DEEP BACKGROUND 09 OF 13 . THE BASE

Why do launches with enormous reach sometimes produce no adoption?

Because some things need to be heard from several people you know before you will do them. In a controlled experiment the same behaviour reached 53.77 per cent of clustered networks against 38.26 per cent of well connected random ones, and spread four times faster.


WHY THIS IS WORTH MONEY

The instinct on every launch is to buy the widest reach the budget allows and connect to as many separate places as possible. Efficiency, in network terms, means not wasting money telling the same group twice.

If some behaviours need to be heard twice from different people before anybody acts, then that efficiency is buying the wrong thing, and the overlap that looks like waste on a plan is the part doing the work.


WHAT THE RECORD SHOWS


Two kinds of spread, not one

Centola and Macy (American Journal of Sociology, 2007) drew the distinction that changes how you read a launch. A simple contagion needs contact with one source: a rumour, a virus, a piece of information. A complex contagion needs "social affirmation or reinforcement from multiple sources", because the behaviour carries cost, risk or social exposure and people need confirmation before they will do it.

Then the finding that overturns received wisdom. Granovetter taught marketing that weak ties and long bridges are how things spread. In their words: "Network topologies that facilitate diffusion through simple contact can have a surprisingly detrimental effect on the spread of collective behaviors that require social reinforcement from multiple contacts."

And: "For simple contagions, too much clustering means too few long ties, which slows down cascades. For complex contagions, too little clustering means too few wide bridges, which not only slows down cascades but can prevent them entirely."

That paper is a simulation. Three years later the same author built it and ran it.


The experiment

Centola (Science, 2010) built an internet health community of 1,528 participants recruited from health interest websites, and assigned them into one of two artificial network structures with identical numbers of neighbours: a clustered lattice with redundant ties linking each person's neighbours to one another, or a random rewiring of the same network. Six trials, twelve independent diffusion processes.

"On average, the behavior reached 53.77% of the clustered networks, whereas only 38.26% of the population adopted in the random networks." The rate of diffusion in the clustered condition was "more than four times faster". Both differences significant at p < 0.01.

And the mechanism was visible directly: "Participants were significantly more likely to adopt after receiving a second signal than after receiving only one signal (P < 0.001)."

The efficient, well bridged network was the worse one. Redundancy, the thing that looks like waste on a network diagram, is what made a behaviour stick, because it meant the same person heard it from several people they knew.


And it has since been tested outside the laboratory

Lee, Lazer and Riedl (Sociological Science, 2025) ran a country-scale randomised field experiment on a mobile data product, with roughly four million focal nodes drawn from fifty million subscribers, randomly exposing people to one friend or two friends who had been encouraged to share a coupon. They report "strong support for complex contagion: the contagion process cannot be understood as independent cascades but rather as a process in which signals from multiple sources amplify each other."

Mønsted and colleagues (PLoS ONE, 2017) ran thirty nine Twitter bot accounts with around 25,000 followers and concluded that "the complex contagion model describes the observed information diffusion behavior more accurately than simple contagion."

Sprague and House (PLOS ONE, 2017) tested twenty six online fads and found complex contagion the better model in twenty two of them.


THE STRONGEST OBJECTION, AND THE ANSWER TO IT

There are three, and the third one is the serious one.

First, weak ties. Bakshy, Rosenn, Marlow and Adamic (WWW, 2012) randomised exposure to friends' sharing across 253 million people and found that "although stronger ties are individually more influential, it is the more abundant weak ties who are responsible for the propagation of novel information."

That is a real result and it does not contradict this page, because it is about information. Complex contagion predicts exactly that: information is a simple contagion and travels on long ties. Behaviour that costs you something does not.

Second, the model. Ghasemiesfeh, Ebrahimi and Gao (EC, 2013) showed that the weakness of long ties depends on how the long ties are distributed. Under a distance-dependent placement rather than a uniformly random one, complex contagion spreads quickly. So the strong form of the 2007 conclusion is partly an artefact of the model's wiring, not a general property.

Third, and this is the one to take seriously: almost nothing spreads at all. Goel, Watts and Goldstein (EC, 2012) looked at seven online domains and found that "the vast majority of instances, ranging from 73% to 95% across domains, show no diffusion at all", and that "very few adoptions (1%-6% across domains) take place more than one degree from a seed node". Goel, Anderson, Hofman and Watts (Management Science, 2016) analysed around a billion Twitter diffusion events and found "about 99% of adoptions are accounted for either by the root nodes themselves or by the immediate followers of root nodes", with viral hits "at a rate closer to one in a million".

If almost nothing cascades, then a mechanism that improves cascading is improving something that rarely happens, and most real reach comes from broadcast.

Our answer, and it costs us the strongest version of the claim. We should not say a dense base is a precondition for spread, because Centola's own random condition reached 38.26 per cent. Diffusion happened without density. What the experiment establishes is that clustering raises both the ceiling and the speed, not that its absence prevents adoption.

The defensible claim is narrower. Where the thing you are asking for carries cost, risk or social exposure, a connected base makes adoption more likely and much faster than the same number of unconnected people. That is enough to change a launch plan, and it is what the evidence supports.


WHERE THE EVIDENCE STOPS

Centola's 2010 participants were recruited strangers assigned anonymous health buddies in networks the researcher built. We looked for a published critique of that artificiality and could not find one, and we are not going to invent it. But it is the obvious question and a reader will ask it.

The 2007 paper is simulation, and its conclusion about long ties is sensitive to how long ties are placed.

The word precondition is not supported. See block 4. If it appears anywhere in our material, it is wrong.

No published work tests whether wide reach without density produces awareness and no adoption. The nearest evidence, from Goel and colleagues, points the other way: most adoption comes from broadcast.


WHERE THE CONE SITS

Start, Build, Grow came from artist development in music, not from reading Centola. Start with the few who lean in, build by giving them reasons to enthuse you to others, then grow into the mainstream.

What the network literature adds is that this stops being sentimentality about superfans and becomes a claim about structure. A base is not just people who like you. It is people who know each other, which is what makes a second signal possible.

And it names the failure mode precisely. Twenty strangers who all like you is a list. Twenty people who like you and know each other is a base.


WHAT TO DO ON MONDAY

Identify the twenty people most likely to lean in first, and work out whether they know each other.

If they are twenty strangers you have a list, and the next thing to build is the connection between them rather than the next twenty names.


Next is what holds a base together once it exists, and the two conditions it has to meet.


SOURCES

Centola, D., & Macy, M. (2007). Complex contagions and the weakness of long ties. American Journal of Sociology, 113(3), 702-734. doi:10.1086/521848
Centola, D. (2010). The spread of behavior in an online social network experiment. Science, 329(5996), 1194-1197. doi:10.1126/science.1185231
Lee, J., Lazer, D., & Riedl, C. (2025). Complex contagion in social networks: Causal evidence from a country-scale field experiment. Sociological Science, 12, 685-714. doi:10.15195/v12.a28
Mønsted, B., Sapiezynski, P., Ferrara, E., & Lehmann, S. (2017). Evidence of complex contagion of information in social media: An experiment using Twitter bots. PLoS ONE, 12(9), e0184148. doi:10.1371/journal.pone.0184148
Sprague, D. A., & House, T. (2017). Evidence for complex contagion models of social contagion from observational data. PLOS ONE, 12(7), e0180802. doi:10.1371/journal.pone.0180802
Bakshy, E., Rosenn, I., Marlow, C., & Adamic, L. (2012). The role of social networks in information diffusion. WWW '12, 519-528. doi:10.1145/2187836.2187907
Goel, S., Watts, D. J., & Goldstein, D. G. (2012). The structure of online diffusion networks. EC '12, 623-638. doi:10.1145/2229012.2229058
Goel, S., Anderson, A., Hofman, J., & Watts, D. J. (2016). The structural virality of online diffusion. Management Science, 62(1), 180-196. doi:10.1287/mnsc.2015.2158
Ghasemiesfeh, G., Ebrahimi, R., & Gao, J. (2013). Complex contagion and the weakness of long ties in social networks: Revisited. EC '13, 507-524. doi:10.1145/2492002.2482550
Aral, S., & Walker, D. (2012). Identifying influential and susceptible members of social networks. Science, 337(6092), 337-341. doi:10.1126/science.1215842

Superfandom . Deep Background . https://superfandom.ai/science/the-base/
Matt Hart . published 2026-09-08 . last changed 2026-09-09