Research of Influence in Offline and Online Social NetworksThe role of tie strength and network degrees in determining the power of social influence.
Jonas SchröderBlockedUnblockFollowFollowingMay 15Almost any article about the analysis of social networks and the flow of information starts by mentioning Stanley Milgram’s famous small-world experiments of the 1960s.
Let’s follow suit.
The first half of this article is about the main findings of classic research on information diffusion through social networks.
Afterwards we will focus on a special kind of social networks: the online kind.
Do we find the same characteristics of offline social networks on social media platforms like YouTube, Twitter, and Facebook?Note: This article will be part of a series on social network analysis and influencer marketing, all based on the research for my master thesis titled “Influencer Fraud on Instagram: A Descriptive Analysis of the World’s Largest Engagement Community”, which is not published yet.
Since almost no one has time these days to read 140+ pages on the subject, I decided to split my master thesis into individual articles of medium length and let the reader decide what interests her.
Feel free to follow my profile to get updates or come back later.
I will add links to the other articles as I go.
It’s a Small World — classic research findings for offline social networksTaken from Milgram (1967)In his 1967 experiment Milgram asked randomly chosen U.
citizens to pass on a letter to random targets using only friends and acquaintances they knew on a first-name basis.
The resulting median path length was 5, meaning that people are generally found to be connected with one another in only a few steps.
This “knowledge” found its way into pop culture and common knowledge (e.
, Six Degrees of Kevin Bacon).
Much more interesting though is his findings regarding the intermediaries’s role.
Half of Milgram’s letters that reached their target went through the same three people: Mr.
Brown, and Mr.
These men were highly connected individuals, or social hubs, bringing together basically the whole network of U.
Granovetter (1973) would call these people bridges, the bottlenecks of information flow through social networks.
He argues that the degree to which social networks overlap depends on the strength of the ties that connect them.
The more similar or homophilous individuals are, the more likely they are to interact and to form strong relationships.
Makes sense: we trust people we know and who tend to have the same background and interest as us.
However, in order to spread new information we need to take a look at weakly tied individuals connecting various social circles with rather different interests.
Thus, weak ties play a special role in the diffusion of novel information, for example a new product or fashion trend.
This assumption was empirically proven by Brown and Reingen (1987).
Weak ties were indeed found to be disproportionally more relevant for the diffusion of new ideas and the spread between sub-communities.
A digital replica of Milgram’s classic experiment supports this finding as well.
In Dodds, Muhamad, and Watts (2003) experiment 60,000 participants were asked to forward an e-mail to 18 targets from several countries.
They found that in successful chains the message was forwarded to people whose relationship the sender described as rather “casual” and “not close”, hence: weak ties.
However, they did not find digital equivalents of Mr.
Jacobs from Milgram’s study.
They found no real bridges.
From the position of any strongly-tied community, outside people appear as outliers.
They are weird.
However, as described above outsiders are important for the spread of new ideas.
We know from many books and articles by about the diffusion of innovation that early adopters of technologies are always perceived to be weird in the beginning.
“Why is this man talking to his hand?”, was on many peoples’ minds in the 90s.
Now everyone has an iPhone and talks to apparently themselves while commuting to work on the subways.
In his often-quoted book The Tipping Point (2000) Gladwell analyzes the spread of any trend or social epidemic — be it technology adoption, fashion trends, or even crime.
He identities three kinds of people who are necessary to spread information like wildfire: connectors, mavens, and salesmen.
Connectors are people like Mr.
Jacobs that bring the world together with their large network.
Mavens are socially motivated people who act as middlemen between individuals and the marketplace through their sharing of information.
When you want to buy a new laptop, you better look for mavens in your social circle.
Salesmen are described as natural persuaders who use subtle and primarily non-verbal cues to influence people’s opinions.
Mavens are data banks.
They provide the message.
Connectors are social glue: they spread it.
Salesmen persuade us when we are unconvinced of what we are hearing.
— Malcom GladwellGladwell assumes that one of the most important function of these people is their acting as translators.
They take the message of innovators and translate it for a broader audience.
Thus, they help to solve Moore’s (1991) chasm problem.
This is important for companies that try to broaden their customer base from early adopters to mass market.
Connectors, Mavens, and Salesmen — or perhaps a more modern term: influencers — can help to do the job.
So much about the old world of interacting with people in person or through analogue means.
Let’s move on to the study of influence in online social networks.
Hello Zuck, I am Tom!.Some research findings regarding online social networksClassic researchers of social influence and the spread of information through offline social networks were not only challenged by computational limits of their time.
They had to deal with many unproven assumptions and limited data as the nature of people’s relationships and their personality was not readily observable but rather addressable through surveys.
Today, thanks to technological advances and the spread of popularity of social media platforms, social network analysis is much more fruitful these days.
People’s connections with one another are readily observable in form of follower-following relationships or graph of friends.
Building and analyzing huge interaction models is not difficult anymore.
Furthermore, people happily share tons of information about themselves and their interests online.
No survey study is necessary to determine heterogenity and homogeneity of groups of individuals — just take a look at who they follow and what they like.
This could potentially solve Manski’s (1993) Reflection Problem often found in network analysis.
Do members of a community share certain characteristics as a result of the group’s influence on them?.Or were people who formed the group similar to begin with?Preview from https://www.
org/stable/2298123Susarla, Oh, and Tan (2011) analyze YouTube’s network structure.
They are able to control for the reflection problem by differentiate between social influence and self-selection of users by systematically separating certain factors.
Their main findings are that power of influence on YouTube comes from a node’s network centrality and that homophily plays a crucial role for social contagion.
Bakshy et al.
(2011) study influence on Twitter which can be defined as the ability to consistently seed cascades that outperform others.
They discover that people with many followers who were influential in the past are indeed more likely to continue their influence.
However, the researchers note that predicting influence on a granular basis is unreliable.
The effect of continued influence is only true for averages, not individuals.
They suggest marketers to use a spread-and-pray portfolio-like strategy instead of betting all on one horse.
Taken from Bakshy et al.
(2011): Everyone’s an Influencer — Quantifying Influence on Twitter.
Despite all the bad news and a turbulent stock performance in 2018, Facebook is still one of the largest social media platforms at present.
Users can connect with friends through bi-directional ties and follow their favorite artists through uni-directional connections.
The NewsFeed algorithm determines which content each individual user gets to see when scrolling through the app or website.
You may get presented a link that your friend Paul shared yesterday and feel like sharing it with your networks, too.
Link-sharing behavior can be seen as one indicator of influence.
Influence within the network is readily observable by Facebook.
However, there are many external sources of influence like in-person contact or E-Mail communication.
Through manipulation of the NewsFeed for 253 million Facebook users, Bakshy et al.
(2012) are able to determine the role of the social media platform in the process of information diffusion.
They create two groups: one in which certain information is filtered out of the News Feed and can only be acquired outside of Facebook (no feed condition), and one where information can be acquired internally or externally (feed condition).
Taken from Bakshy et al.
(2012): The Role of Social Networks in Information DiffusionThe big advantage of an experimental setting like this is that researchers can tackle the reflection problem by controlling for confounding factors related to homophily.
By comparing the behavior of both groups, the researchers find that subjects who are exposed to the sharing behavior of their friends (feed condition) are 7.
37 times more likely to share the same information than those in the no feed group.
Thus, people in our friendlist influence our behavior to some extent.
Additionally, Bakshy et al.
(2012) measured tie strength using several interaction types, e.
, the frequency of private communication through Facebook messages or public communication through comments.
Granovetter’s assumption that weak ties are proportionally more important for the diffusion of new information is again empirically supported.
One more study for Facebook.
Sun et al.
(2009)’s research focuses on diffusion events for Facebook Pages.
When Sally likes the Fanpage of Metallica, this fanning behavior can be broadcasted to some of her network.
Paul sees this and instantly thinks of the metallic snare sound of St.
What a cool band, he thinks, and puts a Like down for Metallica as well.
The result is a huge tree-like diffusion network.
The resulting network below is in contrast to the common assumption that a few people are sufficient to spread information in an epidemic manner (hello, Gladwell!).
Taken from Sun et al.
(2009): Gesundheit!.Modeling Contagion through Facebook News FeedSun et al.
find that Facebook diffusion chains are long in general, but not the result of a single chain-reaction event.
Instead, they are started by a large number of users whose short chains get merged together.
Furthermore, they note that a starting node’s maximum diffusion chain length cannot be predicted with the user’s demographics or Facebook usage characteristics, including the number of friends, after controlling for popularity.
Following this, the identification of trend setters would be hard to do.
ConclusionThe power of influence is not divided equally among individuals of a network.
It can be described as a function of number of connections in a social network (quantity) as well as the strength of each tie (quality).
Knowing many people may not be enough if your connection is not of quality.
Thus, working with influencers that have hundred-thousands of followers is on its own not yet convincing, regardless of whether they are fake or real.
The great appeal of micro influencers is the quality of their relationship with fans which is more important than a broad audience of people who just don’t care.
But this is a topic for another time.
Feel free to follow my profile to get updates about upcoming posts on this topic.
You can contact me on LinkedIn or Twitter in case you want to talk about my research or data science in marketing in general.
Thanks for reading,Jonas SchröderCited LiteratureMilgram, Stanley (1967), “The Small-World Problem,” Psychology Today, Vol.
1, №1, 61– 67.
Granovetter, Mark S.
(1973), “The Strength of Weak Ties,” American Journal of Sociology, Vol.
78, №6, 1360–1380.
Brown, Jacqueline Johnson and Peter H.
Reingen (1987), “Social Ties and Word-of-Mouth Referral Behavior,” Journal of Consumer Research, Vol.
14, №3, 350–362.
Dodds, Peter Sheridan, Roby Muhamad, and Duncan J.
Watts (2003), “An Experimental Study of Search in Global Social Networks,” Science, 301, 827–829.
Manski, Charles F.
(1993), “Identification of Endogenous Social Effects: The Reflection Problem,” The Review of Economic Studies, Vol.
60, №3, 531–542.
Susarla, Anjana, Jeong-Ha Oh, and Yong Tan (2011), “Social Networks and the Diffusion of User-Generated Content: Evidence from YouTube,” Information Systems Research, Articles in Advance, 1–19.
Bakshy, Eytan, Itamar Rosenn, Cameron Marlow, and Lada Adamic (2012), “The Role of Social Networks in Information Diffusion,” Proceedings of the International World Wide Web Conference Committee (IW3C2) 2012, [available athttps://arxiv.
Bakshy, Eytan, Jake M.
Hofman, Winter A.
Mason, and Duncan J.
Watts (2011), “Everyone’s an Influencer: Quantifying Influence on Twitter,” Proceedings of the 4th ACM International Conference on Web Search and Data Mining (WSDM 2011), [available at http://snap.
Sun, Eric, Itamar Rosenn, Cameron A.
Marlow, and Thomas M.
Lento (2009), “Gesundheit!.Modeling Contagion through Facebook News Feed,” Proceedings of the Third International ICWSM Conference (2009), 146–153.
Note: This article will be part of a series on social network analysis and influencer marketing, all based on the research for my master thesis titled “Influencer Fraud on Instagram: A Descriptive Analysis of the World’s Largest Engagement Community”.
Links to related articles will be summarized here in the future.
.. More details