Category Archives: social network

The science of friendship

Flickr/JimBoudThere is an interesting article by Robin Dunbar in The New Scientist: Dunbar’s Number was named after Robin, from his theorizing that humans only had the brain capacity to manage roughly 150 relationships, although depending on gender, social skills and personality, this number could vary from 100-250.  Dunbar observes that communication often breaks down when one exceeds 150 individuals (as evidenced in the Crimean War by the Charge of the Light Brigade) and the modern military and businesses only exceed these limits through strict hierarchies.

Dunbar theorizes that language, laughter and communal music-making evolved as a way to stay connected to a larger group of individuals than possible through physical acts like grooming. Dunbar: “[N]ot only can we speak to many people at the same time, we can also exchange information about the state of our networks in a way that other primates cannot. Gossip, I have argued, is a very human form of grooming.”  Christakis and Fowler (in the excellent book Connected) note that “…language is a less yucky and more efficient way to get to know our peers since we can talk to several friends at once but only groom them one at a time.  In fact, in a conversation with a small group, we can assess the behavior, health, aggressiveness, and altruism of several individuals simultaneously.  Plus, we can talk to someone else while engaged in another activity, like foraging for food in a refrigerator.”  Christakis and Fowler note how radical the idea is that language evolved not primarily as a way to exchange information but to maintain group cohesion.   “Dunbar estimates that language would have to be 2.8 times more efficient than grooming in order to sustain the [average] group size seen in humans” (one speaker per 2.8 listeners).

While language may have originally evolved, as per Dunbar, to maintain a slightly larger group size, once developed it was in principle possible to use language to maintain social relations on a tribal or national level.

A few other excerpts from Dunbar’s article:

Group living needn’t tax your intelligence too much. In a loose herd, cues such as body size or aggressiveness may be enough to judge whether you should challenge or steer clear of another individual. In bonded networks, however, you need to know each member’s personal characteristics and those of the friends and relations that might come to their aid. Keeping track of the ever-changing web of social relationships requires considerable mental computing power.

As a reflection of this, there is a correlation between the size of a species’ brain– in particular its neocortex– and the typical size of its social groups. In other words, brain size seems to place a limit on the number of relationships an individual can have. This link between group size and brain size is found in primates and perhaps a handful of other mammals that form bonded societies such as dolphins, dogs, horses and elephants. In all other mammals and birds, unusually large brains are found only in species that live in pair-bonded (monogamous) social groups.

As group size increases so too does the number of relationships that need servicing. Social effort is not spread evenly. Individuals put most effort into their closest relationships to ensure that these friends will help out when they need them. At the same time they maintain the coherence of the group. As a result, social networks resemble a nested hierarchy with two or three best friends linked into larger groupings of more casual friends, and weaker relationships bonding the entire group. This hierarchy typically has a scaling ratio of three– each layer of decreasing intimacy is three times larger than the one before it….

HUMAN SOCIAL NETWORKS

Our social networks can have dramatic effects on our lives. Your chances of becoming obese, giving up smoking, being happy or depressed, or getting divorced are all influenced by how many of your close friends do these things. A good social network could even help you live longer since laughing with friends triggers the release of endorphins, which seem to “tune” the immune system, making you more resilient to disease. So what factors influence the form and function that our social networks take.

In traditional societies, everyone in the community is related to everyone else, either as biological relatives or in-laws. In post-industrial societies this is no longer true– we live among strangers, some of whom become friends. As a result, our social circles really consist of two separate networks– family and friends– with roughly half drawn from each group.

Because the pull of kinship is so strong, we give priority to family, choosing to include them in our networks above unrelated individuals. Indeed, people who come from large extended families actually have fewer friends. One reason we favour kin is that they are much more likely to come to our aid when we need help than unrelated individuals, even if these are very good friends.

Family and friend relationships differ in other important ways, too. One is that friendships are very prone to decay if untended. Failure to see a friend for six months or so leaves us feeling less emotionally attached to them, causing them to drop down through the layers of our network hierarchy. Family relationships, by contrast, are incredibly resilient to neglect. As a result, the family half of our network remains constant throughout most of our lives whereas the friendship component undergoes considerable change over time, with up to 20 per cent turnover every few years.

More than 60 per cent of our social time is devoted to our five closest friends, with decreasing amounts given over to those in the layers beyond, until at the edge of the 150 layer are people we perhaps see once a year or at weddings and funerals. Nevertheless, the outer reaches of our social networks have a positive role to play. The sociologist Mark Granovetter at Stanford University in California has argued that these weak links in our social networks are especially useful in the modern world. It is through this widespread network of contacts that we find out about job vacancies and other economic or social opportunities. More importantly, perhaps, 70 per cent of us meet our romantic partners through these contacts.

Read “Getting Connected” by Robin Dunbar (New Scientist, 4/3/12)

Having few friends predicts early death as much as smoking or alcoholism

“Low social interaction as high a risk factor for early death as smoking 15 cigarettes daily or being an alcoholic, and twice the risk factor of obesity.”

Julianne Holt-Lunstad, a psychologist at BYU, published a recent meta-analysis with Timothy Smith and J. Bradley Layton (that culls from learning across 148 longitudinal health studies covering over 300,000 individuals). They showed that increased involvement in social networks on average reduces one’s chance of mortality over the period of any particular study by 50+%, a greater effect than either stopping smoking or eliminating one’s obesity/physical inactivity.

The study “Social Relationships and Mortality Risk: A Meta-analytic Review” appears in the journal PLoS Medicine.  They controlled for baseline health status,  and found consistent results for friendships with family, friends, neighbors and colleagues across age, gender, initial health status, cause of death, and follow-up period.

The life-protective benefits of friendship were strongest for complex measures of social integration and lowest for simple measures of residential status (e.g., living alone versus with others) .  In studies that had greater dimensions of social involvement (whether one was in a network, the kinds of social support one got, etc.), the life-protecting benefits of friendships were higher, likely corresponding to the multiple pathways through which friendships provide benefits.

Low social interaction, according to the authors, was as high a risk factor for early death as smoking 15 cigarettes a day or being an alcoholic.  Low social interaction was a higher risk factor than not exercising and twice as high a risk factor for early death as obesity.

Co-author Tim Smith noted: “We take relationships for granted as humans – we’re like fish that don’t notice the water….That constant interaction is not only beneficial psychologically but directly to our physical health.”

The longitudinal studies they analyzed tracked health outcomes and social interaction for a period of seven and a half years on average.

The 50% increased survival rate is quite likely an underestimate: these longitudinal studies don’t track relationship quality but only one’s inclusion in a social network, so they include negative relationships as well. Survival benefits of friendships are likely to be much higher if one could isolate only positive and healthy social relationships.

Holt-Lunstad speculated that the pathways of social relationships to improved longevity stem range from  “a calming touch to finding meaning in life.” She believes that those who are socially connected take greater responsibility for others’ and their own lives and take fewer risks.

Here is key Figure 6 from their study:

Unlike some other work, such as Eric Klinenberg’s Heat Wave, where shut-in elderly were especially at risk of death in Chicago’s 1995 heat wave, the findings of Holt-Lunstad are generalizable to all age groups.

Best friends may provide the most new and valuable info

Flickr photo of BFF from nokapixel

Mark Granovetter in a famous 1973 article “The Strength of Weak Ties” observed that it is our weaker social ties that are most likely to provide us access to information we don’t already know about: job leads, cross-fertilizing information that we can use to great advantage in our jobs, new opportunities, etc.

A recent paper by Sinan Aral and Marshall Van Alstyne says that Granovetter neglected to include frequency of contact.  Yes, our weak ties are more likely per contact to provide us new information, but we contact our strong ties so much more often that a majority of novel information actually comes through those strong and demographically similar friends.

Aral and Van Alstyne analyzed nearly a year of e-mail from an executive recruiting firm (heavily dependent on e-mail for communication and where novel information was critical in finding the right candidates) and found that those with a tighter group of friends (which they define via less network diversity) actually got a higher ratio of new information per unit of time and produced higher revenue for the firm.  As the authors hypothesized, recruiters with more diverse networks suffered a big drop in the volume of communication (what they call “channel bandwidth”).  “Interestingly however, reductions in channel bandwidth associated with greater network diversity do not seem to be driven solely by time and effort costs of network maintenance, but also by the nature of the relationships in sparse networks.” Van Alstyne concludes:  “a smaller number of high-bandwidth relationships can be good for you.”

So no need to jettison your best friends for now…

Read “Buddy System” (WIRED May 2011, by Clive Thompson)

See Sinan Aral and Marshall Van Alstyne, “Networks, Information & Brokerage: The Diversity-Bandwidth Tradeoff” (2010)

Kids vying to be seen as social influencers

Social butterfly; Flickr photo by massdistracton

Excerpt of WSJ piece on what PeerIndex calls the S&P rating of kids’ online social presence:

When Katie Miller went to Las Vegas this Thanksgiving, she tweeted about the lavish buffets and posted pictures of her seats at the aquatic spectacle “Le Reve” at the Wynn Las Vegas hotel.

A week later, the 25-year-old account executive at a public-relations firm got an email inviting her to a swanky holiday party on Manhattan’s West Side.

“At first I was confused,” Ms. Miller said. She read on to learn that she had been singled out as a “high-level influencer” by the event’s sponsors, including the Venetian and Palazzo hotels in Las Vegas, and a tech company called Klout, which ranks people based on their influence in social-media circles. “I was honored,” she said, sipping a cocktail at the $30,000 fete.

So much for wealth, looks or talent. Today, a new generation of VIPs is cultivating coolness through the world of social media. Here, ordinary folks can become “influential” overnight depending on the number and kinds of people who follow them on Twitter or comment on their Facebook pages.

People have been burnishing their online reputations for years, padding their resumes on professional networking site LinkedIn and trying to affect the search results that appear when someone Googles their names. Now, they’re targeting something once thought to be far more difficult to measure: influence over fellow consumers.

Some of the “influence” is real, but other youth are trying to game the system, befriending lots of others on Twitter in “one night stands” in the hopes of upping their own popularity and then dumping these “friends” a day later, or dramatically increasing their number of retweets in the hopes that they get greater attention or credit for Twitter traffic. Others realized that by raising the ratio of “those Twitter accounts following you” to “those Twitter accounts you follow”, they could increase their score. Companies are trying to use these services like Klout, PeerIndex, TweetLevel or Twitalyzer (which processes Twitter, Facebook and LinkedIn data) in an effort to determine which teens are popular and trusted by others.

Read “Wannabe Cool Kids Aim to Game the Web’s New Social Scorekeepers — Sites Use Secret Formulas to Rank Users’ Online ‘Influence’ From 1 to 100; ‘It’s an Ego Thing’ ” (Wall Street Journal, By Jessica E. Vascellaro, Feb. 8, 2011)

While they are in their early days, it’s not clear that any of these companies yet score accurately the true influence of youth or adults, as evidenced by how this can be gamed.

See also, “Web of Popularity Achieved by Bullying” (New York Times, by Tara Parker-Pope, 2/15/11) that notes that “students near the top of the social hierarchy are often both perpetrators and victims of aggressive behavior involving their peers.”

Companies using social capital data for betting on people’s lives

Flickr photo by idletype

The Wall Street Journal recently noted  how insurance companies (Aviva PLC, Prudential Financial, AIG) bet on whom to insure at what rates through data mining.  Much of the info gleaned from online purchases and other digital traces is more lifestyle: is the insurance applicant an athlete? a TV addict? a hunter?

But some of the information is social capital-related:

Increasingly, some gather online information, including from social-networking sites. Acxiom Corp., one of the biggest data firms, says it acquires a limited amount of “public” information from social-networking sites, helping “our clients to identify active social-media users, their favorite networks, how socially active they are versus the norm, and on what kind of fan pages they participate.”

For insurers and data-sellers alike, the new techniques could open up a regulatory can of worms. The information sold by marketing-database firms is lightly regulated. But using it in the life-insurance application process would “raise questions” about whether the data would be subject to the federal Fair Credit Reporting Act, says Rebecca Kuehn of the Federal Trade Commission’s division of privacy and identity protection. The law’s provisions kick in when “adverse action” is taken against a person, such as a decision to deny insurance or increase rates. The law requires that people be notified of any adverse action and be allowed to dispute the accuracy or completeness of data, according to the FTC.

The article also notes that Celent, an insurance consulting division of Marsh & McLennan, indicates that such online social-network data could be mined for policing fraud and in making pricing decisions: “A life insurer might want to scrutinize an applicant who reports no family history of cancer, but indicates online an affinity with a cancer-research group, says Mike Fitzgerald, a Celent senior analyst.  ‘Whether people actually realize it or not, they are significantly increasing their personal transparency,’ he says. ‘It’s all public, and it’s electronically mineable.’  “

We’ve written earlier about other life insurers using social capital data in making insurance decisions, but in those cases, the individual was being asked directly about his social and civic involvement.  [See also this blog post about social capital and healthcare.]

We applaud the life insurers for coming to the late realization that social capital data is strongly related to health, but strongly believe they should be more transparent about what they are doing.  Then it wouldn’t violate privacy concerns and it would have the added benefit of making the insured better aware of the positive health impact of being more involved civicly and socially, which might actually induce those who are less engaged to become more so.

See earlier blog post on loss of digital privacy and digital traces left online.

Read “Insurers Test Data Profiles to Identify Risky Clients” (Wall St. Journal, 11/17/2010, by Leslie Scism and Mark Maremount)

The Friendship Paradox: using social networks to predict spread of epidemics

Nick Christakis and James Fowler (whose research we’ve previously highlighted) is back with research that shows how one can easily use “sensors” in a network to track and get early warning regarding the spread of epidemics.

They took advantage of the “friendship paradox” to do so.  In any real-life network, our friends are more popular than we are.  [This is true mathematically in any group with some loners and some social butterflies.  If you poll members in the group about their friendships, far more of those friends who are reported are going to be the social butterflies.  If far more people reported friendships with the loners, they wouldn't be loners.  See discussion here.]

Thus by asking random people in a network, in this case Harvard students, about their friends, researchers know that their friends are more centrally located in these networks.    Then one can track behavior among the random group and their friends, in this case the spread of H1N1 flu (swine flu) among 744 Harvard students in 2009.

Those more central in these networks (the “friend” group) got the flu a full 16-47 days earlier than the random group.  Thus, for public authorities, monitoring such a “friend” group could give one early indication of a spreading epidemic; they could serve as “canaries in the coal mine”.  If the process of spreading was person-to-person rather than being exposed to some impersonal information (via a website or a broadcast), one could also track the difference between a random group and a friend group to predict other more positive epidemics, like the spread of information, or the diffusion of a product, or a social norm.

We write in general on this blog about the positive benefits of social ties (social capital), but Fowler and Christakis’ study also shows you that having friends and being centrally located has its costs: in this case getting the flu faster.  [In some ways, this is analogous to Gladwell's discussion in the Tipping Point of how Mavens, Connectors and Salesmen may be disproportionately influential in the spread of ideas through networks, although Fowler and Christakis are far more mathematical in identifying who these central folks are.]

The “friends group manifested the flu roughly two weeks prior to the random group using one method of detection, and a full 46 days prior to the epidemic peak using another method.

‘We think this may have significant implications for public health,’ said Christakis. ‘Public health officials often track epidemics by following random samples of people or monitoring people after they get sick. But that approach only provides a snapshot of what’s currently happening. By simply asking members of the random group to name friends, and then tracking and comparing both groups, we can predict epidemics before they strike the population at large. This would allow an earlier, more vigorous, and more effective response.’

‘If you want a crystal ball for finding out which parts of the country are going to get the flu first, then this may be the most effective method we have now,’ said Fowler. ‘Currently used methods are based on statistics that lag the real world – or, at best, are contemporaneous with it. We show a way you can get ahead of an epidemic of flu, or potentially anything else that spreads in networks.’

Christakis also notes that if you provided a random 30% in a population with immunity to a flu, you don’t protect the greater public, but if you took a random 30% of the population, asked them to name their friends, and then provided immunization to their friends, in a typical network the “friend” immunization strategy would achieve as high immunity protection for the entire network as giving 96% of the population immunity shots, but at less than 1/3 the cost.

The following video shows how the nodes that light up first (markers for getting the flu) are more central and far less likely to be at the periphery of the social network.  The red dots are people getting the flu; the yellow dots are friends of people with the flu and the size of the dot is proportional to how many of their friends have the flu.

Good summary of this research and its implications here: Nick Christakis TED talk (June 2010) – How social networks predict spread of flu.  Nick also discusses some of the implications of computational social science, which we’ve previously discussed here under the heading of digital traces.  Nick discusses how one could use data gathered from these networks (either passively or actively) to do things like predict recessions from patterns of fuel consumption by truckers, to communicate with drivers of a road of impending traffic jams ahead of them (by monitoring from cell phone users on the road ahead of them how rapidly they are changing cell phone towers) to asking those central in a mobile cellphone network (easily mapable today) to text their daily temperature (to monitor for impending flu epidemics).  Obviously these raise issues of privacy, which Nick does not discuss.

News release of study

Academic article in PLoS ONE

James Fowler on The Colbert Report discussing the book by Fowler and Christakis called Connected.


Nick Christakis presenting a talk at TED — The Hidden Influence of Social Networks. (February 2010).  In the talk he notes that while almost half of the variation in our number of friends is genetically-based (46%), that another equally large portion (47%) of whether your friends know each other is a function of whether your friends are the type that introduce (“knit”) their friends together or keep them apart (what they call “transitivity”).  About a third of whether you are in the center of social networks or not is genetically inherited.  Christakis believes that these social networks are critically important to transmitting ideas, and kindness, and information and goodness; and if society realized how valuable these networks were, we’d focus far more of our time, energy and resources into helping these networks to flourish.

The social spread of autism diagnoses

Flickr photo by alecani

Peter Bearman, from Columbia University, presented work at the Harvard Inequality Seminar on pathways for the spread of autism.  Bearman is most interested in hypotheses about toxic causes of autism (one of his theories of a likely suspect is pesticides, based on a higher prevalence of autism diagnoses for youth who lived along golf fairways, especially along private golf courses, but he has not been able to prove that yet).  Bearman and collaborators hoped to use incidence of autism among Hispanics in the pesticide-rich Central Valley to prove this, but hispanic autism rates were too highly volatile depending on whether autism diagnoses could put families at risk for deportation or being reported to the INS.

What Bearman did present on was findings resulting from pairing millions of birth records with autism diagnoses in California; he and coauthors found that over 50% of the increase in autism in California in recent years may be spread through social networks and proximity to other autistically-diagnosed youth.

Bearman does not know friendship networks specifically but does know place at birth or during various years of childhood.  He finds, controlling for environmental factors and risk factors (like age of mother at birth, gender, education of mother, etc.) that people who lived within 250 meters (basically the length of a cul-de-sac) of someone with an autism diagnosis who shared a social institution (mall, park or preschool) were 38% more likely to be diagnosed as autistic in the following year whereas those who were the same distance apart from someone diagnosed with autism who shared a non-social institution (cemetary, radiation specialist, dentist, etc.) were not any more likely to be diagnosed with autism.

The only persistent cluster in California for higher incidence of autism from 2000-2005 by place of birth (controlling for all known factors) was one in north Hollywood Hills, near Northridge.  Bearman suspects it was an environmental risk that people were exposed to (a nuclear meltdown that occurred there in 1964) that increased number of autism diagnoses slightly, followed by a four-decade-long cascade caused by social processes: parents who lived close to someone diagnosed with autism were sensitized to these factors and were more likely to diagnose their own child as autistic and work with doctors to verify this diagnosis.  The biggest increases were at opposite ends of the spectrum: both among high-functioning individuals and similarly among low-functioning individuals (who pushed doctors for an autism-mental retardation classification, which offered greater access to services and resources, than a sole mental retardation classification).  The diagnosis of autism was generally done between the ages of 3 and 5 and done only on the basis of self-presentation and parental explanation.  These social networks helped parents find physicians and navigate the California bureaucratic process.  He doesn’t think that it was the influence of doctors since closer distance from a doctor that had diagnosed children as autistic did not predict these children being more likely to be diagnosed as autistic themselves.  (And density of pediatricians did not have an effect.)

Bearman ducked a question of whether this outcome was a desirable one.  For the low-functioning autistic children, they would have gotten special ed services regardless of whether they were classified solely as mentally retarded or as autistic-MR, but for the high-functioning children classified as having autism, they would have gotten more in the way of special ed services, presumably at the expense of all other children (as special ed costs absorbed a higher percentage of the budget).  Bearman focused on the benefit to parents of children who got a high-functioning autistic diagnosis and didn’t address whether this concentrated school resources on a small number of high functioning “autistic” children at the possible expense of other same age children who were not diagnosed with special education needs.  He said that children diagnosed at age 3 with autism do not seem to show any higher final performance outcomes than children diagnosed with autism at age 6; the latter group catches up in outcomes to the earlier-diagnosed autistic children by age 9.  He does not believe there is any kind of objective standard of which of these higher-functioning kids is truly autistic or not.

One person asked whether the Internet would obviate the effect of this physical proximity.  Bearman thinks it will not and that we use the Internet for decisions like choosing a restaurant or finding the cheapest place to buy something but not for picking a doctor or navigating bureaucracy.  That geographic proximity continues to play a role in 2000 or 2004 is testament to the staying power of localized real interactions in the age of the Internet.

Bearman notes that there is no relationship between vaccine use as children and autism, but notes that the spread of Autism Advocacy Organizations (which spread as the number of individuals in a zipcode who are diagnosed with autism) is associated with a higher refusal of vaccines.  Thus he expects that in the future we will see a relationship between these advocacy organizations and the downstream increase in mumps, rubella, measles, etc. as more children opt out of these vaccines.

Bearman noted that in recent years, the rates of increase in California among higher-SES households has slowed, presumably he remarked because such families now see other diagnoses that offer a better array of services for their special need children, and because as the SES-gradient has declined, an autism diagnosis conveys less status.

To read some of this work, see for example “Social Influence and the Autism Epidemic“, American Journal of Sociology, 115(5): 1387-14343 (2010).