Scale-Free Networks: Why Fairness is not the Issue

Networks generally fall into two distinct structures – scale-free and random.   

Random network versus scale-free network

Random networks assume each node has an equal chance of being connected to another node, whereas scale-free networks have a small number of high-degree nodes (hubs) and a large number of small-degree nodes (Hoogduin, 2015).

Examples of scale-free networks are all around us: wealth distribution, STD transmission, and social network popularity.  Artist Erin Gallagher visualized the usage of #metoo on Twitter between October 15 and 17, 2017.  Her image clearly shows a scale-free network, with a small number of highly-connected hubs and the vast majority of users having only a few connections.

This is a common pattern for Twitter and other networks that are based on preferential attachments.  The popular accounts have more users, which amplifies their tweets so that they reach more people.  This higher level of exposure brings them even more followers (Robins, 2015).

Initially, scale-free networks may seem to go against the principles of democracy and fairness.  For example, if power is in the hands of a few instead of many there is danger the powerful will exploit the less powerful.  This is certainly how dictatorships and authoritarian regimes work.  However, there is a difference between power being artificially hoarded and power being unequally distributed.  Artificially rigging systems to disproportionately benefit those already in power is both unequal and unfair.  Disproportionality in blog popularity, as Clay Shirky discusses in Power Laws, Weblogs and Inequality, is unequal but largely fair.    

Scale-free networks are a fact of life – it does no good to argue whether or not they should exist (because they already do) or whether or not they are fair (because fairness depends on the context).  Instead, we should look at the result and impact of network distributions.  Are certain people or groups being disenfranchised or disproportionately hurt?  Is this inequality purposeful, or due to the cumulative impact of preferential attachments?  Purposeful inequality that disproportionately and negatively impacts people should be corrected.  Inequality that puts power in the hands of those who have historically been excluded (such as the citizen journalism Shirky discusses) can enhance freedom and democracy.  When the cause is unintentional preferential attachments, the decisions are more complicated.  Artificial network regulation may be necessary or new networks may need to be created to increase access to missing resources.   We must be careful to maintain a balance between the freedom of individual choices and the collective impact.

References

Hoogduin, L. (2015). Random and scale-free networks. Retrieved from https://www.youtube.com/watch?v=dNbSWsQGHsw&t=57s

Robins, G. (2015). Doing social network research: Network-based research design for social scientists. Thousand Oaks, CA: Sage.

From adults in prison to adolescents using drugs: How SNA helps us understand peer networks

Social network analysis helps researchers and practitioners better understand crime and delinquency networks.  By analyzing the relationships and connections between network members, they have much richer data to help explain complex behaviors, attitudes, and beliefs.  Two articles from different populations demonstrate this point.

Schaefer, Bouchard, Young, and Kreager (2017) used SNA to investigate social structures in an adult prison.  They set relationships between incarcerated persons as the point of analysis instead of individual attributes because they believed it would give a more complete picture of the prison groups.  Specifically, the researchers wanted to determine if the characteristics shown in previous research to influence network structure (race and ethnicity) were the actually the predominant factors in group formation.  The researchers also included other factors past research past research suggested may impact grouping, such as age, religion, gang status, prison tenure, and power.  They collected the data through interviews and department of corrections information.  The incarcerated people were the nodes and the “get along with” relationships were the links.  The “get along with” nomination did not have to be reciprocal to count.  Since there was little existing SNA-based research on prison social structures, the researchers compared the prison network to adolescent friendship networks.

The researchers did find evidence of some race/ethnicity homophily within networks; however, the impact was not as high as prior research suggested (Schaefer, 2017).  In fact, the homophily rates were similar to those in adolescent friendship networks.  The subgroups and cliques showed more racial and religious diversity than segregation.  The participants also had similar numbers of friends as adolescents, although there was less reciprocity in the relationships, making the ties somewhat weaker.

This research advanced the understanding of prison network composition because the analysis showed less racial/ethnic divisions and gang structures than previous research.  This is critical information for people who run programs within prisons and programs targeted toward people recently released from prison. 

Kirke (2004) used SNA to better understand the contributions of peer selection and influence in adolescent substance use.  Researchers have long tried to determine the impact of selection versus influence on adolescent behavior.  The research question for this study was: is there direct evidence of substance abuse similarity in adolescents’ peer networks and is there evidence of influence between similar peers?  Kirke collected data through in-person interviews of 267 teenagers in one district in Ireland.  The teenagers were the nodes and friendship relationships were the links.  She used the name generator technique to construct the links.  

Kirke (2004) found a complex pattern regarding substance use in adolescent peer networks.  Participants chose friends who had both similar and different substance use patterns from them.  Participants were also influenced by peers both within their networks and outside of their networks.

Previous research had focused on influence from within the group and had largely ignored the implications of out-group influence.  Social network analysis allowed the Kirke to see the nuances in the friendship networks and clarify the roles of selection and influence.  This knowledge, in turn, can help professionals create effective prevention and intervention programs.

References

Kirke, D. M. (2004). Chain reactions in adolescents’ cigarette, alcohol, and drug use: Similarity through peer influence or the patterning of ties in peer networks? Social Networks, 26, 3-28. doi: 10.1016/j.socnet.2003.12.001

Schaefer, D. R., Bouchard, M., Young, J. T. N., & Kreager, D. A. (2017). Friends in locked places: An investigation of prison inmate network structure. Social Networks, 51, 88-103. doi: 10.1016/j.socnet.2016.12.006

Victimization and racial attitudes within networks

SNA can provide important insights into social problems.  For example, Swartz, Reyns, Wilcox, and Dunham (2012) used SNA to study the victimization experiences of high school students.  Unlike in similar studies, these researchers focused on patterns within friendship clusters instead of the entire school network.  They looked at whether victimization rates varied between friendship clusters, what those variations looked like, and individual differences found within clusters.  Swartz and colleagues collected data from 541 students from the same high school.  Students completed surveys on past victimization, delinquency, and other personal and family characteristics.  The students also listed the names of their five closest friends.  The researchers constructed friendship clusters using geodesic distances.  In the final analysis, the friendship clusters served as the nodes and the number of friendship ties between clusters linked the nodes.

Peer victimization network. Large squares represent clusters high in victimization; thick edges represent high friendship connections (Swartz et al., 2012).

Overall, the researchers found that high victimization clusters were prominent in the school-wide network; however, the clusters were not highly connected to each other (Swartz et al., 2012).  There was no cluster of clusters to indicate that victimization somehow spread throughout the network.  Within the high-victimization networks, the researchers found that the cluster-based victimization rates were because of the high victimization rates of a few members as opposed to experiences of a majority of the members.  The unique use of clusters as nodes in this study helped simplify the network visualization so that patterns were more apparent.  Additionally, as the researchers reported, existing research and theory holds that the peer group is highly influential in adolescent victimization.  This study provided important information on the presence of victimization within friendship clusters and on the connections between friendship clusters.

SNA has also been used to explore the impact of racial attitudes on socialization.  Tawa, Ma, and Katsumoto (2016) used Second Life (a virtual world) as the context to understand how people who express colorblind attitudes function within diverse networks.  They hypothesized that people higher in colorblind attitudes would have less central roles in diverse networks.  The researchers recruited 64 participants and asked them to create Second Life (SL) avatars that physically resembled themselves.  Participants completed surveys measuring colorblind racial attitudes, outgroup prejudice, previous SL usage, and avatar racial accuracy. The researchers created a specific SL environment for the purpose of the study and observed participant interactions over five 15-minute events.  Researchers collected data on avatars’ friendship statuses, physical distances from each other, and chat participation.  The avatars were the nodes and their friendships were the links (friendship was defined by reciprocal inclusion in SL friendship lists).  

Screen shot of participant interactions (Tawa et al., 2016)

Tawa and colleagues (2016) found that people with higher colorblindness attitudes were less central to the network.  They had lower levels of closeness centrality and of clustering.  This means that while they did have some direct connections, they were not part of subgroups at the same rate as those lower in colorblindedness.  The researchers also showed that the impact of colorblind attitudes was present even when outgroup prejudice attitudes were considered.  Participants high in colorblindedness kept larger physical distances from outgroup members, which could have contributed to their low closeness centrality and clustering scores.  The use of SNA in the Second Life context allowed researchers to to observe interactions and relationships to measure clustering and network positionality in a way that would not have been possible in an offline environment.  

References

Swartz, K., Reyns, B. W., Wilcos, P., & Dunham, J. R. (2012). Patterns of victimization between and within peer clusters in a high school social network. Violence and Victims, 27, 710-729. doi: 10.1891/0886-6708.27.5.710

Tawa, J., Ma, R., & Katsumoto, S. (2016). “All lives matter”: The cost of colorblind racial attitudes in diverse social networks. Race and Social Problems, 8, 196-208.  doi: 10.1007/s12552-016-9171-z

Network density and coalition development

In an article on youth violence prevention coalitions, Bess (2015) investigated the ways that the coalition impacted the broader youth violence prevention (YVP) system.  The researcher analyzed three different network properties: density, hierarchy/centralization, and homophily.  For the sake of clarity, I will be discussing how the network density aspects of the study related to coalition capacity building.

As Figure 1 shows, Bess (2015) conceptualized the broad YVP Intervention System (Network A) as being comprised of two subsystems (Networks B and C).  The official YVP coalition members made up Network B and the other organizations that worked on YVP issues but were not official members comprised Network C.

Research question and hypotheses

How does the density of the YVP Intervention System network change over the five year time period?  In this study, density represents collaboration, with higher levels of network density corresponding to higher levels of collaboration.  Previous research on coalition building has shown that collaboration increases as the coalition forms, and then decreases as the coalition stabilizes and moves into implementation.  Based on that research, Bess (2015) hypothesized that network density would increase during beginning years of the study (coalition formation) and then would decline.

Sample population

The sample included 99 organizations (both public and private non-profits) that participated in YVP work in a mid-sized southeastern city (Bess, 2015).  The sample varied from year to year as organizations came and went.  There were 99 today organizations that participated in at least one year of the study.

Data collection

Investigators collected the data through yearly in-person interviews with organization members who were most familiar with the group’s YVP work (Bess, 2015).  To construct the networks, researchers gave participants a list of the active YVP organizations and asked the participants to identify the organizations they had collaborated with on YVP activities.

Network Structure

The nodes in the networks were the YVP organizations (Bess, 2015).  Official coalition members made up Network B and YVP organizations that were not in the coalition comprised Network C.  Networks B and C combined to form Network A.  The network links were the collaborative relationships between nodes.  A link was formed when agency A listed agency B as an organization with which they collaborated.  The networks were directed because reciprocity was not assumed.

Results

Bess (2015) noted that density is often a function of network size, as larger networks tend to have lower density.  Because the network size varied over the course of the study, analysis focused on overall patterns in increases and decreases instead of discrete values.  As hypothesized, network density increased in the initial years of all three networks and then declined in the later years. Specifically, density peaked in year two and then declined to similar or lower levels as the baseline year.  

These results supported components of coalition development theory (Bess, 2015).  Coalition development theory states that collaboration is highest when coalitions are forming, and then decreases once the coalition has built the necessary capacity for work.  In this context, lower levels of collaboration represent coalition efficiency, not dissolution.  

Network density measures are critical for this type of analysis because they present a picture of of the overall functioning of the network as opposed to information on the nodal level.

Reference

Bess, K. D. (2015). Reframing coalitions as systems interventions: A network study exploring the contribution of a youth violence prevention coalition to broader system capacity. American Journal of Community Psychology, 55, 381-395. doi: 10.1007/s10464-015-9715-1

Public Spheres and Networked Societies: No Matter How Things Change They Always Stay the Same

Coffeehouses, salons, and literary associations were the original public sphere

Imagine a place where people gather to exchange information and discuss ideas as equals. A place that is characterized by civil engagement, is accessible to all, and exists outside of class, government, or business structures.  This is not a scene from a utopian novel; it is what sociologist Jürgen Habermas referred to as the public sphere.  The original public spheres were the literary associations, coffeehouses, and salons frequented by the bourgeoisie.  They engaged in rational debates and generated shared understandings free from the influence of the government.  Habermas believed the ideal public sphere was open and accessible to all and held government and business accountable to the social good. 

In his 1962 book The Structural Transformation of the Public Sphere, Habermas discussed the decay of the public sphere.  Whereas the press used to play an important part in the public sphere by making information freely accessible, Habermas argued that the media was now part of the problem.  In his view, the corporatization of the media blurred the lines between public and private.  Political, business, and media institutions infiltrated the public sphere and usurped the power of the common people.  

Instead of a deteriorating public sphere, sociologist Manuel Castells sees the promise of an open, unbounded network that gives people more access than ever before.  He calls this the network society – a society where technology-based networks are more integral to society’s functioning than institutions. People can have a presence in multiple places at one time and can spend their time on multiple activities at once.  In the network society, people are not dependent on institutions for information or service.  They can mobilize through their technological networks and impact education, governments, and economies.    

Diagram of industrial society versus network society
From Open Network Society

We have certainly seen some of the positive impacts of the network society.  The Occupy movements showed networks of people standing up to capitalist institutions.  In the Movement for Black Lives, Black people are leading demands for structural change to the institutions that have oppressed and killed them for centuries.  However, the network society has also facilitated the public regrowth of hate groups such as the white supremacy movement.  

Black Lives Matter protester holding sign in front of police line
#BlackLivesMatter

Habermas’s critique of the commodification of the press still rings true today, and has grown more relevant as the network society has grown.  News sources are neither independent nor unbiased.  They are increasingly corporate and political.  A large percent of people in the U.S. got at least some of their 2016 presidential election news from social media.  Social media sites, however, are not in the public sphere.  They are institutions that financially profit off of the spread of news.  For example, in addition to the micro-targeted advertising that Facebook routinely sells, the company also embedded Facebook employees within the Trump campaign (Clinton declined a similar offer). 

The network society has changed the public sphere, but the changes began long before the growths in technology.  In fact, Habermas’s original depiction of public spheres was just as idealistic as Castells’ view of the network society.  The working class and other oppressed groups did not frequent the salons.  Access to the public sphere has always been constrained by race, ethnicity, gender, class, and other characteristics.  In the network society, technology has broken down (some) barriers, leveled (some) playing fields, and opened society (for some).  People from minoritized races and ethnicities, trans people, and people with disabilities still face barriers to full societal participation.  As long as our underlying societal structures reward power, any technology we lay on top will replicate disparity and oppression.

(c) the Roosevelt Institute

References

Habermas, J. (1991). The structural transformation of the public sphereBoston, MA: MIT Press.

Van Kreiken, R. (2016). Castells and the network societyUniversity of Sydney.

Using SNA to Understand Social Media Connections Among Red Flag Campaign Campuses

Health promotion and violence prevention campaigns are increasingly using social media to help spread their messages (Potter & Stapleton, 2011).  The Red Flag Campaign (RFC), a project of the Virginia Sexual and Domestic Violence Action Alliance, is one example.  The RFC is a bystander intervention campaign designed to prevent interpersonal violence and increase healthy relationships among college students.  Given the high rates of social media use among young adults, the RFC includes a social media element.  

All consuming possessiveness or suspicion is #ExcessiveJealousy." If you see a #RedFlag in a friend's relationship, #SaySomething. #RedFlagCampaign #HealthyRelationships
Example of Red Flag Campaign content for Instagram
Do you know the #RedFlags? If you see something, #SaySomething! #HealthyRelationships #RedFlagCampaign
Example of Red Flag Campaign content for Instagram

Some aspects of the RFC have been evaluated previously, including the visual appeal, the messaging, and its reach within individual institutions.  No one, however, has looked at the relationships among participating schools.  My research addresses this by asking, How are the RFC partner campuses connected to each other and the Action Alliance on social media?  

I will use a directed, bounded network comprised of 17 campuses selected to work closely with the Action Alliance in implementing and evaluating the campaign (n=18).  To collect the data, I will sort through the Facebook page likes for each node and record when each node likes the page of another node in the network.  I will repeat this process with the list of Instagram accounts and Twitter accounts each node follows.

Spotlight on Tidewater Community College
Example of a connection between a campus and the Red Flag campaign Facebook page.

I am focusing on the social media aspect of the RFC because previous research has shown that social media can be an important tool to fight rape culture on college campuses (Giraldi & Monk-Turner, 2017) and to encourage campus activism (Linder, Riggle, Myers, & Lacy, 2016).  

To the best of my knowledge, social network analysis has not been used to study organizational relationships in bystander intervention campaigns.  I am drawing upon existing social network research related to general nonprofit and public health collaboration to inform the methodology and analysis of this project.  Johnson, Honnold, and Stevens (2010) analyzed four types of relationships among a regional network of nonprofits.  As I am looking at three types of relationships (Facebook, Instagram, and Twitter), their study provides conceptual and analytical approaches I can use.  Valente and colleagues (2015) used social network analysis to understand the implementation process (including sustainability) of health promotion programs.  The study’s inclusion of program sustainability is particularly relevant to my project as the RFC is in the sustainability stage.

The current project will build upon previous research by applying organization level social network analysis to a bystander intervention program.  Social network analysis is not widely used in interpersonal violence prevention work, although it holds great promise.   This project is one step toward a wider adoption of this methodology.

 

References

Giraldi, A. & Monk-Turner, E. (2017). Perception of rape culture on a college campus: A look at social media posts. Women’s Studies International Forum, 62, 116-124.

Johnson, J. A., Honnold, J. A., & Stevens, F. P. (2010). Using social network analysis to enhance nonprofit organizational research capacity: A case study. Journal of Community Practice, 18, 493-512.

Linder, C., Riggle, C., Myers, J. S., & Lacy, M. (2016). From margins to mainstream: Social media as a tool for campus sexual violence activism. Journal of Diversity in Higher Education, 9, 231-244.

Potter, S. J. & Stapleton, J. G. (2011). Bringing in the target audience in bystander social marketing materials for communities: Suggestions for practitioners. Violence Against Women, 17, 797-812.

Valente, T. W., Palinkas, L. A., Czaja, S., Chu, K. H., & Brown, C. H. (2015).  Social network analysis for program implementation. PLOS One. doi: 10.1371/journal.pone.0131712

Using node centrality to understand network dynamics

One of the most helpful aspects of social network analysis, to me, is its use of visuals.  Being able to literally see a picture of a network helps me understand patterns and relationships in a way that neither numerical or textual descriptions can.  Eyeballing the graph is helpful, but social network metrics give us an even deeper picture.  Node centrality measures (betweenness, degree, closeness, and eigenvector) describe specific characteristics of the relationship between nodes.  This infographic briefly describes what these measures can tell us about the network.

(c) Beehaus

Social network analysis is very flexible and can be used to study many different types of networks.  SNA is a promising, albeit underused, methodology in understanding bystander intervention.  Research has shown that norms supporting prosocial bystander behavior increase people’s intent to help, while norms against intervening decrease intent to help (Hoxmeier, Flay, & Ackok, 2016).  Researchers have not, however, used SNA to look at how the group norms are created and spread in the first place.  People who are high in closeness centrality spread new information quickly and efficiently; they could be identified and targeted for specific training.  Identifying people high in betweenness centrality would help prevention educators strategically interject prosocial norms to more isolated parts of the network.  As opposed to training as many people as possible, SNA could help educators efficiently target programming where it would do the most good.

Researchers have used SNA to study gender-based violence response networks.  For example, Rana and Allen (2015) compared the networks of organizations in five separate family violence councils.  They found that the prominent organizations were different in each council.  For example, the domestic violence program was clearly the most important organization in councils A and C as it was highest in all three centralities.  In council D, the prominence of the domestic violence program depended on the role being investigated.  It was highest in betweenness, meaning it was the key broker/bridge in the network.  Child and family services was highest in degree and closeness, meaning it was connected to more organizations and involved in more relationships between organizations.  The researchers discussed the findings in relation to leveraging relationships between organizations to further council goals and initiatives.  The SNA approach allowed for more nuance in understanding the relationships than other analyses would have.

Quinlan and Quinlan (2010) used SNA in a much different way.  Instead of looking at people or organizations, they analyzed institutional and lived experiences of rape.  In the institutional network, the nodes represented the components of the forensic medical exam.  They included physical evidence, survivor history, assault details, and medical professionals.  The links between nodes represented the connections between aspects of the exam.  The lived experience network, however, set the survivor’s feelings, actions, and thoughts as the nodes in the network.  Time (order of what happened when) was represented by the links.

The researchers used node centrality measures to determine the most prominent components of each network.  For the lived experience network, the survivor’s feelings of fear and horror had the highest degree centrality.  This suggested that for this survivor, those feelings were the most powerful aspects of her experience.  For the institutional network, the survivor’s identity had the highest degree centrality and the assault had second highest.  This suggested that the survivor’s identity was more central than any other aspect of the exam process.  In the discussion, the researchers related these findings to the ways that rape is viewed and treated in society.  The use of SNA gave a different and compelling perspective on rape and sexual violence research.

 

References:

Hoxmeier, J. S., Flay, B. R., & Ackok, A. C. (2016). Control, norms, and attitudes: Differences between students who do and do not intervene as bystanders to sexual assault. Journal of Interpersonal Violence, 1-23, doi: 10.1177/0886260515625503

Quinlan, E. & Quinlan, A. (2010). Representations of Rape: Transcending Methodological Divides. Journal of Mixed Methods Research, 4, 127-143.

Rana, S. & Allen, N. E. (2015). Centrality measures to identify key stakeholders in Family Violence Councils. Psychosocial Intervention, 24, 167-176.

Social capital and social networks

Social capital refers to the resources available through relationships and social networks.  It can be built, invested, transferred, and spent.  

Social capital diagram
(c) Coact 2015

One of the most well-known discussions of social capital is in Robert Putnam’s article (1995) and subsequent book (2000), Bowling Alone.  In the book, Putnam explores what he saw as the decline of social capital in the U.S., and the subsequent negative impacts on communities and individuals.  Note the timing of the article and the book – while the internet already existed when Putnam wrote them, it was not as ever-present as it is today.  Cell phones were still used primarily for emergencies and dial up modems were the primary way most people accessed the internet.  The chart below shows the growth in internet and mobile technology usage since 2000.

It is unlikely that Putnam, and other social scientists, could have even imagined the role that the internet and internet-related technology would play in the building and transmission of social capital today.  Current research has tried to tease out that relationship.  While the question of whether the internet increases or decreases people’s social capital has not been definitively answered (see here and here for examples), many researchers agree that its prevalence in our lives has changed how social capital is built and valued.  As Charles Steinfield said, we are not bowling alone, we are bowling online.

Social capital impacts all facets of our lives, from community resilience to health outcomes. Individual social capital is an important, but not often discussed, part of gender-based violence prevention.  Bystander behavior is very much embedded in people’s social networks – network norms shape people’s attitudes and behaviors in ways that can support and/or oppose violence and abuse.  In networks where social norms are tolerant of abuse, individuals who speak out against it pay a price in social capital.  Members of the network who have more social capital (from status, prestige, or other valuable characteristics) can afford to take risks and lose social capital in ways that members with less social capital cannot.  This makes the decision making process of whether or not to intervene, and in what way to intervene, different for those with high versus low social capital.

People who occupy prominent roles in the network and have high social capital are also more effective at changing the group norms around bystander intervention than those with low social capital.  This is why prevention programs such as Green Dot and Huddle Up target high status individuals.  To my knowledge, social network analysis has never been a part of the evaluation of these types of programs.  Although methodologically challenging, it would be helpful to see how bystander intervention information actually spreads through sample networks.  Were the individuals and groups who were identified as high status by the program organizers (usually administrators or professional staff) the ones who 1) had high social capital and 2) were willing to activate their social capital for this issue?  

Social capital is also an important area of research around institutions’ response to gender-based violence. Research that answers the following questions would provide critical information to improve institutional response to survivors.  Does social capital impact survivors’ choices report assaults or participate in institutional or criminal hearings? If so, how?  What do response systems look like at institutions where advocates have high social capital versus where they have low social capital? How does social capital impact collaborations between departments, and what impact does that have on institutional response?

Is it a small world, after all?

Two weeks ago, I blogged about Six Degrees of Kevin Bacon.  At the time, I used the game as an example to introduce the concept of social network analysis in general.  This week, I am referring back to it as an example of a specific social network principle: the small world theory.  The small world theory, made famous by Stanley Milgram’s research, posits that people can be connected to each other in less than 6 degrees.  The Six Degrees of Kevin Bacon game replicated that theory within the acting world.  While you may not know if you are connected to Kevin Bacon, you have likely encountered a stranger whom you later learned you were connected to in some way.  For me, this happened for the first time in the 7th grade.  I had started at a new school, and we had a project to trace our family trees back as far as we could.  The social studies teacher had done this project for many years, and had found familial connections between students who did not know they were related.  He discovered that I was distantly related to two of my friends (and Warren Beatty)!

My distantly related childhood friends

Before we dive too much into the small world theory, let’s review some key SNA concepts.

Nodes (sometimes called actors) represent individuals, organizations, or other entities.

Ties are the links between two actors in a social network; ties are also referred to as links or relationships.

The purpose of SNA is to describe the patterns of relationships (ties, links) between nodes (individuals, organizations).

 

The strength of ties is dependent on a variety of factors, including the type of network, the research question, and the important relationship characteristics in the analysis (Kadushin, 2012).  In general though, strong ties are relationships where people have multiple, important, deep, or long connections to each other.  Weak ties, on the other hand, are much easier to break and may be based on fewer or more superficial characteristics.  Stronger does not mean better in the context of networks, however.  Weak ties are key to the functionality of the small world theory.  Nodes with large numbers of strong ties tend to be insulated within their own network and disconnected from people in other networks.  As a node’s number of weak ties increases, so does the number of networks to which it can belong.  In his often-cited article on weak ties, sociologist Mark Granovetter (1973) wrote that weak ties serve as bridges between network clusters.

A weak tie serves as a bridge

Resources (such as information, behaviors, or money) are able to spread throughout the larger network due to these weak ties.  Without weak ties, there would be no small world phenomenon.  There would only be a big world, consisting of many separate networks with nothing passing between.  In this regard, weak ties are integral to network diffusion.

Also, without weak ties, my Bacon number would not be 3!

images from imdb.com

SNA and My Bamboo Infestation – How Tracking Unseen Connections Leads You to the Bigger Picture

My backyard is infested with bamboo.  The people who owned the house before us thought it would be a great idea to plant bamboo in the backyard (it does, after all, have many helpful environmental properties).  They neglected to fully account for the growing patterns of bamboo, and when we moved in it had choked out much of the grass.  As bamboo newbies, we tried to get rid of it by chopping down the mature bamboo and by knocking down new shoots as they came up.

We thought we were dealing with this kind of problem:

bamboo shoots at ground level
Image from http://www.bamboogarden.com/

We were actually dealing with this kind of problem:

The root system of bamboo
Image from http://www.bamboobotanicals.ca

Social network analysis operates on the assumption that human behavior is like running bamboo – what you see on the surface (individuals or bamboo shoots) is connected by a complicated path that run just below the surface (relationships between people or the root system of the bamboo plant).  Addressing what is visible is not as effective as addressing the entire system.

Social network analysis (SNA for short) is an integrated way to understand how relationships and interactions impact all facets of our lives (Robins, 2015).  Traditional social science research generally centers on individuals, groups, or systems.  Gender-based violence researchers, for example, often focus on increasing bystander behavior (individual), groups that are at higher risk for perpetration or victimization (groups), and rape culture (systems).  These lines of research are important and have shaped a great deal of current response and prevention work.  Some research questions, however, cannot be effectively answered through traditional methodologies.  For example, why might bystander intervention be more prevalent on some college campuses than others?  Measuring bystander behaviors can tell us there is a difference, but not much more.  Focusing on individuals, groups, or institutional levels will not help us understand the mechanisms that lead to different behaviors in different places.

To do this, we need to change our perspective.

Perspective GIFs - Find & Share on GIPHY

Instead of having individuals, groups, or systems as the center of analysis, SNA looks at the relationships (connections) between them.  These connections form networks, and it is the structure and nature of these networks that are analyzed.  According to Yang, Keller, and Zheng (2017), SNA uses two datasets instead of one.  The node dataset is similar to what is used in other social science research – it includes the individuals or organizations (called nodes in SNA).  The second dataset holds the relationship information – how the nodes are connected to each other.  SNA offers the unique capability to analyze the two types of data together.  This analytic shift offers a more nuanced, dynamic picture of how information, behaviors, norms, and even goods spread throughout networks.

Tracking back to our example above, traditional gender-based violence research may ask people to list the number of times that they have intervened in potentially abusive situations.  SNA, however, would ask people to list those in their social circles who have intervened in abusive situations, people they have talked to about bystander intervention, or people whose opinion on intervening is important to them.  Often we assume (or assert) that social groups have influence over people’s behaviors.  SNA actively investigates those connections to understand for whom and in what ways the influence occurs (Robins, 2015).

Connection GIFs - Find & Share on GIPHY

This relational data is what makes SNA powerful.  It also, however, makes it hard to predict anything.  Experimental studies are the most valid ways to predict outcomes because they control for as many variables as possible except for the ones in question.  SNA does not control for or exclude these other variables, it welcomes them in to the analysis!  In their book, Yang, Keller, and Zheng (2017) review new SNA designs that increase the predictive validity of the research.  The strength of SNA, however, still lies in its ability to tease out and describe network behavior.

 


Information cited above:

  • Robins, G. (2015). Doing social network research: Network-based research design for social scientists. Los Angeles, CA: Sage Publications.
  • Yang, S., Keller, F. B., & Zheng, L. (2017). Social network analysis: Methods and examples. Los Angeles, CA: Sage Publications.