Chapter 6 – Diffusing Innovations that Face Opposition

Consider a public health intervention in which a small number of individuals in a population are given personal incentives to lose weight. How should these individuals be chosen? Can they be chosen so that their change in behavior will influence others to also lose weight? What if there are existing norms that create resistance to the new weight-loss behavior? This chapter shows how the insights from the previous chapters might be applied in these kinds of situations. Specific network strategies for selecting a small number of “seed” individuals may be able to greatly increase the number of people who are ultimately reached by an intervention.


The approach adopted here is of someone who is responsible for developing an intervention in which several people are “treated” in the hopes of spreading a sustainable change in behavior. From this interventionist point of view, the goal is to seed a small segment of a population in a way that stimulates change in the greatest number of people. The results presented below are from computational experiments that were conducted on the three empirical health networks.  The first two networks are from the Add Health data set, which contain (A) 1,082 people and (B) 1,525 people, respectively, and the final one, C, is from the Framingham Heart Study and contains 2,033 people.  In every case, intervention strategies were tested to see how introducing a challenging new health behavior into the social network—for instance, condom use or regular exercise—would translate into sustained changes in behavior.

In each simulation, a small number of seeds were chosen for treatment. Similar to the computational experiments from chapters 2 and 3, the seeds were created by exogenously activating nodes in the network who could then transmit the behavior to their neighbors. Unlike the previous computational experiments, however, here the focus is not just on adoption but also on long-term engagement. Condom use and regular exercise are not one-and-done behaviors; they require maintenance. In these simulations, as in many public health interventions, individuals could abandon the behavior at any time if they did not receive sufficient social reinforcement to maintain it.

The experiments began with the activation of the seeds, who were the “treated” individuals. Other actors in the population decided whether or not to adopt the behavior based on whether their thresholds for activation were triggered by their activated neighbors. In the first two experiments, using the Add Health networks, actors required at least 40% of their neighbors to be activated in order to adopt. The last experiment, using the Framingham network, tested the effects of higher thresholds for adoption—actors required at least 60% of their neighborhood to be activated in order to adopt.

In these simulations, actors resisted the intervention. Non-adopters put pressure on the adopters to abandon the intervention behavior. This pressure also affected the treated seed nodes. After the behavior was seeded and began to spread through the network (that is, after five rounds), if the countervailing pressures from the seeds’ neighbors were too great, then the seeds themselves would abandon the behavior. Thus, starting on round six of each simulation, the seeds were just as susceptible to social influence as everyone else.

These simulations compare two different seeding strategies.  For each public health network, the panel on the left shows a “viral” seeding strategy, in which seeds are randomly placed all over the network to maximize exposure.  Correspondingly, in each simulation, the panel on the right shows a “clustered” seeding strategy in which the same number of seeds are clustered together into a small number of neighborhoods.

In each simulation, the seeds are shown in YELLOW, the unactivated people (who have not adopted the innovation) are shown in GREEN, and the newly activated adopters are shown in RED.

(A) AddHealth Network (N=1082)

Random Seeding                                                              Clustered Seeding


(B) AddHealth Network (N=1525)

Random Seeding                                                              Clustered Seeding


(C) Framingham Network (N=2033)

Random Seeding                                                              Clustered Seeding


The main lesson from these computational experiments is that when a population is inclined to resist an innovation, clustering the seeds together can be a useful strategy for initiating diffusion. As the innovation spreads, reinforcement among the seeds and their shared neighbors creates local lock-in on the behavior in each new part of the network. The more locally entrenched an intervention becomes, the greater reach it can have throughout a population, and the more successful it is likely to be.


A similar kind of example outside of the health domain is to consider a setting in which an entrepreneur attempts to diffuse an innovation that challenges an incumbent technology. As with a public health intervention, even a beneficial innovation can face strong opposition from an entrenched competitor. This simulation shows how the structure of the social network can be used to initiate the successful spread of a “challenger” technology.

Consider, for instance, the spread of a new social media platform, for which the value of the technology is determined primarily by the number of other people who are using it. We can think of this in terms of a coordination problem in which actors are deciding between two plat- forms, A and B, where there is a high level of complementarity in their choices.

Let’s assume that option A is the better platform. It offers a better interface and it is easier to use. However, we can also assume that every- one in the population is already using option B. Since B is the universally adopted incumbent, it is the desirable choice. Let’s also suppose that individuals in the population are adventurous. They occasionally make brief “experiments” in which they try out new social media platforms. While it is easy to try out a new platform, it can also be costly because time spent learning how to use one platform is also time away from participating in the other one.  How does the structure of the neighborhoods in the social network determine the ability to diffuse option A?

While this problem of diffusion is similar to what we saw above in the seeding experiments, there is an important difference here. This time, everyone has unlimited access to both A and B and can try either option at any time. Both technologies are freely available and known to everyone from the start. Thus, while the seeding experiments showed how to use an exogenous seeding strategy to initiate diffusion, this example shows how the structure of the social network can be used to initiate diffusion endogenously.

In these simulations, actors begin with option B, shown in WHITE, and occasionally experiment with option A, shown in RED.  Actors can switch back and forth between options, and they only stick with an option if there is enough support for it to encourage them to keep using it.

In this simulation, the panel on the left shows the dynamics of innovation diffusion in a random network, and the panel on the right shows diffusion in a clustered network.

Innovation Diffusion in Random and Clustered Networks

Random Network                                                             Clustered Network


Clustered neighborhoods act as social incubators that protect new ideas from being swamped early on by countervailing influences from the rest of the population. Contrary to the lessons learned from decades of research on the diffusion of simple contagions, incubator neighborhoods accelerate the diffusion of a contested innovation precisely because they limit early adopters’ exposure to the rest of the network.

The diffusion of a contested innovation in a hostile environment bears an interesting similarity to the spread of cooperative behavior in a population of defectors. For instance, when altruists are clustered together in social networks, the mutual benefits that they provide each other can allow them to outcompete the surrounding defectors, eventually transforming a population mired in mutual defection into a co- operative social regime. For both contested innovations and cooperative behaviors, the same logic applies.13 Network clustering allows social innovators to work together to reinforce each other’s behavior, while protecting them from the countervailing influences of the rest of the population. This process falls apart, however, when there are too may long ties in the network. The less clustering there is in the neighborhood, the more exposed social innovators become to the countervailing influences around them, and the harder it is for them to coordinate their efforts to challenge entrenched social norms.