occupy bridging positions, particularly among kin and non-kin, are more likely to exhibit depressive symptoms than their older counterparts over time, partly because of more strain associated with having poorly connected kin and non-kin during stressful later life-course transitions. Note that the primary goal here is not to provide holistic results regarding age or cohort disparities in the effect of social networks on depression. Rather, the present study situates the link between different bridging circumstances under the later life-course contexts and age cohort to gain a better understanding of how social relationships are structured to affect mental health among older adults.

Data and Methods

The study uses data from the National Social Life, Health, and Aging Project (NSHAP), a nationally representative study of community-dwelling older adults interviewed in 2005–06, 2010–11, and 2015–16. In 2005-06, the NSHAP team began by surveying 3,005 noninstitutionalized older Americans (born between 1920 and 1947 and aged 57–85) belonging to the Silent Generation on their health and social lives using a multistage area probability design (Waite et al. 2021). Surviving respondents were re-interviewed in 2010–11 (n = 2,261) and again in 2015–16 (n = 1,561).

To harness the NSHAP data’s longitudinal aspect for within-person estimates, I prepared data in long form and only included respondents who participated in all three rounds of the survey. I further excluded respondents with fewer than two confidants and zero kin in the network, which led to missing data on kin-related network bridging measures (n = 337 person-years). I also excluded those with missing data on important covariates such as perceived social support (n = 93 person-years). These criteria resulted in a final analytic sample of 4,106 person-years.

Dependent Variable: Depressive Symptoms

The 11-item version of the Center for Epidemiologic Studies Depression Scale (CES-D) (Radloff 1977) is used to measure depressive symptoms. Respondents rated their feelings over the past week on a scale from 0 to 3, reporting the frequency from “rarely or none of the time”, “some of the time”, “occasionally”, to “most of the time.” This scale includes items (1) “I did not feel like eating,” (2) “I felt depressed,” (3) “I felt that everything I did was an effort,” (4) “My sleep was restless,” (5) “I felt lonely,” (6) “People were unfriendly,” (7) “I felt sad,” (8) “I felt that people disliked me,” (9) “I could not get ‘going,’” (10) “I was happy,” and (11) “I enjoyed life.” This scale demonstrated strong reliability across all survey rounds, with Cronbach’s alpha scales ranging from .78 to .80 from Round 1 to 3. I calculated a total score by summing responses to these 11 items (Range: 0 to 30), where higher scores indicated more depressive symptoms (York Cornwell and Waite 2009).

Independent Variables: Social Network Bridging

The NSHAP asked respondents to name up to five confidants “with whom they most often discussed things that were important” over the last 12 months and recorded them in the network Roster A (see Cornwell, Laumann, and Schumm 2008). Respondents were asked to describe the nature of their relationship with each confidant, how often they interacted with each of these confidants, and how frequently each one interacted with each of the others. I used this alter-level network data to construct network bridging measures.

Social Network Bridging

The main predictors in the analyses are different types of social network bridging. Bridging refers to the extent to which a person is tied to people who are not directly, or only poorly, connected to each other (Burt 1992, 2017). Following this definition, I measured the overall network bridging as the number of pairs of network members who are disconnected from each other, meaning that the respondent reported that the two network members talk to each other once a year or less (Cornwell 2009).¹

Kin-Related Network Bridging

To classify different types of unconnected alters that respondents bridge, I combined data on alters’ relationships (kin or non-kin) to respondents and the frequency of interaction. Kin relationships are defined as those related by marriage or blood, while all others are considered non-kin. My primary bridging measures consist of three count variables recording the number of pairs of network members of a given type (kin or non-kin) that were completely unconnected or only poorly connected to each other (i.e., talk to each other once a year or less). These measures categorize network bridging into three distinctive types: kin-kin bridging (pairs of unconnected kin), kin-non-kin bridging (pairs of unconnected kin and non-kin), and non-kin bridging (pairs of unconnected non-kin).

I also assessed alternative bridging measures for validity checks: a dichotomous indicator of bridging at least one type of alter pair and the effective size of a person’s network as one’s exposure to structural holes (Burt 1992, 2017). These two alternative measures are highly correlated with the count measure of bridging (γ = .75, .82).² I used the count measure in the study because it is intuitive to interpret. Overall, there is evidence that the findings do not depend on how the network bridging is measured.

Age and Cohort

To distinguish the time-stable aspect of age (i.e., cohort) from the time-varying one, I first constructed a constant categorical cohort variable in ten-year intervals based on respondents’ ages in NSHAP Round 1 in 2005-06. 55–64-year-old cohort in 2005 (late midlife), 65–74-year-old cohort in 2005 (young-old), and 75+-year-old cohort in 2005 (mid-old and oldest old).³ I included time-varying age in years to control the aging effect.

Support and Quality of Relationship

Following the evidence that the quality of social relationships intervened with mental health (York Cornwell and Waite 2009), I included time-varying perceived social support and difficult ties to investigate the extent to which these two variables mediate or confound the association between bridging and depression. A scale of perceived support was created by standardizing responses to six ordinal items, including “How often can you open up to your spouse/family members/friends?” and “How often can you rely on your spouse/family members/friends for help?” (α = .60). A scale of difficult ties was also created, including questions “How often do you feel your spouse/family member/friend is demanding?” and “How often does your spouse/family member/friend criticize you?” (α = .61).

Health

Two measures of health were introduced due to their profound implications for later life-course experiences. Chronic diseases were measured using the Modified Charlson Comorbidity Index (Modified CCI, Charlson et al. 1987), summing respondents’ affirmative responses to the following illnesses: heart attack, heart failure, coronary procedure, stroke, diabetes, leukemia, metastatic cancer, asthma, arthritis, dementia, and Alzheimer’s (range: 0 - 7). Functional disability was assessed using the Activities of Daily Living Index by averaging responses to seven questions that assess the degree of difficulty that respondents have performing daily tasks (i.e., bathing by oneself) (α = .85).

Later Life-Course Experiences, Social Networks, and Demographic Covariates

People tend to occupy bridging in their networks if they have a large-size network and fewer kin ties; hence the models adjusted for time-varying social network size and proportion of kin. Pivotal later-life transitions such as marital status (married/partnered, separated/divorced, widowed, and never married) and retirement (retired or not) were entered as time-varying variables.⁴ Time-invariant demographic covariates were obtained from the Round 1 survey, including gender (female versus male), race/ethnicity (White, Black, Hispanic, or other race), and education attainment (college degree or not). Detailed descriptive statistics across surveys are given in Table 1.

Table 1.

Sample Characteristics and Comparison of Key Variables across Rounds ^a

	Round 1 (2005–06)	Round 2 (2010–11)	Round 3 (2015–16)	Total
	Mean (SD)/Percent				Range
Aged 57–64	45.94%	16.14%	0.00%	45.94%
65–74	38.77%	51.24%	45.78%	38.77%
75+	15.29%	32.63%	54.22%	15.29%
Female				56.20%	[0,1]
Race/Ethnicity
White (Ref.)				72.90%
African American				15.07%
Hispanic				9.90%
College education (Ref. = no)				60.08%	[0, 1]
Marital status
Married/partnered	73.84%	60.53%	55.2%	64.58%*
Separated/divorced	11.48%	13.54%	14.62%	12.97%
Widowed	12.36%	23.07%	27.68%	19.88%*
Never married	2.32%	2.86%	2.50%	2.58%
Currently Retired (Ref. = no)	53.24%	71.05%	81.56%	66.32%*	[0, 1]
Overall network bridging	1.64	1.83	2.08	1.77*	[0, 10]
	(2.13)	(2.33)	(2.37)	(2.25)
Kin bridging	0.13	0.14	0.17	0.14	[0, 7]
	(0.55)	(0.58)	(0.59)	(0.57)
Kin-non-kin bridging	0.96	1.05	1.20	1.03*	[0, 6]
	(1.48)	(1.60)	(1.64)	(1.55)
Non-kin bridging	0.54	0.64	0.70	0.60*	[0, 10]
	(1.22)	(1.49)	(1.44)	(1.38)
Social network size	3.90	4.02	4.12	3.98*	[2, 5]
	(1.13)	(1.12)	(1.09)	(1.13)
Proportion of kin	0.66	0.64	0.61	0.64*	[0, 1]
	(0.32)	(0.33)	(0.33)	(0.33)
Total Scores of depressive symptoms	4.80	4.84	4.91	4.82	[0, 30]
(CES-D)	(4.87)	(4.81)	(4.56)	(4.81)
Chronic illness	1.13	1.03	0.95	1.07*	[0,7]
	(1.04)	(1.11)	(1.03)	(1.07)
Functional disability	−0.19	−0.11	0.037	−0.13*	[−0.42, 5.79]
	(0.46)	(0.63)	(0.74)	(0.58)
Perceived support	0.089	0.12	0.076	0.10	[−3.63, 1.36]
	(0.60)	(0.55)	(0.63)	(0.58)
Perceived difficult ties	0.068	−0.046	−0.068	0.0020	[−0.79, 2.53]
	(0.63)	(0.58)	(0.59)	(0.61)*
N	1,354	1,413	1,339	4, 106

^aEstimates are weighted using NSHAP person weights and adjusted for attrition and selection using propensity scores. All models account for a multistage, clustered survey design. Number of total person-years = 4,106.

^bmeans a significant trend over rounds at p < 0.05 in this variable using the trend test or chi-square test.

Analytic Strategy

I employed two-level negative binomial models to examine the variations in depressive symptoms across respondents (between-person level) and “within” respondents (round level) over a ten-year period. The multilevel approach leverages all available data from each panel despite missing values in the longitudinal measurement and allows for estimating the round level (within-person) and person level (between-person level) simultaneously (Raudenbush and Bryk 2002). I used negative binomial regression models because of evidence of overdispersion (the variances of these counts are higher than their means) in the outcome variable—the total score of depressive symptoms (McCullagh and Nelder 1989). The multilevel negative binomial models take into account both person-level clustering and overdispersion.

The analysis involved three steps. First, I examined the relationship between social network bridging and depressive symptoms net of social network size and round indicators. Each bridging measure was split into a time-constant person-specific mean at the between-person level (e.g., a person’s average pairs of unconnected alters across three rounds), and a time-varying bridging at the within-person level—deviations in a particular round from a person’s own mean over time (Hoffman and Stawski 2009). This within-person bridging indicates how many more or fewer pairs of unconnected alters are in a given round relative to the average over time. I also included time-constant socio-demographic characteristics, time-varying age, health conditions, and later-life factors, as well as relationship quality to the models to examine the extent to which later life-course experiences account for the association between network bridging and depression.

Regarding the role of the respondents’ cohort, I introduced cohort interactions (57–64-year-old cohort, 65–74-year-old cohort, and 75+-year-old cohort in 2005) with network bridging at both within- and between-person levels. I also accounted for the time-varying age, and later life-stage factors such as retirement and widowhood in the models.

Finally, I conducted similar analyses for three distinctive types of bridging to evaluate whether the specific relationship type of alter pairs matters. Focal interactions were then added to examine the variations by respondents’ cohort. Throughout the analyses, the round indicator and network size were included in the models. Model selection was guided by the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC).

Attrition and Selection

It is important to note that the analyses only included respondents who participated in all three rounds and those who named at least two confidants and one kin member in their networks because the possibility of bridging kin and non-kin depends on having at least one kin and one non-kin alter. ⁵ It is possible that those respondents included in the analyses differ systematically from those who are missing data on important covariates, who are excluded on the basis of naming no kin or non-kin, or not answering all three surveys. To account for this possibility, I followed the inverse probability weighting adjustments to consider non-random attrition between rounds and potential selection bias (Morgan and Todd 2008; see Schafer and Koltai 2015 for an example). ⁶ First, I employed a logistic regression model to predict whether a baseline respondent is included in the final analytic sample, using several socio-demographic, health, and network-related covariates.⁷ Then, I took the inverse of this predicted probability and multiplied it by the NSHAP person-level weights that adjust differential probabilities of selection into the sample with post-stratification adjustments for non-responses at each round, and finally, applied these weights to the models. The final weight attenuates the selection bias by giving greater weight to those cases that are more likely to be excluded and allowing the models to generate estimates that better approximate those that would have been derived if all respondents were included in the final sample. All models used the NSHAP sample clustering and stratification to account for complex survey designs (e.g., strata).

Results

Table 1 displays summary statistics for key variables across each round. All types of network bridging grew over the study period from 2005-06 to 2015-16, with more notable trends in bridging involving unconnected non-kins, particularly among the younger age cohorts (See Appendix A3 for details). Respondents’ depressive symptoms remained relatively stable, with around 17% and 14% reporting no symptoms in the first and last survey, respectively. In 2005-06, 45.94% of the sample belonged to the 57–64 age cohort, 38.77% were aged 65–74, and 15.29% were 75 years old or older in 2005-06.

The variation partition coefficient (VPC)⁸ in the unconditional model shows that 40% of the variance in depression pertains to differences between respondents, while 60% is attributed to variations within the same person over time. The likelihood ratio test confirmed the significance of these variations at both levels, providing justifications for the multilevel modeling approach.

Bridging, Depression, and Cohort Differences

Table 2 provides evidence regarding the link between social network bridging and worse depressive symptoms in older adults, particularly among the youngest cohort transitioning from late midlife to later life (57–64-year-old cohort in Round 1). Model 1 shows that when adjusting for person-average network bridging, social network size, and round indicators, having an additional pair of unconnected network members than a person’s own average is associated with a 3% increase in depressive symptoms (IRR = 1.03, p < 0.01). Model 2 demonstrates that this association remains significant after considering socio-demographic, health, and later-life experiences. However, in Model 3, when accounting for perceived support and difficult ties, the association becomes insignificant (IRR = 1.02, p = 0.12), suggesting that the adverse impact of bridging on depressive symptoms might depend on the quality and support of respondents’ family members and friends.

Table 2.

Two-level Negative Binomial Models Predicting Depressive Symptoms from Overall Social Network Bridging^{a b}

	Model 1	Model 2	Model 3	Model 4	Model 5
Overall network bridging (WP) c	1.03**	1.02*	1.02	1.05**	1.04*
	[1.01,1.05]	[1.00,1.04]	[1.00,1.04]	[1.02,1.09]	[1.01,1.08]
Overall network bridging (BP)d	1.02	1.04*	1.03+	1.06*	1.07*
	[0.99,1.06]	[1.00,1.07]	[1.00,1.06]	[1.01,1.11]	[1.01,1.12]
Age 65–74 at R1 (Ref. 57–64)		1.07	1.08	1.19+	1.19+
		[0.91,1.25]	[0.93,1.25]	[0.98,1.45]	[0.99,1.43]
Age 75 + at R1		1.31+	1.28+	1.62**	1.49*
		[0.97,1.77]	[0.97,1.71]	[1.16,2.27]	[1.08,2.06]
Overall network bridging (WP) × Age 65–74 at R1				0.95+	0.96+
(Ref. 57–64)				[0.90,1.00]	[0.91,1.00]
Overall network bridging (WP) × Age 75+ at R1				0.96	0.96*
				[0.92,1.01]	[0.92,1.00]
Overall network bridging (BP) × Age 65–74 at R1				0.94+	0.95+
(Ref. 57–64)				[0.89,1.01]	[0.89,1.01]
Overall network bridging (BP) × Age 75+ at R1				0.92+	0.92*
				[0.84,1.01]	[0.85,1.00]
Time-varying age		0.99	0.99	1.00	0.99
		[0.98,1.01]	[0.98,1.01]	[0.98,1.02]	[0.98,1.01]]
Female		1.13**	1.16**		1.15**
		[1.04,1.23]	[1.06,1.27]		[1.05,1.26]
African American (Ref. White)		1.13	1.03		1.03
		[0.97,1.31]	[0.89,1.19]		[0.88,1.19]
Hispanic		1.07	0.99		1.00
		[0.87,1.33]	[0.82,1.20]		[0.83,1.21]
Marital status (Ref. married/partnered)
Separated/divorced		0.97	1.01		1.02
		[0.84,1.12]	[0.88,1.16]		[0.89,1.16]
Widowed		1.24***	1.26***		1.27***
		[1.10,1.39]	[1.12,1.42]		[1.13,1.42]
College education (1 = Yes)		0.80***	0.81***		0.81***
		[0.72,0.89]	[0.73,0.90]		[0.73,0.90]
Currently Retired (1 = Yes)		0.95	0.95		0.95
		[0.88,1.03]	[0.88,1.03]		[0.88,1.03]
Chronic illness		1.08***	1.08***		1.08***
		[1.04,1.11]	[1.04,1.11]		[1.04,1.11]
Functional disability		1.34***	1.34***		1.34***
		[1.22,1.48]	[1.23,1.47]		[1.22,1.47]
Social network size	0.94***	0.96*	0.97	0.94***	0.97
	[0.91,0.97]	[0.92,0.99]	[0.94,1.01]	[0.91,0.97]	[0.94,1.01]
Proportion of kin		0.93	0.94		0.94
		[0.81,1.06]	[0.82,1.07]		[0.83,1.07]
Perceived support			0.81***		0.81***
			[0.76,0.87]		[0.76,0.87]
Perceived difficult ties			1.28***		1.28***
			[1.21,1.35]		[1.21,1.35]
Fixed intercept (lnalpha)	0.25***	0.24***	0.24***	0.24***	0.24***
	[0.21,0.29]	[0.20,0.28]	[0.20,0.29]	[0.20,0.29]	[0.20,0.28]
Random effect: variance of the intercept	1.87***	1.62***	1.49***	1.84***	1.49***
	[1.72,2.03]	[1.51,1.74]	[1.40,1.59]	[1.70,1.99]	[1.40,1.59]
AIC	21900.2	21547.5	20992.2	21881.4	20993.9
BIC	21950.9	21693.3	21150.2	21976.5	21177.2

^aN of total person-years = 4,106. Estimates are weighted using NSHAP person weights and adjusted for attrition and selection using propensity scores. All models account for a multistage, clustered survey design. 95% confidence intervals in brackets.

⁺p < 0.10,

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

^bAll models report incidence rate ratio and include round dummy indicators where t = 1–3. R1 = Round 1

^cPerson-mean across waves (i.e., time-varying). WP=within-person.

^dPerson-mean centered scores (i.e., time-invariant). BP=between-person.

Model 4 and 5 in Table 2 further introduce age cohort interactions at the between- and within-person levels. Both models reveal significant main effects of bridging, with Model 5 showing a declining impact of bridging on depression for older age cohorts (65–74-year-old and 75+-year-old cohort in Round 1) relative to the youngest (57–64-year-old cohort), even after accounting for health conditions, later-life events, and relationship quality. This suggests that the negative association between bridging and depression may be more pronounced in the youngest age cohort, especially when considering older adults’ perceived support and strain.

Comparisons by Bridging Type

I also explored whether the association between risks of depression and network bridging depends on the type of alter pairs that respondents bridge. Figure 2 summarizes additional analyses that compare predicted depressive symptoms based on within-person bridging status (whether a person bridges more than their own average) among the specific type of alter pair (kin-kin, kin–non-kin, non-kin) while accounting for respective between-person bridging. Notably, there is a 9.8% increase in predicted depressive symptoms for those who bridge more kin and non-kin pairs than their own averages, and this is the only significant association among the three distinct types (IRR = 1.10, p < 0.01). For the other two types of alter pairs, the differences between above-average bridging and below-average bridging are not statistically significant (IRR = 0.97, p > 0.05 for bridging more kins; IRR = 1.02, p > 0.05 for bridging non-kins).⁹

Figure 2. Marginal Effects of Bridging Status by Types of Alter Pair for Depressive Symptoms.

Note: Each type of bridging status indicates whether there are more pairs of unconnected network members of a specific relationship type (i.e., kin bridging, kin-non-kin bridging, non-kin bridging) than the person’s average. The marginal effects reflect the predicted scores of depressive symptoms associated with within-person bridging. Three separate models are used (one model for each type of within-person bridging), each weighted using NSHAP person weights (adjusted for attrition and selection using propensity scores) and accounted for the multistage, clustered survey design. 95% confidence intervals are applied. All models included between-person bridging, social network size, and round dummy indicators. ⁺ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001

Table 3 presents associations between kin–non-kin bridging and depression symptoms. As Model 1 and 2 show, bridging unconnected kin and non-kin remains associated with elevated depressive symptoms after controlling for later-life factors such as health issues, retirement, and widowhood. Again in Model 3, within-person kin–non-kin bridging becomes marginally associated with depression risks when net of perceived support and difficult ties (IRR = 1.02, p < 0.10). Given that these two factors are significant predictors of depressive symptoms, it is likely that the effect of kin–non-kin bridging on depressive symptoms is contingent on the support and quality of older adults’ social relationships. ¹⁰

Table 3:

Two-level Negative Binomial Models Predicting Depressive Symptoms from Kin-Non-Kin Bridging ^ab

	Model 1	Model 2	Model 3	Model 4	Model 5
Kin-non-kin bridging (WP)c	1.03**	1.03*	1.02+	1.08**	1.06**
	[1.01,1.06]	[1.00,1.05]	[1.00,1.05]	[1.03,1.12]	[1.02,1.11]
Kin-non-kin bridging (BP) d	1.05*	1.06*	1.05*	1.08*	1.09*
	[1.00,1.09]	[1.01,1.10]	[1.00,1.09]	[1.00,1.17]	[1.02,1.17]
Age 65–74 at R1 (Ref. 57–64)		1.07	1.08	1.12	1.13
		[0.91,1.26]	[0.93,1.25]	[0.93,1.35]	[0.96,1.34]
Age 75 + at R1		1.30+	1.28+	1.60**	1.48*
		[0.96,1.76]	[0.96,1.70]	[1.14,2.24]	[1.07,2.06]
Kin-non-kin bridging (WP) × Age 65–74 at R1 (Ref. 57–64)				0.92*	0.92*
				[0.86,0.99]	[0.87,0.98]
Kin-non-kin bridging (WP) × Age 75 + at R1				0.94+	0.93*
				[0.88,1.01]	[0.88,0.99]
Kin-non-kin bridging (BP) × Age 65–74 at R1 (Ref. 57–64)				0.96	0.95
				[0.86,1.06]	[0.87,1.05]
Kin-non-kin bridging (BP) × Age 75 + at R1				0.88+	0.87*
				[0.76,1.02]	[0.77,0.99]
Time-varying age		0.99	0.99	1.00	0.99
		[0.98,1.01]	[0.98,1.01]	[0.98,1.02]	[0.98,1.01]
Female		1.13**	1.15**		1.15**
		[1.04,1.23]	[1.05,1.26]		[1.05,1.26]
African American (Ref. White)		1.13	1.03		1.02
		[0.97,1.31]	[0.89,1.19]		[0.88,1.18]
Hispanic		1.08	1.00		1.01
		[0.88,1.33]	[0.83,1.20]		[0.84,1.21]
Marital status (Ref. married/partnered)
Separated/divorced		0.97	1.01		1.02
		[0.84,1.13]	[0.89,1.16]		[0.90,1.16]
Widowed		1.23***	1.26***		1.26***
		[1.10,1.39]	[1.12,1.42]		[1.13,1.42]
College education (1 = Yes)		0.80***	0.81***		0.81***
		[0.72,0.89]	[0.73,0.90]		[0.74,0.90]
Currently Retired (1 = Yes)		0.96	0.95		0.95
		[0.88,1.03]	[0.88,1.03]		[0.88,1.02]
Chronic illness		1.08***	1.08***		1.08***
		[1.04,1.11]	[1.04,1.11]		[1.04,1.11]
Functional disability		1.34***	1.34***		1.34***
		[1.22,1.48]	[1.23,1.47]		[1.23,1.47]
Social network size	0.94***	0.96*	0.98	0.95***	0.98
	[0.91,0.97]	[0.93,0.99]	[0.94,1.01]	[0.92,0.98]	[0.94,1.01]
Proportion of kin		0.89+	0.91		0.90
		[0.78,1.01]	[0.79,1.04]		[0.79,1.03]
Perceived support			0.81***		0.81***
			[0.76,0.87]		[0.76,0.87]
Perceived difficult ties			1.28***		1.28***
			[1.21,1.35]		[1.21,1.35]
Fixed intercept (lnalpha)	0.25***	0.24***	0.24***	0.24***	0.24***
	[0.20,0.29]	[0.20,0.28]	[0.20,0.29]	[0.20,0.29]	[0.20,0.28]
Random effect: variance of the intercept	1.87***	1.62***	1.49***	1.84***	1.49***
	[1.72,2.03]	[1.52,1.74]	[1.40,1.59]	[1.70,1.99]	[1.40,1.59]
AIC	21900.0	21549.8	20990.6	21882.0	20990.9
BIC	21950.8	21695.6	21148.6	21977.1	21174.2

^aN of total person-years = 4,106. Estimates are weighted using NSHAP person weights and adjusted for attrition and selection using propensity scores. All models account for a multistage, clustered survey design. 95% confidence intervals in brackets.

⁺p < 0.10,

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

^bAll models report incidence rate ratio and include round dummy variables where t = 1–3. R1 = Round 1

^cPerson-mean centered scores (i.e., time-invariant). BP=between-person.

^dPerson-mean across rounds (i.e., time-varying). WP=within-person.

I further introduced cohort interactions with within- and between-person kin–non-kin bridging.¹¹ Model 4 and 5 show that cohort differences in the effect of bridging kin and non-kin pertain to the within-person level. For example, when compared to the youngest cohort (57–64-year-old cohort in Round 1), bridging additional above-average kin and non-kin pairs is associated with an 8% and a 7% decrease in depressive symptoms for those aged 65-74 and 75+ in Round 1 (IRR = 0.92, p < 0.05 for 65–74; IRR = 0.93, p < 0.05 for 75+). Additionally, compared with Model 3, Model 4 and 5 present both significant main effects of kin–non-kin bridging net of perceived social support and difficult ties after introducing the cohort interactions. This suggests that when the support and quality of older adults’ relationships with their family and friends are considered, the association between kin–non-kin bridging and increased depressive symptoms may be particularly relevant to the youngest cohort.

Figure 3 displays the visuals of the interactions. The left panel presents within-person variations in depressive symptoms, highlighting a positive trend linking above-average kin-non-kin bridging and depression risks, especially among the youngest cohort. In the right panel, the cohort differences are not significant at the person-mean level. These findings imply that it is the changes in bridging status within a person’s network over time that induce depressive symptoms.

Figure 3. Adjusted Relationship Between Kin-Non-Kin Network Bridging and Depressive Symptoms by Age Groups (Left: Within-Personal Level; Right: Between-Person Level).

Note: This Figure is derived from Model 5 in Table 3. The Y-axis is the marginal mean of the depressive symptoms score. 95% confidence intervals are applied. Between-person kin-non-kin bridging is the person mean calculated by averaging the number of unconnected kin and non-kin across Round 1–3. Within-person kin-non-kin network bridging is centered around person mean, such that 0 indicates the mean, and the positive/negative values indicate the above/below average number of pairs of unconnected kin and non-kins. The group differences in the right panel are not significant at p < 0.05.

Sensitivity Analyses

These findings are robust against several sensitivity tests and alternative modeling strategies. To validate the findings, I employed multilevel logistic regression models using self-reported mental health status, where respondents reporting “poor” or “fair” mental health were coded “1,” and “0” otherwise (“good,” “very good,” or “excellent”). I also used linear mixed-effect models to predict a standardized scale of depressive symptoms (α = 0.80). Furthermore, I conducted analyses on less strict samples that included respondents who participated in more than two survey rounds (R1 and R2, R2 and R3, or at least once in R1, R2, or R3). Among these alternative specifications, models, and samples, the key estimates (e.g., bridging) are consistent with the primary analyses in terms of both direction and statistical significance in predicting depressive symptoms. Hence, I am encouraged that the findings do not depend on the specific dataset, measure specifications, or modeling approaches.

Conclusions and Discussion

Building on prior research examining the link between social network structure and health, this study emphasizes the significance of (absent) connections among network members in shaping older adults’ depressive symptomatology. Bridging distinct social circles, such as kin and non-kin, is significantly associated with elevated depressive symptoms over time, and this association is contingent on social support and strain. More so for the youngest cohort (57–64-year-old cohort in Round 1) than older cohorts in NSHAP in 2005, a higher number of poorly connected kin and non-kin pairs is associated with severe depressive symptoms over time.

These findings underscore the importance of connections among one’s closest kin and non-kin in yielding vital health resources that influence older adults’ mental health. Bridging kin and non-kin may compromise older adults’ access to joint support, including ordinated health regulation and caregiving, which is usually provided within a tightly integrated support system encompassing interconnected family and informal care. Notably, informal support like friendship, alongside predetermined family relationships, is increasingly vital for enhancing older adults’ mental health, buffering against life stressors, and re-affirming self-worth (Ang 2018; Huxhold et al. 2014). In this regard, bridging kin and non-kin might signal poor caregiving patterns during later life events (e.g., health decline) that contribute to stress and social strain. Non-kin relationships, owing to their voluntary nature, are often less enduring than family bonds when faced with the ongoing health needs of older adults due to higher commitments on reciprocal obligations (Perry and Pescosolido 2012; Qu 2023). Another possible explanation is that the growing prospects of bridging kin and non-kin stems from a lack of social cohesion and reflects that older adults might be poorly embedded in their networks or positioned on the periphery of their family structure. in such a situation, older adults might experience diminished levels of affection, trust, and support from their peers, which fail to mitigate the negative impacts of sparse relationships on depressive symptoms.

What’s more, the association between bridging and elevated depression risks is unique to bridging kin and non-kin, partly explained by a lack of support from relationships. This finding aligns with previous research suggesting that perceived social disconnectedness predicts mental distress (e.g., Song et al. 2021; York Cornwell and Waite 2009). Although not a focus of this study, it would be interesting to explore the psychosocial mechanisms in which bridging might be a structural source of strain and inadequate support that elevate the risks of depression (Bearman and Moody 2004). Social network theory posits that social relationships are structured around social contexts (Feld 1981). Kin and non-kin may lack familiarity or have fewer reasons to interact. With bridging tasks requiring physical and cognitive capabilities (Cornwell 2009), the strain may arise from cognitive challenges of coordinating support and mediating interests from multiple social domains, especially when they adhere to distinct sets of behavioral norms. Specifically, transitioning from kin to disconnected non-kin circles may reduce cognitive reliance on pre-existing scripts and norms within family realms, thereby increasing uncertainty in interactions (Giddens 1984). In contrast, bridging kin pairs, or non-kin pairs, may involve less cognitive shifts in thinking modes associated with context-specific scripts and norms. Additionally, bridging kin could relax the stringent family pressure, while bridging non-kin could facilitate the ego’s integration into diverse social domains, both of which to some extent might benefit mental health (Falci and McNeely 2009; Goldman 2016).

Respondents with depressive symptoms also bridge a variety of kin and non-kin pairs (details in Appendix A1). In this study, most unconnected kin and non-kin pairs include a respondent’s friend and a core family member, usually their child, representing one’s immediate sources of support within and beyond family. This scenario might be especially true when adult children are geographically dispersed, for which older adults become increasingly reliant on friends for healthcare needs such as doctor visits or caregiving. While existing literature on intergenerational solidarity focuses on parent-child bonds in supplying social support (e.g., Connidis and McMullin 2002), increasing research points out that friendship also plays an indispensable role in providing companionship and fostering solidarity in later life, especially in a time of increasing complexity and reconfigurations in family structure (e.g., Huxhold et al. 2020, 2014). Active involvement of both children and friends in important life matters is essential for older adults, helping them confirm information sources, seek emotional validation, and transfer beliefs and values around solidarity (Baumeister and Leary 1995). As a result, this study suggests exploring the role of friendships and other non-kin relationships in reinforcing family solidarity across generational lines.

Furthermore, the effect of kin-non-kin bridging on depression is most pronounced in the youngest cohort (57-64 years old in Round 1). This finding appears to slightly contradict studies following the socioemotional selectivity theory in that bridging seems to matter less in late adulthood. Recall that bridging increases over time, especially involving non-kin (details in Table 1 and Appendix A3). This finding suggests that older adults may move towards network structures prioritizing kin-based strong bonds while having non-kin at the periphery. Contextual changes brought on by later-life transitions would be responsible for such network reconfiguration. Specifically, younger cohorts in 2005 were transitioning into later life; they may face increased turbulence due to stressful life events, like retirement and the onset of illness, thus experiencing more strain due to losing and establishing relationships across different social domains (Alwin et al. 2018; Cornwell 2009). Alternatively, this situation could be explained by some unobservable generational differences in which younger generations came of age dealing with more complicated and varied family structures and relying less on kin structures and community organizations (Sassler 2010; Waite et al. 2021), making them more dependent on informal networks for mental health support. These issues resonate with the importance of nurturing close, non-kin relationships in the community (e.g., friends) in promoting mental well-being mentioned above and also with existing work on increasing the stress-buffering effect of informal activities with friends in older ages (Ang 2018, 2021; Huxhold et al. 2014). For these reasons, this study underscores the importance of early intervention in proactively enforcing interconnected kin and non-kin relationships to address mental well-being around later-life challenges and encourages further research on generational disparities.

This study is not exempt from limitations. First, NSHAP’s name-generator instrument often elicits frequently accessed ties, which renders bridging more of a perceived position in a confidant group, potentially resulting in conservative estimates. Another legitimate concern is that the lack of alter-level information in the NSHAP project makes it hard to assess the extent to which respondents were on the periphery of their networks in contrast to occupying bridging potential, which could have different health implications. Additionally, previous research suggests that individuals’ sentiments can collectively shape network structures, and their depressive moods also spread through social ties, both of which relate to the focal person’s mental health (Ellwardt et al. 2020; Schaefer, Kornienko, and Fox 2011). Therefore, future research may consider exploring these possibilities in a socio-centric network or incorporating alter-level characteristics when investigating how network structure affects health.

More longitudinal analysis is needed to investigate how heterogeneous network structures affect mental health over later life course and across generations. While this study employs panel data to reduce concerns like reverse causality, it is crucial to exercise caution in drawing causal conclusions. I attempted to address selection effects through propensity weighting, but this did not eliminate issues around the selection, particularly in an aging population sample with considerable attrition over the 10-year survey period. Although consistent results were obtained with a less restrictive sample (including respondents who took the survey at least twice), future research should consider using more robust panel data, perhaps with additional rounds, to explore the evolving dynamics of social networks and their influence on mental health in later life.

Moreover, while the interaction “effect” is significant, its size is marginal. One reason for this might be that average within-person estimates result from sizeable heterogeneity in older adults’ susceptibility to the effects of social network bridging on depression. Specifically, the average within-person effect size can be considered as the aggregate of numerous individuals’ within-person effect sizes ranging from highly positive to negative (Beyens et al. 2020). Thus, it is necessary to consider the possibility that the influence of bridging may vary from person to person in different groups. Future research could allow heterogeneous variances in the model intercept and slope to examine the within-person fluctuations/variability across different groups (Hoffman and Stawski 2009).

Limitations aside, this study contributes to the emerging nexus of social gerontology and network analysis by exploring how structural factors are associated with mental well-being. Specifically, we consider the increasing absence of connections between kin and non-kin alters as relational disadvantages and investigate its implications for mental health. While bridging kin and non-kin facilitates access to non-redundant information and resources, it also yields relationship costs and strain that result in mental distress. More importantly, this work is especially germane for understanding how social relationships are structured around generational dynamics and later-life transitions in producing relationship benefits and deficits for older adults’ health needs. Lastly, this study suggests proactive interactions that foster diverse and interconnected social networks from an early age, especially for young cohorts, and emphasizes the need for further research on generational disparities in family network structure and its impact on health.

Highlights.

1. The paper discusses how bridging disparate network members produces strain that elevates depression risks in later life.

2. Older adults’ bridging potential in personal networks is classified into three distinct types.

3. The association between bridging and elevated depressive symptoms is unique to bridging unconnected kin and non-kin.

4. The effect of bridging is more so for the youngest age cohort than older cohorts.

Appendix A2. Adjusted Relationship Between Social Network Bridging and Depressive Symptoms by Age in Years across Different Age Cohorts.

Note: This figure is derived from fixed-effect models that analyze the interaction between within-person network bridging and age, specifically in three age cohorts (57-65, 65-75, 75+ years old at Round 1 in 2005-06). These models account for time-varying health and life-course factors, and demographics. The Y-axis represents the marginal mean of total depressive symptom scores with 95% confidence intervals. Within-person network bridging is centered around the individual mean, with positive/negative values indicating above/below average numbers of unconnected network members. The lines in each panel represent the age within the respective age cohort. Notably, the impact of network bridging on depression varies with age, primarily in the 57-65 age cohort, where the lines are steeper for younger ages, indicating a stronger effect of within-person network bridging on depression among the youngest cohort aged 57-64 in at Round 1 (2005-06).

Appendix A3. Predicted Social Network Bridging over Time.

Note: This figure shows predicted values for the number of pairs of unconnected network members over time across various age cohorts, based on a negative binomial model adjusting for time-varying age to distinguish aging from cohort effects.

Appendix A1:

Matrix Showing Frequency of Unconnected Pairs of Kin and Non-Kin by Alters’ Relationship to Respondent Who Report Depressive Symptoms (Unweighted) ^a

Alter Relationship to Respondent Who Report Depressive Symptoms	Romantic Partner/Ex-spouse	Friend	Neighbor	Colleagues	Minister, Priest, or Other Clergy	Psychiatrist, Psychologist, Counselor or Social Worker	Housekeeper/Home Health Care Provider	Other non-kin member
Spouse	0 | 0 | 0	42 | 25 | 43	1 | 1 | 0	17 | 12 | 7	1 | 0 | 2	6 | 2 | 1	1 | 0 | 1	2 | 4 | 5
Parent	0 | 0 | 0	27 | 9 | 5	0 | 0 | 0	7 | 4 | 2	0 | 2 | 0	0 | 0 | 0	0 | 0 | 1	2 | 2 | 1
Child or Grandchild	5 | 2 | 8	241 | 275 | 305	14 | 36 | 40	42 | 20 | 27	12 | 20 | 19	8 | 4 | 5	1 | 0 | 5	9 | 19 | 15
Sibling	3 | 3 | 4	146 | 153 | 150	19 | 22 | 20	20 | 20 | 10	13 | 10 | 9	2 | 3 | 2	1 | 0 | 2	7 | 6| 11
In-law	0 | 0 | 2	46 | 66 | 64	9 | 12 | 8	15 | 7 | 4	4 | 6 | 1	0 | 0 | 1	0 | 0 | 0	2 | 4 | 1
Other relatives	0 | 0 | 3	60 | 81 | 102	8 | 12 | 18	10 | 6 | 8	0 | 4 | 8	3 | 4 | 1	0 | 0 | 2	3 | 5 | 5

^aN = 1, 467. Frequencies are based on respondents who are included in Model 5 in Table 3 and who reported any depressive symptoms (depressive symptoms score > 0). Numbers from left to right in each cell represent the total number of unconnected pairs of that type of kin-non-kin dyad at the time of Round 1, 2, and 3, respectively.