2.
Social Network Analysis of 2016 US Presidential Election Candidates 1
Introduction and Background
Social Media’s presence is ubiquitous today. Its extent can be realised with Facebook’s 1.55
billion monthly active users. As Social Media grew from its embryonic state in 2004, it gradually
transformed into a communication platform on which governance, news, protests and friendships
originated and sustained. Around the world, social media consistently plays a major role from
managing natural disasters to winning election campaigns. As many political candidates started
using social media to mobilize and influence voters, the social media companies also started to
realize the opportunities of using data created in this communication platform to predict and
understand the outcome of an election campaign. With less than a year remaining for the US
Presidential Election 2016, Election campaigns are in full swing and yet again social media is
poised to become a game changer. Social Media’s major role came into effect during Obama’s
2008 presidential victory, It was not a surprise that one of Obama’s key strategist was Facebook
cofounder Chris Hughes (Soumitra Dutta & Matthew Fraser, 2008). Even in current 2016 US
presidential election social media’s role has been noteworthy. Twitter has rolled out special
features like $Cashtags to aid candidates in raising funds for their election campaigns (Reuters,
2015). Bernie Sanders who started the year 2015 with 0 followers in Twitter now has almost 1.4
million followers, more than triple the followers of Jeb Bush and Chris Christie combined. Social
media’s excessive use as a part these campaign strategy is due to the fact that it is able to reach a
variety of audience and get the attention of the millennials.
Objective
The impact of social media in elections has been studied extensively. Our objective here is to
study the social network of the two prominent election candidates Hillary Clinton and Donald J.
Trump. The main aim of the paper is to understand the reach of Hillary Clinton and Donald J.
Trump among Twitter users and how their social media strategy is helping or affecting their
election campaign.
Literature Review
Even though social media’s history started as early as 1997 with the emergence of Six Degrees
(Drew Hendricks, 2013), research on social media’s impact on election results were not studied
or researched until it grew exponentially from 2004 with the launch of Facebook and Twitter.
Research on social media’s impact on election results started with social networking sites that
were predominantly blogs. Initial research, after studying the impact of blogging and
hyperlinking in 2004 U.S Presidential Campaign, stressed the need for userfriendly and
interactive websites with more options for user participation, (Andrew Paul Williams et al,
2006).
In 2008 Presidential election, Obama’s strategy of using his socialnetworking website
mybarackobama.com, known as MyBO enabled more user interaction, grassroot mobilization
and ultimately played a major role in his victory (Williams and Gulati 2008). Usage of social
media in election campaigns evolved from websites like blogs and hyperlinking in 2004 to social
media heavyweights like Twitter and facebook in 2008, as the results of Obama’s victorious
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
3.
Social Network Analysis of 2016 US Presidential Election Candidates 2
online strategy (Tumasjan, Andranik, et al, 2010). This led to multiple types of research on
predicting election results based on Twitter data. Simple Sentiment Analysis on Twitter data
correlated with public opinions collected from various survey results from 2008 to 2009 strongly
indicated that social media analysis can supplement time intensive polling during Presidential
Elections (O'Connor, Brendan, et al, 2010). On the other hand, social media’s predictive analysis
on 2010 presidential election did not correlate with the results of the election (Gayo Avello, D.,
Metaxas, P. T., & Mustafaraj, E., 2011). Conflicting claims from the two researches were based
on data collected before 2010 and also took into consideration only the tweets to predict the
results.
Studying social networks and its impact in the political arena opened up new ways of analysis
and claims and also overcame the shortcomings of using just text for predicting results.
Experimental studies to measure the effects of social influence online led to many debates on
whether political influence were facilitated by “weak ties relationship” or “strong ties
relationship”. During 2010 US Congressional elections a randomized and controlled trial of
political mobilization messages were delivered to 61 million Facebook users and the results were
studied. The reports showed that the messages not only influenced the persons who received the
message directly (Strong ties), but also their friends and friends of their friends (weak ties).
Furthermore, the influence of these messages transmitted among close friends were four times
the effect of the messages itself, indicating the importance of strong ties in influencing behaviour
change (Bond, Robert M., et al, 2012). Combining these results with decades of research on
understanding the impact of Interpersonal, Media, and Organizational Influences on Presidential
Election indicates that political choices are affected to a significant degree by the flow of
information from the decisionmaker’s immediate social network (Beck, Paul Allen, et al, 2002).
Further, diving deep into social network to analyze the political discussion practices has
produced interesting results. Studies have shown that large egocentric networks were more
politically engaged (Lake and Huckfeldt 1998) and also discussed politics more often (Moy and
Gastil 2006). Study focused on individual’s role in a social network and its authority over
political influence indicates that people with high betweenness centrality were less inclined to be
neutral and were more inclined to attempt political persuasion.
Research questions and Setup
As discussed above, social media has started acting as a game changer in Presidential Elections.
This is evident from the amount of expenditures being done on social media which
approximately accounts for more than half of the $1 billion budget for digital media (R. Kay
Green, 2015). As a part of our project, we will be doing research and analysis of the below
questions.
1. Do people use social media as a source of information?
2. How does the information flow through the candidate’s social network?
3. Do people with high betweenness centrality acts as political influencers?
4. Can a candidate's primary opponent from the opposition party as well as the same party
be identified from their social network?
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
4.
Social Network Analysis of 2016 US Presidential Election Candidates 3
The data for our analysis has been collected from Twitter for two presidential election
candidates: Donald Trump (Republican) and Hillary Clinton (Democrats) using multiple popular
and current hashtags. Like: #Trump, #HillaryClinton, #Trump2016, #Hillary2016. Twecoll and
NodeXL were used to collect the data from Twitter API and Gephi was used to visualize the
network.
Results and Discussions:
In order to answer our first research question, we conducted an online survey to gather
information about user’s views on social media as an information source. An online survey was
created and shared with people in our social network to record their responses. The respondents
of the survey were predominantly from the age group of 2230 spanning across two countries
(India and United States of America). In Total, 62 people responded to the survey, the survey
provided an overview of what young population thought about the social media, how dependent
they are on social media and how social media influences their decisions.
1. How would people like to be informed about global news?
Fig. 1
67.74% of the people relied on social media to get daily updates about the world as opposed to
32.26% of the people who still followed the traditional news channel viewership to gain
information.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
5.
Social Network Analysis of 2016 US Presidential Election Candidates 4
2. Which social media website do people visit frequently?
Fig. 2.
After the advent of social media, many people were hooked onto it and started using it almost
every day. We can clearly infer from the bar chart that many people use Facebook more
frequently as compared to other social networking sites like Twitter, Tumblr etc. But we cannot
overlook the fact that of all the people who took the survey almost 80% are from India and since
Twitter is not that widely used in India, these stats can be deceiving.
3. How many Twitter users people generally follow?
Fig. 3.
This result was expected, almost 80% of the people who took our survey follow less than 100
users. 18.03% people follow 100500 people and 4.92% people follow >500 Twitter users.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
6.
Social Network Analysis of 2016 US Presidential Election Candidates 5
4. Visitors of social media site consumer or producer of information?
Fig. 4.
In this question we try to categorize our sample into two groups ie Consumer of Information
the person who visits various social networking sites to grasp information about the world and
Producer of Information the person who voices his opinion by sharing, commenting, tweeting
etc. From the above bar graph, it is evident that people basically use social media to gain
information. Thus social media plays a pivotal role in keeping people informed.
5. Does social media influence people’s decision?
Fig. 5.
From the above graph, we infer that only 29.03% people are influenced by social media.
However since we conducted this survey on a sample size of 62 we can expect this % to increase
further, Many people do take into consideration other user’s opinions expressed on social media
eg. movie reviews. Thus, we can conclude from our survey that social media is used extensively
by many users as a source of information.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
8.
Social Network Analysis of 2016 US Presidential Election Candidates 7
Observations:
From the network visualization, three prominent communities were identified and color coded
Blue, Red, and Green. Primarily, Blue community constitutes of Hillary Clinton and her
supporters, Red community consists of Ted Cruz and his supporters and Green community
consists of Donald Trump and his supporters. ‘ibegoodnow’ of green community and
‘thegreatfeather’ of the pink community have the highest betweenness centrality. Three
communities identified form three clusters. Hillary Clinton is connected to the other clusters via
only one user, ‘MiltonMiller14’ while Ted Cruz clusters and Donald Trump clusters are
connected via multiple users indicating they belong to the same party.
Analysis:
The presence of the blue community which comprises of Hillary Clinton and her followers is
very interesting, as she belongs to the Democrats as opposed to Trump who is a part of the
Republicans. Thus, we can clearly observe the two opposing parties and their emerging
contenders in this network. Another interesting point to notice is that people who are tweeting
about Donald Trump are also tweeting about Hillary Clinton. This can be interpreted as an
indication that Trump supporters consider Hillary as his biggest competition and thus, she is
mentioned in most of the tweets directed towards Trump. The important node in the red cluster is
that of Ted Cruz, who is the member of the Republican party. His presence in this network might
be due to the fact that he is competing against Trump to emerge as the presidential candidate for
Republicans. The important node in the green cluster is that of Trump himself and his followers
who tweet about him and his ongoing campaigns.
The highest betweenness centrality is of Trump supporters who are responsible for the
propagation of information from one part of the network to the other. For example, the user
‘thegreatfeather’ has the highest weight i.e. he has been tweeting extensively in the support of
Donald Trump and also has a high betweenness centrality. Some of his tweets are “That’s Right
we’re for Mr Trump”, “Mr Trump, make America great again!” etc.
@hillaryclinton analysis
Layout : Yifan Hu
Node Size : Betweenness Centrality
Color : Modularity Class
Density : 0.011
Degree Range : 3 537
Node : 204
Edges : 474
Average Degree : 4.64
On January 18th, 2016, NodeXL was used to extract tweets with keyword, ‘hillaryclinton’. Limit
of 10,000 tweets was set and a network was created with people who mentioned
‘@hillaryclinton’ in their tweets. The network was then loaded and visualized in Gephi. To
remove users who are loosely bound in the network, nodes with degree range < 3 were filtered to
result in the above network. The directed edges between the nodes represent that the users
mentioned each other or replied to each other in their tweets. The node size was adjusted as per
the betweenness centrality and the color was set as per the modularity class.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
9.
Social Network Analysis of 2016 US Presidential Election Candidates 8
Fig. 7. Hillary Clinton’s Network showing the Information propagation
Observations:
From the network visualization, three prominent communities were identified and color coded as
Purple, Blue, and Green. Primarily, purple community constitutes of Hillary Clinton and her
supporters, blue community constitutes of Bernie Sanders and his supporters and the green
community consists of Trump and his followers. ‘Redrising11’ and ‘Dyniace’ have the highest
betweenness centrality in this network and are tweeting extensively in the support of Hillary
Clinton. Another Interesting find is that both ‘CNN’ and ‘CNN_politics’ are part of Donald
Trump’s network.
Analysis:
As observed earlier in Trump’s network where Hillary Clinton was part of his network; similarly
Donald Trump has a fair share in the network of Hillary Clinton which is a strong indication that
both the party’s supporters consider them as the front runners in the upcoming US Presidential
Elections. The edge weight indicates the number of times a person has tweeted about Hillary,
surprisingly in this network it was observed that Barack Obama has a direct edge with Hillary
Clinton and has been tweeting about her lately.
Also, the presence of Bernie Sanders in this network indicates that he is another candidate trying
to emerge as the single most powerful leader for Democrats and has his own cluster and set of
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
10.
Social Network Analysis of 2016 US Presidential Election Candidates 9
supporters. ‘Redrising11’ comes across as the biggest supporter of Hillary Clinton with various
tweets in support of her election campaign. Another interesting observation from the network
was the presence of ‘CNN’ and ‘CNN_politics’ node in Bernie Sanders community. Many
people in Bernie Sanders community not only replied or mentioned Bernie Sanders in their
tweets along with Hillary Clinton but also mentioned or replied to CNN. This also shows that
when CNN carried out information on Hillary Clinton, many Bernie Sanders followers replied to
CNN’s tweet and information propagated to Bernie Sanders supporters network as a result.
Visualizing social network of politically active users
#Trump2016 Visualizing Donald Trump’s Social Network
Layout : Yifan Hu
Node Size : Betweenness Centrality
Color : Modularity Class
Density : 0.073
Degree Range : 10 557
Node : 402
Edges : 11742
Average Degree : 58.418
Fig. 8. Social Network of active Donald Trump followers
On January 19th, 2016, Twecoll was used to extract list of tweets with ‘#Trump2016’ from
Twitter. 10,000 tweets were collected after this process. From the dataset a subset of 998 random
unique users was parsed and saved as a new file. To the above list two users were added,
‘@realdonaldtrump’ and ‘@hillaryclinton’. The newly created file was then passed to Twecoll
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
11.
Social Network Analysis of 2016 US Presidential Election Candidates 10
to generate a network from the 1000 nodes. The network was then loaded and visualized in
Gephi. To remove users who are loosely bound in the network, nodes with degree range < 10
were filtered to result in the above network. The directed edges between the nodes represents that
the source user follows the target user. The node size was adjusted as per the betweenness
centrality and the color was set as per the modularity class.
Observations:
From the network visualization, three prominent communities were identified and color coded
Purple, Blue, and Green. ‘katrinapierson’ of purple community, ‘sandiv11’ of blue community
and ‘theratzpack’ of the green community have the highest betweenness centrality. The network
has a density of 0.073.
Analysis:
Donald Trump’s Twitter feed acts as a critical way of communication with his 5.81M followers.
This intense network collaboration is evident in Fig. 8. Such kind of network enables rapid
communication across all the followers of Trump. Follower’s profile consists largely of retweets.
‘Retweeting’ assists in social interaction, expanding the network and contributes immensely in
solidifying Trump’s brand. Such a strongly connected social network is a rare find in social
network analysis, in order to further confirm that this network was not formed by biased nodes
with the maximum edge the whole process was repeated twice with random 1000 nodes and the
results were consistent. The degree distribution chart also followed Power law distribution
indicating the results were not an anomaly.
Fig. 9. Degree Distribution of Donald Trump Social Network (Random Sampling)
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
12.
Social Network Analysis of 2016 US Presidential Election Candidates 11
Fig. 10. Social Network of Active Donald Trump Followers (Filtered)
For the above visualization, nodes with degree < 100 have been filtered out to sight a clearer
picture. It can be clearly observed that the blue and the green community are well connected. The
purple community consists of a lot of politically important people, like realdonaldtrump official
profile for Donald Trump, katrinapierson National Spokeswoman for Donald Trump's
presidential campaign, michaelcohen212 an executive at the Trump Organization,
waynedupreeshow Writer for The Political Insider. Owing to their importance these individuals
have a high number of tweets and a large number of followers. The blue and green community
primarily consists of Trump’s followers who follow these individuals. These nodes are well
connected as followers have a tendency to connect with people who are supporting the same
cause, in this case supporting Donald Trump. These nodes are mostly spread across Nevada,
Pennsylvania, New York and few other states where Trump enjoys maximum support.
Below figures shows the analysis of the three communities. Random nodes from green, blue and
purple community were selected from the network and their bio, followers count and number of
tweets and recent tweets from Twitter were analysed to understand the three communities in the
network.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
15.
Social Network Analysis of 2016 US Presidential Election Candidates 14
more in number. Amongst the blue and green community, there are few users who have a high
number of followers from both the communities. These are those individuals who are significant
in transferring of information within a community, like ‘philmonaco67’ of the blue community.
Egocentric network analysis of Hillary Clinton.
Fig. 15. Egocentric network of Hillary Clinton in Donald Trump’s Network
As per the above egocentric network of Hillary Clinton, we can clearly see that Hillary Clinton is
followed by lots of unconnected individuals (black dots). Her connections with the green and
blue community, who prominently belong to Trump’s followers are less as expected. The few
connections with the purple community are with those users who are not ardent supporters of
Trump and do not tweet about him with as much frequency, like oraztex, ranamayle, paulrdube1.
Comparing this network with the sentiment analysis which is discussed in this paper it is clear
the nodes connected to Hillary Clinton are the people who use “#Trump2016” in their tweets to
convey a negative message about Donald Trump.
#Hillary2016 Visualizing Hillary Clinton’s Social Network
Layout : Yifan Hu
Node Size : Betweenness Centrality
Color : Modularity Class
Density : 0.048
Degree Range : 10 492
Node : 452
Edges : 9767
Average Degree : 43.217
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
16.
Social Network Analysis of 2016 US Presidential Election Candidates 15
Fig. 16. Social Network of active Hillary Clinton followers
On January 19th, 2016, Twecoll was used to extract list of tweets with ‘#Hillary2016’ from
Twitter. 10,000 tweets were collected after this process. From the dataset a subset of 998 random
unique users was parsed and saved as a new file. To the above list two users were added,
‘@realdonaldtrump’ and ‘@hillaryclinton’. The newly created file was then passed to Twecoll to
generate a network from the 1000 nodes. The network was then loaded and visualized in Gephi.
To remove users who are loosely bound in the network, nodes with degree range < 10 were
filtered to result in the above network. The directed edges between the nodes represents that the
source user follows the target user. The node size was adjusted as per the betweenness centrality
and the color was set as per the modularity class.
Observations:
From the network visualization, three prominent communities were identified and color coded
Purple, Blue, and Green. ‘hillary2016press’ of purple community, ‘lostdiva’ of blue community
and ‘night_rider2014’ of the green community have the highest betweenness centrality. In purple
community ‘hillaryclinton’, ‘stylistkavin’ and ‘hillary2016press’ had a high betweenness
centrality as compared to other nodes and formed a clique. There was a clear divide between the
green community and the rest of the communities.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
17.
Social Network Analysis of 2016 US Presidential Election Candidates 16
Analysis:
From the above visualization it can be clearly seen that all the communities are internally well
connected. The green community is primarily dominated by Republicans, like ‘libertarianWE’,
‘KevniChang’ etc. The purple community primarily consists of Hillary Clinton’s supporters, like
‘Hillary_HQ’, ‘777sjr’, ‘renayws’ etc. Blue community consists of Bernie supporters, like
‘bernieartists’, ‘adamnguy’ etc. The purple and blue community are much more closely knit as
compared to the green community which is apparent as they support the Democrat party.
‘stylistkavin’, ‘hillary2016pres’ and ‘hillaryclinton’ are the most prominent users in the purple
community and forms a clique within themselves indicating they are strongly connected.
Fig. 17. Egocentric network analysis of Donald J. Trump in Hillary Clinton’s network
Egocentric network analysis of Donald J. Trump reveals interesting observation, There is clear
divide between the green community and the rest of the network in Hillary Clinton’s network,
Interestingly these people are bridged by Donald Trump to the rest of the network indicating
these people who are tweeting with hashtag #Hillary2016 are not following Hillary directly and
are using the hashtag to tweet negative information about Hillary, Further analysis indicates
these people tweeting about her decision in benghazi attacks by including #HillaryforPrison in
their tweets.
Sentiment Analysis:
In our attempt to understand the public reaction garnered by Donald Trump and Hillary Clinton,
we performed sentimental analysis in R using a few popular hashtags like #Trump2016 and
#Hillary2016.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
18.
Social Network Analysis of 2016 US Presidential Election Candidates 17
#Trump2016 #Hillary2016
Fig. 18. Fig. 19.
At first we collected 1500 tweets using Twitter API’s and various packages in R. The tweets
collected were then processed i.e. made ready for sentiment analysis by removing spaces,
hyperlinks, punctuations and other unnecessary content. The final result is displayed in Fig. 18.
and Fig. 19.
From the graphs we can infer that there is not a significant difference between the public
reactions received by Donald Trump and Hillary Clinton. This is a clear indication as to how
fierce this battle is and there is not much which separates the two. On examining closely we
found out that the negative comments received by Hillary are briefly more in number as
compared to Trump. This can impact the results of the upcoming elections if the two candidate
emerge as the winners from their respective parties.
Combining these results with the social network analysis provides fruitful insights, Hillary
Clinton’s network had a huge chunk of people following Donald Trump and Bernie Sanders.
Sentiment analysis of tweets containing #Hillary2016 also showed 1315% tweets with negative
rating indicating in Twitter even though Hillary is the favourite among the Democrats, she has
many people who are against her. Alternatively Trump has more tweets which has neutral
polarity and very tightly connected social network indicating Trump has good social media
strategy and it’s working well with less than 10% people expressing view against Donald Trump.
Conclusion:
From our survey results, we found out that social media plays a huge role in dissipating
information across the globe. Many people are hooked onto social networking sites like Twitter,
Facebook, Tumblr, etc. and share their views on varied subjects around the world. Thus, it acts
as a source of information for many people and is therefore used extensively in the upcoming
U.S. Presidential elections.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
19.
Social Network Analysis of 2016 US Presidential Election Candidates 18
Social network analysis of Donald Trump and Hillary Clinton gave a strong indication that they
are the frontrunners from Republicans and Democrats respectively. We also found out certain
Twitter users in their network who are responsible for spreading information about them and
tweet extensively to show their support. For example, thegreatfeather is an ardent follower of
Donald Trump with tweets like “That’s Right we’re for Mr Trump”, “Mr Trump, make America
great again!” etc. Visualizing information propagation for both the candidates indicates the
speed in which information spreads about Trump is more compared to Hillary Clinton. Twitter
statistics also indicates the same with ~10,000 tweets per 15 minute for Donald Trump while
Hillary had only ~3000 tweets for the same period.
Trump’s presence on social media has been very boisterous since he joined the presidential race.
His crude and loud comments on Twitter attract a lot of attention and evoke reaction amongst the
public. On the other hand, Clinton’s record so far has been very professional and
politicallycorrect which doesn’t provide much entertainment. This inevitably results in him
being the ‘social media darling’ with a powerful presence.
Comparing Hillary Clinton’s and Donald Trump’s social network indicates Donald Trump as a
clear winner in his social media strategy with his well formed and strongly connected social
network which propagates positive sentiments across his network. The strongly connected social
network provides a perfect medium to reiterate his view across his social network multiple times
and establish interaction among his loyal followers. Downside of Donald Trump’s social
network is its lack of weak ties that spread information outside the social network. While Hillary
Clinton’s social network contained diversified communities Donald Trump had only strong
followers in his social network. This also provides a medium for Hillary Clinton to reach out to
different communities.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
21.
Social Network Analysis of 2016 US Presidential Election Candidates 20
Lake, Ronald La Due, and Robert Huckfeldt. 1998. “Social Capital, Social Networks, and
Political Participation.” Political Psychology 19 (3): 567–84.
Moy, Patricia, and John Gastil. 2006. “Predicting Deliberative Conversation: The Impact of
Discussion Networks, Media Use, and Political Cognitions.” Political Communication 23 (4):
443–60.
Miller, P. R., Bobkowski, P. S., Maliniak, D., & Rapoport, R. B. (2015). Talking Politics on
Facebook Network Centrality and Political Discussion Practices in Social Media. Political
Research Quarterly, 1065912915580135.
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
22.
Social Network Analysis of 2016 US Presidential Election Candidates 21
Acknowledgment
For analysis, open source python code ‘Twecoll’ was used to collect tweets and form a network
from the nodes parsed from the tweets: https://github.com/jdevoo/twecoll
All the datasets mentioned in this paper and Gephi files can be downloaded from Github
repository at this link: https://github.com/Rameshbumd/SocialNetworkAnalysis
Apoorva Ajmani, Karan Kashyap, Ramesh Balasekaran
