twitter and the senatorial race 1 - brownblogs · social media has become a ubiquitous part of 20th...
TRANSCRIPT
1
Twitter and the Senatorial Race
Anisa Holmes
Brown University
December 20, 2014
Abstract
This study examines the causal link between U.S. Senatorial Candidates’ use of Twitter
and subsequent voteshare in the election results. Election results from the 2008 and 2010
Senatorial Race were used to compare candidates who used Twitter as a campaign tool
with candidates who did not have Twitter accounts. Controlling for other variables
including the candidate’s party affiliation, the state’s affiliation, campaign finances and
incumbency, the findings show that twitter had a positive effect on voteshare for
Senatorial Candidates.
Introduction
Social media has become a ubiquitous part of 20th century life. From sharing photos to
forging business connections, most Internet users are involved in some aspect of social media.
Facebook paved the road for the age of Social Media and while it still remains the frontrunner,
other social media websites like LinkedIn and Pinterest have gained considerably in usership.
Today there are thousands of options for Internet users to engage in. This paper attempts to
look specifically at what role political engagement has in this new age of Social Media using
Twitter as an analytic tool.
The Pew Research Center estimates 60% of American adults use social media or
2
networking sites. Of that 60%, 19% use twitter. Overall, there are about 255 million twitter
users world wide. With the sheer numbers it represents, and the open forum format, Twitter is
an amazing tool for politicians with which they can personally address their constituency. This
study is concerned with candidate’s use of Twitter while campaigning, and whether the active
use of Twitter while campaigning actually contributes to a rise in voteshare. The hypothesis,
based on results from other studies and intuition is that twitter usage will have a positive
correlation with voteshare. See figures 1, 2 & 3 for the Pew Research Center’s Internet usage
statistics.
Data concerning the preferences and behaviors of the masses are more readily available
than ever, and resultantly there are many studies in the past 5 years that have aimed to measure
correlation between political engagement and social media. Twitter is particularly useful in that
it not only has data based on numbers of tweets or followers, but tweets can also be mined for
specific types of text and content. With these capabilities, many recent studies have focused on
examining tweet content as a measure of public opinion, or as a way to predict election results.
One such study, Political Polarization on Twitter, has shown that Twitter users are more
partisan and ideologically oppositional than the general public. More along the vein of my
study, Getting connected: The effects of online political communication on
Citizens’ political involvement uses number of followers to examine correlation between the
use of twitter and the number of preferential votes in the Netherland’s presidential elections. It
also uses Twitter communication styles to determine whether it is more beneficial for
candidates to use informal or formal language in their tweets.
This study attempts to develop a more robust strategy to determine the correlation
between twitter use in campaigning and voteshare through fairly straightforward methodology.
3
Using data from two years of Senatorial races in the United States, twitter usage while
campaigning is determined by a candidate’s twitter join date and is compared with candidates
who had not joined prior to the November 4th election date. The study controls for factors
including campaign budget, state and year fixed effects, and state and candidate’s party
affiliations. Twitter was founded in March 21, 2006 and only began to gain significant usage
among political candidates in the US from around 2008 to 2010; So, the use of two Senatorial
races, from 2008 and 2010, attempts to explore the marginal benefit to being among the first to
adopt Twitter. This study is relevant in that it examines whether politicians’ use of social
media as a campaign tool is justified, and to what extent is it a useful tool. Additionally it has
implications as to the benefit of risk-taking or being among the first to adopt burgeoning
technological or social media platforms as a campaign strategy.
Figure 1.
http://www.pewinternet.org/2013/12/30/social-media-update-2013/social-media-sites-2012-2013/
4
Figure 2.
http://www.pewinternet.org/2013/04/25/part-2-political-engagement-on-social-networking-sites/
Figure 3.
http://www.pewinternet.org/2013/12/30/social-media-update-2013/twitter-users/
5
Data
Data for this study was obtained from several sources. The election results including
candidates’ names, party affiliation, state, incumbency and voteshare was obtained from the
Federal Election Commission’s (FEC) public records. Because some states have different
policies for Senatorial elections, a few outliers were eliminated. This includes candidates who
received less than 3% voteshare in the final election, and states with more obscure voting
policies where a candidate might win with 100% voteshare in an uncontested race. The
candidate’s party affiliation uses categorical dummy variables with “Other” parties (Green
Party, Libertarian Party) as the base to avoid perfect multicollinearity.
Using the candidates’ information from the FEC, Twitter join date was obtained from
the Twitter website through a simple Google search of each candidate’s name (and state if
needed to differentiate for common names). Unfortunately the Twitter join date data is based
on the assumption that any candidate running did not delete their account at any point after its
creation. The join date is most frequently in the form of a month and a year, which is sufficient
information to determine if they were actively using twitter to campaign before the November
election date. Additionally, some candidates had more than one twitter account, for example
one account for campaigning, and a new account if they won the election. In this case, the
original campaigning account was used.
Data for the state party affiliation came from the voting records of each state for the
past 4 Presidential elections. Red, Blue and Purple (common terminology for Republican,
Democrat, and Neutral) state affiliation was obtained through an infographic aggregating
state’s voting records from the past 4 presidential elections, which was verified with FEC data.
Red states were determined as those who have voted for strictly Republican presidential
6
candidates in the past 4 elections. Blue states were determined as those who have voted for
only Democrat presidential candidates in the past 4 elections. Purple states were states that
have voted for any combination of Republican and Democratic candidates in the past 4
elections. Only records from the past 4 Presidential Elections were used because states often
change party affiliation gradually. The state’s party affiliation uses categorical dummy
variables with purple states as the base to avoid perfect multicollinearity. See figure 4 for the
State Party affiliation infographic.
Campaign funding was determined by the FEC’s Campaign Finance Disclosures. A
search for each candidate and election year yields a Financial Summary of the candidate’s
campaign expenditure. Because there was an abundance of financial information, Net
Operating Expenditures, was used as the metric to compare each candidate’s campaign
finances. The net operating expenditures are measured by millions of dollars and controls for
inflation from 2008 to 2010 were included in the dataset. Additionally, there are some missing
finance reports; candidates opt not to report campaign finances if it is a small to negligible
sum.
Altogether the data amounted to a total of 163 candidates spanning the 2008 and 2010
Senatorial elections. Only 17.7% of candidates used Twitter to campaign during the 2008
elections while in 2010 the amount jumped to 78.57%. See figure 5 for the data summary.
7
Figure 4.
http://en.wikipedia.org/wiki/Red_states_and_blue_states
Figure 5.
8
Methods
In this study the primary source of comparison is between Senatorial candidates who
used twitter as a campaign tool and those who did not have a twitter account while
campaigning. Using the data outlined above, Twitter Join date was used incorporated as a
dummy variable, 1 if joined prior to the Senatorial elections and 0 if not joined. In addition to
this base comparison, campaign finance, incumbency, party affiliation and candidate party
affiliation are incorporated as controls to narrow down the causal effect of twitter on voteshare.
To analyze the hypothesis of Twitter usage with Senatorial candidates’ voteshare, multiple
regressions were used. In the first round of regression analysis, data from the 2008 and 2010
elections are used separately. The controls for these initial rounds are incumbency, (log
adjusted) budget, the interaction between the candidate’s party affiliation and the state’s
affiliation, and the candidate’s party. In the second round of testing the same controls and
comparisons are used, however year fixed effects are also included to account for differences
in trends between the two election cycles.
The methodology of the study relies on the assumption that Twitter usage is uniform
among candidates. In other words, the use of a dummy variable fails to account for variation
among Twitter users like number of followers, tweet content, or tweet frequency. However, the
variation of candidate’s Twitter behavior shouldn’t be very relevant to the main hypothesis.
Additionally, because Senatorial races are staggered, it’s difficult to gauge comparisons for
twitter usage within each state. But, because this study is concerned with the effect of twitter
on Senatorial voteshare nation-wide, state-level differences are less relevant. Finally, while
campaign expenditure attempts to control for unobservables like campaign strategy (ex. youth-
oriented), it may not account for all factors that may contribute to voteshare like public
9
opinion, general recognition, or prior experience. Also, there may be unobservables linked to a
candidate’s likelihood to adopt Twitter that the regression may capture as part of the causal
effect between Twitter and voteshare.
Results
The regression results from the three different models yielded interesting results that
confirmed the hypothesis of this study. In the 2008 Model, Twitter usage was statistically
significant on Voteshare at 95% level and above for each regression. In other words, the study
confirms an associated rise in voteshare of ~.1 (1=100% voteshare) for candidate’s who used
Twitter as a campaign tool. Incumbency played a significant role in voteshare, as expected
based on findings from Fundamental Models for Forecasting Elections. Additionally, in the
2008 elections, incumbency played a much larger role in voteshare than traditional
state/candidate affiliation. Interestingly, the candidates’ net operating expenditures proved
statistically insignificant and negligible across all models. In the second Model, there was an
associated increase in voteshare of ~.07, but the association was not statistically significant.
The third model of combined results confirmed an association of around .1.
However, when natural log-controlled budget was introduced as a control variable
(which was introduced later), it dwarfed the effect of the other variables, rendering them at
times statistically insignificant. This may suggest that if campaign finance weren’t such a huge
factor in election results, Twitter usage would play a more significant role than it does in
reality. However, campaign finance is a partially a product of public support and donations, so
this could be encompassing unobserved effects including popularity, notoriety or higher
institutional support which could be leading to skewed results.
10
See figure 6, 7 & 8 for the regression results.
Figure 6, 7, & 8
11
12
Discussion
The results of the regressions in the three models showed a strong and significant
association between Twitter usage and increase in a candidate’s voteshare. Unfortunately, it is
13
a limited data set, and the staggered senatorial elections provide little room to control for
differences in states. Additionally, overall social media presence, or news presence is not
accounted for in the Model, the effect of which could problematically be captured within the
Twitter-Voteshare correlation leading to omitted variable bias. However, despite these
concerns, the connection between Twitter and voteshare is mirrored by the results of
Kruikemer’s Getting connected study which showed a correlation of around ~.348 to ~.1 in
two different regression results for political candidates in the Netherlands. This suggests that
perhaps the use of social media is in fact a very powerful political tool.
With additional time and resources, it would be prudent to introduce more explanatory
variables into the Models in order to get a clearer estimate of the causal impact of Twitter
usage on candidate voteshare. These variables might include voter demographics, economic
climate, or the candidate’s aggregated social media presence. Additionally, doing a comparison
across countries might also be interesting; Is twitter more beneficial to political candidates in
democratic or participatory governments as opposed to more censored governments or vice
versa? Using elections from multiple countries could provide a solution to the limited data set
and yield very interesting results. Examining social media’s effect on voteshare beyond the
Twitter platform is another avenue for further explorations.
Conclusion
In a broad sense, the goal of this study was to take a closer look at the role of social
media in political engagement. Using Twitter and the US Senatorial elections as a proxy to
answer greater questions about this topic, substantial findings confirmed the hypothesis that
social media as a campaign tool has a positive effect on voteshare. The study confirmed a
positive effect of twitter usage on Candidate voteshare in Senatorial elections of ~.1, which
14
could be enough to tip the scales in a highly contested race. However, because most US
candidates actively use Twitter today, this study provides insight as to the consequences of
ignoring social media as a campaign strategy, and the benefits of being among the first to adopt
new social media platforms.
15
References
Kruikemer, S. 2014. Getting connected: The effects of online political communication on
citizens’ political involvement. Digital Academic Repository of the University of
Amsterdam.141105: 48-67.
Conover, M. D., et al. 2011. Political Polarization on Twitter, 89-96.
Rothschild, David. 2014. Fundamental Models for Forecasting Elections, <http://www.researchdmr.com/RothschildForecast12> 1-24.