1. Introduction
Previous research has highlighted the significance of social networks not only for the transmission of communicable diseases but also for non-communicable health outcomes, including lower risk of cognitive decline (Kuiper et al., Reference Kuiper, Zuidersma, Zuidema, Burgerhof, Stolk, Oude Voshaar and Smidt2016), dementia (Kuiper et al., Reference Kuiper, Zuidersma, Voshaar, Zuidema, van den Heuvel, Stolk and Smidt2015), depression (Reiner & Steinhoff, Reference Reiner and Steinhoff2024; Reiner et al., Reference Reiner, De Gioannis and Steinhoff2025), and premature mortality (Holt-Lunstad et al., Reference Holt-Lunstad, Smith and Layton2010). A key mechanism linking social networks to health is the mobilization of resources embedded in social relationships that support the prevention of and recovery from illness. These resources include various forms of social support, such as advice, information, emotional support, affirmation, and attitudes from others with regard to managing individual health issues (Abbott et al., Reference Abbott, Bettger, Hanlon and Hirschman2012; Schafer, Reference Schafer2013). Thereby, communication about health issues is a central mechanism through which social networks exert their influence on health outcomes, and communication often leads to the activation of social support and the transmission of valuable information. People routinely share concerns, seek advice, and shape one another’s health decisions through everyday conversations (Berkman et al., Reference Berkman, Glass, Brissette and Seeman2000; Smith & Christakis, Reference Smith and Christakis2008). Seeking health-related advice becomes particularly important with aging, as morbidity progresses (Thoits, Reference Thoits2011).
Previous research on health advice among middle-aged and older adults primarily investigated egocentric network data and found that it is mainly exchanged within close relationships, such as family and friends, but also others, like coworkers (Perry & Pescosolido, Reference Perry and Pescosolido2010, Reference Perry and Pescosolido2015). However, less attention has been paid to social settings that encompass both close and more distant relationships despite the potential benefits of the latter in providing nonredundant information (Granovetter, Reference Granovetter1973). As it is notoriously difficult to record conversations among participants in real time (for a rare exception, see McFarland et al., Reference McFarland, Jurafsky and Rawlings2013), survey questions usually ask respondents who they confide in related to personal matters (e.g., Marsden, Reference Marsden1987). However, only asking for instances of seeking advice overlooks dormant social capital in social ties that might only get activated if a problematic situation, such as a health issue, arises (Small, Reference Small2017).
Rather, the belief that social support, including advice, is available if needed has continuously shown to be more predictive of positive health outcomes than the actual support received (Uchino, Reference Uchino2009; Wills & Shinar, Reference Wills, Shinar, Cohen, Underwood and Gottlieb2000). The perception of available advice shapes an individual’s willingness to seek advice or support during times of acute illness (Thoits, Reference Thoits2011). Accordingly, we examine the perceived structural opportunities for obtaining health advice. In addition, most previous work lacks complete network data, which prevents researchers from disentangling the interplay between health advice and close relationships in bounded settings for interaction, such as voluntary associations. Our study complements existing literature by investigating how health advice networks (HANs), particularly those of older adults, are structured and extend beyond immediate social circles.
To address this gap, we aim to present a first case study analyzing HANs outside the family or institutional context, more specifically within voluntary associations. Similar to urban communes (e.g., Martin et al., Reference Martin, Yeung and Zablocki2001), voluntary associations represent a naturally bounded, yet informal social setting where social ties form and evolve organically. Formal volunteering is frequently used by older adults as an active strategy to expand their networks and combat social isolation and loneliness (Donnelly & Hinterlong, Reference Donnelly and Hinterlong2010; Jongenelis et al., Reference Jongenelis, Jackson, Newton and Pettigrew2022). Studying voluntary associations present a suitable research site to study the interplay between health and multiple types of networks as they include—but are not restricted to—family, close friends, and colleagues in addition to health advice networks. Importantly, beyond their primary purpose, such associations may serve as sites where individuals gain access to unanticipated social capital (Small, Reference Small2009), including valuable health-related information and support. We argue that voluntary associations are a fruitful breeding ground for health advice and that HANs exhibit a distinct structure in comparison to close ties.
We draw on two parallel streams of literature. The first investigates the effects of HANs on individuals’ health outcomes, largely independent of social context (Perry & Pescosolido, Reference Perry and Pescosolido2010, Reference Perry and Pescosolido2015; Schafer, Reference Schafer2013). A growing number of studies have used statistical network models to examine the spread of health information in specific domains, such as HIV prevention (Young et al., Reference Young, Mayaud, Suen, Tambe and Rice2020), vaccination attitudes (Salathé & Bonhoeffer, Reference Salathé and Bonhoeffer2008), or misinformation (Dunn et al., Reference Dunn, Surian, Leask, Dey, Mandl and Coiera2017; Surian et al., Reference Surian, Nguyen, Kennedy, Johnson, Coiera and Dunn2016), treating health advice as an implicit transmission mechanism. However, this work tends to focus on specific behaviors, online settings, or only implicitly model health advice and rarely examines the structural features of complete HANs in naturally bounded offline environments. In line with a growing stream of network research, we argue that studying network endogenous processes, such as transitivity and reciprocity, is crucial to better understanding how social networks shape the life outcomes of individuals (Christakis & Fowler, Reference Christakis and Fowler2007; Perkins et al., Reference Perkins, Subramanian and Christakis2015).
The second stream investigates the structure of complete networks, particularly in naturally bounded contexts, such as schools (Bearman et al., Reference Bearman, Moody and Stovel2004; Moody, Reference Moody2001) or science (Newman, Reference Newman2001; Wittek et al., Reference Wittek, Bartenhagen and Berthold2023). Even though there exists some research on specific types of ties, such as negative (Berger & Dijkstra, Reference Berger and Dijkstra2013; Isakov et al., Reference Isakov, Fowler, Airoldi and Christakis2019), gossip (Ellwardt et al., Reference Ellwardt, Steglich and Wittek2012), or romantic ties (Bearman et al., Reference Bearman, Moody and Stovel2004), much of this work has emphasized close relationships, which are typically represented as strong ties, defined by frequent contact, emotional intensity, and mutual investment (Granovetter, Reference Granovetter1973). However, scholars also show a growing recognition of the important role that weaker ties can play in structured social environments. Recent research has continued to explore the role of weak ties in social networks, noting their potential relevance for information diffusion and access to diverse resources under certain conditions (Aral, Reference Aral2016; Kim & Fernandez, Reference Kim and Fernandez2023), as well as their occasional presence within individuals’ core discussion networks (Small, Reference Small2013). This underscores that function-specific relationships, like health advice ties, need not align neatly with close relationships. Rather, such networks may rely on a mix of strong and weak connections, depending on trust, accessibility, and expertise (Perry & Pescosolido, Reference Perry and Pescosolido2010).
Despite growing interest in the consequences of weak ties for health outcomes, function-specific ties, particularly those involving the exchange of health-related advice, have received far less attention, and there are no studies investigating the structure of HANs with statistical network models. We argue that this gap limits our understanding of the structural conditions that shape how individuals identify potential health advisors within social environments beyond the family context—an essential process for effectively addressing health-related challenges. Our study addresses this lacuna by analyzing complete networks of health advice and close relationships embedded in three voluntary associations with exponential random graph models (ERGMS, Lusher et al., Reference Lusher, Koskinen and Robins2013) for the first time. Recent advances in ERGMs allow us to build models taking into account how health advisors and close ties are intertwined and to compare the presence and strength of social processes in these tie types by using average marginal effects (Duxbury, Reference Duxbury2023).
Our results indicate that, on average, individuals identify two health advisors in their voluntary organization. Crucially, health advice and close relationships overlap only by 34%, and the network structure of health advice differs starkly from that of close relationships. This indicates that voluntary associations play a vital role in broadening access to diverse health information beyond the individual’s immediate social circles. Additionally, we observed that homophily in sociodemographic traits and individual health status influences the likelihood of seeking health-related advice. As a result, individuals’ efforts to seek health advice are shaped not only by their personal characteristics but also by the social dynamics of their relationships and local communities.
We argue that combining insights on HANs with a social network lens to study networks and health in voluntary associations offers a fruitful extension of the existing literature. If researchers and practitioners are better able to understand the self-organizing principles of HANs that shape an individual’s opportunities to receive, share, and exchange health advice with others, more effective interventions can be tailored toward the promotion of health information exchange and, consequently, toward the improvement of community and public health (cf. Small, Reference Small2013, Reference Small2017).
2. Study context: voluntary associations
Prior research on HANs has mostly taken an egocentric approach and found that HANs are comprised of family and other close individuals (Perry & Pescosolido, Reference Perry and Pescosolido2010, Reference Perry and Pescosolido2015). Less attention has been paid to broader contexts that include both close and distant ties, with the latter offering access to nonredundant information (Granovetter, Reference Granovetter1973). Particularly in contexts beyond the family, people may unintendedly access information that offer unanticipated gains (Small, Reference Small2009). For middle-aged and older adults, local voluntary social settings—distinct from family, close friends, and work—are particularly relevant. These settings become increasingly important as aging, retirement, and health changes lead to shrinking social networks (Wrzus et al., Reference Wrzus, Hänel, Wagner and Neyer2013).
Older adults often engage in formal social activities, particularly volunteering, as a way to combat social isolation, strengthen their networks, and reduce loneliness (Donnelly & Hinterlong, Reference Donnelly and Hinterlong2010; Jongenelis et al., Reference Jongenelis, Jackson, Newton and Pettigrew2022). Defined as unpaid and non-mandatory work for a community or organization (Donnelly & Hinterlong, Reference Donnelly and Hinterlong2010) volunteering is widely recognized for its role in supporting healthy aging. Studies have associated it with numerous health advantages, such as improved self-rated health, enhanced life satisfaction, a lower risk of mortality, reduced depressive symptoms, and greater functional independence (Greenfield & Marks, Reference Greenfield and Marks2004; Webster et al., Reference Webster, Ajrouch and Antonucci2021).
Here, we examine HANs within the contexts of voluntary associations, specifically carnival clubs in a metropolitan region in Germany. These clubs are responsible for organizing cultural festivities during Carnival season (yearly starting on November 11th until February/March), a vibrant and long-standing tradition marked by parades, music, costumes, and social gatherings. Strongly intertwined with the region’s cultural identity, these festivities celebrate local dialects, customs, and community ties.
However, carnival clubs serve a purpose beyond the seasonal celebrations, as they facilitate year-round engagement (Niekrenz, Reference Niekrenz and Niekrenz2011). Members participate in various social activities, including summer festivals, monthly informal gatherings, and charitable initiatives, fostering continuous interaction within the group. Prior qualitative research on carnival club members (Steinhoff et al., Reference Steinhoff, Ellwardt and Wermeyer2024), suggests that participation is driven more by the sense of community than by the festival itself. Members find it easy to establish and maintain social ties within the clubs, as active participation is not a strict requirement. For retirees, these clubs serve as a means to regain a sense of purpose and mitigate the loss of role and status often associated with retirement. The sense of being needed and valued through involvement in the club contributes positively to overall well-being (Steinhoff et al., Reference Steinhoff, Ellwardt and Wermeyer2024).
Unlike institutional settings—such as workplaces or retirement homes—where social ties are often shaped by structural constraints and limited choice, carnival clubs are characterized by self-selection and greater individual agency in forming social connections (Rawlings et al., Reference Rawlings, Smith, Moody and McFarland2023). This makes carnival clubs an ideal case for studying HANs of middle-aged and older adults exceeding the family context. First, these associations provide a voluntary leisure setting in which informal socializing takes place in a heterogeneous group, exceeding the contexts of family, neighborhood, and work organizations (cf. Granovetter, Reference Granovetter1973; Niekrenz, Reference Niekrenz and Niekrenz2011). Second, they often include a disproportionate share of adults in the second half of life. Third, because membership is formally defined, they offer a clearly demarcated network boundary, a crucial requirement for employing sociometric social network analysis. Furthermore, these associations remained active during Covid-19-related social distancing measures and are open to all individuals, with no prerequisites for joining.Footnote 1 Unlike other voluntary settings such as sports clubs or retirement homes, these clubs are less selective regarding members’ health. The health demographics of our study sample closely reflect those of the general German population (Robert Koch-Institut, 2018), reducing the likelihood of selection bias related to health and making them a valuable context for studying health-related network effects.
3. Theory
Close relationship networks and HANs are not mutually exclusive; rather, they coevolve. For example, in their study on clients using mental health services for the first time, Perry and Pescosolido (Reference Perry and Pescosolido2010) found HANs to be particularly comprised of close, strong, and frequently contacted relationships, in additional to more specialized associates. Given the sensitivity of health-related topics, individuals may be reluctant to share medical experiences or seek advice from those with whom they lack emotional closeness and thus, prefer close confidants. From the long research tradition on close relationships in other domains, such as school settings (Coleman, Reference Coleman1968; McFarland et al., Reference McFarland, Moody, Diehl, Smith and Thomas2014; Moody, Reference Moody2001), universities (Vörös et al., Reference Vörös, Boda, Elmer, Hoffman, Mepham, Raabe and Stadtfeld2021; Wimmer & Lewis, Reference Wimmer and Lewis2010), and workplaces (Ellwardt et al., Reference Ellwardt, Steglich and Wittek2012; Kilduff & Krackhardt, Reference Kilduff and Krackhardt1994), we know that networks of close relationships are typically structured by multiple organizational principles, such as transitive closure and segregation along social categories. Based on their coevolution, we expect that some of these principles will also apply to HANs. Additionally, as Small (Reference Small2017) highlights, individuals do not always anticipate the sources of support, e.g., advisors, they will rely on. More recent findings further suggest that individuals commonly and intentionally avoid confiding in close friends and family depending on the conjunction of network member and topic (Small et al., Reference Small, Brant and Fekete2024). Conversely, they may seek advice from more distant or even unexpected ties (Small, Reference Small2013, Reference Small2017). These patterns underscore the importance to consider the possible conditions across a range of stronger and weaker ties that make such discussions more likely. Voluntary associations create opportunities for unanticipated gains by exposing individuals to diverse social interactions, including weak ties that may become crucial for health-related exchanges. In the following sections, we discuss organizational principles for networks of close relationships and present hypotheses regarding whether and to what extent we expect to observe these principles in HANs. Thereby, we look at the structural mechanisms (e.g., transitivity and homophily) that shape how health-related advice unfold, as well as the conditions that make health-related advice more likely—particularly when involving weak ties. Table 1 provides an overview of the hypotheses and the modeled terms.
Table 1. Overview of hypotheses
3.1. Transitivity
Transitive closure or clustering is a common feature in many social networks, which is to say that actors with shared contacts tend to establish relationships (e.g., Granovetter, Reference Granovetter1973), ranging from more emotionally distant networks (e.g., work advice-seeking networks, Bunger et al., Reference Bunger, Doogan, Hanson and Birken2018) to emotionally close networks, such as friendships (McFarland et al., Reference McFarland, Moody, Diehl, Smith and Thomas2014; Moody, Reference Moody2001). This tendency can be explained by the fact that common contacts act as foci for interactions (Feld, Reference Feld1981) and that actors prefer balanced social relationships (Heider, Reference Heider1958; Yap & Harrigan, Reference Yap and Harrigan2015). Theory and empirical studies suggest that trust, repeated interaction, and shared norms in close relationships often amplify transitive closure (McFarland et al., Reference McFarland, Moody, Diehl, Smith and Thomas2014; Moody, Reference Moody2001). Close-tie networks are generally dense, and the alters in these networks are strongly interconnected (Granovetter, Reference Granovetter1973). This density results in relationships that mutually reinforce one another (Small et al., Reference Small, Deeds Pamphile and McMahan2015), which reflects a high level of transitive closure, or in quantitative terms, a greater number of closed relationship triangles.
In the context of voluntary associations, we expect to observe transitive closure in HANs as well, but to a lesser extent. Voluntary associations foster diverse interactions that include both strong and weak ties. While members may develop recurring interactions, their engagement is often structured around shared interests or group activities rather than deep, long-standing personal connections. Additionally, by definition, HANs can involve a broader and more heterogeneous range of social ties, often including instrumental, topic-specific interactions that may be less committed, more sporadic, more targeted, and less reciprocated than deeper, multifunctional connections characterized by long-standing emotional bonds and enduring relational histories. Because of these reasons, individuals do not necessarily prioritize balance to the same extent as in close relationship triangles (Kawachi & Berkman, Reference Kawachi and Berkman2001), especially in the setting of voluntary associations. In other words, if one actor seeks advice from a second, and a third seeks advice from the same second actor, there may be little expectation that the first and third actors must also reciprocate advice-seeking between themselves. Advice relations in HANs should thus be less constrained by the need for balance, and advice imbalance should be less disruptive to the stability of these ties. In short, imbalance should more tolerable in HANs. In addition, individuals in HANs may actively seek advice from beyond their immediate ties to gain diverse perspectives and avoid redundancy (Perry & Pescosolido, Reference Perry and Pescosolido2010), resulting in greater numbers of open triangles with bridges to adjacent clusters.
Furthermore, advice exchanges are not confined to regular face-to-face interactions and can even flow between strangers (Small, Reference Small2017), although this seems to be more the exception than the rule when it comes to private conversations about sensitive health issues. Studies have demonstrated that discussion networks—with whom people discuss important matters with, which may include advice relationships—often form around immediate needs and availability, rather than preexisting strong ties. For example, in a series of studies, Small (Reference Small2017) argued that the assumption that discussion networks closely mirror networks of close ties is likely to be incorrect. Through an extensive study of a cohort of graduate students at an elite university, Small (Reference Small2017) illustrated that individuals seek out others who are readily available in their daily lives to discuss important matters, rather than solely turning to friends and family. In addition, discussion networks were found to adapt rapidly to new environments, due to the quick transformation of respondents’ obligations and routine activities (Small et al., Reference Small, Deeds Pamphile and McMahan2015). This observation underscores the fluid and dynamic nature of discussion networks, which we assume is also the case for advice networks. This is particularly important in the health context, where lower transitive closure allows HANs to be more adaptable and responsive to changing health needs and information. Such networks can quickly disseminate important health information or advice without being constrained by the rigid structures of highly transitive networks.
H1: HANs in voluntary associations are characterized by transitive closure, but to a lesser extent than networks of close relationships.
3.2. Homophily
A second recurring structural feature of close relationships is network segregation along multiple social categories, such as gender, education, and age (McPherson et al., Reference McPherson, Smith-Lovin and Cook2001). Wimmer and Lewis (Reference Wimmer and Lewis2010) argue that network segregation is constituted by several factors, such as the opportunity structure for tie formation, network endogenous processes, and a genuine preference for others from the same social category (i.e., homophily). Previous studies have provided evidence for homophilous tie formation in various settings, for example, gender homophily in school children’s friendships (Shrum et al., Reference Shrum, Cheek and Hunter1988; Stehlé et al., Reference Stehlé, Charbonnier, Picard, Cattuto and Barrat2013), racial homophily in online dating platforms (Bruch & Newman, Reference Bruch and Newman2019), or educational homophily in parental networks (Lenkewitz & Wittek, Reference Lenkewitz and Wittek2022).
In general, homophilous ties are more likely to be activated for support and advice because similarity facilitates communication, increases predictability, promotes trust and reciprocity, and reduces conflict (McPherson et al., Reference McPherson, Smith-Lovin and Cook2001; Suitor & Keeton, Reference Suitor and Keeton1997). Here, homophily is most beneficial when two individuals are similar in terms of characteristics that are relevant to the challenges or circumstances they are trying to overcome. Particularly with regard to health issues that are tied to social categories (e.g., age and gender), similarity may facilitate communication and promote trust in seeking health-related advice. For instance, in qualitative research, women were found to be more likely to turn to women than men to talk about menopause (Edwards et al., Reference Edwards, Shaw, Halton, Bailey, Wolf, Andrews and Cartwright2021). Similarly, network members of the same age may be more inclined to seek advice on the topic of an upcoming hip surgery—a relatively common treatment in aging adults.
We expect that both HANs and networks of close relationships in voluntary associations will exhibit homophily. Although voluntary associations bring together diverse individuals, health advice are likely to be more common among members who share relevant social characteristics. While some studies suggest that health advice in acute health crises may transcend social categories (Perry & Pescosolido, Reference Perry and Pescosolido2010), we test for homophilous tie formation in HANs within voluntary associations to assess its significance. As an important organizational principle of close relationship networks, homophily within voluntary associations may provide members with a sense of belonging and trust, further reinforcing the role of these institutions in facilitating health-related support and advice exchange.
H2: HANs and networks of close relationships in voluntary associations exhibit homophily with regard to gender, age, and education to a similar extent.
3.3. Network structure and health
In general, networks reflect competing preferences to associate with the most desirable individuals (e.g., Martin, Reference Martin2009). Particularly in bounded settings, such as voluntary associations, such preferences may be directed towards the most successful, the most physically attractive, or the healthiest individuals (Centola & Van De Rijt, Reference Centola and Van De Rijt2015), or more generally speaking, those with the highest status in a social group. Poor health is a stigmatized condition (Link & Phelan, Reference Link and Phelan2001), and research has found that poor health—especially if a condition is both stigmatized and visible—influences friendship choices among adolescents (Ali et al., Reference Ali, Amialchuk and Rizzo2011; Crosnoe et al., Reference Crosnoe, Frank and Mueller2008). Also, multiple studies report that older adults with depression have smaller networks (for a review, see Reiner & Steinhoff, Reference Reiner and Steinhoff2024). For several reasons, this stigmatization may result in the social isolation of those who are perceived as unhealthy.
First, unhealthy individuals might not be desirable as friends, as they cannot participate regularly in group activities (Galenkamp & Deeg, Reference Galenkamp and Deeg2016). Second, people may be reluctant to associate with those who are unhealthy and stigmatized, due to concerns about the potential impact on their own social reputation (Crosnoe et al., Reference Crosnoe, Frank and Mueller2008; Haas et al., Reference Haas, Schaefer and Kornienko2010). Third, people with poor health use strategies such as concealment and withdrawal to hide their medical condition which can also be a pathway into social isolation (Link, Reference Link1987; Link et al., Reference Link, Cullen, Struening, Shrout and Dohrenwend1989). Those in poor health may anticipate negative interactions and stigmatization, which makes them withdraw from social relationships (Link & Phelan, Reference Link and Phelan2001). Whether driven by the avoidance of others or self-withdrawal, individuals with poor health are likely to both receive and send fewer nominations for close friends in voluntary associations.
However, voluntary associations also provide opportunities for seeking health-related advice and support, which conversely may lead individuals with poor health to perceive more health-related advice and receive more nominations as health advisors. According to the Network Episode Model (Perry & Pescosolido, Reference Perry and Pescosolido2010), social ties are often activated during times of illness to provide health-related attitudes, information, and access to health services. This activation may increase individuals’ perception of available support. In addition, their experience with health issues may make them valuable sources of health advice, perhaps even facilitating expert status in the group. Thus, people in poor health are expected to be more engaged in HANs, both sending and receiving more health-related nominations as compared with close relationship nominations.
However, the sender and receiver effects in HANs and close-tie networks are likely to differ based on the type of health condition. Physical health limitations are expected to have a stronger influence on individuals’ activity in both networks. Despite general reluctance to discuss poor health (Small et al., Reference Small, Brant and Fekete2024), people with physical health problems may still be more inclined to seek advice or share experiences. In contrast, mental health conditions, often associated with stigma, withdrawal, and a reduced ability to engage in social interactions (Cacioppo & Cacioppo, Reference Cacioppo and Cacioppo2014; Link & Phelan, Reference Link and Phelan2001) may lead to a diminished capacity or desire to seek advice or support from others. Moreover, the distorted thought patterns associated with poor mental health can lead to a systematic underestimation of available social support (Beck, Reference Beck1967, Reference Beck1979). Consequently, individuals with physical health issues are expected to be more active in HANs compared to those experiencing mental health challenges.
H3a: In networks of close relationships individuals with poor health receive fewer nominations compared to individuals in good health. In contrast, individuals with poor health should receive more nominations in health advice networks. This effect is expected to be stronger among those with poor physical health than among those with poor mental health.
H3b: In networks of close relationships individuals with poor health nominate fewer nominations compared to individuals in good health. In contrast, individuals with poor health should nominate more nominations in health advice networks. This effect is expected to be stronger among those with poor physical health than among those with poor mental health.
Furthermore, several studies have demonstrated homophily in relation to health. Scholars have found that depressed adolescents often face peer avoidance, leaving them with limited friendship options aside from others who are experiencing similar mental health issues (Hogue & Steinberg, Reference Hogue and Steinberg1995; Schaefer et al., Reference Schaefer, Kornienko and Fox2011). Researchers have observed similar patterns for obese adolescents (Crosnoe et al., Reference Crosnoe, Frank and Mueller2008). In addition, Schafer (Reference Schafer2016) provides evidence that retirement residents are more likely to interact with those who share similar health statuses.
Prior research has also found that homophily yields the most benefits when it involves characteristics directly relevant to the challenges or situations people are facing. In keeping with this notion, experiential homophily (i.e., having encountered similar difficulties or situations, such as cancer) plays a bigger role in the selection of discussion partners (Thoits, Reference Thoits1986). Perry and Pescosolido (Reference Perry and Pescosolido2010) further support the idea of experiential homophily in HANs, finding that people are more likely to seek health advice from those who have faced similar mental health challenges.
H3c: Networks of close relationships and HANs show experiential homophily in voluntary associations. This effect should be more pronounced in HANs than in networks of close relationships.
4. Methods
4.1. Data
We used sociometric survey data collected from three voluntary associations in a region in Germany. Research staff initially recruited professional contacts and further used snowball sampling to gain access to three voluntary associations. We deemed only active members eligible to ensure that every member had a nonzero chance of meeting and talking to every other member. Therefore, after debriefing the association’s head of management, we excluded five permanently inactive members, as well as people who were living in institutions, far away, or abroad, and people who were unable to participate due to severe health condition. This resulted in a target sample of 143 members, ranging from 45 to 53 members per association. None of the participants were members of multiple participating associations; thus, the sample yielded three entirely nonoverlapping networks.
After the manager of each voluntary association contacted the participants, we invited the respondents to complete an online questionnaire. A digital survey was feasible because the participating associations had shifted much of their correspondence to internet-mediated communication during the COVID-19 pandemic, and nearly all participants were experienced using computers or smartphones. We offered home visits for assistance where appropriate; one participant provided their answers in a Computer Assisted Personal Interview visit. Filling in the online survey took 25.8 min, on average.
High response rates are a prerequisite for social network analysis that investigates complete networks. Therefore, we incentivized study participation with a monetary donation to the voluntary association, contingent on its members’ response rate. Specifically, each association could earn a maximum of 500 €: for an 80% response rate, an association would receive 80% of that maximum (i.e., 0.8 * 500 = 400 €). As an additional incentive, we offered to include several customized questions at the end of the survey that allowed associations to gather information regarding their topics of interest in an anonymized setting.
Data were collected between May and October 2023, with a total of 114 participants and a resulting mean response rate of 80%. Two of the three clubs consisted exclusively of men. Within the third club, 44% of the members were male. The mean age ranged from 50 to 58 years (total age range = 23–86 years), and 97% of the respondents were born in the territory of present-day Germany.Footnote 2 According to the CASMIN classification (Federal Institute for Vocational Education and Training, 2024), 24% of the respondents had low education, 38% middle education, and 38% higher education. Most of the respondents (72%) were engaged in paid work for at least 19 hours per week, net of retirement status. 17% lived on their own. Others either lived with their (marital) partner, children, parents (or in-laws), and/or another nonrelated person.
We received a positive vote from the ethics committee (University of Cologne; reference: 220036LE) prior to our data collection. We followed strict data protection guidelines and ensured informed consent.
4.2. Measures
4.2.1. Network variables
All network variables used a roster design such that respondents could select individual members from a roster of all members. To reduce respondent burden and the time required to fill in the survey, respondents were initially asked to indicate those members with whom they had ever had contact. Only members selected in that initial question were then presented in a respondent’s subsequent rosters; members who were not personally known to the respondent were filtered out. The composition of the HAN was assessed by asking respondents with whom they would be likely to talk if they had a health problem they were concerned about or if they had to make an important decision about their own medical treatment. This is a validated item from the National Social Life, Health, and Aging Project (Waite et al., Reference Waite, Laumann, Levinson, Lindau and O’Muircheartaigh2007). This was a directed network in which respondents could nominate others as advisors (i.e., they could send a tie), and they themselves could be nominated as an advisor by others (i.e., they could receive a tie).
Close relationships were operationalized as the presence of recent informal contact and positive emotion. Two binary network items were combined: respondents had also met each other outside of voluntary association events within the previous 6 months, and the other person brought them great joy or great happiness (Engstler et al., Reference Engstler, Stuth, Lozano Alcántara, Luitjens, Klaus, Schwichtenberg-Hilmert, Behaghel, Kortmann, Martin, Drewitz and Körber2022). This was a directed network with sent and received nominations as well. Other measurements of close-tie relationships, measured by combining the two network items of giving them great joy or great happiness and that they had had contact at least several times per month, did not yield to different results (see Table A3).
Kinship ties indicated whether network members were related by blood or married. Kinship was coded as present when at least one person indicated being related; hence, it was coded as an undirected network.
4.2.2. Individual variables
Age was captured with three categories: less than 45 years, 45–64 years, and 65 years and older. Gender was constructed as a binary measure, with males as the reference category. Education consisted of three categories: low, middle, and high education, in keeping with the CASMIN classification (Federal Institute for Vocational Education and Training, 2024). Poor physical health was measured with a single item regarding whether respondents had, in the previous 6 months, experienced limitations on activities they usually engage in due to a health problem. We operationalized not being strongly restricted and being severely restricted as poor physical health, whereas not being restricted served as the reference category. Poor mental health was based on the index of the Negative Affect Subscale of the Positive and Negative Affect Schedule (Crawford & Henry, Reference Crawford and Henry2004). Scores ranged from 1 to 5, with higher values indicating poorer mental health. Individuals with scores of 3 or higher were classified as having poor mental health. We controlled for respondents’ occupation, as working in the health sector and being perceived as a professional are likely to attract health advice partners. Respondents indicated whether they currently worked or had ever worked in healthcare. This resulted in a binary measure, with not having worked in healthcare being the reference category.
4.3. Analytic strategy
4.3.1. Exponential random graph models (ERGMs)
Using the R-package statnet (Handcock et al., Reference Handcock, Hunter, Butts, Goodreau and Morris2008), we modeled the structure of HANs and networks of close relationships with ERGMs, which compare the relational patterns in a network with those found in a set of simulated random networks (Lusher et al., Reference Lusher, Koskinen and Robins2013), and we tested the interplay of these two networks through entrainment effects (Yap & Harrigan, Reference Yap and Harrigan2015). In ERGMs, the more an observed network structure deviates from what would be expected by chance given all terms included in a model specification, the larger the effect and the higher its significance. These models provide a valuable method for dissecting the global structure of networks, offering insights into the underlying generative processes for tie formation while considering the influence of related factors (Lusher et al., Reference Lusher, Koskinen and Robins2013). In our study, this method allowed us to examine the formation of health advice ties while taking into account other network endogeneous processes, such as transitivity or mutual nominations, as well as individual-level characteristics, such as gender.
To capture social network mechanisms in close relationships and health advice networks, we decided to model each tie type independently and pool all three voluntary associations for each tie type. We decided to model both tie types independently because applying ERGMs for multilayer networks led to problems with convergence. Also, modeling separate ERGMs for each voluntary association was difficult due to poor model convergence. Similarly, fitting separate ERGMs and combining the results in a meta-regression was not suitable because the group-level sample size of three was small. We, therefore, combined the three respective networks into a single block-diagonal adjacency matrix prior to fitting one ERGM. This facilitated the estimation of a pooled ERGM, with the added benefit of greater statistical power (Duxbury & Wertsching, Reference Duxbury and Wertsching2023; Vega Yon et al., Reference Vega Yon, Slaughter and De La Haye2021) and ease of interpretation. To account for missing data, we applied multiple imputation techniques throughout all analytic steps, using chained equations (van Buuren & Groothuis-Oudshoorn, Reference van Buuren and Groothuis-Oudshoorn2011).
4.3.2. Average marginal effects
We used Average Marginal Effects (AMEs) to enhance the interpretability of results and to reduce bias induced by scaling (Duxbury & Wertsching, Reference Duxbury and Wertsching2023). Crucially, AMEs ensure a valid comparison of estimates of HANs with networks of close relationships and allow for a substantial interpretation of coefficients on an absolute probability scale (Duxbury, Reference Duxbury2023). To accurately compare effect sizes between HANs and networks of close ties, we interpreted AMEs in relation to the baseline probability of forming a tie. Kreager et al. (Reference Kreager, Young, Haynie, Schaefer, Bouchard and Davidson2021) recently pointed out that “AMEs differ from odds ratios in that they are on a probability scale and so their magnitudes should be interpreted relative to the baseline tie probability (i.e., network density)” (p. 59). Here, we used the average density weighted by network size, as the block diagonal estimation underestimates the overall density. Consequently, we present AMEs that have been adjusted by dividing them by the baseline probability of forming a tie, which can be interpreted as the change in the baseline tie probability when a network variable increases by one unit.
4.3.3. Goodness of fit and sensitivity analyses
We examined the goodness of fit (GOF) using statnet’s built-in GOF command for ERGMs. This procedure simulates networks based on the modeled coefficients and compares the simulated values for the edgewise-shared partner, degree distribution, and geodesic distance statistics with the respective observed values.
Further sensitivity analyses include other operationalizations of health variables and the network of close relationships. We alternatively operationalized physical health as self-rated health. Further, we built an index of emotional and social loneliness as an alternative measure for mental health. Similar to the procedure used for the mental health variable in the main analyses, we combined information on whether respondents miss the pleasure of the company of others, miss emotional security and warmth, often feel rejected, whether there are many people they can trust completely, whether there are plenty of people they can rely on when they have a problem, and whether there are enough people they feel close to. To effectively capture lonely people, we dichotomized the index, ranging from 1 to 4, using 2 as a cut-off point. We alternatively used a stricter definition of close ties that required respondents to indicate that the other person brought them great joy or great happiness (Engstler et al., Reference Engstler, Stuth, Lozano Alcántara, Luitjens, Klaus, Schwichtenberg-Hilmert, Behaghel, Kortmann, Martin, Drewitz and Körber2022) and that they had to have contact at least several times per month, whether in person, by phone, mail, email, or other means. In the sensitivity analyses, we also explored the effect of being employed at least 19 hours a week, testing for incoming and outgoing ties across both networks (M8, see Table A4 and Table A5). All sensitivity analyses suggested that the results were generally robust; these are discussed in the Appendix (see Appendix, Sensitivity analyses, Table A1, Table A2, Table A3).
4.3.4. Model specifications
Following an iterative modeling strategy (Wimmer & Lewis, Reference Wimmer and Lewis2010, p. 625), we estimated a variety of specifications under different settings for the estimation process. Through this iterative procedure, we aimed to find convergence for a given specification and aimed to achieve satisfactory GOF. Table 2 provides an overview of the different model specifications we estimated to study the structural anatomy of HANs and close-tie networks. Generally, the model with the smallest Bayesian Information Criterion (BIC) should be preferred. Ultimately, we chose M1 for both networks, as it demonstrates the best convergence and model fit, given the inclusion of all theoretically relevant parameters. The results of the other models are displayed in the Appendix, Table A4 for HANs and Table A5 for close-tie networks.
Table 2. Summary of exponential random graph model specifications
Note. X signifies whether a term was included in the respective model specification; - signifies that the model did not converge under the given specification.
Structural effects are part of every model and control for endogenous compositions. The edges term models the general tendency of respondents to nominate network members. This term counted all ties present in a network, thus representing the network’s density (cf. Smith et al., Reference Smith, McFarland, Van Tubergen and Maas2016). Because most close relationships are marked by a preference for reciprocity (Gould, Reference Gould2002), all models included the mutual term, which captured the general tendency of respondents to reciprocate the nominations they received from others. In addition, we included the geometrically weighted edgewise shared partner (gwesp) term and the geometrically weighted dyadic shared partner (gwdsp) term. The gwesp term captured transitivity, which is the tendency of actors to befriend their friends’ friends (Hunter, Reference Hunter2007). The gwdsp term captured how often pairs of nodes shared connections to the same other nodes in the network. The likelihood of a tie increased with each additional edgewise/dyadic shared partner, but the magnitude of this increase diminished with each additional shared partner. This diminishing return of additionally shared friends is represented by the gwesp/gwdsp alpha term, both of which we fix to 0.5. Throughout our iterative modeling procedure, we included the geometrically weighted indegree effect (gwideg) for all tie types to account for different activity levels between actors.
Entrainment effects modeled exogeneous effects of other tie characteristics on tie formation (i.e., whether a tie of one type predicted ties of another type; Robins & Pattison, Reference Robins and Pattison2006). To address the coevolving relationship between networks of close relationships and HANs, we introduced a close relationship entrainment effect into our model of HANs and vice versa. These effects quantified the extent to which close relationships and health advice ties co-occurred by counting directed ties of one type that coincided with nominations of another type between two actors. Furthermore, all models included a kin entrainment effect to account for being related by blood or marriage.
Node-level characteristics (dyad) modeled exogeneous effects of individual attributes on dyadic tie formation (i.e., whether the attributes combined from two individuals predicted a tie between them). Homophily included a count statistic that enumerated all same-attribute ties, with all cross-attribute ties serving as reference categories (e.g., same-gender ties vs. cross-gender ties). We included homophily terms for education, age, gender, poor physical health, and poor mental health.
Node-level characteristics (individual) modeled exogeneous effects of individual attributes on tie formation in general (i.e., whether an attribute was associated with the individual’s activity and popularity in the network). To test our theoretical expectations of individual health status, we included terms that captured whether members with poor health sent and received more or fewer nominations. These main effects were included for physical health and mental health. Furthermore, we included the same terms for each level of age, gender, and education to test for the overrepresentation of possible ties between nodes that shared an attribute (i.e., homophily). Finally, we controlled for the received nominations for those who were, at the time of data collection, or had been employed in the health sector.
5. Results
5.1. Descriptives
Table 3 presents the individual-level descriptive statistics for our analysis sample. Education and age were roughly equally distributed across clubs. Only the second voluntary association had a mixed-gender composition.
Table 3. Summary statistics for analysis sample
Our descriptive results (see Table 4) provide the first evidence for the notion that health advice and close relationships are distinct relational processes: the overlap between both network types was modest, with a Jaccard index of 0.34 (n = 188). Whereas 51% (n = 284) of all ties were exclusively close ties, 16% (n = 90) of all ties were characterized by health advice but no close relationship.
Table 4. Network descriptives: health advice networks and close relationship networks
a Numbers are weighted by sizes of the voluntary associations, except for the Jaccard Index.HAN, Health advice network; CRN, Close relationship network.
Also, health advice were sparser than close relationships, as people—on average—perceived 1.94 members of the voluntary association as health advisor (SD = 3.06) and indicated 3.3 close relationships (SD = 4.75). Transitivity was higher in close relationship networks compared to HANs across all voluntary associations (see Table 4). This descriptive finding is also confirmed visually, as clear differences in the structures between HANs and close relationship networks can be found (see Figure 1). The networks of close ties seemed denser and clustered more than did the networks of health advice. These descriptive findings support H1, which expects transitive closure to be more pronounced in networks of close relationships as compared with HANs.
Figure 1. Visual comparison network of close relationships and health advice.
The descriptive patterns (see Table 4) show age homophily to be stronger in HANs than in networks of close relationships, albeit varying degrees between the voluntary associations. The overall education homophily is stronger in close relationships networks compared to HANs, with some variability between the associations. In the mixed-gender voluntary association, gender homophily is stronger in HANs than networks of close ties. Descriptive findings barely suggest experiential homophily to be apparent in both networks, as the homophily measures based on mental or physical health are all close to zero.
5.2. Hypothesis testing
Table 5 shows the results of the ERGMs, the average marginal effects, their corresponding delta standard errors, and the scaled average marginal effects (Duxbury, Reference Duxbury2023). The theoretically relevant coefficients of the scaled AMEs are visually presented in Figure 2. Note that the confidence intervals refer to testing the predictions to be equal to 0, rather than referring to the significance level of the comparisons.
Table 5. Average marginal effects (AME) of exponential random graph models (ERGMs) for HAN and network of close ties
Note. Delta standard errors (Duxbury, Reference Duxbury2023) are reported in parentheses. Scaled AME are AME divided by the weighted network density and can be interpreted as relative changes in tie probability if a network variable increases by one unit. We multiplied scaled AME by 100 to provide a measure capturing the percentage change of the baseline probability.
† p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
5.2.1. Transitivity
In keeping with our theoretical expectations, the results indicated that both network types were marked by transitive closure. As expected and descriptively suggested, transitive closure was more pronounced in networks of close relationships as compared with HANs (H1). This indicates that potential information sharing reaches beyond immediate, local interactions in HANs.
Figure 2. Scaled AME of the health advice network and network of close ties; Note: Only theoretically relevant coefficients of M1 are displayed here; confidence intervals refer to testing the predictions to be equal to 0 and do not refer to the significance.
5.2.2. Homophily
We expected that both networks would be characterized, to a similar extent, by homophily with respect to gender, age, and education (H2). Our results indicated that HANs exhibited gender and age homophily, but not educational homophily, whereas networks of close ties did not seem to be segregated along any social category.Footnote 3 Networks of close ties initially appeared to be segregated with respect to gender; however, when examining the only mixed-gender club separately, no such effect was evident (see Table A6). This initial finding (see Table 5) was an artefact driven by the two other male-only networks.
More specifically, having the same gender increased the probability of forming a health advice tie by 31% (see Table A6), whereas being in the same age group increased the probability of forming a health advice tie by 7% (see Table 5). The gender and age homophily effects were constant, albeit varying model specifications (see Appendix, Table A4, Table A5). No educational homophily was evident in HANs.
Interestingly, age was predictive of receiving nominations in both network types. Whereas older people were more likely to be perceived as health advisor, they were less likely to be nominated as a close tie. Being 45–65 years old or older than 65 increased the probability of being perceived as health advisor by 12% or 17%, respectively, and it decreased the probability of being nominated as a close tie by 5% or 8%, respectively (see Table 5). Furthermore, women did not perceive significantly more network members as health advisors than men did, but they had a 40% higher probability of being nominated as health advisor, compared to men (see Table A6).
5.2.3. Network structure and health
When focusing on the conditions that make perceptions of health-related advice more likely, we expected individuals with poor health to have fewer network partners in close relationships but more in health advice, and that this effect would be more pronounced among those with poor physical health than those with poor mental health (H3a). Contrary to our expectations, neither individuals with poor physical nor those with poor mental health were more or less likely to be perceived as health advisor, compared to healthy individuals. However, those with poor physical health had a 4% increased probability of being nominated as a close tie. Furthermore, we expected that individuals with poor health would nominate fewer close relationships but more health advisors, although the degree of this effect would vary according to health condition (H3b). There was no association with close ties, and we found people with poor mental health to be not more or less likely to perceive others as health advisors than those in good mental health. Contrary to our expectations, less physically healthy respondents perceived significantly fewer health advice partners. Poor physical health decreased the probability of nominating health advisors by 7% (see Table 5). Additionally, we found suggestive evidence for experiential homophily among those in poor physical health (H3c). Sharing the same physical health status increases the probability of forming a health advice tie by 5% (see Table 5). However, this evidence does not necessarily hold across model specifications and should thus be interpreted as suggestive rather than definite evidence (see Table A4).
5.3. Goodness of fit and alternative model specifications
We evaluated GOF for all models by simulating networks based on estimated ERGMs and comparing their degree, edgewise-shared partner, and geodesic distance statistics with the observed statistics in the corresponding network (Hunter et al., Reference Hunter, Handcock, Butts, Goodreau and Morris2008). Figure A1 shows the model fit for the HANs and Figure A2 for the network of close relationships, respectively. In summary, results indicated that the GOF for the degree distribution, edgewise-shared partners, and geodesic distances was sufficient.
For the estimation process, we also estimated a variety of specifications under different settings. The effects were largely stable across models with different model specifications (see Appendix, Table A4, Table A5). However, models that did not account for network endogenous effects overestimated homophily effects in close-tie networks. Once we accounted for higher structural factors, the effects become insignificant. This discrepancy highlights that network-endogenous effects in sociometric data—such as mutual ties and triadic closure—play a significant role in explaining the observed patterns of homophily among close-tie networks.
6. Discussion
This study aimed to describe the self-organizing principles of HANs through a comparison with close relationship networks. Previous research has highlighted the importance of HANs to health outcomes (Perry & Pescosolido, Reference Perry and Pescosolido2010, Reference Perry and Pescosolido2015; Schafer, Reference Schafer2013) and emphasized their similarities with close relationship networks. The structural anatomy of HANs, however, has received little theoretical and empirical consideration until now.
Our study demonstrates that perceptions of health advice constitute a distinct relational process that exhibits different structural patterns than networks of close relationship. Similar to previous studies (Small, Reference Small2013), we found that a substantial share (16%) of all ties is exclusively characterized by health advice, without the presence of a close relationship. This supports the notion that advice relationships are function specific and goal oriented (Perry & Pescosolido, Reference Perry and Pescosolido2010; Small, Reference Small2013), which is to say that people would also seek advice from others with whom they have no strong personal connection. Additionally, this finding extends Small’s (Reference Small2009) argument that also non-institutional settings, such as voluntary associations, can serve as unexpected conduits for valuable resources. In this regard, voluntary associations bear similarities to urban communes (Martin et al., Reference Martin, Yeung and Zablocki2001), which provide structured yet informal social environments where relationships evolve organically and serve multiple functions beyond their explicit purpose. Individuals do not always actively seek health-related advice, yet they perceive the possibility to obtain health-related advice also in casual or situational interactions within these associations. This suggests that voluntary associations play a crucial role in expanding access to diverse health information, beyond the boundaries of close personal networks. Moreover, we found that homophily in sociodemographic characteristics and individual health is associated with variations in the tendency to perceive others as health advisors. People’s perceptions in obtaining health advice are thus shaped by their personal attributes, as well as by the social structure inherent to dyadic ties and local communities.
6.1. Theoretical implications
Based on the transitive closure common to various networks (Coleman, Reference Coleman1968; Ellwardt et al., Reference Ellwardt, Steglich and Wittek2012; Kilduff & Krackhardt, Reference Kilduff and Krackhardt1994; McFarland et al., Reference McFarland, Moody, Diehl, Smith and Thomas2014; Moody, Reference Moody2001; Vörös et al., Reference Vörös, Boda, Elmer, Hoffman, Mepham, Raabe and Stadtfeld2021; Wimmer & Lewis, Reference Wimmer and Lewis2010), we expected transitive closure in HANs, albeit to a lesser extent than in networks of close relationships, due to the broader scope of interactions (Small, Reference Small2017). In keeping with our expectations, we find that perceived health advice extends beyond close social circles. Hence, advice networks may form around needs and availability, rather than preexisting strong ties (Small, Reference Small2017; Small et al., Reference Small, Deeds Pamphile and McMahan2015). In contrast to clustered close relationship networks—reinforcing existing knowledge through tightly knit connections (Burt, Reference Burt1992; Granovetter, Reference Granovetter1973)—lower transitive closure in HANs connects a broader range of people and thereby facilitates the flow of novel and diverse advice. This is particularly beneficial where access to up-to-date, accurate, and specialized information and advice can meaningfully impact health outcomes.
Contrary to theoretical expectations (McPherson et al., Reference McPherson, Smith-Lovin and Cook2001) and previous descriptive analyses, which suggest no segregation along social categories in people experiencing an acute health crisis (Perry & Pescosolido, Reference Perry and Pescosolido2010), we found homophily with respect to gender and age in HANs. These mostly salient characteristics may serve as a proxy for shared experiences and increase comfort in seeking health advice, as previous research has indicated (Edwards et al., Reference Edwards, Shaw, Halton, Bailey, Wolf, Andrews and Cartwright2021). In addition, in our study, women were more likely to be perceived as health advisors. This is in line with previous research that identified women as more willing and effective advisors and sources of social support than men (Beutel & Marini, Reference Beutel and Marini1995; Fischer, Reference Fischer1982; Perry & Pescosolido, Reference Perry and Pescosolido2010; Wellman & Frank, Reference Wellman, Frank, Lin, Cook and Burt2001).
Our results indicating no homophily in networks of close relationships contrast with previous research on gender homophily in school settings (McMillan, Reference McMillan2022; Shrum et al., Reference Shrum, Cheek and Hunter1988; Stehlé et al., Reference Stehlé, Charbonnier, Picard, Cattuto and Barrat2013) and the workplace (Mollenhorst et al., Reference Mollenhorst, Völker and Flap2008). However, the dynamics of social network segregation may vary by context. The voluntary nature of the associations in our sample comes with less formalized, self-selected social environments, with greater individual agency in forming social connections (Rawlings et al., Reference Rawlings, Smith, Moody and McFarland2023), and—also based on the older age—members may be more open to mingling across gender boundaries. Similar to scholars who argue that ethnic segregation is an unintended byproduct of opinion homophily in schools (Stark & Flache, Reference Stark and Flache2012), in this context, too, close relationships may be driven by shared interests rather than demographic similarities. Voluntary associations may promote a more inclusive environment in which members connect through joint activities, rather than segregating along the lines of gender, age, and education.
Further, we were interested in the conditions that make health advice perceptions more likely—particularly when involving weak ties. Based on the Network Episode Model (Pescosolido, Reference Pescosolido1992), we expected individuals with poor health to receive fewer nominations and nominate fewer network partners as close relationships but perceive and be perceived more as health advisors. Our results support the notion that social integration into different networks varies by health condition. Surprisingly, we found individuals in poor health to be less likely to nominate health advisors, and this effect was more pronounced among those with poor physical health than those with poor mental health. This suggests that obtaining health advice when in poor health is not as common contexts of voluntary associations, perhaps because of the fear of stigmatization that visible illnesses carry (Link & Phelan, Reference Link and Phelan2001). Moreover, individuals with specific health problems may not perceive others as knowledgeable about their condition or may have already experienced unhelpful advice. In other words, shared activity does not automatically imply willingness—or social openness—to obtain advice about sensitive matters, even perceiving advisors.
Further, the findings imply that stigma operates differently for physical versus mental health. Individuals with poor physical health seem to be more likely to be nominated as a close tie as compared with healthy individuals, whereas there is no difference in likelihood of being nominated as a close tie between those with poor mental health and those with good mental health. This contrasts with research on adolescents, which found that health factors, particularly those that are both stigmatized and visible, influence friendship formation (Ali et al., Reference Ali, Amialchuk and Rizzo2011; Crosnoe et al., Reference Crosnoe, Frank and Mueller2008). This discrepancy in findings suggests that unlike adolescents, older adults may not view poor health as a relevant determinant of close relationships. Older individuals may be less concerned about the implications of poor health for their reputation, perhaps because physical limitations are more prevalent and socially normalized within this population. Again, voluntary associations may serve as important venues for social participation, even for those with poor health, providing a sense of inclusion and community, despite physical health challenges.
Contrary to previous research focusing on adolescents (Crosnoe et al., Reference Crosnoe, Frank and Mueller2008; Hogue & Steinberg, Reference Hogue and Steinberg1995; Schaefer et al., Reference Schaefer, Kornienko and Fox2011), retirement communities (Schafer, Reference Schafer2016), and egocentric HANs (Perry & Pescosolido, Reference Perry and Pescosolido2010), we did not find experiential homophily in the HANs. This could be a byproduct of the lower tendency of individuals in poor health to perceive health advisors. Another explanation relates to how HANs are measured—as reflecting perceived rather than received informational support. When considering perceived sensitive exchanges, individuals may not differentiate between others based on shared health status. Also recall that participation in our study and membership in these associations required a minimal level of mobility, meaning that severely impaired people were excluded. This sample selectivity may have led to a more homogeneous group in terms of health and fewer shared critical experiences.
6.2. Limitations and future research
A limitation that our investigation shares with other network studies is the fact that it is bound to a particular setting (Ellwardt et al., Reference Ellwardt, Steglich and Wittek2012; Schafer, Reference Schafer2016; Vörös et al., Reference Vörös, Boda, Elmer, Hoffman, Mepham, Raabe and Stadtfeld2021; Yap & Harrigan, Reference Yap and Harrigan2015), thus limiting the generalizability of our results. Case studies, by design, offer rich contextual insights but often do so at the expense of broad applicability. In our case, we examine members of voluntary associations, a group that is likely to be more socially integrated than the general population. Furthermore, carnival clubs may attract individuals who identify closely with local cultural and linguistic traditions, potentially reinforcing a distinctive social composition. Notably, only three percent of the study population was born outside present-day Germany, suggesting a marked underrepresentation of migrants relative to national demographics (Zensus 2022, 2024). While ethnicity is often a network-segregating factor (Glitz, Reference Glitz2014; Hu et al., Reference Hu, Moayyed and Frank2022; Kroneberg & Wittek, Reference Kroneberg and Wittek2023; Wittek et al., Reference Wittek, Kroneberg and Lämmermann2020), it appears to be a negligible factor within these carnival clubs. Although our focus was on complete networks, we lack information about other perceived advisors outside these networks, including spouses, children, or friends. Previous research has shown that when examined egocentrically, HANs often consist of core supporters (Perry & Pescosolido, Reference Perry and Pescosolido2010), suggesting that close ties are key. However, we believe in the added benefit of researching local communities beyond the personal network, because in our data, both types of ties coincided in only in 34% of cases.
Another limitation lies in the quantitative study design, which provided no data on why some people are more likely to be perceived as health advisors than others. Integrating qualitative evidence in future research may contribute to a more nuanced understanding of perceptions about, and ultimately, whom to turn to to receive informational support among older and middle-aged adults.
A third limitation is the relatively small sample size—only three voluntary associations with 45–56 members each—which affects statistical power. However, it is important to emphasize that the ERGM method relies on ties as the primary data unit. Additionally, in the early phase of social network analysis, studies with similar sample sizes successfully tested hypotheses (Breiger, Reference Breiger1974; Burt, Reference Burt1973; Freeman, Reference Freeman1978; White et al., Reference White, Boorman and Breiger1976), reinforcing the validity of our approach. Furthermore, because smaller samples make it more challenging to achieve statistical significance, any significant findings are likely to be robust, reflecting a conservative bias rather than an overestimation of effects.
A fourth limitation concerns other important mechanisms in social networks—specifically, reciprocity and popularity—that we did not explore in depth in this study (Rivera et al., Reference Rivera, Soderstrom and Uzzi2010). While we did include reciprocity in our models and considered popularity as part of our iterative modeling strategy (see Table 2), these mechanisms were not central to our analysis. Descriptively, we observed lower levels of reciprocity in HAN compared to close-tie networks, although results from more advanced models (e.g., ERGMs) were less conclusive. Ultimately, our focus was the mechanisms transitivity and homophily that showed consistency across descriptive statistics and multivariate modeling. Future research should more systematically examine how other network mechansisms shape HANs.
A fifth limitation concerns the cross-sectional nature of our analyses. The present study identified structural features of HANs and compared them with the structural anatomy of close relationships. Future research may take our study as a starting point and investigate the relational dynamics between networks and health status. For example, longitudinal models would allow to disentangle whether people influence each other in their health behaviors and outcomes or whether they select each other as advisors based on their health status. We know from previous research that changes in social networks shape individual health, and vice versa (Haas et al., Reference Haas, Schaefer and Kornienko2010; Smith & Christakis, Reference Smith and Christakis2008). Understanding these temporal dynamics could help to identify members at risk of social exclusion or unhealthy behaviors, as well as to design interventions to support healthy aging and social integration in later life.
Importantly, carnival clubs exemplify a compelling yet understudied form of voluntary, community-based participation. Although this research focuses on one specific setting, the findings likely extend to many types of voluntary associations. With tens of millions involved in Germany and the U.S. alone (AmeriCorps, 2024; Priemer et al., Reference Priemer, Bischoff, Hohendanner, Krebstakies, Rump, Schmitt and Krimmer2019), voluntary engagement affects a substantial portion of the population across the world. This widespread involvement highlights the importance of voluntary contexts for studying social networks, health, and aging—particularly as older adults often engage in formal volunteering to sustain social ties and reduce loneliness (Donnelly & Hinterlong, Reference Donnelly and Hinterlong2010; Jongenelis et al., Reference Jongenelis, Jackson, Newton and Pettigrew2022). Given their meaningful, long-term, and self-selected nature, such associations may serve as valuable sites for public health initiatives.
Taken together, we conclude that voluntary associations may exhibit unanticipated gains (Small, Reference Small2009), as they provide inclusive spaces where individuals can engage socially with both close and distant confidents without fear of being marginalized based on their health. Putnam (Reference Putnam2001) has emphasized the role of civic engagement in fostering social trust and community bonds. Voluntary associations like those in our study may help transcend traditional demographic divides and ultimately contribute to public health and social cohesion.
Acknowledgements
We would like to thank all the participants of the study for their time and contributions. Further, we would like to thank Scott Duxbury, Robert Hellpap, and Sebastian Pink for their helpful comments and suggestions.
Funding statement
This work has been funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) − 454899704 and 454899823.
Competing interests
The authors report there are no conflicts of interest to declare.
Data availability statement
The data that support the findings of this study are not publicly available due to data protection regulations. Access to these data is restricted to the research team. Requests for access to the data may be considered on a case-by-case basis and should be directed to Prof. Dr. Lea Ellwardt.
Appendix
Goodness of fit
Figure A1. Goodness of fit: Health advice network.
Figure A2. Goodness of fit: Network of close relationships.
Sensitivity analyses
A first set of sensitivity analyses used other operationalizations of health variables and the network of close relationships but the same modeling approach as the main analyses. The results suggest the results to be largely robust (see Table A1, Table A2, Table A3). The second set of sensitivity analyses tests for different model specifications, as explained in the section concerning model specifications (see Table A4 for health advice network, and Table A5 for close-tie network). The results of the mixed-gender voluntary association can be found in Table A6.
Table A1. Sensitivity analysis: Operationalization of physical health as self-rated health
Note. Delta standard errors (Duxbury, Reference Duxbury2023) are reported in parentheses. Scaled AME are AME divided by the weighted network density and can be interpreted as relative changes in tie probability if a network variable increases by one unit. We multiplied scaled AME by 100 to provide a measure capturing the percentage change of the baseline probability.
† p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
Table A2. Sensitivity analysis: Operationalization of mental health as loneliness
Note. Delta standard errors (Duxbury, Reference Duxbury2023) are reported in parentheses. Scaled AME are AME divided by the weighted network density and can be interpreted as relative changes in tie probability if a network variable increases by one unit. We multiplied scaled AME by 100 to provide a measure capturing the percentage change of the baseline probability.
† p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
Table A3. Sensitivity analysis: Operationalization of close-tie network as being in contact at least once a month and respondents indicated the other person to give them great joy or great happiness
Note. Delta standard errors (Duxbury, Reference Duxbury2023) are reported in parentheses. Scaled AME are AME divided by the weighted network density and can be interpreted as relative changes in tie probability if a network variable increases by one unit. We multiplied scaled AME by 100 to provide a measure capturing the percentage change of the baseline probability.
† p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
Table A4. Health advice network: AME estimation results of other model specifications
Note. Delta standard errors (Duxbury, Reference Duxbury2023) are reported in parentheses.
† p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
Table A5. Close-tie network: AME estimation results of other model specifications
Note. Delta standard errors (Duxbury, Reference Duxbury2023) are reported in parentheses.
† p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
Table A6. Mixed-gender voluntary association: AME estimation results
Note. Delta standard errors (Duxbury, Reference Duxbury2023) are reported in parentheses. Scaled AME are AME divided by the weighted network density and can be interpreted as relative changes in tie probability if a network variable increases by one unit. We multiplied scaled AME by 100 to provide a measure capturing the percentage change of the baseline probability.
† p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
Open access
