Influence or be influenced: the cause and effect of your social network

Social networks weave the societal fabric through which information, influence and behaviors can spread.

We hear about new ideas, products and juicy gossip from our friends, family and colleagues. But perhaps more importantly, we tend to trust these contacts more than we would a political campaign message or product advertisement. Not only do our friends have more influence over us, they know our preferences and are therefore likely to know who in their social circle would be most interested in or persuaded by a particular idea or message. It’s not surprising that new social media companies like Klout, PeerIndex and Social Amp are trying to harness the power of social influence to help brands reach new consumers. The rise of Facebook, Twitter and other social networking technologies has automated and supercharged naturally occurring social influence processes for over a billion people already and the reach of basic SMS and mobile phone networks is even greater.

But helping consumers find products is not the only application of social network influence. The same theories and methods can also be applied to promote social good. Imagine if social influence was systematically activated and channeled to encourage friends to take an HIV test, to de-escalate a violent event, or to contribute to a local charity. These types of applications could spread positive influence and create more effective and long lasting social change.

We’re currently working on just such applications. In collaboration with the Praekelt Foundation, my lab at NYU is planning a large-scale effort harnessing the power of peer referral to spread HIV testing in South Africa using mobile messaging. The idea is to encourage friends to encourage their friends to take an HIV test with the hope that a message from a friend is more influential than a government billboard or radio spot. We are also planning to apply the same techniques to violence prevention and the promotion of exercise and healthy living.

But there are some challenges that need to be overcome in order to give ourselves the best chance of success.

HIV virus
Social network graph
  1. Human Immunodeficiency Virus (HIV-1)
    (Image: Microbe World/Flickr)
  2. Social network graph (Image: Sinan Aral)

Dan Ariely on what makes unethical behavior contagious. (Video produced by Duke University)

A television commercial produced as part of the U.S. Department of Transportation's infamous "Friends Don't Let Friends Drive Drunk" campaign.

First, it’s not clear whether, when, and to what extent different behaviors are truly “contagious.”

The main problem is separating correlation from causation in data on the peer to peer transmission of behaviors. There is abundant evidence that human behaviors tend to cluster in social networks over time – friends tend to adopt the same behaviors or purchase the same products at approximately the same time. But is this because one friend influenced another to adopt the behavior or are friends simply more likely to like the same things and behave in similar ways? We tend to make friends with those who are like ourselves – people with similar tastes, preferences and lifestyles. This process, known as “homophily”, can create the same pattern of behavioral change in a population as a social contagion, but without any social influence. Friend after friend will adopt the behavior simply because they are similar, not because they are influencing one another. Peers are also more likely to be exposed to the same external stimuli. We tend to make friends with people we work with or who live nearby. As a result, our exposure to changes in health benefit plans at work or new restaurants opening in our neighborhoods is correlated with that of our friends, and our common exposure to such stimuli can drive patterns of correlation in our behaviors over time.

Separating correlation from causation in data on the spread of a behavior in a social network is critical to policy because the underlying dynamic process of the spread of a behavior through a population implies a) different diffusion properties for the behavior (where will it spread next and so who should we target) and b) different optimal promotion or containment policies.

For example, consider two hypothetical scenarios in which data show that the adoption of a behavior like smoking is significantly correlated amongst friends. In one scenario 90% of this correlation is explained by peer influence — friends convincing friends to smoke — and only 10% is explained by correlated preferences. In an alternative scenario, 10% of the correlation is explained by peer influence while 90% is explained by correlated preferences. In the first scenario, a peer-to-peer intervention that creates incentives for friends to prevent their friends from smoking may be quite effective. In the second scenario, a traditional market segmentation strategy based on people’s observable characteristics may be much more effective and a peer-to-peer strategy may not work at all.

The same logic applies to whether the National Institutes of Health should allocate substantial funding to peer-to-peer obesity prevention programs, whether the CDC should focus on peer promotion of HIV testing and whether advertisements such as the Department of Transportation’s ‘Friends Don’t Let Friends Drive Drunk’ campaign are likely to succeed. Developing methods that robustly identify causal estimates of peer-to-peer influence in social networks will therefore have dramatic implications for policies aimed at spreading positive social change.

Second, we know little about how to create cascades of positive social behavior.

What types of policies would work best and under what conditions? We need a better understanding of “viral incentive systems,” the policy mechanisms that most effectively spread a behavior from friend to friend. Should people be incentivized to send messages to their friends? Should message recipients be incentivized to change their behavior? Should we try to do both? To what extent is one more effective than the other? Do incentives eliminate the positive influence of peer-to-peer messages altogether?

Answers to these questions will help us understand which behaviors to dedicate resources to changing. Toward this end, we are now undertaking large scale scientific studies of viral incentive systems and social messaging techniques to understand what works and what doesn’t. More importantly, we are trying to figure out the conditions under which the techniques work best.

Finally, we need a better understanding of who is influential and who is susceptible to influence in order to target behavior change programs toward the right parts of a network.

In the end, we believe we can harness the power of social contagion for social good. But we also believe in robust science driving the development of effective policies. Using out-of-the-box thinking and new methods to promote positive behaviors and contain destructive ones, we hope we can spread a little love… virally.

Sinan Aral

Sinan Aral studies how behavioral contagions spread through social networks — from products to productivity to public health. His work has been published in leading journals such as Science, and the Proceedings of the National Academy of Sciences, Marketing Science, and has been mentioned in The New York Times and The Economist. Sinan was a PopTech Science Fellow in 2010.