Nicholas A. Christakis (MD, PhD) and James H. Fowler (PhD) hold a high opinion about the potential value of their own field of expertise: "If we do not understand social networks, we cannot hope to fully understand either ourselves or the world we inhabit." Having read their book Connected I am generally inclined to agree with them, although I remain skeptical of much they have to say, including the validity of some of their most attention-grabbing conclusions. The book exhibits many of the merits that accrue when scientific authors are skilled at writing for a popular audience, but it also illustrates some of the perils that arise when complex and technical research details are truncated to make the product palatable for non-specialists. Nevertheless, anyone with a serious interest in the social sciences, public health, or public policy generally, but not previously fully-versed in social network analysis, should find Connected very instructive.
That is my summary judgment and you can stop here if you just want to understand why I assigned four stars. Connected is rich in content and I apologize that to summarize the book fairly and further justify my evaluation requires considerably more words, quite likely more than you may want to read at this point.
Social networks consist of humans and the connections between them. Most of us are members of "multiplex" networks involving different kinds of connections such as family, close friends, coworkers, neighbors, acquaintances, and so on. We can be either directly connected to others (first degree of separation), or indirectly so, through the second degree (a friend of a friend, for instance) up to about six degrees of separation to cover the globe. Networks have structure and shape: people have specific locations within networks, sometimes forming clusters that are themselves distinct network components. Christakis and Fowler do a commendable job of explaining all of this, plus they include several very clever color "maps" of different kinds of networks to provide visual reinforcement (credit the software).
Networks have the capacity for contagion, for influences to flow through the connections. We all understand how this works in a health epidemic when germs are spread, but Christakis and Fowler contend that it also applies to certain behaviors as well. There are several plausible explanations for why we might often behave or feel like others in our network, or vice-versa, including genetics (for directly related family members), our tendency to associate with others like ourselves (homophily), "emotional contagion" (the authors present several striking accounts of epidemic hysteria), shared environmental exposures, our propensity for imitation, and our desire to conform to social norms.
We can see readily enough how such influences might often operate at the first degree of separation, through ties with people to whom we are directly connected. But the authors are especially interested in "hyperdyadic spread," the tendency of behavioral influences to extend beyond the first level to generate effects at higher degrees of separation (just like germ contagion). It is not immediately intuitive that people we don't even know (aside from media personalities and the like) can significantly influence how we behave or feel. Yet Christakis and Fowler report on research where they claim to have found just that, including studies of network effects on happiness, obesity, and smoking, for example.
They propose a "Three Degrees of Influence Rule." For instance, they found in one of their studies that a person is about 15 percent more likely to be happy if someone to whom he or she is directly connected is happy; ten percent more likely if a connection two degrees removed is happy; and six percent more likely for happy persons three degrees removed. Beyond three degrees "intrinsic decay" (like in the telephone tree game) and network instability (social ties change over time) diminish the influence to negligible levels. Christakis and Fowler report similar findings from their obesity study, where they conclude (somewhat sensationally) that, "You may not know him personally, but your friend's husband's coworker can make you fat." Such network effects were observed supposedly after controlling for genetic influence, the tendency of people to befriend similar people, and the possibility of shared exposures that contributed to weight gain or loss.
The authors summarize many social network studies in addition to their own, addressing such subjects as employment search, romantic matches, sexually transmitted disease, currency circulation, financial market activity, voting behavior, and suicide. Much of the discussion is provocative and enlightening. For instance, readers will learn that there are circumstances where weak ties to others may be more helpful than strong ties, that higher status (college educated) persons influenced the spread of smoking in the 1930s and now influence the cessation of it, that networks might help explain why the rich are getting richer (the rich attract more friends, and having more friends is an aid in getting rich), and that political polarization (a bad thing in the eyes of many) increases political participation (a good thing to most).
I have various reservations about Connected, however. Some are methodological. For instance, the authors obtained the data for their principal research from the Framingham Heart Study, which has collected health information about a large inter-generational group of people since 1948, but was not designed as a social network study. Christakis and Fowler turned it into one by using the researchers' contact notes to reconstruct social connections among the participants in the more recent cohorts. But because the available information was limited the average number of assigned direct "friends" (not counting family, coworkers, neighbors or other kinds of connections) for the study subjects was only 0.7 (I know the number only because I consulted one of the authors' journal articles, not because they provided such detail in the book). Thus to me it seems their network data were at best incomplete, or worse, possibly selection biased, leaving out a large majority of the real-life friends of the study subjects. Surely many persons who were actually closely connected at the first degree were not recognized by the study as connected at all, and possibly many actual first-degree connections were linked inaccurately only at higher degrees of separation.
That is just one example. Peer scientists, persons much more qualified than myself, have raised other methodological concerns. For instance, Christakis and Fowler have been criticized for the specification of one of their mathematical models, with one adjusted replication finding no significant social network effects on obesity.
While it seems clear enough that network analysis has been and can be applied in socially beneficial ways, the authors at times appear overly optimistic about the prospects. One of their themes is that "positional inequality" (persons placed in less advantageous positions in networks) can be mitigated if policy makers better take into account the role of social networks. For example, they suggest that a friends-of-friends strategy may sometimes be preferable to one that involves just the direct subjects and their one-degree friends -- if you join your friends to lose weight you might succeed, they say, but if their friends are overweight and not working on it, your friends might relapse, affecting you too. But, at least in this example, their strategy does not seem to offer much practical promise, since with each additional tier of friends engaged the cumulative cost goes up exponentially and there are diminishing marginal returns in the effectiveness pay-off for the first-tier direct subjects.
One big factor that makes Christakis and Fowler upbeat about the future of network analysis is the advance of communications technology. In part they are excited because they see treasure mines of data. They observe, for instance, that cell phone information allows researchers to study where connected people are minute by minute. The authors do not pause to consider the potential privacy intrusions and risks that expanded access to such data might enable.
Christakis and Fowler believe that technology now makes physical distance less of a constraint for certain networks (such as scientific collaborations, for instance) and they think the Internet has facilitated match-making networks of various kinds. Yet interestingly, they suggest that modern technologies do not take us away from our prehistoric past, but rather move us closer toward it, in the sense that "our desire to form connections depends partly on our genes." They say that, "Overall, the evidence from real-world networks suggests that online networks can be used to enhance what flows between real-world friends and family, but we do not know yet whether the Internet will increase the speed or scope of social contagions in general."
The bottom line is that there is still much about social networks that social scientists and others do not understand. For instance, I do not think that we can yet say much with great confidence about the magnitude and diffusion of behavioral influences (as distinct from pathogenic agents) beyond the first degree of separation, in spite of the assertions of Christakis and Fowler here. However, what they have done should help enrich future research.