Difficult People: Who Is Perceived to Be Demanding in Personal Networks and Why Are They There?

Abstract

Why do people maintain ties with individuals whom they find difficult? Standard network theories imply that such alters are avoided or dropped. Drawing on a survey of over 1,100 diverse respondents who described over 12,000 relationships, we examined which among those ties respondents nominated as a person whom they “sometimes find demanding or difficult.” Those so listed composed about 15 percent of all alters in the network. After holding ego and alter traits constant, close kin, especially women relatives and aging parents, were especially likely to be named as difficult alters. Non-kin described as friends were less, and those described as co-workers more, likely to be listed only as difficult alters. These results suggest that normative and institutional constraints may force people to retain difficult and demanding alters in their networks. We also found that providing support to alters, but not receiving support from those alters, was a major source of difficulty in these relationships. Furthermore, the felt burden of providing support was not attenuated by receiving assistance, suggesting that alters involved in reciprocated exchanges were not less often labeled difficult than were those in unreciprocated ones. This study underlines the importance of constraints in personal networks.

In the development of social network analysis since the 1950s, researchers and popularizers have stressed the importance to individuals of egocentric networks as sources of emotional support, information, and material assistance in both mundane situations and emergencies. Also recognized, but trailing far behind in attention, is the importance of egocentric networks as sources of demands and difficulties. Scholars have noted that sometimes l’enfer, c’est les autres by showing that individuals are also connected to people who burden and stress them (Bertera 2005; Durden, Hill, and Angel 2007; Lee and Szinovacz 2016; Rook 1984, 1989; Thomas 2010), but they have paid far less attention to difficult ties than to supportive or even weak ties (for an overview, see Kadushin 2012). The modest literature on difficult ties within networks, although alerting us to the complexities of relationships, insufficiently identifies who tends to be felt as difficult by whom and, more generally, insufficiently explains why individuals maintain such burdensome relationships. Studies of network formation, from either an agentic or a structural perspective, typically assume that difficult ties will be avoided and eventually dissolved (Harrigan and Yap 2017), thus making their presence and persistence in networks a puzzle.

In this article, we use data from the first wave of the University of California Social Network Study (UCNets) to learn about the prevalence, attributes, and correlates of difficult ties. UCNets is an extensive egocentric network survey of some 1,150 adults in the greater San Francisco Bay Area. Respondents described their relationships with over 11,000 alters drawn from their answers to six name-eliciting questions asking about the people with whom they were involved in different spheres of activity. A major advantage of UCNets is that, in addition to these questions, the survey also asked respondents to name the people whom they found “demanding or difficult,” thus allowing us to learn about the sources of burdens and difficulties in personal networks. We argue that individuals experience constraints that press them to continue engaging with others whom they would have preferred to avoid or to disengage from. Our study focuses on two types of constraints, role- and interaction-based, and examines their association with the likelihood that different alters will be named as participants in difficult ties in a network.

PREVIOUS RESEARCH ON DIFFICULT TIES

The observation that individuals may have ties that are partly or even predominantly costly to them would not be striking to scholars who analyze whole networks, such as classrooms or work settings, where dislike, competition, and conflict are expected (e.g., Everett and Borgatti 2014; Lyons and Scott 2012). The bulk of theory and research on personal, or egocentric, networks, however, focuses on the supportive function of personal relationships (Cohen and Janicki-Deverts 2009; Fischer 1982a; House, Umberson, and Landis 1988; Thoits 2011; Wellman and Wortley 1990). These studies typically stress how family, friends, and even acquaintances assist respondents, connect them to various resources, and contribute to their physical and mental health. (For a representative overview of a vast literature, see Kawachi and Berkman 2001.) Considerably less attention, however, has been given to the role and implications of difficult relationships.

Karen Rook is one of the few scholars to have addressed this issue. In her study of elderly women, Rook (1984, 1989) found that the number of relationship burdens they reported affected the women psychologically more than did the number of supportive relationships they described (see also Lincoln 2000). Importantly, she found that the number of difficult and helpful ties the women reported were uncorrelated; indeed, specific friends and relatives could be sources of both help and strain. More recently, several surveys have asked respondents to report support or strain from specific types of alters, such as their spouses or daughters (e.g., Bertera 2005; Birditt and Antonucci 2007), or from general categories of ties, such as family and friends (e.g., Chen and Feeley 2014; Durden et al. 2007; Shaw et al. 2007; Walen and Lachman 2000). Other researchers have estimated how many people provide support versus problems or how often they do so (e.g., how many of the people you listed “get on your nerves?” [Thomas 2010]).

Additionally, studies of “psychological ambivalence” in the family literature emphasize that relationships with kin often involve simultaneously positive and negative content (e.g., Birditt and Fingerman 2013; Connidis and McMullin 2002; Fingerman, Hay, and Birditt 2004). In a recent example, Lee and Szinovacz (2016), using the 2008 Health Retirement Study, found that respondents reported the greatest mix of positive and negative evaluations for their spouses and children, then next greatest for other relatives, and least for friends. In addition, they found that negative reports about spouses and children, but not about relatives or friends, correlated with various psychological well-being measures.

Overall, these studies commonly find that respondents who give more negative reports also tend to report higher levels of stress and loneliness and lower levels of physical health and psychological well-being (see Ailshire and Burgard 2012; Lee and Szinovacz 2016; Lund et al. 2014; Rook 2003, 2015; for some complexity, see Antonucci et al. 2010). Although the association between difficult ties and health and well-being appears to be robust, the causal paths linking them have not been sufficiently studied. Difficult ties may impair well-being by increasing a person’s sense of stress and burden; or, psychologically distressed people and those with a negative outlook on life may tend to perceive others in an unpleasant way (see Vinokur, Schul, and Caplan 1987).

Despite the consensus on the correlates of difficult ties, the existing literature barely addresses the question of which types of people and what kinds of ties are felt to be difficult That is, it is important to understand not just whether individuals can identify stressful ties in their networks, but also to understand which specific alters and which specific ties—defined by relationship type and other features, such as alters’ interactions with the individual—seem to provoke stress and feelings of burden. To our knowledge, the only approximate precedent to our analysis is Leffler, Krannich, and Gillespie (1986), which asked residents of four Utah villages to name people with whom they had various positive interactions and to name those who were overly demanding, would let them down, and made them angry. Our study uses a larger and more diverse sample and expands what we know about these relationships, allowing us to address the issue of who is viewed as creating stress and burden, and then permitting us to infer answers to the broader question of why individuals would maintain such relationships. Is it, as an exchange model would suggest, because people gain more than they lose in these ties, or is it because people cannot avoid these ties?

THEORIES OF TIE FORMATION AND THE QUESTION OF DIFFICULT TIES

The question of why individuals sustain connections to people whom they consider difficult and demanding directs us to the broader, and relatively unexplored, issue of how people build and maintain personal networks. The social science literature has proposed two main complementary approaches. The first emphasizes individuals’ agentic nature and treats them as active and purposive builders of their social worlds. The second approach focuses on the structural locations in which individuals are embedded and how these locations provide opportunities for social connections. These two approaches, however, have not been applied to explain the existence and persistence of difficult ties in people’s networks. We begin by briefly reviewing the two main approaches to tie formation, address their limitations, and then discuss the idea of constraints to help resolve the puzzle of lasting difficult ties.

Tie Formation as a Purposive Process

The approach that treats tie formation as a purposive and deliberate process rests on the assumption that individuals are agentic actors who make strategic decisions regarding whom to include and whom to exclude from their networks. This approach highlights the mental process individuals engage in as they evaluate and screen potential network members. Explanations typically focus on the gains obtained from the connections or the utility of potential associates for fulfilling individual needs and interests.

The idea that people form and invest in relationships with an eye to the benefits, both extrinsic and intrinsic, they can accrue from them is most explicit in early exchange theory approaches (e.g., Blau 1986; Homans 1958; Thibaut and Kelly 1959) and more recently in the investment model put forward by theorists of social capital. As Bourdieu (1986:249) explains, “the network of relationships is the product of investment strategies, individual or collective, consciously or unconsciously aimed at establishing or reproducing social relationships that are directly usable in the short or long term” (see also Coleman 1990; Lin 2001). Tie formation and activation is thus treated as an instrumental process whose main motivation is driven by the utility of the tie. Consequently, ties based on unreciprocated exchanges, or those in which costs outweigh benefits, are likely to be discontinued (Ikkink and van Tilburg 1999).

In research on social support, this approach resonates with models of “targeted mobilization” (Small 2013) and “functional specificity” (Perry and Pescosolido 2010), which contend that individuals assess their own needs and then selectively turn to specific people to access their resources based on evaluations of how useful these people are. Consistent with this idea, research points to specialization in the provision of support by showing that different types of ties are mobilized for different types of support (e.g., Fischer 1982a; Pescosolido 1992; Wellman and Wortley 1990).

Other studies, mostly ethnographies of family and community relationships, use language that is not explicitly actor-based but that also assumes people build networks in a deliberate and purposive manner. Nelson (2005), for example, shows how the low-income single mothers she interviewed in Vermont carefully select network members by evaluating who in their social environments is most suited to fulfill their needs. Nelson refers to this screening process as “the work of sociability.” Similarly, Hansen (2005) describes a process of sifting and sorting network members, referred to as “staging networks,” that parents continuously go through in their attempt to build networks of care for their children. Hansen (2004) shows how parents assess network candidates based on their past behaviors, values, and childrearing philosophy and how they follow relatively strict rules regarding what to ask from whom, when, and under what circumstances. These studies nicely demonstrate how individuals meticulously seek to gain information about potential network members and weigh the benefits and costs of drawing someone into or excluding someone from the network (see also Domínguez and Watkins 2003; Menjívar 2000).

Tie Formation and the Context of Opportunity

The second approach focuses on the more contingent and idiosyncratic character of tie formation. It stresses that individuals often create connections with the people in their social surroundings who are accessible and available to them. Social contexts (or foci, see Feld 1981) matter much more for network inclusion than agentic search models suggest. Simple inaccessibility, for example, rules out all but a small sliver of hypothetical relationships. Consistent with recent developments in network research, this approach assumes that circumstances play a crucial role in shaping the compositional and structural features of social networks (e.g., Doreian and Conti 2012; Entwisle et al. 2007; Mollenhorst, Völker, and Flap 2014; Small 2013; Small, Pamphile, and McMahan 2015). For example, studies show how people often form quite intimate and supportive ties with others whom they barely know and have recently met, sometimes unexpectedly, in different places, such as childcare centers (Small 2009), food pantries, neighborhood clinics, homeless shelters (Desmond 2012), college classrooms (Kossinets and Watts 2009; Small 2013), beauty salons (Furman 1997), and diners (Torres forthcoming).

The broader social and institutional context is thus conceptualized as an opportunity structure for social meetings and the formation of new ties mainly because it determines the pool of potential associates and allows for the emergence of interactions (Blau and Schwartz 1984; Feld 1982; Fischer 1982a; Lazarsfeld and Merton 1954). This idea is key to research on homophily, which shows that the degree of similarity between associates is affected by the characteristics of the social contexts in which they meet and interact (e.g., Marsden 1990; McPherson and Smith-Lovin 1987; McPherson, Smith-Lovin, and Cook 2001). In support of this view, Small (2013) finds that students discuss important matters not just with others they feel close to, but also with acquaintances whom they meet in various groups and associations. Mollenhorst and colleagues (2014), using longitudinal personal network data in the Netherlands, find that many social ties are discontinued over time due of a lack of meeting opportunities.

The opportunity approach does not discredit the idea that within specific contexts individuals may select ties purposively, but its focus is elsewhere. It assumes that access, physical or social proximity, is key to the process of tie formation and maintenance.¹ Experimental studies provide evidence showing how proximity contributes to the endurance of ties by facilitating regular interactions and the development of positive sentiments and trust between individuals (Lawler 2001). Further support is provided by survey research suggesting that even in today’s digital world, propinquity is important for social interaction and the receipt of support (Mok, Wellman, and Carrasco 2010).

TOWARD UNDERSTANDING DIFFICULT TIES IN EGOCENTRIC NETWORKS

People form ties in many different ways: some are purposeful, others incidental, or even spontaneous (Small and Sukhu 2016). The two approaches outlined in the previous section help distinguish analytically between the different mechanisms but, and most importantly for the present study, neither has been used to explain the maintenance of difficult ties in networks. Both approaches implicitly assume that difficult and demanding individuals will be avoided, dropped from the network, or not recruited into it in the first place. Nelson (2005), for example, reports that mothers tend to socially disengage from people who make burdensome demands, fail to reciprocate their gestures, or are judgmental. Similarly, Desmond (2012) finds that the “disposable ties” evicted tenants create with new acquaintances, and from whom they obtain important support needed for their daily survival, are typically short-lived and frequently dissolve following unexpected crises, emergencies, and mounting relational tension (for other examples, see review in Offer 2012).

The prediction that difficult ties will be avoided or dropped is based on the idea that people can freely choose with whom they want, or do not want, to associate. The focus on individual choice is explicit in the agentic model, but it is also part of the structure of opportunity approach, which specifies that people can freely choose their associates from the pool of available candidates. Little account, however, is given to how the social environment can restrict individual freedom and constrain behavior (Emirbayer and Goodwin 1994; Granovetter 1985, 2002).² Yet people often feel pressure to continue engaging socially with others whom they would prefer to avoid or disengage from, thus further souring the relationship. Scholars often overlook this aspect of personal relationships. For example, in his study of childcare centers in New York City, Small (2009) describes how the centers serve an important brokerage function by giving mothers the opportunity to connect with each other and with other organizations through routines and activities, such as drop-off and pick-up times, parties, and fieldtrips, but he only briefly mentions instances when parents feel coerced by the centers to cooperate with parents whom they find annoying.

Seeking to understand which ties are perceived as difficult and why they are part of a network, we elaborate on the agentic and structure of opportunity approaches of tie formation by articulating two distinct types of constraints on individual choice and behavior. Consistent with the structural approach, the first type is role-based and refers to the source or context (“focus” in Feld’s terms) of the relationship between the individual and the alter, and the normative, institutional, and material limitations imposed by this relationship. Consistent with the agentic model, the second type is interaction-based and refers to the patterns of social exchange between the individual and the alter and the potential utility derived from their relationship.

STUDY HYPOTHESES

Role-Based Constraints

Individuals’ roles in a network emerge from the contexts in which they participate, and these contexts vary in their level of constraint. According to Feld (1981), highly constraining contexts produce highly interconnected networks, which are likely to engender not only positive but also negative sentiments because they tend to force people to interact with other members. Hence, and consistent with the structural model, the contexts in which individuals are embedded determine not only the pool of potential associates whom individuals can choose, but also people whom individuals have no choice but to engage with, even if they would rather not. As Brashears and Brashears (2016:23) explain, in the absence of mechanisms to “eliminate” them, such ties are likely to endure: “[N]egative ties are more likely to be found in a persistent form when interaction is unavoidable. By extension this suggests that stable negative ties will rarely exist outside of some overarching framework (e.g. a larger family grouping, a workplace).”³

Contexts can be bounding for different reasons. For example, normative pressures exist in the context of the family, which Feld identifies as one of the most constraining foci. Most family-of-origin ties are inherited and typically characterized by high emotional closeness, strong commitment, and norms of care (Bengtson 2001; Silverstein, Gans, and Yang 2006; Swartz 2009; Wellman and Wortley 1990). The high level of interconnectedness among family members also constitutes an important mechanism of norm reinforcement and social control. Families often pressure their members to comply, cooperate, and share resources and sanction those who do not abide (Coleman 1988; Portes and Sensenbrenner 1993; Stack 1974). Hence, strong normative pressures within families may force people to retain ties to relatives, especially close ones, who are viewed as difficult and burdensome (see Brashears and Brashears 2016).

Other contexts likely impose practical and material constraints that force people to preserve difficult ties. This is most evident in hierarchical contexts that create power asymmetries and where people occupy subordinate positions that make them highly dependent on others (Burt 2000; Emerson 1962). The workplace provides an excellent example of such a context: people cannot easily disengage from their supervisor or manager nor limit interactions with difficult workmates unless they quit their jobs (Levine 2013). Similarly, the complexities of moving may make it difficult for people to avoid annoying and disturbing neighbors (Goering and Feins 2003; Lee, Oropesa, and Kanan 1994).

Greater freedom, however, exists in ties originating in voluntary associations, including religious congregations. In these contexts, people can turn away relatively easily from bothersome fellow-members, or leave the church or even the denomination (Fischer and Hout 2006; Wuthnow 1988). Similarly, friendship ties are less constraining (Blieszner and Roberto 2004), particularly those that are “just” friendships and are not tangled in other role relationships (Fischer 1982b). Friendship ties may therefore be more easily disbanded if they are distressful or burdensome. Nevertheless, even though they are more voluntary in nature than kinship ties, friendships may also be subject to normative constraints. What will people say if a person drops his friend when she needs him? Here, too, normative pressures may be heightened if the friendship is part of a larger web of relationships (Feld and Carter 1998), as people care for their self-image and reputation and seek approval from others in their surroundings (Blau 1986; Podolny 1993).

In summary, the structural model predicts that for normative, institutional, material, and other reasons, people may feel pressure to retain difficult ties. We thus predict that the probability of being perceived as difficult will be greater for alters in contexts that most constrain an individual’s ability to dissolve the tie:

• Hypothesis 1: Ties to kin are more likely to be perceived as difficult than ties to non-kin.

• Hypothesis 2: Ties to non-kin in more constraining contexts are more likely to be perceived as difficult than ties to non-kin in less constraining contexts.

Interaction-Based Constraints

A second type of constraint arises from social exchanges between the individual and the alter, with some exchanges being more constraining than others. Based on the agentic model, individuals consider the utility derived from interactions with the alter—in our data, various forms of social support—and what would be lost by terminating the tie. Moreover, the receipt of support engenders a sense of obligation, commitment, and indebtedness to the alter, which pressures her to reciprocate and sustain the tie (Blau 1986; Mauss [1923] 1990; Simmel [1950] 1978). These pressures can constitute a source of constraint that further bounds the receiving individual to giving alters and restricts her ability to disengage from the difficult ones. In Emerson’s (1962) terms, obtaining support from an alter creates power-dependence relations that do not allow individuals to “overcome resistance” by simply severing ties to alters perceived to be difficult. Dependence on a tie for unequal exchanges may thus generate tension, rivalry, and conflict and lead to what Blau (1986) calls a process of “differentiation,” in which failure to reciprocate establishes a status hierarchy and claims of superiority (see also Mauss [1923] 1990).

We expect these interaction-based constraints to be stronger if the utility of the good or service rendered is high, such as in emergency situations. By contrast, the constraint will be of lower magnitude in cases that involve a lower sense of obligation, such as engaging in leisure activities. This leads to the following prediction:

• Hypothesis 3: Ties to alters who provide critical types of support are more likely to be perceived as difficult than ties to alters who provide less critical types of support.

Nevertheless, the constraining effect associated with the sense of commitment and obligation that the receipt of support creates may be attenuated if an individual also provides support to the alter. Research shows that for normative and practical reasons, people strive to maintain reciprocal relationships with others by returning favors and avoiding social debts (Blau 1986; Gouldner 1960; Plickert, Côté, and Wellman 2007; Roberto and Scott 1986; van Tilburg 1992), and failing to do so can lead to feelings of distress and guilt (Menjívar 2000; Nelson 2005; Offer 2012) and tie dissolution (Blau 1986; Ikkink and van Tilburg 1999).⁴ That is, when a person reciprocates an alter’s support, or provides more than the alter does, she is less dependent on the alter and may more easily disengage if that alter is viewed as difficult. Hence, we hypothesize the following:

• Hypothesis 4: Non-reciprocal exchange ties providing unilateral benefits to ego are more likely to be perceived as difficult than reciprocal exchange ties.

DATA AND MEASURES

UCNets Data

We use the University of California Social Networks Study (UCNets), a longitudinal egocentric network survey on personal relationships, life events, and well-being,⁵ to examine who is considered difficult and why they are part of personal networks. UCNets participants were drawn from two distinct age groups in the greater six-county San Francisco Bay Area: 50- to 70-year-olds and 21- to 30-year-olds. The study focused on these two age groups to maximize the number of key transitions and life events respondents would likely experience between survey waves. Using address-based sampling, people in the eligible age range were solicited to participate (for pay) in the three-wave survey of personal networks. This procedure sufficed with the older cohort, but it fell short with the 21- to 30-year-olds. To increase their number, we added about 300 complete surveys of respondents in that age group largely by recruiting through Facebook. As part of a mode experiment, respondents were randomly assigned to either an in-person or a web version of the survey at a 3:1 ratio. All Facebook-recruited respondents did the survey online. The two instruments were substantively identical and we control for mode effects in all analyses.

The data we use in this study are based on the first wave of UCNets, which includes 666 respondents age 50 to 70 who completed the survey and named 6,689 alters, and 480 respondents age 21 to 30 who completed the survey and named 5,064 alters ⁶ (see Appendix Table A1 for a full description of respondents’ demographic and socioeconomic characteristics). All analyses use weights that adjust for combinations of gender, age, race, Hispanic ethnicity, marital status, and education to match the corresponding age-specific population of the region.⁷

Methods: The Extended Egocentric Survey

UCNets used an extended egocentric name-eliciting method to draw a detailed map of respondents’ personal networks and collect information about their social connections. The first stage in this procedure, name-eliciting, asked respondents to name the people to whom they were connected. A major advantage of the present study is that unlike most other egocentric network studies, which are mainly based on the General Social Survey “discuss important matters” question (e.g., Brashears 2014; Marsden 1987; Small 2013), the UCNets survey generated names using seven distinct name-eliciting questions. Research shows this extended procedure yields greater reliability than do methods that use a single or restricted number of name-eliciting questions (Killworth, Shelley, and Robinson 1990; Marin 2004; Marin and Hampton 2007; McCallister and Fischer 1978).

The first six name-eliciting questions asked respondents to name the people with whom they engaged in various spheres of social activity (see details below). The last name-eliciting question asked respondents to name the people whom they “sometimes find demanding or difficult.” This question, which has rarely been used in previous egocentric network research, is the major focus of our study: it allowed us to examine the prevalence and role of difficult ties in personal networks.

The second stage in the procedure applied several name-interpreting questions to obtain descriptions of the named alters and of the ties respondents had with them, including role relationship categories, felt closeness, geographic proximity, and homophily in gender, age, religion, and race and ethnicity. In addition, the UCNets survey collected detailed information about participants’ demographic, socioeconomic, and health characteristics.

Variables

Difficult tiess

Based on respondents’ responses to the “difficult” name-eliciting question and to the six name-eliciting questions tapping engagement in various social activities, we created two distinct measures of difficult ties that serve as our dependent variables. (1) Difficult-only ties: names respondents mentioned only in response to the difficult name-eliciting question and which did not appear on any of the other six name-eliciting questions tapping social exchanges. (2) Difficult engaged-in-exchange ties: names respondents mentioned in response to both the difficult question and at least one of the other six social exchange name-eliciting questions.

Role-based constraints

We use role-relationship categories as proxies for role-based constraints. Using prefixed categories, respondents were asked to specify how they were connected to each alter whom they named. We distinguish between kin and non-kin. The kin category includes wife, husband, female romantic partner, male romantic partner, mother, father, adult daughter, adult son, sister, and brother, (other relative is the reference category). The non-kin variables include a series of dummies referring to different role relationships specifying whether the alter is a housemate, neighbor, workmate, schoolmate, churchmate, friend, or acquaintance. Note that the non-kin variables are not mutually exclusive. A person could, for example, be mentioned as both a workmate and a friend or as a brother, housemate, and friend.

Interaction-based constraints

The first six name-generating questions compiled a list of the people with whom respondents engaged in different types of social exchanges. We use these six types of exchanges as proxies for interaction-based constraints: (1) socializing—the people with whom respondents usually get together and do social activities such as visiting for meals, going to cultural events, or just hanging out; (2) confiding in—the people in whom respondents confide about relationships, important things in life, or difficult experiences; (3) advice—the people respondents turn to when seeking advice for making important decisions; (4) practical help—the people who, in the previous few months, had given respondents practical help such as doing repairs, looking after a child, or providing a ride; (5) emergency help—the people whom respondents would ask if they were seriously injured or sick and needed help for a couple of weeks with things such as preparing meals and getting around; and (6) providing support—the people whom respondents help out in different ways. Respondents could mention up to nine names in answering the socializing question and up to six names for all other questions.

Table 1 shows the distribution of all alters named in a network by role relationship and type of social exchange. Overall, respondents’ networks were quite varied. Most notably, the majority of alters (close to 60 percent) were labeled as friends. More than half of alters mentioned in a network were people with whom respondents socialized. This may be due, however, to allowing respondents to name up to nine names in this item, compared to only six names in the other name-eliciting items. About a third of alters were named as confidants, advisors, and sources of emergency help for respondents. Forty percent of alters were named as people to whom respondents provided support. Finally, we see expected age differences, for example, in the presence of spouses, parents, housemates, and schoolmates.

Table 1.

Percentages of Alters in the Network, by Cohort (Weighted)

	21- to 30-Year-Olds (n = 5,022 alters)	50- to 70-Year-Olds (n = 6,613 alters)
Role Relationship
Kin
Wife	1.2	3.1***
Husband	1.3	3.1***
Female romantic partner	1.7	.8***
Male romantic partner	3.1	.6***
Mother	8.0	2.7***
Father	5.2	1.5***
Daughter		5.2
Son		4.5
Sister	5.1	5.4
Brother	3.7	3.0*
Other female relative	4.6	7.2***
Other male relative	3.7	4.9**
Non-kin
Housemate	8.7	3.3***
Neighbor	3.1	7.3***
Workmate	9.0	8.9
Schoolmate	12.0	1.7***
Churchmate	4.9	5.7
Friend	58.9	53.9***
Acquaintance	4.1	3.5
Type of Social Exchange
Socialize	55.8	56.3
Confide	30.5	29.4
Advice	30.0	25.3***
Practical help	22.8	16.5***
Emergency help	31.5	29.3**
Provide support to alter	42.0	38.8***

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed tests) for differences between cohorts.

Controls

At the tie level, we include controls for alters’ sociodemographic characteristics, most of which are measured in terms of homophily with the respondent. These include gender (for non-kin alters), age, religion, race and ethnicity, and political orientation. Other controls refer to relationship characteristics between the respondent and the alter, such as whether the respondent had met the alter in the previous year, feels emotionally close to alter, lives with alter in the same household, lives within five minutes from alter, or lives more than one hour away from alter. Descriptive statistics for all tie-level controls are presented in Appendix Table A2. Finally, we control at the individual-level for respondents’ sociodemographic background and health.

PLAN OF ANALYSIS

The major objective of this study is to reveal which ties are perceived as difficult and why they are part of the network. To examine these questions, we use two distinct dependent variables: (1) difficult-only ties and (2) difficult engaged-in-exchange ties. We begin with basic descriptive statistics by calculating the prevalence of the two types of difficult ties and examining their distribution by role relationship and types of exchange.

In the next stage, we estimate a series of multilevel models to test the study hypotheses, positing that difficult ties are maintained in networks because of role- and interaction-based constraints (we use HLM 7.01 [Raudenbush, Bryk, and Congdon 2013]). The advantage of multilevel modeling is that, rather than using aggregated measures at the individual level or simply examining all ties together as if they were independent of each other, this method accounts for the nested structure of the data (i.e., alters or ties nested within individual networks) and the non-independence of observations within individuals, which allows us to simultaneously estimate variables at different levels (Raudenbush and Bryk 2002; van Duijn, van Busschbach, and Snijder 1999; Wellman and Frank 2001).⁸ Because our outcome is a binary variable (i.e., whether the alter is named as difficult or not), we use the binomial sampling distribution with the logit link function.

We estimate separate sets of models to predict the log-odds of the two outcomes using the role-relationship variables as proxies for role-based constraints, and the types of exchange variables as proxies for interaction-based constraints. All models control for sociodemographic characteristics of both the tie and the respondent.⁹ The first set of models predicts the log-odds that an alter would be named as a difficult-only tie as opposed to all other ties (i.e., ties that are either not difficult or difficult but otherwise engaged in the network). The latter set of models excludes the difficult-only ties and estimates the log-odds that an alter would be named as a difficult engaged-in-exchange tie as opposed to a tie that is not difficult. We estimate these two outcomes separately not only for practical reasons (i.e., the second set of models includes additional variables, measures of social exchanges), but also because we believe they are conceptually distinct. The difficult-only ties are more similar to what the literature calls “negative” ties, that is, ties characterized by antagonism and dislike that typically lead to disconnected networks (Everett and Borgatti 2014), whereas difficult engaged-in-exchange ties are more ambivalent in nature and constitute a source of both positive affect and hardship (Connidis and McMullin 2002).

THE PREVALENCE AND DISTRIBUTION OF DIFFICULT TIES

How many difficult ties do respondents have in their networks? The upper panel in Table 2 shows the overall proportion of difficult ties and then distinguishes between difficult-only and difficult engaged-in-exchange ties. These estimates are calculated at the aggregated individual-level out of the total number of ties in a respondent’s network. The vast majority of participants, about three-quarters in the young cohort and two-thirds in the older cohort, nominated at least one person in their network as difficult or demanding. The proportion of difficult ties in each network, however, is relatively small. About 16 percent of ties in young respondents’ networks, and 13 percent in older respondents’ networks, were labeled as difficult. Additionally, relatively few alters appeared only in answer to the “difficult” name-eliciting question—only 5 percent for both cohorts. Nearly 70 percent of respondents did not report any such tie. Hence, most alters named as difficult were reported as engaged in the network in some other way. Overall, these alters account for 12 and 8 percent of all ties among the younger and older cohorts, respectively.

Table 2.

Mean Proportion of Difficult Ties in the Network, by Cohort (Weighted)

	21- to 30-Year-Olds (n = 480 respondents)		50- to 70-Year-Olds (n = 666 respondents)
	Mean Proportion	% Reporting No Difficult Alters	Mean Proportion	% Reporting No Difficult Alters
Difficult Tiesa	.162 (.149)	25.3	.131*** (.129)	33.0
Difficult-only ties	.046 (.077)	69.1	.052 (.092)	67.1
Difficult engaged-in-exchange ties	.117 (.141)	42.0	.079*** (.112)	54.3

Difficult Ties by Role Relationshipb	% Difficult-Only	% Difficult Engaged-in-Exchange	% Difficult-Only	% Difficult Engaged-in-Exchange
Kin
Wife	.0	27.0	.0	13.0*
Husband	.0	20.0	.0	15.0
Female romantic partner	.0	14.0	.0	28.0
Male romantic partner	.0	19.0	.0	14.0
Mother	4.5	24.0	12.7**	29.0*
Father	5.6	13.0	2.1	24.0
Daughter	.0	.0	6.0	16.0
Son	.0	.0	5.2	19.0
Sister	5.6	30.0	10.2*	8.0***
Brother	13.1	13.0	5.7*	5.0*
Other female relative	11.0	6.0	6.5	8.0
Other male relative	5.1	7.0	5.8	5.0
Non-kin
Housemate	3.3	18.0	.7*	24.0
Neighbor	.7	5.0	2.7	3.0
Workmate	11.6	6.0	11.7	5.0
Schoolmate	2.5	9.0	3.6	4.0*
Churchmate	3.5	5.0	2.9	5.0
Friend	2.3	7.0	1.9	6.0
Acquaintance	11.7	9.0	15.5	6.0
Difficult Ties by Type of Exchangeb
Socialize		9.5		8.0**
Confide		11.8		9.0**
Advice		14.8		8.0***
Practical help		9.9		7.0*
Emergency help		15.2		11.2***
Provide support to alter		17.1		14.9**

^aAggregated person-level measures.

^bName-level measures.

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed tests) for differences between cohorts.

The middle panel in Table 2 shows, at the tie-level, the proportion of alters who were named difficult by role relationship. The first column in each cohort refers to difficult-only ties. Within the family, among 21- to 30-year-olds, brothers were the most likely to be labeled solely as difficult (13 percent). Among 50- to 70-year-olds, mothers and sisters were most likely to be labeled solely as difficult (13 and 10 percent, respectively). Interestingly, no spouses or romantic partners were mentioned in the network solely in response to the difficult name-eliciting question. This may be due to selection bias—intimate relationships that become highly strained are often terminated. Unlike other close kin relations, most notably parents and adult children, ties to spouses and intimate partners are more voluntary in nature and, despite the stress involved, may be easier to dissolve. Another possible explanation is that in a culture emphasizing marital satisfaction, people may be less inclined to report that their relationship with their intimate partner is stressful and burdensome (Hackstaff 1999). Among non-kin, workmates and acquaintances were the most likely to be named as difficult.

The second column in each cohort refers to the difficult engaged-in-exchange ties. By and large, the percentage of difficult ties among alters who engaged in social exchanges was much higher for kin, and especially close female kin, than for non-kin. This finding points to the complex and perhaps ambivalent role these alters play in personal networks. The percentage of difficult engaged-in-exchange ties was particularly high among wives (27 percent), mothers (24 percent), and sisters (30 percent) for respondents in the young cohort, and among parents (29 and 24 percent for mothers and fathers, respectively) and female romantic partners (28 percent) for respondents in the older cohort.

Finally, the lower panel in Table 2 indicates how the difficult engaged-in-exchange ties are involved in the networks. We calculated these numbers at the name-level, and they refer to the proportion of all alters mentioned in answer to each social exchange question who were also named as difficult. Overall, we find similar patterns for the two age groups. About 10 percent of alters whom respondents named as socializing partners, confidants, and providers of practical help were subsequently named as difficult. Alters who provided emergency support were slightly more likely than those who did not provide such assistance to be viewed as difficult (15 and 11 percent of alters in the young and older cohorts, respectively). This finding may reflect the type of alters who typically provide help in emergency situations (i.e., close kin). The percentage of alters considered difficult among those who provided advice was also relatively high (nearly 15 percent), but only in the young cohort. Ties in which respondents provided support to an alter were the most likely to be considered difficult: nearly 17 percent of alters to whom respondents provided support in the young cohort, and 15 percent in the older cohort, were considered difficult.

WHO ARE THE DIFFICULT TIES AND WHY DO THEY APPEAR IN NETWORKS?

Testing for Role-Based Constraints

Our first two hypotheses predict that difficult ties are more likely to be present in contexts where individuals have limited ability to exercise choice in selecting their associates or they are pressured to socially engage with them. Table 3 presents the models testing these hypotheses and shows, for each cohort, results of the two sets of multilevel regressions. The first two columns show the models predicting the log-odds that a tie would be named as difficult-only (Analyses 1 and 2 for the young and old cohorts, respectively); the last two columns show the analyses predicting the log-odds that a tie would be named as difficult engaged-in-exchange (Analyses 3 and 4 for the young and old cohorts, respectively).

Table 3.

Multilevel Results Predicting the Log-Odds an Alter Would Be Named a Difficult Tie: Coefficients (Odds Ratios in Parentheses), by Cohort (Weighted)

	Difficult-Only Tie		Difficult Engaged-in-Exchange Tie
	21- to 30-Year-Olds	50- to 70-Year-Olds	21- to 30-Year-Olds	50- to 70-Year-Olds
	Analysis 1	Analysis 2	Analysis 3	Analysis 4
Intercept	–2.265** (.104)	–2.599*** (.074)	–4.214*** (.015)	–3.728*** (.024)
Alter-Level Variables
Role Relationship
Kin
Wife			1.937** (6.935)	.658 (1.932)
Husband			1.273 (3.573)	.498 (1.645)
Female romantic partner			1.036 (2.819)	1.736** (5.676)
Male romantic partner			.856* (2.353)	.964 (2.623)
Mother	–.703 (.495)	.963* (2.620)	1.541*** (4.669)	1.848*** (6.350)
Father	–.724 (.482)	–.190 (.304)	.827 (2.287)	1.716*** (5.563)
Daughter		.914* (2.493)		1.036*** (2.819)
Son		.404 (1.497)		1.231*** (3.424)
Sister	–.120 (.887)	.820* (2.271)	2.025*** (7.576)	.691*** (1.995)
Brother	.899 (2.458)	.083 (1.087)	.875* (2.400)	.062 (1.064)
[other relative]
Non-kin
Housemate	–.587 (.556)	–1.488 (.226)	.278 (1.321)	.547 (1.727)
Neighbor	–2.265** (.104)	–.398 (.672)	–.210 (.811)	–.701 (.496)
Workmate	.844** (2.327)	.620* (1.858)	–.251 (.778)	–.149 (.862)
Schoolmate	–.403 (.668)	–.412 (.662)	.170 (1.185)	–.214 (.807)
Churchmate	–.355 (.701)	–.359 (.698)	–.182 (.834)	–.030 (.971)
Friend	–1.392*** (.249)	–1.756*** (.173)	–.244 (.784)	–.090 (.914)
Acquaintance	.453 (1.574)	.316 (1.372)	.515 (1.673)	.317 (1.373)
Type of Social Exchange
Socialize			.041 (1.042)	–.109 (.897)
Confide			–.403 (.668)	.038 (1.039)
Advice			.324 (1.383)	–.625*** (.535)
Practical help			–.001 (.993)	–.131 (.877)
Emergency help			.057 (1.058)	.198 (1.219)
Provide support to alter			1.476*** (4.376)	1.421*** (4.143)
Alter Descriptors
Female (for non-kin)	.290 (1.337)	.333 (1.395)	.519 (1.681)	–.164 (.849)
Same age	–.685 (.509)	.238 (1.268)	.563 (1.756)	.479** (1.615)
Older	–.246 (.782)	.471 (1.602)	.896 (2.449)	.387 (1.472)
Met in last year	–.173 (.841)	–.043 (.958)	–.421 (.656)	–.628 (.534)
Emotionally close	–2.242*** (.106)	–1.576*** (.207)	–.531* (.588)	–.434** (.648)
Share household	.768 (2.156)	–.115 (.891)	.276 (1.318)	.732** (2.080)
Live within five min.	–.100 (.904)	–.738** (.478)	–.068 (.934)	–.125 (.822)
Live over one hour away	.636* (1.889)	.417* (1.516)	–.129 (.879)	–.309 (.734)
Same religion	.048 (1.049)	–.311 (.733)	.110 (1.117)	.213 (1.238)
Same race	–.091 (.913)	.158 (1.171)	–.168 (.846)	–.815 (.831)
Different political opinion	.910*** (2.485)	.521** (1.683)	.602*** (1.826)	.262 (1.299)
Respondent-Level Variables
Male	–.206 (.813)	–.168 (.845)	–.377 (.686)	–.283 (.754)
Age 50 to 60		.061 (1.063)		–.205 (.815)
Asian	.279 (1.322)	–.660* (.517)	–.244 (.783)	–.282 (.754)
Latino	.248 (1.281)	.374 (1.453)	–.168 (.845)	.031 (1.031)
Black and other	.549 (1.731)	.701* (2.016)	–.201 (.818)	–.307 (.736)
Married	.329 (1.390)	.066 (1.068)	–.696* (.499)	.018 (1.018)
Partnered	.054 (1.056)	.258 (1.294)	–.278 (.758)	–.242 (.785)
Foreign-born	.473 (1.604)	.550* (1.775)	–.007 (.993)	.272 (1.312)
New town	–.191 (.826)	.086 (1.09)	–.050 (.952)	–.392 (.676)
Educ. less than BA	–.300 (.741)	–.429* (.651)	.437 (1.547)	.498** (1.646)
Educ. BA	–.115 (.891)	.055 (1.057)	.251 (1.285)	.034 (1.034)
Income low	.466 (1.594)	.315 (1.370)	.118 (1.125)	.258 (1.295)
Income med.	–.096 (.908)	.161 (1.175)	.428 (1.535)	.117 (1.124)
Health good	.215 (1.239)	–.004 (.996)	.062 (1.064)	.482** (1.619)
Health fair/bad	1.129 (.138)	.372 (1.451)	–.109 (.897)	.160 (1.173)
No health problem	–.267 (.766)	–.067 (.936)	–.147 (.863)	–.126 (.882)
Network size	.075** (1.078)	.036 (1.037)	–.065** (.937)	.007 (1.007)
Prop. of kin in network	.098 (1.103)	–.722 (.486)	–.207 (.813)	.038 (1.039)
Web	.547 (1.728)	.154 (1.166)	.453 (1.573)	.152 (1.164)
Facebook	–.366 (.693)		–.120 (.896)
Personal reference	–.284 (.752)		.156 (1.168)
N alters	5,022	6,602	4,772	6,238
N respondents	480	666	480	666
Variance component
Between-person variance	.515	.363	.208	.517
Chi-square (intercept)	596.63***	745.524**	560.077***	745.344**

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed tests).

Hypothesis 1 postulates that within a family, close kin are more likely to be perceived as difficult than are more distant relatives. Overall, our results support this prediction. Analysis 2 shows that among 50- to 70-year-olds, the log-odds of close relatives, and especially of female relatives, being named only in response to the “difficult” name-eliciting question were significantly higher than for more distant relatives. Mothers, adult daughters, and sisters were over twice as likely as more distant relatives to be named as difficult-only ties. To better illustrate this effect, for each kin category we calculated the predicted probability of being named a difficult-only tie while holding all other alter- and individual-level variables constant at their mean levels. Figure 1 presents these probabilities and shows that mother, daughters, and sisters have an approximately 10 percent probability of being perceived as a difficult-only tie. We found no such effect for respondents in the young cohort (see Analysis 1). This important age-group difference may reflect the higher demands imposed by elderly relatives on advanced middle-aged respondents, and these respondents’ lower ability to respond to them. The older cohort may thus view relatives as exclusively difficult and burdensome. Or perhaps among young respondents, close kin are more often a source of support.

Figure 1.

Predicted Probabilities of Kin Being Named as Difficult-Only Ties (50- to 70-Year-Olds)

Note: All alter- and individual-level variables are held constant at their mean levels.

*p < .05 (two-tailed test) for difference from “other relative.”

The results in Table 3 further indicate that close relatives were substantially more likely to be named as both difficult and otherwise engaged in social exchange than were more distant relatives. In the young cohort, wives, male romantic partners, mothers, brothers, and sisters had higher log-odds of being named as difficult engaged-in-exchange alters than did other relatives (Analysis 3). In the older cohort, female romantic partners, mothers, fathers, daughters, sons, and sisters had higher log-odds of being named as difficult engaged-in-exchange alters (Analysis 4). To ease interpretation, for each kin category we calculated the predicted probability of the alter being named as a difficult engaged-in-exchange tie. Figure 2 presents the results. The probability of being named a difficult engaged-in-exchange tie was highest in the young cohort for sisters (14 percent) and wives (12 percent), and in the old cohort for aging parents (approximately 12 percent for both mothers and fathers).

Figure 2.

Predicted Probabilities of Kin Being Named as Difficult Engaged-in-Exchange Ties, by Cohort

Note: All alter- and individual-level variables are held constant at their mean levels. In younger cohort: effects for mother and sister significant at p < .001; wives significant at p < .01; male romantic partner and brother significant at p < .05 (compared to other relative). In older cohort: effects for mother, father, daughter, son, and sister significant at p < .001; female romantic partner significant at p < .01 (compared to other relative).

Taken together, and consistent with Hypothesis 1, these findings suggest that relationships to close kin, perhaps due to the familial obligations these relationships entail and their level of intensity and embeddedness, are more likely to be viewed as difficult or ambivalent (i.e., involve both supportive and burdensome experiences). The results also highlight the particular role of female kin as a source of difficulty in networks.

Hypothesis 2 makes a similar prediction with respect to ties outside the family: ties in more constraining and less voluntary contexts are more likely to be named difficult. The results in Table 3 provide some support for our prediction. As expected, in both the young and old cohorts, workmates had higher, and friends lower, log-odds of being labeled as difficult-only ties (Analyses 1 and 2, respectively). Figure 3 presents the predicted probabilities we calculated for each non-kin role relationship category. Among 21- to 30-year-olds (panel A), workmates had a 15 percent probability of being named as difficult-only ties, whereas the probability for non-workmates was reduced by more than half. We find a similar trend, but of smaller magnitude, among the 50- to 70-year-olds (panel B). By contrast, the probability of friends being named as difficult-only ties was as low as 4 percent in the young cohort and 2 percent in the old cohort. The probability of alters who were not considered friends being named difficult-only was substantially higher (16 and over 10 percent in the young and old cohorts, respectively). Contrary to our expectation, neighbors were rather less likely to be named as difficult-only ties (and only in the young cohort, see Analysis 1). Furthermore, we found no effect for any of the other non-kin role relationships.

Figure 3.

Predicted Probabilities of Non-Kin Being Named as Difficult-Only Ties

Note: All alter- and individual-level variables are held constant at their mean levels.

*p < .05; **p < .01; ***p < .001 (two-tailed test).

Testing for Interaction-Based Constraints

Hypothesis 3 tests the interaction-based constraints. It posits that alters providing more critical forms of support are more likely to be named difficult than are alters who provide less critical forms of support. Overall, however, the results do not support this expectation. As Analyses 3 and 4 in Table 3 indicate, none of the types of support received from alters was significantly associated with the log-odds of an alter being labeled difficult, except for giving advice to the respondent in the old cohort (Analysis 4). By contrast, alters who were recipients of respondents’ help were important sources of difficulty in networks. Alters whom respondents helped had substantially higher log-odds of being named difficult ties compared to alters whom respondents did not name as recipients of their help. Figure 4 displays the calculated probabilities for these effects: they suggest that much of the burden in networks is driven by helping others.

Figure 4.

Predicted Probabilities of Alter Being Named a Difficult Engaged-in-Exchange Tie by Type of Social Exchange

Note: All alter- and individual-level variables are held constant at their mean levels.

***p < .001 (two-tailed test).

In additional analyses, we examined which other types of social exchange are associated with alters being perceived as difficult by excluding alters to whom respondents provided support. Because of the overall low percentages of difficult ties in these categories, we estimated a logistic regression model at the alter-level rather than a multilevel model. Figure 5 presents the results and shows the predicted probabilities that an alter would be named difficult by type of social exchange (for full results see the online supplement). In line with Hypothesis 3, the results indicate that in the older cohort (panel A), providing emergency help is significantly associated with being labeled as difficult. This finding suggests that individuals may keep critical helpers in their network even if they are difficult because of the support they may receive from them in times of need. In the younger cohort (panel B), confiding is associated with lower, and advice with higher, log-odds that an alter will be named difficult.

Figure 5.

Predicted Probabilities of Alter Being Named a Difficult Engaged-in-Exchange Tie by Type of Social Exchange for Restricted Sample (Excluding Alters Who Provided Support to the Respondent)

Note: All alter- and individual-level variables are held constant at their mean levels.

**p < .01; ***p < .001 (two-tailed test).

According to Hypothesis 4, reciprocated relationships (i.e., those in which the respondent both receives support from and provides support to the alter) should be less likely to be considered difficult than non-reciprocated relationships (i.e., those in which the respondent only receives support from the alter). Although none of the types of support received from alters were statistically significant, we were able to test this hypothesis about reciprocation because we found that providing support to an alter is associated with an increased likelihood of the alter being named a difficult tie. Thus, in the next stage, we examined whether the effect of providing support to an alter was counterbalanced by help received from the alter. We tested this by adding to the model interaction terms between the variable providing support to the alter and the five exchange variables tapping support received from the alter (see Table 4). By and large, the results do not support Hypothesis 4. For the younger cohort, receiving support from an alter did not matter much—none of the interaction terms were statistically significant. In the older cohort, only the interaction effect with receiving advice from an alter was significant. Alters to whom respondents provided support who also gave advice had lower log-odds of being labeled difficult than did alters who did not give advice but received support from respondents. None of the other interaction effects were significant.

Table 4.

Multilevel Results Predicting the Log-Odds an Alter Would Be Named a Difficult Engaged-in-Exchange Tie with Interactions between Providing Support to Alter and Support Received from Alter: Coefficients (Odds Ratios in Parentheses), by Cohort (Weighted)

	21- to 30-Year-Olds	50- to 70-Year-Olds
Providing support to alter	1.426*** (4.164)	1.732*** (5.652)
x socialize	.242 (1.273)	.033 (1.034)
x confide	–.140 (.869)	–.051 (.950)
x advice	–.171 (.843)	–.869*** (.420)
x practical help	–.001 (.999)	–.035 (.966)
x emergency help	–.008 (.992)	–.075 (.928)

Note: Controlling for alter descriptors and social exchange variables at the name-level and for sociodemographic variables, network size, and proportion of kin in network at the person-level.

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed tests).

Considering the strong effect of providing support to alters, in a post-hoc analysis we examined whether this effect varies by who the alter is. That is, we examined whether providing support to certain kinds of alters was considered especially burdensome and difficult. We tested this possibility by adding interaction terms between providing support to alter and role relationship. The results, presented in Table 5, reveal several interesting results. Among the younger cohort, providing support to wives and sisters, compared to others, was associated with higher log-odds of that woman being seen as difficult. Providing support to acquaintances was also associated with higher log-odds that the alter would be viewed as difficult, perhaps due to the low social expectations involved in relationships with acquaintances. In the older cohort, we found significant effects for parents and sons. To better illustrate these effects, we calculated their predicted probabilities (see Figure 6). Among the 50- to 70-year-olds, both mothers and fathers to whom respondents provided support had a 25 percent probability of being viewed as difficult, compared to only 8 percent for other relatives whom respondents helped. The predicted probability for female romantic partners was also 25 percent, but the effect was not statistically significant. We found similar probabilities for providing support to wives and sisters in the younger cohort. We thus conclude that the perception of difficulty associated with providing support varies by who the person is who receives that support.

Table 5.

Multilevel Results Predicting the Log-Odds an Alter Would Be Named a Difficult Engaged-in-Exchange Tie with Interactions between Providing Support to Alter and Role Relationship: Coefficients (Odds Ratios in Parentheses), by Cohort (Weighted)

	21- to 30-Year-Olds	50- to 70-Year-Olds
Providing support to alter	1.120*** (3.034)	1.414*** (4.111)
Kin
x wife	1.836*** (6.274)	–.437 (.646)
x husband	.727 (2.069)	–.438 (.645)
x female romantic partner	.483 (2.198)	1.283 (3.607)
x male romantic partner	.788 (1.621)	–.713 (.490)
x mother	.688 (1.989)	1.243** (3.465)
x father	–.444 (.641)	1.313* (3.717)
x daughter		.535 (1.708)
x son		.788** (2.199)
x sister	1.708*** (5.520)	–.089 (.915)
x brother	.378 (1.460)	–.539 (.583)
Non-kin
x housemate	–.399 (.712)	.274 (1.315)
x neighbor	–.055 (.946)	–.408 (.665)
x workmate	–.072 (.931)	.009 (1.009)
x schoolmate	.364 (1.439)	–1.048 (.351)
x churchmate	–.856 (.425)	.601 (1.824)
x friend	.020 (1.021)	–.257 (.773)
x acquaintance	1.250* (3.492)	–.025 (.976)

Note: Controlling for alter descriptors and social exchange variables at the alter-level and for sociodemographic variables, network size, and proportion of kin in network at the person-level.

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed tests).

Figure 6.

Predicted Probabilities of Alter Being Named a Difficult Engaged-in-Exchange Tie for Providing Support by Kin Categories

Note: All alter- and individual-level variables are held constant at their mean levels. In the younger cohort: effects for wife and sister significant at p < .001. In the older cohort: effects for mother and son significant at p < .01; father significant at p < .05 (compared to other relative).

DISCUSSION AND CONCLUSIONS

Previous research points to the detrimental effect of difficult ties for individuals’ health and well-being (Ailshire and Burgard 2012; Bertera 2005; Durden et al. 2007; Lee and Szinovacz 2016; Lund et al. 2014; Rook 2003, 2015), yet little is known about who people find to be difficult and burdensome, what makes relationships with these individuals difficult, and why these relationships are present in networks. The extensive multifaceted egocentric method used in UCNets allowed us to address these issues and expand on previous studies by shedding new light on the sources and mechanisms of difficulty in personal networks.

Our findings show that the vast majority of participants (about three-quarters in the younger cohort and two-thirds in the older cohort) nominated at least one person in their network as difficult or demanding. The proportion of difficult alters in a network, however, was relatively small, and even a smaller fraction were named solely as difficult alters who did not take part in any type of exchange with respondents. This finding, that most alters were not considered difficult, can be explained by people’s tendency to avoid or exclude relationships that weigh on them (Harrigan and Yap 2017; Ikking and van Tillburg 1999; Nelson 2005; Offer 2012). However, even though they constituted a minority, difficult ties did exist in networks. Theories of tie formation, in either the agentic or the structure of opportunity approach, have not been applied to explain the existence and persistence of difficult ties in personal networks. The major contribution of this study is our examination of the different types of constraints that may pressure people to interact with others whom they would otherwise prefer to avoid or disengage from.

With respect to role-based constraints, the results support Hypothesis 2: difficult alters are likely to be found in contexts where individuals have relatively low levels of freedom and where associations are less voluntary. Friends were substantially less likely, and workmates more likely, to be mentioned in networks as solely difficult. Contrary to Hypothesis 2, however, neighbors were not more likely to be seen as difficult. This result may seem surprising, given the popular view that depicts neighbors as noisy and nosy. In fact, respondents in the younger cohort viewed their neighbors as less difficult than other alters. This result may reflect younger participants’ greater physical mobility or their greater selectivity in dealing with neighbors.

As Hypothesis 1 postulates, close family appear to be a particularly constraining context. Close kin were substantially more likely to be considered difficult yet also engaged in exchange, compared to more distant kin and non-kin. Many ties with close kin include negative elements, perhaps because close family ties may, due to their nature and long duration, generate more multifaceted and intensive interactions, or because such ties are hard for individuals to drop due to strong normative constraints, or both (see the ambivalence literature, e.g., Birditt and Fingerman 2013; Connidis and McMullin 2002; Fingerman et al. 2004; Lee and Szinovacz 2016). This finding suggests that despite demographic and cultural changes of the past four decades and the prevalent discourse about the weakening of the family, expectations from kin and a sense of familial responsibility remain strong (Bengtson 2001; Connidis 2015; Johnson 2000; Silverstein et al. 2006; Swartz 2009). Analyses we conducted with a subsample from the UCNets¹⁰ support this interpretation by showing that respondents felt substantially more obligated to close kin than to more distant kin and non-kin (see results in the online supplement).

Another possible explanation is related to the level of embeddedness in networks. People may find it difficult to avoid or disengage from burdensome ties in highly interconnected contexts because, as Feld (1981) explains, the amount of constraint in such contexts makes each tie highly dependent on the entire set of relationships (for empirical support, see Burt 2000). This is most typical of close family. We found some preliminary support for this view using a subsample of names: alters named as difficult were more likely to know well most of the other alters in the subsample (i.e., they were more highly embedded) than alters who were not named as difficult. This effect was especially prevalent among close kin (see results in the online supplement).¹¹ Future research will benefit from a more comprehensive investigation of the role of network structure, including embeddedness, in the maintenance of difficult ties.

We also found that older respondents were more likely than the younger cohort to name close kin as difficult. This finding may reflect the particular status of 50- to 70-year-olds, who, sandwiched between two generations, face competing demands for their time and resources from adult children and aging parents. Studies show that today middle-aged parents continue to support their adult children for long periods of time (Furstenberg 2010; Settersten and Ray 2010; Swartz 2009). Simultaneously, because of the increase in longevity, these mature adults may also need to provide care for their elderly and often frail parents (Birditt and Fingerman 2013; Grundy and Henretta 2006). Such dual obligations to the generations above and below can be a major source of stress, conflict, and strain for 50- to 70-year-olds, especially considering that at this age they themselves may start experiencing health, economic, and other challenges that make it harder to support others.

Additionally, respondents in the older cohort were almost two times more likely to label as difficult their aging parents than their adult children, suggesting that dealing with aging parents is more burdensome than dealing with adult children. This interpretation is consistent with the intergenerational stake hypothesis, that parents are more emotionally invested in their children and report better quality relationships with them than with their parents (Giarrusso, Feng, and Bengtson 2004). The finding from the model with interaction effects in the older cohort, showing that the association between difficulty and providing assistance to aging parents is stronger than the association with providing assistance to adult children, further supports this interpretation.

Another important family-related finding indicates that female relatives were generally more likely than male relatives to be named as either difficult-only ties or as difficult engaged-in-exchange ties. This gendered pattern may be explained by women’s greater involvement in kinship networks (Fischer 1982a; Roschelle 1997; Sarkisian and Gerstel 2004; Wellman and Wortley 1990). Women typically assume the role of household managers and kin-keepers who bear the major responsibility for maintaining relationships with relatives (Gerstel and Gallagher 1993) and planning and organizing family activities and events (Daly 2002; Shaw 2008). Women’s more intensive contact and interaction with kin, and greater sense of obligation to kin, may make them more vulnerable to criticism and stress and provide fodder for tension and conflict (Connidis and McMullin 2002; Gerstel and Gallagher 1993; Offer 2014).

This study also examined interaction-based constraints for alters who were engaged in a network. The results do not support Hypothesis 3, that alters providing more critical types of support are more likely to be perceived as difficult than alters providing less critical support. Almost none of the variables tapping support received from an alter were a significant predictor of being named in the “difficult” question. By contrast, providing support to an alter was a major source of difficulty in relationships (for similar results, see Durden et al. 2007; Lincoln 2000; Rook 2015). Furthermore, we did not find evidence to support Hypothesis 4, that reciprocated exchanges would be related to reduced difficulty compared to unreciprocated exchanges. Receiving support from alters did not attenuate the effect of just providing support to them. These results seem to deviate from previous studies, which, based on the norm of reciprocity approach, find that people generally feel more distressed by ties from which they overbenefit (i.e., gave less than they received) than by ties from which they underbenefit (i.e., gave more than they received) (see review in Uehara 1995). Rather, our findings seem to suggest that underbenefiting is a more negative experience.

This interpretation, however, should be taken with much caution because our data reflect only the respondent’s perspective and not that of the alter. Considering people’s tendency to report that they give more help than they receive (Phan, Blumer, and Demaiter 2009; Uehara 1995), self-reported data can lead to biased results. Additionally, assessing the effect of reciprocity in a relationship based on self-reported information and cross-sectional data is problematic because it does not account for generalized forms of reciprocity¹² and the sometimes long time lag that takes place between receiving and returning support (Ekeh 1974; Yamagishi and Cook 1993; Sahlins 1972; Stack 1974). These two considerations are especially important when dealing with familial relationships (Hansen 2004; Nelson 2005; Plickert et al. 2007; Uehara 1990).

Several other study limitations are noteworthy. The question asking respondents to name the people they find difficult appeared last in the name-eliciting battery and was preceded by questions about positive exchanges. This may have led to the underestimation of the number of alters perceived to be difficult in a network. Additionally, although UCNets used an extensive name-eliciting methodology tapping a variety of exchanges, several social domains were not addressed. Thus, an alter named as a difficult-only tie might have, had we asked more name-eliciting questions, been named elsewhere, for example, as someone consulted about work issues. This, in turn, may have led to overestimating the number of difficult-only ties. Finally, the cross-sectional nature of the data did not allow us to examine issues of causality and track difficult ties over time. Longitudinal data, which will be available in the next waves of the UCNets, will allow us to examine which difficult ties tend to persist in networks over time and which eventually disappear.

Nevertheless, despite these limitations, our study makes an important contribution to the literature on personal networks. By focusing on role- and interaction-based constraints, an insufficiently addressed issue in previous research, we help identify who is viewed as difficult in personal networks and better understand why they are there. As such, this study provides a more comprehensive and complex view of personal networks, which is important for understanding the functioning of networks and their enduring role in the lives of individuals.

Biographies

Shira Offer is an Associate Professor at the Department of Sociology and Anthropology at Bar-Ilan University, Israel. She received her Ph.D. in Sociology from the University of Chicago. Her research focuses on personal networks, social support and reciprocity, time use and family dynamics, and work and family. Her work appears in various journals, including the American Sociological Review, Journal of Marriage and Family, Social Forces, Social Science Research, and Gender & Society. For more information and downloads visit https://biu.academia.edu/ShiraOffer

Claude S. Fischer is Professor of the Graduate School in Sociology at the University of California, Berkeley. His recent books deal with social change over the 20th century (Fischer and Hout, Century of Difference, 2006), the social history of American culture (Made in America, 2010), and developments in Americans’ personal ties (Still Connected: Family and Friends in America Since 1970 Still Connected: Family and Friends in America Since 2011). A collection of his columns for the Boston Review appeared in 2014 as Lurching Toward Happiness in America. His major current project, funded by the National Institute of Aging, is a panel study of how personal ties change.

APPENDIX

Table A1.

Percentages for Respondent-Level Variables, by Cohort (Unweighted)

	21- to 30-Year-Olds (n = 480)	50- to 70-Year-Olds (n = 666)
Male	31.3	35.7
Age 50 to 60		44.6
Race/ethnicity
White	50.0	75.2***
Asian	29.0	10.0***
Latino	14.8	6.0***
Black and other	6.3	8.7
Married	11.3	46.6***
Partnered	51. 7	16.7***
Foreign-born	19.0	13.1**
New resident in towna	59.8	5.9***
Education
Less than bachelor’s degree	23.5	29.9**
Bachelor’s degree	54.2	24.8***
More than bachelor’s degree	22.3	35.3***
Incomeb
Low (up to $35,000)	49.0	18.3***
Medium ($35,000 to $75,000)	27.1	26.9
High ($75,000 and higher)	24.0	54.8***
Self-reported health
Excellent	64.8	58.3*
Good	24.4	24.9
Fair or bad	10.8	16.8**
No health problemc	78.0	67.0***
Network sized	11.24 (4.25)	10.33*** (4.47)
Prop kin in networkd	35.62 (19.43)	40.25*** (24.02)
Web	71.0	24.0***
Facebook	59.4
Personal reference	7.3

^aLiving in current town for two or fewer years.

^bTotal household income (before taxes) for married or partnered respondents; individual income for respondent living alone or with otherwise unrelated roommates.

^cHealth problem refers to serious illness, recent hospitalization, and disability.

^dMean and standard deviation (in parentheses) calculated at the aggregated person-level.

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed tests) for differences between cohorts.

Table A2.

Alter Descriptors, Percentages by Cohort (Weighted)

	21- to 30-Year-Olds (n = 5,022 alters)	50- to 70-Year-Olds (n = 6,613 alters)
Alter Descriptors
Female (for non-kin)	32.6	35.9***
Same age	62.5	44.4***
Older	33.0	13.3***
Met in past year	15.5	5.4***
Emotionally close	45.0	46.3
Share household	13.0	9.9***
Live within five min.	20.8	22.3*
Live over one hour away	30.3	21.9***
Same religion	49.2	45.9***
Same race	65.9	71.7***
Different political opinion	22.9	18.1***

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed tests) for differences between cohorts.