|Home | About | Journals | Submit | Contact Us | Français|
The discrepancy between an individual’s loneliness and the number of connections in a social network is well documented, yet little is known about the placement of loneliness within, or the spread of loneliness through, social networks. We use network linkage data from the population-based Framingham Heart Study to trace the topography of loneliness in people’s social networks and the path through which loneliness spreads through these networks. Results indicated that loneliness occurs in clusters, extends up to three degrees of separation, is disproportionately represented at the periphery of social networks, and spreads through a contagious process. The spread of loneliness was found to be stronger than the spread of perceived social connections, stronger for friends than family members, and stronger for women than for men. The results advance our understanding of the broad social forces that drive loneliness and suggest that efforts to reduce loneliness in our society may benefit by aggressively targeting the people in the periphery to help repair their social networks and to create a protective barrier against loneliness that can keep the whole network from unraveling.
Social species do not fare well when forced to live solitary lives. Social isolation decreases lifespan of the fruit fly, Drosophilia melanogaster (Ruan & Wu, 2008); promotes the development of obesity and Type 2 diabetes in mice (Nonogaki, Nozue, & Oka, 2007); delays the positive effects of running on adult neurogenesis in rats (Stranahan, Khalil, & Gould, 2006); increases the activation of the sympatho-adrenomedullary response to an acute immobilization or cold stressor in rats (Dronjak, Gavrilovic, Filipovic, & Radojcic, 2004); decreases the expression of genes regulating glucocorticoid response in the frontal cortex of piglets (Poletto, Steibel, Siegford, & Zanella, 2006); decreases open field activity, increased basal cortisol concentrations, and decreased lymphocyte proliferation to mitogens in pigs (Kanitz, Tuchscherer, Puppe, Tuchscherer, & Stabenow, 2004); increases the 24 hr urinary catecholamines levels and evidence of oxidative stress in the aortic arch of the Watanabe Heritable Hyperlipidemic rabbit (Nation et al., 2008); increases the morning rises in cortisol in squirrel monkeys (Lyons, Ha, & Levine, 1995); and profoundly disrupts psychosexual development in rhesus monkeys (Harlow et al., 1965).
Humans, born to the longest period of abject dependency of any species and dependent on conspecifics across the lifespan to survive and prosper, do not fare well, either, whether they are living solitary lives, or whether they simply perceive they live in isolation. The average person spends about 80% of waking hours in the company of others, and the time with others is preferred to the time spent alone (Emler, 1994; Kahneman, Krueger, Schkade, Schwarz, & Stone, 2004). Social isolation, in contrast, is associated not only with lower subjective well-being (Berscheid, 1985; Burt, 1986; Myers & Diener, 1995) but with broad based-morbidity and mortality (House, Landis, & Umberson, 1988).
Humans are an irrepressibly meaning-making species, and a large literature has developed showing that perceived social isolation (i.e., loneliness) in normal samples is a more important predictor of a variety of adverse health outcomes than is objective social isolation (e.g., (Cole et al., 2007; Hawkley, Masi, Berry, & Cacioppo, 2006; Penninx et al., 1997; Seeman, 2000; Sugisawa, Liang, & Liu, 1994). In an illustrative study, Caspi et al. (Caspi, Harrington, Moffitt, Milne, & Poulton, 2006) found that loneliness in adolescence and young adulthood predicted how many cardiovascular risk factors (e.g., body mass index, waist circumference, blood pressure, cholesterol) were elevated in young adulthood, and that the number of developmental occasions (i.e., childhood, adolescence, young adulthood) at which participants were lonely predicted the number of elevated risk factors in young adulthood. Loneliness has also been associated with the progression of Alzheimer’s Disease (Wilson et al., 2007), obesity (Lauder, Mummery, Jones, & Caperchione, 2006), increased vascular resistance (Cacioppo, Hawkley, Crawford et al., 2002), elevated blood pressure (Cacioppo, Hawkley, Crawford et al., 2002; Hawkley et al., 2006), increased hypothalamic pituitary adrenocortical activity (Adam, Hawkley, Kudielka, & Cacioppo, 2006; Steptoe, Owen, Kunz-Ebrecht, & Brydon, 2004), less salubrious sleep (Cacioppo, Hawkley, Berntson et al., 2002; Pressman et al., 2005), diminished immunity (Kiecolt-Glaser et al., 1984; Pressman et al., 2005), reduction in independent living (Russell, Cutrona, De La Mora, & Wallace, 1997; Tilvis, Pitkala, Jolkkonen, & Strandberg, 2000), alcoholism (Akerlind & Hornquist, 1992), depressive symptomatology (Cacioppo et al., 2006; Heikkinen & Kauppinen, 2004), suicidal ideation and behavior (Rudatsikira, Muula, Siziya, & Twa-Twa, 2007), and mortality in older adults (Penninx et al., 1997; Seeman, 2000). Loneliness has even been associated with gene expression -- specifically, the under-expression of genes bearing anti-inflammatory glucocorticoid response elements (GREs) and over-expression of genes bearing response elements for pro-inflammatory NF-κB/Rel transcription factors (Cole et al., 2007),
Adoption and twin studies indicate that loneliness has a sizable heritable component in children (Bartels, Cacioppo, Hudziak, & Boomsma, 2008; Mcguire & Clifford, 2000) and in adults (Boomsma, Cacioppo, Slagboom, & Posthuma, 2006; Boomsma, Willemsen, Dolan, Hawkley, & Cacioppo, 2005; Boomsma, Cacioppo, Muthen, Asparouhov, & Clark, 2007). Social factors have a substantial impact on loneliness, as well, however. For instance, freshman who leave family and friends behind often feel increased social isolation when they arrive at college even though they are surrounded by large numbers of other young adults (e.g., (Cutrona, 1982; Russell, Peplau, & Cutrona, 1980). Lower levels of loneliness are associated with marriage (Hawkley, Browne, & Cacioppo, 2005; Pinquart & Sorenson, 2003), higher education (Savikko, Routasalo, Tilvis, Strandberg, & Pitkala, 2005), and higher income (Andersson, 1998; Savikko et al., 2005), whereas higher levels of loneliness are associated with living alone (Routasalo, Savikko, Tilvis, Strandberg, & Pitkala, 2006), infrequent contact with friends and family (Bondevik & Skogstad, 1998; Hawkley et al., 2005; Mullins & Dugan, 1990), dissatisfaction with living circumstances (Hector-Taylor & Adams, 1996), physical health symptoms (Hawkley et al., In press), chronic work and/or social stress (Hawkley et al., In press), small social network (Hawkley et al., 2005; Mullins & Dugan, 1990), lack of a spousal confidant (Hawkley et al., In press), marital or family conflict (Jones, 1992; Segrin, 1999), poor quality social relationships (Hawkley et al., In press; Mullins & Dugan, 1990; Routasalo et al., 2006), and divorce and widowhood (Dugan & Kivett, 1994; Dykstra & De Jong Gierveld, 1999; Holmen, Ericsson, Andersson, & Winblad, 1992; Samuelsson, Andersson, & Hagberg, 1998).
The discrepancy between an individual’s subjective report of loneliness and the reported or observed number of connections in their social network is well documented (e.g., see (Berscheid & Reis, 1998), but few details are known about the placement of loneliness within or the spread of loneliness through a social network. The association between the loneliness of individuals connected to each other, and their clustering within the network, could be attributed to at least three social psychological processes.
First, the induction hypothesis posits that the loneliness in one person contributes to or causes the loneliness in others. The emotional, cognitive, and behavioral consequences of loneliness may contribute to the induction of loneliness. For instance, emotional contagion refers to the tendency for the facial expressions, vocalizations, postures, and movements of interacting individuals to lead to a convergence of their emotions (Hatfield, Cacioppo, & Rapson, 1994). When people feel lonely, they tend to be shyer, more anxious, more hostile, more socially awkward, and lower in self esteem (e.g., (Berscheid & Reis, 1998; Cacioppo et al., 2006)). Emotional contagion could therefore contribute to the spread of loneliness to those with whom they interact. Cognitively, loneliness can affect and be affected by what one perceives and desires in their social relationships (Peplau & Perlman, 1982; Rook, 1984; Wheeler, Reis, & Nezlek, 1983). To the extent that interactions with others in an individual’s social network influences a person’s ideal or perceived interpersonal relationship, that person’s loneliness should be influenced. Behaviorally, when people feel lonely they tend to act toward others in a less trusting and more hostile fashion (e.g., (Rotenberg, 1994); cf. (Berscheid & Reis, 1998; Cacioppo & Patrick, 2008)). These behaviors, in turn, may lower the satisfaction of others with the relationship or lead to a weakening of loss of the relationship and a consequent induction of loneliness in others.
Second, the homophily hypothesis posits that lonely or non-lonely individuals choose one another as friends and become connected (i.e., the tendency of like to attract like) (Mcpherson, Smith-Lovin, & Cook, 2001). Byrne (Byrne, 1971)’s law of attraction specifies that there is a direct linear relationship between interpersonal attraction and the proportion of similar attitudes. The association between similarity and attraction is not limited to attitudes, and the characteristics on which similarity operates move from obvious characteristics (e.g., physical attractiveness) to less obvious ones (social perceptions) as relationships develop and deepen (e.g., (Neimeyer & Mitchell, 1988)). Although feelings of loneliness can be transient, stable individual differences in loneliness may have sufficiently broad effects on social cognition, emotion, and behavior to produce similarity-based social sorting.
Finally, the shared environment hypothesis posits that connected individuals jointly experience contemporaneous exposures that contribute to loneliness. Loneliness, for instance, tends to be elevated in matriculating students because for many their arrival at college is associated with a rupture of normal ties with their family and friends (Cutrona, 1982). People who interact within a social network may also be more likely to be exposed to the same social challenges and upheavals (e.g., co-residence in a dangerous neighborhood, job loss, retirement).
To distinguish among these hypotheses requires repeated measures of loneliness, longitudinal information about network ties, and information about the nature or direction of the ties (e.g., who nominated whom as a friend) (Carrington, Scott, & Wasserman, 2005; Fowler & Christakis, 2008). With the recent application of innovative research methods to network linkage data from the population-based Framingham Heart Study, these data are now available and have been used to trace the distinctive paths through which obesity (Christakis & Fowler, 2007), smoking (Christakis & Fowler, 2008), and happiness (Fowler & Christakis, 2008) spread through people’s social networks. We sought here to use these methods and data to determine the role of social network processes in loneliness, with an emphasis on determining the topography of loneliness in people’s social networks, the inter-dependence of subjective experiences of loneliness and the observed position in social networks, the path through which loneliness spreads through these networks, and factors that modulate its spread.
The Framingham Heart Study (FHS) is a population-based, longitudinal, observational cohort study that was initiated in 1948 to prospectively investigate risk factors for cardiovascular disease. Since then, it has come to be composed of four separate but related cohort populations: (1) the “Original Cohort” enrolled in 1948 (N=5,209); (2) the “Offspring Cohort” (the children of the Original Cohort and spouses of the children) enrolled in 1971 (N=5,124); (3) the “Omni Cohort” enrolled in 1994 (N=508); and (4) the “Generation 3 Cohort” (the grandchildren of the Original Cohort) enrolled beginning in 2002 (N=4,095). The Original Cohort actually captured the majority of the adult residents of Framingham in 1948, and there was little refusal to participate. The Offspring Cohort included offspring of the Original Cohort and their spouses in 1971. The supplementary, multi-ethnic Omni Cohort was initiated to reflect the increased diversity in Framingham since the inception of the Original Cohort. For the Generation 3 Cohort, Offspring Cohort participants were asked to identify all their children and the children’s spouses, and 4,095 participants were enrolled beginning in 2002. Published reports provide details about sample composition and study design for all these cohorts (Cupples & D'agnostino, 1988; Kannel, Feinleib, Mcnamara, Garrison, & Castelli, 1979; Quan et al., 1997).
Continuous surveillance and serial examinations of these cohorts provide longitudinal data. All of the participants are personally examined by FHS physicians and nurses (or, for the small minority for whom this is not possible, evaluated by telephone) and watched continuously for outcomes. The Offspring study has collected information on health events and risk factors roughly every four years. The Original Cohort has data available for roughly every two years. Importantly, even participants who migrate out of the town of Framingham (to points throughout the U.S.) remain in the study and, remarkably, come back every few years to be examined and to complete survey forms; that is, there is no necessary loss to follow-up due to out-migration in this dataset, and very little loss to follow-up for any reason (e.g., only 10 cases out of 5,124 in the Offspring Cohort have been lost).
For the purposes of the analyses reported here, exam waves for the Original cohort were aligned with those of the Offspring cohort, so that all participants in the social network were treated as having been examined at just seven waves (in the same time windows as the Offspring, as noted in Table 1a).
The Offspring Cohort is the key cohort of interest here, and it is our source of the focal participants (FPs) in our network. However, individuals to whom these FPs are linked – in any of the four cohorts – are also included in the network. These linked individuals are termed linked participants (LPs). That is, whereas FPs will come only from the Offspring Cohort, LPs are drawn from the entire set of FHS cohorts (including also the Offspring Cohort itself). Hence, the total number of individuals in the FHS social network is 12,067, since LPs identified in the Original, Generation 3, and Omni Cohorts are also included, so long as they were alive in 1971 or later. Spouses who list a different address of residence than the FP are termed non-co-resident spouses. There were 311 FP’s with non-co-resident spouses in exam 6 and 299 in exam 7.
The physical, laboratory, and survey examinations of the FHS participants provide a wide array of data. At each evaluation, participants complete a battery of questionnaires (e.g., the CES-D measure of depression and loneliness, as described below), a physician-administered medical history (including review of symptoms and hospitalizations), a physical examination administered by physicians on-site at the FHS facility, and a large variety of lab tests.
To ascertain the network ties, we computerized information from archived, handwritten documents that had not previously been used for research purposes, namely, the administrative tracking sheets used by the FHS since 1971 by personnel responsible for calling participants in order to arrange their periodic examinations. These sheets record the answers when all 5,124 of the FPs were asked to comprehensively identify relatives, friends, neighbors (based on address), co-workers (based on place of employment), and relatives who might be in a position to know where the FPs would be in two to four years. The key fact here that makes these administrative records so valuable for social network research is that, given the compact nature of the Framingham population in the period from 1971 to 2007, many of the nominated contacts were themselves also participants of one or another FHS cohort.
We have used these tracking sheets to develop network links for FHS Offspring participants to other participants in any of the four FHS cohorts. Thus, for example, it is possible to know which participants have a relationship (e.g., spouse, sibling, friend, co-worker, neighbor) with other participants. Of note, each link between two people might be identified by either party identifying the other; this observation is most relevant to the “friend” link, as we can make this link either when A nominates B as a friend, or when B nominates A (and, as discussed below, this directionality is methodologically important and might also be substantively interesting). People in any of the FHS cohorts may marry or befriend or live next to each other. Finally, given the high quality of addresses in the FHS data, the compact nature of Framingham, the wealth of information available about each participant’s residential history, and new mapping technologies, we determined who is whose neighbor, and we computed distances between individuals (Fitzpatrick & Modlin, 1986).
The measure of loneliness was derived from the Center for Epidemiological Studies Depression Scale (CES-D) administered between 1983 and 2001 at times corresponding to the 5th, 6th, and 7th examinations of the Offspring Cohort. The median year of examination for these individuals was 1986 for exam 5, 1996 for exam 6, and 2000 for exam 7. Participants are asked how often during the previous week they experienced a particular feeling, with 4 possible answers, 0–1 days, 1–2 days, 3–4 days, and 5–7 days. To convert these categories to days, we recoded these responses at the center of each range (0.5, 1.5, 3.5, and 6). Factor analyses of the items from the CES-D and the UCLA loneliness scales indicate they represent two separate factors, and the “I felt lonely” item from the CES-D scale loads on a separate factor from the depression items (Cacioppo et al., 2006). The face-valid nature of the item also supported the use of the “How often I felt lonely” item to gauge loneliness.
Table 1b shows summary statistics for loneliness, network variables, and control variables we use to study the statistical relationship between feeling lonely and being alone.
To distinguish among the induction, homophily, and shared environment hypotheses requires repeated measures of loneliness, longitudinal information about network ties, and information about the nature or direction of the ties (e.g., who nominated whom as a friend) (Carrington et al., 2005; Fowler & Christakis, 2008). For the analyses in Table 2, we averaged across waves to determine the mean number of social contacts for people in each of the four loneliness categories. For the analyses in Table 3–Table 4, we considered the prospective effect of LPs, social network variables, and other control variables on FP’s future loneliness. For the analyses in Table 5–Table 8 we conducted regressions of FP loneliness as a function of FP’s age, gender, education, and loneliness in the prior exam, and of the gender and loneliness of an LP in the current and prior exam. The lagged observations for wave 7 are from wave 6 and the lagged observations for wave 6 are from wave 5. Inclusion of FP loneliness at the prior exam eliminates serial correlation in the errors and also substantially controls for FP’s genetic endowment and any intrinsic, stable tendency to be lonely. LP’s loneliness at the prior exam helps control for homophily (Carrington et al., 2005), which has been verified in monte carlo simulations (Fowler & Christakis, 2008).
The key coefficient in these models that measures the effect of induction is on the variable for LP contemporaneous loneliness We used generalized estimating equation (GEE) procedures to account for multiple observations of the same FP across waves and across FP-LP pairings (Liang & Zeger, 1986). We assumed an independent working correlation structure for the clusters (Schildcrout & Heagerty, 2005). These analyses underlie the results presented in Figure 1–Figure 4.
The GEE regression models in the tables provide parameter estimates that are approximately interpretable as effect sizes, indicating the number of extra days of loneliness per week the FP experiences given a one unit increase in the independent variable. Mean effect sizes and 95% confidence intervals were calculated by simulating the first difference in LP contemporaneous loneliness (changing from 0.5 days feeling lonely to 1.5 days) using 1,000 randomly drawn sets of estimates from the coefficient covariance matrix and assuming all other variables are held at their means (King, Tomz, & Wittenberg, 2000). We also checked all results using an ordered logit specification and none of these models changed the significance of any reported result; we therefore decided to present the simpler and more easily interpretable linear specifications.
The regression coefficients have mostly the expected effects, such that, for example, FP’s prior loneliness is the strongest predictor for current loneliness. The models in the tables include exam fixed effects, which, combined with age at baseline, account for the aging of the population. The sample size is shown for each model, reflecting the total number of all such ties, with multiple observations for each tie if it was observed in more than one exam, and allowing for the possibility that a given person can have multiple ties. As previously indicated, repeated observations were handled with GEE procedures.
We evaluated the possibility of omitted variables or contemporaneous events explaining the associations by examining how the type or direction of the social relationship between FP and LP affects the association between FP and LP. If unobserved factors drive the association between FP and LP friendship, then directionality of friendship should not be relevant. Loneliness in the FP and the LP will move up and down together in response to the unobserved factors. In contrast, if an FP names an LP as a friend but the LP does not reciprocate, then a causal relationship would indicate that the LP would significantly affect the FP, but the FP would not necessarily affect the LP.1 The Kamada-Kawai algorithm used to prepare the images in Figure 1 generates a matrix of shortest network path distances from each node to all other nodes in the network and repositions nodes so as to reduce the sum of the difference between the plotted distances and the network distances (Kamada & Kawai, 1989). The fundamental pattern of ties in a social network (known as the “topology”) is fixed, but how this pattern is visually rendered depends on the analyst’s objectives.
In Figure 1, we show a portion of the social network, which demonstrates a clustering of moderately lonely (green nodes) and very lonely (blue nodes) people, especially at the periphery of the network. In the statistical models, the relationships between loneliness and number of social contacts proved to be negative and monotonic, as illustrated in Figure 1 and documented in Table 2.
To determine whether the clustering of lonely people shown in Figure 1 could be explained by chance, we implemented the following permutation test: we compared the observed network to 1,000 randomly generated networks in which we preserved the network topology and the overall prevalence of loneliness but in which we randomly shuffled the assignment of the loneliness value to each node (Szabo & Barabasi, 2007). For this test, we dichotomized loneliness to be 0 if the respondent said they were lonely 0–1 days the previous week, and 1 otherwise. If clustering in the social network is occurring, then the probability that an LP is lonely given that an FP is lonely should be higher in the observed network than in the random networks. This procedure also allows us to generate confidence intervals and measure how far, in terms of social distance, the correlation in loneliness between FP and LP reaches. As described below and illustrated in Figure 2, we found a significant relationship between FP and LP loneliness, and this relationship extends up to three degrees of separation. In other words, a person’s loneliness depends not just on his friend’s loneliness, but also extends to his friend’s friend and his friend’s friend’s friend. The full network shows that participants are 52% (95% C.I. 40% to 65%) more likely to be lonely if a person they are directly connected to (at one degree of separation) is lonely. The size of the effect for people at two degrees of separation (e.g., the friend of a friend) is 25% (95% C.I. 14% to 36%) and for people at three degrees of separation (e.g., the friend of a friend of a friend) is 15% (95% C.I. 6% to 26%). At four degrees of separation, the effect disappears (2%, 95% C.I. –5% to 10%), in keeping with the “three degrees of influence” rule of social network contagion that has been exhibited for obesity, smoking, and happiness (e.g., (Christakis & Fowler, 2007; Christakis & Fowler, 2008; Fowler & Christakis, 2008).
The first model in Table 3, depicted in the first three columns, shows that: (1) loneliness in the prior wave predicts loneliness in the current wave; and (2) current feelings of loneliness are much more closely tied to our networks of optional social connections, measured at the prior wave, than to those that are handed to us upon birth or to demographic features of the individuals. People with more friends are less likely to experience loneliness in the future, and each extra friend appears to reduce the frequency of feeling lonely by 0.04 days per week. That may not seem like much, but there are 52 weeks in a year, so this is equivalent to about two extra days of loneliness per year; since, on average (in our data) people feel lonely 48 days per year, having a couple of extra friends decreases loneliness by about 10% for the average person. The same model shows that the number of family members has no effect at all.
Analyses also showed that loneliness shapes social networks. Model 2 in Table 3, depicted in the middle three columns, shows that people who feel lonely at an assessment are less likely to have friends by the next assessment. In fact, compared to people who are never lonely, they will lose about 8% of their friends on average by the time they take their next exam in roughly four years. For comparison, and not surprisingly, the results depicted in the third model in Table 3 (last three columns) show that loneliness has no effect on the future number of family members a person has. These results are symmetric to both incoming and outgoing ties (not shown – available on request) – lonely people tend to receive fewer friendship nominations, but they also tend to name fewer people as friends. What this means is that loneliness is both a cause and a consequence of becoming disconnected. These results suggest that our emotions and networks reinforce each other and create a rich-gets-richer cycle that benefits those with the most friends. People with few friends are more likely to become lonelier over time, which then makes it less likely that they will attract or try to form new social ties.
We also find that social connections and the loneliness of the people to whom these connections are directed interact to affect how people feel. Figure 3 shows the smoothed bivariate relationship between the fraction of a person’s friends and family who are lonely at one exam, and the number of days per week that person feels lonely at the following exam. The relationship is significant and adds an extra quarter day of loneliness per week to the average person who is surrounded by other lonely people compared to those who are not connected to anyone who is lonely. In Table 4, we present a statistical model of the effect of lonely and non-lonely LPs on future FP loneliness that includes controls for age, education, and gender. This model shows that each additional lonely LP significantly increases the number of days a FP feels lonely at the next exam (p<0.001). Conversely, each additional non-lonely LP significantly reduces the number of days a participant feels lonely at the next exam (p=0.002). But these effects are asymmetric: lonely LPs are about two and a half times more influential than non-lonely LPs, and the difference in these effect sizes is itself significant (p=0.01). Thus, the feeling of loneliness seems to spread more easily than a feeling of belonging.
To study person-to-person effects, we examined the direct ties and individual-level determinants of FP loneliness. In the GEE models we present in Table 5–Table 9 we control for several factors as noted earlier, and the effect of social influence from one person on another is captured by the “Days/Week LP Currently Lonely” coefficient in the first row. We have highlighted in bold the social influence coefficients that are significant. Figure 4 summarizes the results from these models for friends, spouses, siblings, and neighbors. Each extra day of loneliness in a “nearby” friend (who lives within a mile) increases the number of days FP is lonely by 0.29 days (95% C.I. 0.07 to 0.50, see first model in Table 5a). In contrast, more distant friends (who live more than a mile away) have no significant effect on FP, and the effect size appears to decline with distance (second model in Table 5a). Among friends, we can distinguish additional possibilities. Since each person was asked to name a friend, and not all of these nominations were reciprocated, we have FP-perceived friends (denoted “friends”), “LP-perceived friends” (LP named FP as a friend, but not vice versa) and “mutual friends” (FP and LP nominated each other). Nearby mutual friends have a stronger effect than nearby FP-perceived friends; each day they are lonely adds 0.41 days of loneliness for the FP (95% CI: 0.14 to 0.67, see third model in the third column of Table 5a). In contrast, the influence of nearby LP-perceived friends is not significant (p=0.25, fourth model in the fourth column of Table 5a). If the associations in the social network were merely due to confounding, the significance and effect sizes for different types of friendships should be similar. That is, if some third factor were explaining both FP and LP loneliness, it should not respect the directionality or strength of the tie.
We also find significant effects for other kinds of LPs. Each day a coresident spouse is lonely yields 0.10 extra days of loneliness for the FP (95% CI: 0.02 to 0.17, fifth model in Table 5a), while non-coresident spouses have no significant effect (sixth model). Next-door neighbors who experience an extra day of loneliness increase FP’s loneliness by 0.21 days (CI 0.04 to 0.38, third model in the third column of Table 5b), but this effect quickly drops close to 0 among neighbors who live on the same block (within 25M, fourth model in Table 5b). All these relationships indicate the importance of physical proximity, and the strong influence of neighbors suggests that the spread of loneliness may possibly depend more on frequent social contact in older adults. But siblings do not appear to affect one another at all (even the ones who live nearby, see first model in Table 5b), which provides additional evidence that loneliness in older adults is about the relationships people choose rather than the relationships they inherit. And spouses appear to be an intermediate category; Table 6 shows that spouses are significantly less influential than friends in the spread of loneliness from person to person (as indicated by the significant interaction term in the first row).
Analyses separated by gender suggested that loneliness spreads more easily among women than among men, and that this holds for both friends and neighbors. As shown in the coefficients in the first row of Table 7 and Table 8, women are both more likely to be affected by the loneliness of their friends (Table 7) and neighbors (Table 8), and their loneliness is also more likely to spread to other people in their social network. The coefficients in bold show that social influence is greatest when the FP or the LP is female. Women also reported higher levels of loneliness than men. We are reporting estimates from a linear model, however, so the baseline rate of loneliness should not affect the absolute differences that we observed. (We would be more concerned about this possible effect if we were reporting odds ratios or risk ratios that are sensitive to the baseline.) In a linear model, any additive differences in baseline should be captured by the sex variable in the model, which does show a significantly higher baseline for women. However, since we include this control, the baseline difference in men and women should not affect the interpretation of the absolute number of days each additional day of loneliness experienced by an LP contributes to the loneliness experienced by an FP.
Finally, our measure of loneliness was derived from the “I feel lonely” item in the CES-D. To address whether our results would change if depression were included in the models, we created a depression index by summing the other 19 questions in the CES-D (dropping the question on loneliness). The Pearson correlation between the indices in our data is 0.566. If depression is causing the correlation in loneliness between social contacts, then the coefficient on LP loneliness should be reduced to insignificance when we add depression variables to the models in Table 5. Specifically, we add a contemporaneous and lagged variable for both FP’s and LP’s depression. The results in Table 9a and Table 9b show that there is a significant association between FP current depression and FP current loneliness (the eighth row in bold), but other depression variables have no effect and adding them to the model has little effect on the association between FP and LP loneliness. Loneliness in nearby friends, nearby mutual friends, immediate neighbors, and nearby neighbors all remain significantly associated with FP loneliness.
The present research shows that what might appear to be a quintessential individualistic experience – loneliness – is not only a function of the individual but is also a property of groups of people. People who are lonely tend to be linked to others who are lonely, an effect that is stronger for geographically proximal than distant friends yet extends up to three degrees of separation (friends’ friends’ friend) within the social network. The nature of the friendship matters, as well, in that nearby mutual friends show stronger effects than nearby ordinary friends. If some third factor were explaining both focal and linked participants’ loneliness, then loneliness should not be contingent on the different types of friendship or the directionality of the tie. These results, therefore, argue against loneliness within networks primarily reflecting shared environments.
Longitudinal analyses additionally indicated that non-lonely individuals who are around lonely individuals tend to grow lonelier over time. The longitudinal results suggest that loneliness appears in social networks through the operation of induction (e.g., contagion) rather than simply arising from lonely individuals finding themselves isolated from others and choosing to become connected to other lonely individuals (i.e., the homophily hypothesis). The present study does not permit us to identify the extent to which the emotional, cognitive, and behavioral consequences of loneliness contributed to the induction of loneliness. All three contagion processes are promoted by face-to-face communications and disclosures, especially between individuals who share close ties, and can extend to friends’ friends and beyond through a chaining of these effects. The social network pattern of loneliness and the inter-personal spread of loneliness through the network therefore appear most consistent with the induction hypothesis.
If loneliness is contagious, what if anything keeps the contagion in check? An observation by Harlow and colleagues in their studies of social isolation in rhesus monkeys offers a clue. When the isolate monkeys were reintroduced into the colony, (Harlow et al., 1965) noted that most of these isolate animals were driven off or eliminated. Our results suggest that humans may similarly drive away lonely members of their species, and that feeling socially isolated can lead to one becoming objectively isolated. Loneliness not only spreads from person to person within a social network, but it reduces the ties of these individuals to others within the network. As a result, loneliness is found in clusters within social networks, is disproportionately represented at the periphery of social networks, and threatens the cohesiveness of the network. The collective rejection of isolates observed in humans and other primates may therefore serve to protect the structural integrity of social networks.
The findings in the present study that loneliness spreads more quickly among friends than family further suggest that the rejection of isolates to protect social networks occurs more forcibly in networks that we select rather than in those we inherit. This effect may be limited to older populations, however. The mean age in our sample was 64 years, and elderly adults have been found to reduce the size of their networks to focus on those relationships that are relatively rewarding, with costly family ties among those that are trimmed (Carstensen, 2001). Although a spouse’s loneliness was related to an individual’s subsequent loneliness, friends appeared to have more impact on loneliness than spouses. The gender differences we observed may contribute to this finding. Wheeler et al. (Wheeler et al., 1983) reported that loneliness is related to how much time male and female participants interact with women each day, and we found that the spread of loneliness was stronger for women than for men. Research is needed to address whether the absence of an effect of spouses and family members on the loneliness is more typical of older than younger adults and women than men.
Fowler and Christakis (Fowler & Christakis, 2008) found that happiness also occurred in clusters and spread through networks. Several important differences have emerged in the induction of happiness and the induction of loneliness, however. First, Fowler and Christakis (2008) found happiness to be more likely to spread through social networks than unhappiness. The present research, in contrast, indicates that the spread of loneliness is more powerful than the spread of nonloneliness. Negative events typically have more powerful effects than positive events (i.e., differential reactivity; (Cacioppo & Gardner, 1999), so Fowler and Christakis’ (2008) findings about the spread of happiness through social networks is distinctive. Whereas laboratory studies are designed to gauge differential reactivity to a positive or negative event, the Fowler and Christakis (2008) study also reflects people’s differential exposure to happy and unhappy events. Thus, happiness may spread through networks more than unhappiness because people have much more frequent exposures to friends expressing happiness than unhappiness.
Loneliness does not have a bipolar opposite like happiness, but rather is like hunger, thirst, and pain in that its absence is the normal condition rather than an evocative state (Cacioppo & Patrick, 2008). Furthermore, as an aversive state, loneliness may motivate people to seek social connection (whatever the response of others to such overtures), which has the effect of increasing the likelihood that those proximal to a lonely individual will be exposed to loneliness. Together, these processes may make loneliness more contagious than nonloneliness.
A second difference between the spread of happiness and loneliness concerns the effect of gender. Fowler and Christakis (2008) found no gender differences in the spread of happiness, whereas we found that loneliness spreads much more easily among women than men. Women may be more likely to express and share their emotions with, and be more attentive to, the emotions of others (Hatfield et al., 1994), but the spread of happiness as well as loneliness should be fostered similarly among women were this a sufficient cause. There is also a stigma associated with loneliness, particularly among men; women are more likely to engage in intimate disclosures than men; and relational connectedness is more important for women than men (Brewer & Gardner, 1996; Hawkley et al., 2005; Shaver & Brennan, 1991). These processes may explain the greater spread of loneliness among women relative to men. The present results, however, clearly show that gender, like proximity and type of relationship, influences the spread of loneliness.
A limitation of all social network analyses is that the studies are necessarily bound their sample. The compact nature of the Framingham population in the period from 1971 to 2007 and the geographical proximity of the influence mitigate this constraint, but we nevertheless considered whether the results might have changed with a larger sample frame that includes all named individuals who were themselves not participants in the Framingham Heart Study. For instance, we calculated the statistical relationship between the tendency to name people outside the study and loneliness. A Pearson correlation between the number of contacts named outside the study and loneliness is not significant and actually flips signs from one exam to another (exam 6: 0.016, p=0.39; exam 7: −0.011, p=0.53). This result suggests that the sampling frame is not biasing the average level of loneliness in the target individuals we are studying.
A second possible limitation is that we included all participants in the analysis. It is possible that the death or loss of certain critical social network members during the study systematically affect how lonely FPs felt across time. To address this possibility, we restricted analysis to those individuals (both FPs and LPs) who remained alive at the end of the study. If death is the only or most important source of network loss that causes the association between FP and LP loneliness, then removing observations of people who died during the study should reduce the association to insignificance. Results of these analyses show that the restriction has no effect on the association between FP and LP loneliness. Loneliness in nearby friends, nearby mutual friends, spouses, and immediate neighbors all remain significantly associated with FP loneliness. The death of critical network members, therefore, does not appear to account for our results.
Prior research has shown that disability is a predictor of loneliness (Hawkley et al., In press). A related issue, therefore, is whether the disability status of FPs factor into our findings. To address this issue, we created a disability index by summing five questions from the Katz Index of Activities of Daily Living about the subjects’ ability to independently dress themselves, bathe themselves, eat and drink, get into and out of a chair, and use the toilet. The Pearson correlation between the indices in our data is 0.06 (n.s.). If disabilities are affecting the correlation in loneliness between social contacts, then the coefficient on LP loneliness may be reduced to insignificance when we add disability variables to the models in Table 5. Specifically, we add a contemporaneous and lagged variable for both FP’s and LP’s disability index. The results of these ancillary analyses indicated that loneliness in nearby friends, nearby mutual friends, immediate neighbors, and nearby neighbors all remain significantly associated with FP loneliness. Thus, disability does not appear to account for our findings.
In conclusion, the observation that loneliness can be passed from person to person is reminiscent of sociologist Emile Durkheim’s famous observation about suicide. He noticed that suicide rates stayed the same across time, and across groups, even though the individual members of those groups came and went. In other words, whether people took their own lives depended on the kind of society they inhabited. Although suicide, like loneliness, has often been regarded as entirely individualistic, Durkheim’s work indicates that suicide is driven in part by larger social forces. Although loneliness has a heritable component, the present study shows it also to be influenced by broader social network processes. Indeed, we detected an extraordinary pattern at the edge of the social network. On the periphery, people have fewer friends, which makes them lonely, but it also drives them to cut the few ties that they have left. But before they do, they tend to transmit the same feeling of loneliness to their remaining friends, starting the cycle anew. These reinforcing effects mean that our social fabric can fray at the edges, like a yarn that comes loose at the end of a crocheted sweater. An important implication of this finding is that interventions to reduce loneliness in our society may benefit by aggressively targeting the people in the periphery to help repair their social networks. By helping them, we might create a protective barrier against loneliness that can keep the whole network from unraveling.
The research was supported by National Institute on Aging Grants No. R01AG034052-01 (to JTC) and P01AG031093 and R01AG24448 (to NAC). Address correspondence to John T. Cacioppo, Department of Psychology, University of Chicago, Chicago, IL, 60637, ude.ogacihcu@oppoicaC; James H. Fowler, Department of Political Science, University of California, San Diego, CA 92093, ude.dscu@relwofhj; or Nicholas A. Christakis, Department of Health Care Policy, Harvard Medical School, and Department of Sociology, Harvard University, Cambridge, MA 02138, ude.dravrah.dem.pch@sikatsirhC.
1We explored the sensitivity of our results to model specification by conducting numerous other analyses each of which had various strengths and limitations, but none of which yielded substantially different results than those presented here. For example, we experimented with different error specifications. Although we identified only a single close friend for most of the FPs, we studied how multiple observations on some FPs affected the standard errors of our models. Huber-White sandwich estimates with clustering on the FPs yielded very similar results. We also tested for the presence of serial correlation in the GEE models using a Lagrange multiplier test and found none remaining after including the lagged dependent variable (Beck, 2001).
John T. Cacioppo, University of Chicago.
James H. Fowler, University of California, San Diego.
Nicholas A. Christakis, Harvard University.