Diffusion of Innovations

How Things Spread in Networks

What Is Diffusion?

Diffusion is the spread of objects, ideas, information, concepts, fashions, tastes, or practices within a social network.

The process has five components:

  • Innovation: the object that diffuses
  • Actors: the people or groups that adopt, or fail to adopt, the innovation
  • Flow: movement of the innovation from a source actor to an adopting actor
  • Mechanism: the process linking sources and adopters, such as imitation, contact, proximity, influence, or comparison
  • Outcome: the end state of the adopter, such as opinion change, conversion, participation, or infection

Why Do Things Diffuse?

Three factors account for why some things spread and others do not.

  • The object: familiar, low-risk innovations with a favorable cost-benefit ratio diffuse more easily
  • The source and adopter: people adopt from high-status sources, or from those they see as similar to themselves
  • The environment: geographic or social proximity, shared social foci such as workplaces or clubs, and overall network structure

Four Network Models of Diffusion

  • Direct cohesion: the connectivity of adopters to their sources through network ties
  • Two-step: the social distance between adopters and influential sources, and the traits of those influentials
  • Threshold: the susceptibility of adopters to the behavior of multiple sources
  • Positional equivalence: indirect influence and comparison between sources and structurally similar adopters

Diffusion via Cohesion

  • The simplest model: innovations “travel” along direct network links
  • The main mechanism is imitation, often driven by the similarity heuristic: “I adopt the innovations that people like me adopt” (homophily)
  • A second mechanism is contagion, where an innovation spreads like a virus, so more direct connections to adopters raise the chance of adopting

The Medical Innovation Study

Coleman, Katz, and Menzel (1957) followed doctors’ adoption of a new drug over about 15 months in four cities.

  • The team mapped advice and discussion ties among doctors, and merged them with prescription records that fixed each doctor’s first prescription month
  • Months 1–2: professional advice and discussion ties were most effective
  • Months 3–4: friendship ties took over
  • Months 4–5: the network reached less integrated, “isolated” doctors
  • Months 6 and later: the network was exhausted, and late adopters relied on ads and salesmen instead

Two-Step Models

  • Innovations reach opinion leaders first, and potential adopters then look to them for what to adopt
  • Opinion leaders are visible, well-connected early adopters, such as celebrities or social media influencers
  • Network terms for leaders: high degree (popular), high eigenvector centrality (connected to popular people), or high betweenness (bridging groups)
  • The two-step model can combine with cohesion: leaders receive the innovation from a source, and then it spreads among adopters through their ties

Diffusion via Positional Equivalence

  • Cohesion models assume influence flows only between connected actors
  • Diffusion can also operate between disconnected actors who occupy similar structural positions
  • The mechanism is comparison: people adopt to keep their status or competitiveness relative to similarly positioned others

The S-Curve

Cumulative adoptions typically follow an S-shaped pattern with three phases:

  • Initial slow uptake: early adopters and innovators take up the innovation while most of the population waits
  • Rapid spread: adoption accelerates, and the number of new adopters per unit time peaks during the “take off” period
  • Final slow saturation: adoption levels off as most potential adopters have joined

The S-Curve Shape

Measured 50% adoption midpoints; curve shapes are illustrative. Sources on the next slide.

Sources for the S-Curve Figure

Technology Midpoint Measured points
Telephone 1945 36.9% (1940), 61.8% (1950), Census
Electricity 1925 34.7% (1920), 68.2% (1930), Census
Automobile 1948 50% (1948)
Radio 1932 40% (1930), 60% (1934)
Color TV 1971 48% (1971), 50%+ (1972)
Air conditioning 1976 35.8% (1970), 58.5% (1980)
Personal computer 2000 51% (2000, CPS)
Internet 2001 50.4% (2001, CPS)

Denominators are in the figure legend. Cellphone is omitted: the Pew series starts at 62% of adults in 2002.

The Tipping Point

  • The tipping point comes after early adopters and before the early majority fully engage
  • Innovations that never reach it often die out
  • Many historical innovations, from electricity to the VCR to the internet, follow the S-shaped pattern
  • Social movements can show a long uptake period followed by rapid mobilization

Global Threshold Model

Granovetter and Soong (1983) proposed that each person has a threshold \(q(i)\), with \(0 < q(i) < 1\).

  • Person \(i\) adopts when the proportion of people who have adopted, \(p\), reaches \(q(i)\)
  • A person with \(q = 0.25\) adopts once a quarter of the population has adopted; one with \(q = 0.75\) waits until three quarters have
  • The model assumes everyone knows the population-wide adoption rate, which fits public, observable behaviors

Thresholds Explain the S-Curve

  • Slow start: only people with very low thresholds adopt
  • Rapid spread: when thresholds cluster in the middle, the majority joins
  • Slow finish: only people with high thresholds remain

The shape of the threshold distribution matters. A distribution skewed toward risk-averse late adopters makes full diffusion less likely. One with plenty of high-risk early adopters makes it more likely.

Rogers’s Classification of Adopters

Rogers (1962) groups adopters by their time of adoption relative to the average.

Category Share of adopters Timing
Innovators 2.5% More than 1 SD before average
Early adopters 13.5% More than 1 SD before average
Early majority 34% Within 1 SD before average
Late majority 34% Within 1 SD after average
Laggards 16% More than 1 SD after average

Early adopters are the opinion leaders who influence the early majority. Failed innovations die before reaching the tipping point.

Valente’s Network Threshold Model

  • Granovetter’s model assumes global knowledge; Valente’s model uses each person’s local network
  • Each person has an exposure \(e(i)\), the proportion of their friends who have adopted
  • A person adopts when \(e(i)\) reaches their personal threshold \(q(i)\)
  • In the textbook example, a person with \(q = 0.25\) adopts after one of three friends does, while a person with \(q = 0.75\) waits until three of four friends have adopted

Valente’s Four Types of Adopters

Two thresholds, personal and whole-network, yield four types:

  • True innovators (influentials): low personal and low whole-network thresholds
  • Isolates: low personal threshold, high whole-network threshold
  • Connected bandwagon: high personal threshold, low whole-network threshold
  • Resistant rearguard: high personal and high whole-network thresholds, largely matching Rogers’s laggards

Adopter categories can also be relative. You can be an innovator in the system and a laggard among your friends, or the reverse.

Local and Cosmopolitan Diffusion

  • Cosmopolitan innovators tend to be early adopters relative to the system, with low thresholds relative to their own network
  • They often adopt through media or weak ties
  • Late adopters often rely on friends and strong ties

Simplex and Complex Contagion

Feature Simplex Complex
Threshold Low; one exposure often enough High; multiple sources needed
Risk or cost Low High; risk, ostracism, or resource loss
Primary ties Weak ties (bridges) Strong ties (cliques)
Information Simple, novel Complex, interdependent

Information, gossip, and disease are candidates for simplex contagion. Social movement participation, religious conversion, and migration are candidates for complex contagion.

Which Ties Speed Diffusion?

  • Simplex contagion can spread faster through weak ties, which bridge otherwise separate parts of a network (Granovetter 1973)
  • Complex contagion can spread faster through cliques and strong ties, because it needs repeated reinforcement
  • Weak ties carry an innovation across a bridge only once, so they often fail to supply the multiple exposures complex contagion requires

Key Takeaways

  • Diffusion follows an S-shaped path, with slow uptake, rapid spread, and saturation
  • Direct cohesion, two-step, threshold, and positional models each explain a different part of that path
  • Whether an innovation spreads through weak bridges or strong cliques depends on whether it is simple or complex to adopt
  • Local networks can explain adoption patterns that global models cannot