CHALLENGE SUMMARY
In the twentieth century, many social scientists were focused on individual behavior, and the models they used to describe it colored their view of the world. Statisticians modeled individuals under the assumption that the actions of multiple individuals were independent. Economists assumed that individuals were self-interested. But increasingly we have come to realize that these models are deficient. We are influenced by and care about others.
An old way of dealing with this dependence is to conceptualize abstract collections of individuals as “groups” and to assume that individuals behave the same within these groups. Categories of race, class, and nationality served as proxies for individual interests and behavior in some of the most popular social theories of the twentieth century. But these models were also deficient because they ignored within group variation and had little to say about how and why the group influenced individual behavior.
The new science of social networks looks at the world in a different way. Rather than focusing on individuals or groups, it focuses on the relationships between individuals. This third way of seeing the world maintains individuals at the center (the “nodes” of a network), but it recognizes their interdependence by including explicit information about their interactions with other individuals (the “connections” in a network). If “groups” exert influence on individual behavior, they do so via the direct relationships between individuals. In fact, groups are primarily abstract representations of the “communities” that can be identified as parts of the network that have
many connections within a set of individuals and few connections between that set and the rest of the network.
Although the science of social networks can be applied at many scales and to many organisms, it is especially in individualized societies based on memory of past interactions that we see mutual interdependence and long-term partnerships, often based on reciprocity, such as in the primates, elephants, dolphins, and other large-brained mammals (de Waal & Tyack, 2003). Here we focus on how to use social network theory to understand and improve outcomes for humans, even while recognizing that some of the issues are not limited to our species. A wide variety of research is making it clear that behaviors spread in social networks. There are already many specific models in sociology, economics, social psychology, and related fields on social influence, some of which address the flow of influence in networks. There is also work in computer science and what is now being called computational social science on networks and influence. However, this work suffers from (at least) six important problems.
1. It remains unclear which methods are best for measuring social influence.
Network science is a fast-growing field, and it is clear that perfect methods, free of any limitations or assumptions, do not exist for every sort of question one might want to ask with observational (or even experimental) data. The classic problem of distinguishing selection and contextual effects from influence remains, though recent advances suggest that sensitivity and bounds analysis may hold promise. Additionally, basic issues in coping with missing data (missing nodes, ties, covariates, waves), sampling (design effects and incomplete network ascertainment), and computation of standard errors are still being addressed.
2. Only a limited number of long-term longitudinal social network data sets exist.
A plethora of studies are based on available data from the Framingham Heart Study and National Longitudinal Study of Adolescent Health, but it is not clear if these networks are representative. Are there systematic differences in the strength of network effects or the processes by which they occur when studying networks of different sizes and composition? Are there important problems when defining the boundaries of these networks that may interfere with valid inferences?
3. The literatures that address social networks are rooted in very different fields of biology, sociology, economics, statistics, epidemiology, and physics, and a common model and language has not yet emerged.
Scholars in various fields continue to look inward toward models and solutions proposed by their close colleagues rather than reaching out to join forces across disciplinary boundaries. A large-scale effort to systematically explain the differences and similarities of the most used social influence models in various fields has not been conducted. As a result, it is difficult to know which models are working best and to transmit advances across the sciences.
4. Most studies focus on documenting network effects rather than identifying and testing their mechanisms.
If we want to alter the dynamics we observe in social networks, we need to understand what drives them. Current work is moving in this direction, but not quickly enough. We need to encourage scholars to identify the key aspects of the social processes or mechanisms involved and how network phenomena play out in different applications (e.g., to the spread of ideas, attitudes, norms, behavioral change) and different settings (especially online versus real-world networks). A unifying theory that explains when to expect various mechanisms to matter would help in improving our capacity to analyze novel phenomena. And ideally, one would have randomized experiments or rigorously designed quasi-experiments to allow the strongest tests of causal direction and provide opportunities to test specific mechanisms.
5. We do not yet understand how the transmission of behaviors through a network itself alters the structure of that network.
It is well known that network structure influences many human outcomes, but much less is known about how human behavior alters network structure. For example, the social networks of smokers changed dramatically over the past 40 years as social pressure campaigns marginalized them and public policies forced them into smaller spaces where they were more likely to connect to other smokers. Understanding the effect of influence on structure may thus be a critical element in any effort to use networks to change behavior.
6. In spite of recent advances in understanding the processes underlying social influence, there has been very little work showing how to apply
this understanding in interventions to improve our health, our wealth, our global environment, and our democratic institutions.
There are, of course, some notable exceptions like real-world network interventions that target central actors to prevent smoking and online network interventions to spread voter participation. But these are the exception that proves the rule. We need more experiments designed to test theories derived from observational studies, and more large-scale tests of interventions based on successful observational and experimental studies.
Key Questions
What models do we currently have that identify the role of social networks in this process, and where do we go from here? Is there a way to unify the proposed models or adjudicate among them so that we can agree on a central methodology? It might be possible to take a “top-down” approach by facilitating interchange among theorists representing the different models in an attempt to forge agreement on shared concepts, definitions, and methods. Another way would be to ask such a group of theorists first to design on their own how they would study the spread of a given phenomenon and then ask them, in the context of creating a single design, to capture emergent concepts and methods.
What mechanisms result in the spread of influence, when? We must move beyond a mere demonstration of spread to studies that explain how and why things spread and the conditions that hasten or slow transmission.
What are the differences between modes of transmission and outcomes with respect to behavior change in particular?
In light of the geometric growth of online communities (Facebook, LinkedIn, Patients Like Me, and so on), will the same principles determine spread in online networks that determine spread in face-to-face networks?
How early in life can network effects be seen? For example, we are finding effects of social hierarchies in kindergarteners—are these due to their networks?
How do we model reciprocal effects of networks and behavior change? In particular, how do behaviors themselves alter the shape of social networks?
Over time, people who engage in stigmatized or illicit behaviors become more tangential within larger networks, but do they form new networks, and what effect do these have on their behavior? For example, there
may be enhanced communication among smokers huddled in smoking areas outside of large office buildings.
What role do individual characteristics play in network structure and transmission? For example, is transmission facilitated in networks with more homogenous members (same gender, ethnicity, age, social class)?
How can the principles and approaches of social network research be applied to understand social interactions and cooperation in nonhuman species and groups involving multiple species?
How does the spread of information change a social network?
Reading
Bond RM, Fariss CJ, Jones JJ, Kramer ADI, Marlow C, Settle JE, and Fowler JH. A 61-million-person experiment in social influence and political mobilization. Nature 2012;489:295-298.
Centola D. The spread of behavior in an online social network experiment. Science 2010; 329:1194-1197.
Christakis NA and Fowler JH. Social contagion theory: Examining dynamic social networks and human behavior. Statistics in Medicine 2012;32:566-577.
de Waal FBM and Tyack PL. Animal Social Complexity: Intelligence, Culture, and Individualized Societies. Harvard University Press, Cambridge, MA, 2003.
Dodds PS and Watts DJ. Universal behavior in a generalized model of contagion. Physical Review Letters 2004;92(21):218701.
Lazer D, Pentland A, Adamic L, Aral S, Barabási A-L, Brewer D, Christakis N, Contractor N, Fowler J, Gutmann M, Jebara T, King G, Macy M, Roy D, and Van Alstyne M. Computational social science. Science 2009;323:721-723.
Salathé M, Kazandjieva M, Lee JW, Levis P, Feldman MW, and Jones JH. A high-resolution human contact network for infectious disease transmission. Proceedings of the National Academy of Sciences of the United States of America 2010;107(51):22020-22025.
Valente TW. Network interventions. Science 2012;337:49-53.
Because of the popularity of this topic, two groups explored this subject. Please be sure to review the other write-up, which immediately follows this one.
IDR TEAM MEMBERS—GROUP A
Fahmida N. Chowdhury, National Science Foundation
Harry Dankowicz, University of Illinois at Urbana-Champaign
Stephen M. Fiore, University of Central Florida
Lindsey Johnston, University of Georgia
Amy A. Kruse, Intific
Kalev H. Leetaru, Georgetown University
Gloria J. Mark, University of California, Irvine
Shelie A. Miller, University of Michigan
Ioannis Paschalidis, Boston University
Iraj Saniee, Bell Labs, Alcatel-Lucent
Lindsey Johnston, NAKFI Science Writing Scholar University of Georgia
IDR Team 2A was asked to answer the question “how would you spend $100 million over the next five years to understand and harness the power of social networks?” The overall theme of the conference was collective behavior. Social networks are built on collective behavior. There is a wide range of challenges facing the area of social network research. The team started the conversation by addressing some of those challenges. The team also discussed where the money would be coming from and picked three main areas of study where it could focus research: (1) understanding the fundamentals, patterns, and structure of social networks; (2) dealing with multiplicity of scales, networks, and species and their interactions; and (3) harnessing the power of social networks for the public good.
Why Are Social Networks Important?
Social networks exert a powerful influence on human thoughts and behavior. They consolidate memory, shape emotions, cue investigative thinking and biases in judgment, influence in-group/out-group distinctions, and may affect the fundamental contents of personal identity. Because of these influences social networks are important in a variety of societal contexts: for example, they have been observed to change the course of insurgencies, and frame negotiations; they have played a role in political radicalization, and influence the methods and goals of social movements.
By studying social networks in the animal and insect kingdoms, i.e., ants, prairie dogs, bees, and butterflies, one may be able to infer how human social networks function. Principles from animal and insect networks have been shown to be applicable to other species, including humans. In
addition studying the collective behavior of these species can provide insight into how to understand patterns in their collective behavior.
Research Challenges
Some of the challenges discussed were privacy, data access and standardization, ethics, and defining the scientist’s role in social network studies. The first question the team posed was about how to make sure research subjects remained anonymous in any kind of social media or Internet-based research. Along with this issue of privacy is the lack of knowledge of how the Institutional Review Boards (IRBs) work among researchers who do not normally use it in their field. The team decided these researchers would need to be educated about how IRBs function before researchers could conduct studies with social scientists. Not only is there an issue of privacy, but it is also difficult to get data from companies. Large for-profit companies, i.e., Facebook and Twitter, have significant amounts of social network data, but it is difficult for researchers outside these companies to access these. If researchers can get access to the privately held data, it is often at great expense. The question of the scientist’s role in using data from social networks was raised during the discussion. Are scientists supposed to just observe social networks or should they also intervene in the networks’ activity?
Who Has the $100 Million?
After the team talked about the challenges with social network research, the participants discussed where the hypothetical $100 million would come from. Who provides the funds is important because that would influence what direction the research and spending would take. They discussed how projects would differ if the funding came from the National Science Foundation versus the Defense Advanced Research Projects Agency, National Institutes of Health, or a private organization. However, despite these differences, the team understood that they could narrow down the scope of potential expenditure to maximize the ultimate impact; for that, they needed to focus on the fundamental of social networks.
Fundamentals, Patterns, and Structure of Social Networks
One of the main questions the team focused on when discussing the fundamentals, patterns, and structures of social networks was how social
network research can be leveraged to increase resilience to negative outside influences, such as rumors and general misinformation.
Another question raised about resilience was whether resilience was a reflection of confidence in society and the government. In trying to answer this question, the team discussed people’s tendency to act based on emotional response rather than on facts—i.e., believing propaganda instead of looking for facts. The team also considered the propagation of rumors and information that alters society in major ways, as well as how this information is generated and how it spreads through social networks. An example of this is the supposed identity of the Boston Marathon bomber being spread throughout Reddit, which had led to a false arrest. Adding to the team’s original definition of resiliency, they decided that a network could be more resilient if it could distinguish between perception and reality, which could help the network be resilient to the spread of rumors. Exactly how this would be accomplished technologically remains an important matter for study. An example of resilience that the team came up with was how networks can help people survive; during the Iraq war, people relied on their established social networks, as well as newly created networks with people living outside Iraq, to help them determine when it was safe to leave their homes. These networks helped keep people alive.
In addition to discussing resilience in networks the team also explored how social networks learn by adapting to change as well as how networks change over time.
Dealing with Multiplicity of Scales, Networks, and Species and Their Interactions
The second topic the team examined was how to look at different kinds of social networks on multiple scales and how those networks interact with each other. The team acknowledged that there was not sufficient research to examine social networks through multiple scales. The team decided they would want to use a multiscale, multilayered study that would look at social networks on a range of scales: on the individual level, on the group level, the societal/ecosystems level, and on a global scale.
They did not just want to focus on human social networks, however. There are social networks among insects and members of the animal kingdom as well. Social network studies can be performed with any animal that has a communication system, including ants, prairie dogs, and bees. The team discussed the need for more research on species-to-species communi-
cation. An example of this would be how pollination patterns of bees and butterflies affect each other and how these patterns interact. If the team were to do a study about something like cross-species pollination they could look at whether these behaviors resulted in competition, collaboration, or a combination of both.
The team also thought it would be interesting to look at multilayered networks that are interconnected, such as humans interacting with robots and other systems of artificial intelligence. In order to carry out this kind of study they would want to design a machine, using computer algorithms that would work symbiotically with human networks to drive the reward system and increase the resilience of the group. By building a relationship with the computer, the individuals involved would indirectly build relationships with other people. This computer-mediated resilience could allow the group to fact-check itself, thus building resilience to rumors and misinformation as well as a way to adapt in times of change and conflict. The team then tied this back to earlier discussion about creating a research infrastructure by exploring the question of how ecosystems achieve resilience through networks between species and also within species. All of these issues relate to societal impacts and public good.
Harnessing the Power of Social Networks for the Public Good
When they first examined this topic, the IDR team discussed using social networks to improve communication during crises such as natural disasters, mitigating the spread of epidemics, and using the power of social networks for counterterrorism, conflict resolution, and education.
After more discussions, the team came up with the idea of creating a social network that people would feel committed to, encouraging them to remain a part of the network over a long period of time. Part of the incentive structure would be that people could observe the consequences of their actions. Participants in the network could see how the ideas they contributed to the network were advancing knowledge on a certain topic.
The team discussed how they could build a network where people would feel connected, and they decided that some kind of nonmonetary incentive structure could be useful. The network would be similar to Wikipedia in that it would allow people to exchange knowledge and ideas. The team looked at how networks like Galaxy Zoo, a place where people can name new galaxies, works. They also proposed platforms similar to those used for crowdsourcing and the idea of citizen science to answer
important scientific questions, as in the recent progress on twin primes made by the larger mathematical community. Anyone could participate in the network and post their answer to the questions that the team posed. Through this kind of social network, participants’ cognitive surplus could be used to advance knowledge. A network like this could also help generate ideas about how to encourage society to engage in sustainable behavior. For example the team could pose a question about how people would solve the tragedy of the commons.
Conclusion
The team decided that they would spend $100 million over the next five years to harness the power of social networks for public good, and in order to achieve that goal, there must be research on understanding multiscale, temporal, complex networks that included humans and nonhumans (computers, other species).
IDR TEAM MEMBERS—GROUP B
Hans A. Hofmann, University of Texas at Austin
Barbara R. Jasny, Science/AAAS
Challa Kumar, Louisiana State University
Chris McGee, University of Georgia
Ziad W. Munson, Lehigh University
Wenying Shou, Fred Hutchinson Cancer Research Center
Jennifer Vanos, Texas Tech University
Raffaele Vardavas, RAND
Christopher McGee, NAKFI Science Writing Scholar University of Georgia
“. . . We must learn to treat comparative data with the same respect as we would treat experimental results. . . .”
- J. Maynard Smith & R. Holliday, 1979
IDR Team 2B was asked to theoretically spend $100 million over the next five years to understand and harness the power of social networks.
The team collectively decided that the discussion should focus on the ideas behind the concept rather than the $100 million itself. Research, in general, should be funded on how bold or innovative ideas are, rather than simply funding something expensive.
A four-question model was developed by the IDR team to address the research question: (1) How should the boundaries of networks be defined? (2) How can data mining tools be applied to networks to elicit information? (3) Should dynamic and static networks be differentiated? (4) Should naturally occurring networks and intentionally built networks be differentiated for purposes of study?
The size of the networks was defined as small (10-12 members), intermediate (1000 or more members), and large (millions of members). Three types of network model systems were identified: human, nonhuman animals, and artificial or inanimate systems.
The team saw the need to craft a question broad enough to capture the imaginative investors or funders in the event that there are enough good ideas to justify spending $100 million on research on social networks. The question developed would be a counter question to the research challenge question and intentionally tackle the key words that drive the research. In this case, the key words are “harnessing” and “understanding” the power of social networks. The team came up with this counter question:
Can unifying principles be identified for how information flows through networks and how networks respond to perturbations?
The IDR team members additionally framed individual counter questions:
What is the difference between information-based and material-based networks? If any?
How do we optimize networks when disaster strikes?
Are the appropriate technologies available to approach research on this question? If they are, how will those technologies be developed and adapted in using $100 million to understand and harness the power of social networks?
What are the tipping points and criticality that control a network response?
How is the initial condition important to understand the system?
Identifying Social Network Complexities
Each member of the IDR team provided a specific example of a social network, applicable to the counter research question, to achieve a clear focus. Six examples—chosen from many—could potentially be included in a $100 million social network research project.
The first example of a social network for study was about parenting—the interplay between the structure of a network and how individual members behave. With this problem, it is important to understand how social networks reach critical mass, or tipping points, and at what point amplification occurs, causing a phase transition and affecting the system. The team highlighted two very different specific topics. The first was societal factors that influence whether parents choose to vaccinate their children. The second example focused on tipping points in terrorist organizations. Terrorist organizations operate on interplay between familial, ethnic, and sociopolitical conventions that lead to the mobilization of a violent organization.
The third example of the complexity of social networks was disaster response systems. The IDR team wanted to understand the permeation of information in these networks to further study the behavior of individuals who respond to disasters. Disaster response networks, for purposes of this discussion, refer to only weather or naturally occurring hazards, like hurricanes or tornadoes, and what would drive individuals to flee from town or seek shelter in a basement.
The fourth example was about understanding how nanomaterials combine and interact in a microenvironment and how their properties change as a consequence (e.g., how nanomaterials act in the process of building self-driving cars).
The fifth example focused on microbial communities and how they change in group-level behavior and interaction, causing feedback to individual members. This discourse will help delineate the differences between macro- and microsystems and how it helps to make the track universal principles of networks better.
The sixth example was to identify and understand the process that social networks must go through to achieve a favorable end. One IDR team member used the example of setting up a global virus network for scientists. Labs of this kind have previously been established with the intent of streamlining communication among labs. Though the technologically advanced labs are readily available, they have not been used to their full potential.
Thus, this member is interested in researching what is needed to get nodes to engage in a network from which they would benefit.
Project Analysis
At the end of the sessions, the IDR team came up with several goals to achieve success in a $100 million research project with six examples of relevant social networks.
The IDR team also found it necessary to define why artificial networks should be studied when measuring the power of social networks to produce data that are particularly relevant to humans. The team finds that human systems are unpredictable and the complexity and attributes of human behavior need to be measured precisely. Therefore, measuring artificial systems is needed to wield more control for the researcher. It is understood that artificial systems do not necessarily recapitulate human systems, but measuring artificial networks first is useful to produce the generality needed to narrow down what should be measured in human networks.
The team set forth the basic principle that the understanding and harnessing of the power of social networks is not unidirectional, but iterative, or mutually informative. The study networks at various scales and levels (human, nonhuman animals, artificial/inanimate) as an iterative process will capitalize on the advantages of each approach.
To achieve optimal results, it is important to build a community of researchers across disciplines who work as collaborative teams to truly understand the basic principles of a social network, and most important, identify commonalities across networks. To achieve this goal an additional workshop, outside of NAKFI, is needed to get all the project researchers in one room and strive to reach a common goal.
Conclusions
With robust conversation and a complex problem, IDR Team 2B left the NAKFI conference with loose ends on the research question. Rather than itemizing the $100 million dollars toward harnessing and understanding of social networks, the IDR team identified several examples of present challenges to social networks where the money could be beneficial in aiding research.
Outlook
The IDR team collectively decided to apply for a seed grant to fund a small, 10- to 15-person workshop which would delve further into harnessing and understanding social networks. The interdisciplinary workshop team would host researchers with work relevant to the power of social networking. The primary goal of the workshop would be to find commonalities among the different systems of social networks (human, nonhuman animals, and artificial/inanimate) to produce content analyses. Researchers in the workshop would hope to develop a possible research program. The workshop will be the first step in building an interdisciplinary community of researchers who can effectively propose research that aims to understand and harness the power of social networks.
