You are not a free-floating individual making choices in a vacuum. You sit inside a web of connections, and the shape of that web quietly decides which ideas reach you, which jobs you hear about, and how the things around you spread. Some of what network science claims about that is rock solid. Some of the most famous parts are not.
The move that starts everything is to stop looking at people one at a time and start looking at the pattern of connections between them. Draw people as nodes and their relationships as links, and behavioural and informational consequences fall out of the shape of the web, independent of anyone's personality.
The founding insight is counterintuitive. When Granovetter asked how people actually found jobs, the useful leads came less from close friends than from acquaintances, the people you barely keep up with. The reason is structural: your close friends tend to know each other and to know what you know, so their information is redundant. A weak tie, an acquaintance, is far more likely to be plugged into a different circle and to carry news you would never otherwise hear. This is the strength of weak ties: novel information travels along the loose, bridging connections, not the tight ones (Granovetter, 1973).
Burt generalised this into a theory of structural holes. A "hole" is a gap between two clusters that do not otherwise talk. Someone who bridges that gap, a broker, sits between worlds, sees opportunities before either side does, and controls what passes across. Studying managers, Burt found that those whose networks spanned more structural holes were rated as having better ideas and were promoted faster; brokerage, not just effort, predicted who got ahead (Burt, 2004; the book-length argument is Burt, 1992).
Structure also explains spread. Watts and Strogatz showed that adding just a few long-range links to an otherwise clustered network collapses the distance between any two points, producing the small-world effect, the "six degrees" phenomenon, where everyone is reachable in a short chain (Watts and Strogatz, 1998). Barabási and Albert showed that many real networks are scale-free: because new nodes prefer to attach to already-popular ones, a few enormous hubs emerge, which is why some things race across a population while others stall (Barabási and Albert, 1999). Together these gave a structural vocabulary for why networks behave as they do.
The last, most eye-catching step was contagion: the claim that not just information but behaviours, states, even moods spread through ties. Using decades of data from the Framingham Heart Study, Christakis and Fowler reported that obesity, smoking, and happiness clustered and appeared to propagate through the social network up to three degrees of separation, your friends' friends' friends (Christakis and Fowler, 2007; Fowler and Christakis, 2008). It is a striking picture: your behaviour rippling outward through people you have never met.
This is where a careful reader has to slow down, because the fame of the findings runs ahead of their certainty, and the pattern is instructive: the structural results are robust, while the most viral claim, three-degree contagion, is the shakiest.
The problem is identification. In observational network data, three different things produce the same picture, friends resembling each other. There is genuine contagion (your friend's behaviour changes yours). There is homophily (you befriended people like you in the first place, so you were always going to end up similar). And there is shared environment (a fast-food outlet opens near a cluster of friends, and they all gain weight together, with no influence passing between them at all). Lyons argued the Framingham analyses could not statistically separate these, and Shalizi and Thomas proved the deeper point: homophily and contagion are generically confounded in observational network studies, meaning that without an experiment or a natural break, you essentially cannot tell them apart (Lyons, 2011; Shalizi and Thomas, 2011). So the headline that behaviours spread three degrees is not established as cause; it is a correlation that could be several things.
The interesting turn is that the critique did not kill contagion, it disciplined it. When Centola ran a controlled experiment, randomly assigning people to different network structures and watching a health behaviour spread, he could rule out homophily by design, and behaviour genuinely spread. Crucially, it spread further and faster in clustered networks than in ones wired for short paths (Centola, 2010). That fits the distinction he and Macy had drawn between simple and complex contagion. Information is a simple contagion, one exposure can pass it on, so weak, bridging ties are ideal. But many behaviours are complex contagions: adopting them takes seeing several people you know do it, for reinforcement and social proof. Complex contagions need the redundancy that weak ties lack, so they crawl along dense, clustered neighbourhoods and stall on long-range links, the "weakness of long ties" (Centola and Macy, 2007). The naive picture of one super-connector broadcasting a new behaviour to the masses gets this backwards.
One more boundary: weak ties are not universally superior. They win for searching out novel information, but for high-stakes, trust-laden, risky transfers, the strong ties, the people who will vouch for you and take a chance on you, matter more, the "strength of strong ties" (Krackhardt, 1992; and see L4-06 on trust).
Most of the crisp contagion evidence outside experiments is observational, and inherits the confound above. Complete network data is genuinely hard to collect, so many studies work from partial maps. And "influence" is measured very differently online (clicks, shares) than offline (behaviour change), so results do not automatically transfer between them.
How much true social contagion of behaviour exists once homophily and environment are properly stripped out, and for which behaviours? Do the rules worked out for face-to-face networks hold on platforms engineered to maximise spread? And can diffusion be reliably engineered, seeded deliberately, or is it too sensitive to structure and timing to steer?
The usable core: to move information, find the bridges; to change behaviour, saturate a cluster so people see several others doing it; and never mistake "people near each other are similar" for "one caused the other."
This is the evidence base under influencer marketing and diffusion strategy, so the honest lessons matter. First, split your goal. If you want awareness to travel, weak ties and bridges do the work, so seeding across many loosely connected pockets beats concentrating in one (Granovetter, 1973). If you want adoption, a behaviour change, that is a complex contagion: people need to see multiple people they trust doing it, so saturate clusters densely rather than buying one giant influencer and hoping it broadcasts outward (Centola and Macy, 2007; Centola, 2010). Second, brokerage is a real edge: sitting between disconnected worlds is where novel ideas come from (Burt, 2004). Third, and this is the discipline, social-listening and correlation in your customer graph cannot prove contagion; similar customers cluster for a dozen reasons (Shalizi and Thomas, 2011), so do not sell, or believe, a "it went viral because of X" story your data cannot support.
Mobilisation moves through networks, but turning out a vote or shifting a view is a complex contagion, it needs reinforcement from several trusted contacts, not one viral clip. That argues for dense, local, peer-to-peer organising over a single big broadcast. The manipulative shortcut, manufacturing the appearance of grassroots spread through coordinated fake accounts (astroturf), is increasingly detectable, and when it is exposed the trust cost is severe (see L4-06).
Public-health and behaviour campaigns should be built on network structure, not on lone messengers: seed clustered communities so people see neighbours adopting the behaviour, and use bridging ties to carry information across groups. And be honest about the evidence. Policy built on the uncritical claim that a behaviour "spreads three degrees" rests on a contested, confounded finding (Lyons, 2011); the experimental, structure-dependent version of contagion is the sturdier foundation (Centola, 2010).
Three habits. To spread an idea, look for the acquaintances and brokers who bridge to circles you cannot reach, not the friends who already know what you know. To spread a behaviour, make it visible and repeated inside a tight group, because people adopt what they see several trusted others doing. And whenever you see clustering, of customers, voters, symptoms, ask the L0 question before claiming influence: did they catch it from each other, were they alike to begin with, or are they all standing in the same rain?
.png)