Given at MPSA 2012
Public officials’ communication has been explored at length in terms of how such their statements are conveyed in the traditional media, but minimal research has been done to examine their communication via social media. This paper explores the kinds of statements U.S. officials are making on Twitter in terms of the actions they are trying to achieve. We then analyze the correlation between these statements, Congressional communication network structures, and voting behavior. Our analysis leverages over 29,000 tweets by members of Congress in conjunction with existing DW-NOMINATE voting behavior data. We find that pro-social and self-promoting statements correlate with Congressional voting records but that position within the Congressional communication network does not correlate with voting behavior.
2. PROJECT TEAM
• Matt Shapiro
• Libby Hemphill
• Jahna Otterbacher
• Drexler James
• W. David Work
Illinois Institute of Technology
info@casmlab.org
http://www.casmlab.org/projects/publicofficials/
April 12, 2012
Shapiro, Hemphill, and Otterbacher
3. OVERVIEW
• Twitter overview
• Communication networks on Twitter
• Coding for action
• Using Twitter for prediction
April 12, 2012
Shapiro, Hemphill, and Otterbacher
9. LEGEND FOR GRAPHS
Edge Properties
Color Gray = same party
Yellow = different parties
Node Properties
Color Red = Republican
Blue = Democrat
Yellow = Independent
Shape Solid square = House
Solid circle = Senate
Size In degree
Opacity Out degree
April 12, 2012
Shapiro, Hemphill, and Otterbacher
10. CONGRESS MENTIONING EACH OTHER:
EXCLUDING SELF-LOOPS
April 12, 2012
Shapiro, Hemphill, and Otterbacher
11. CONGRESS MENTIONING EACH OTHER:
INCLUDING SELF-LOOPS
April 12, 2012
Shapiro, Hemphill, and Otterbacher
14. NETWORK
PROPERTIES
• Low transitivity
• Low density
• High distance
• No evidence of higher-order structure
April 12, 2012
Shapiro, Hemphill, and Otterbacher
15. INTERPRETING
RESULTS
• Low density indicates low cohesion (Livne et al.
2011)
• Congress much like the public
• Conservatives mention each other more (Adamic
& Glance 2005)
• Explicitly engage small subset of those under
surveillance (Bakshy et al. 2011)
• New medium, not new behavior
• Avoiding issue dialogue (Huckfeldt et al. 1995)
• No real role of third parties (Xenos & Foot 2005)
April 12, 2012
Shapiro, Hemphill, and Otterbacher
16. CODING FOR ACTION
Code Definition N Cohen’s
kappa
Narrating Telling a story about their day, 173 0.83
describing activities
Positioning Situating one's self in relation to 405 0.87
another politician or political issue
Directing to Pointing to a resource URL, telling 465 0.70
information you where you can get more info
Requesting Explicitly telling followers to go do 15 0.70
action something online or in person
Thanking Says nice things about or thanks 57 0.90
someone else
April 12, 2012
Shapiro, Hemphill, and Otterbacher
17. MAKING
PREDICTIONS
Item Measure
Size of audience Followers
Surveillance Friends
Frequency Tweets
Polarizing DW-NOMINATE
April 12, 2012
Shapiro, Hemphill, and Otterbacher
18. MAKING
PREDICTIONS
Item Measure
Size of audience Followers
Surveillance Friends
Frequency Tweets
Polarization DW-NOMINATE
April 12, 2012
Shapiro, Hemphill, and Otterbacher
19. PREDICTING
AUDIENCE
Measure Coefficient Measure Coefficient
Narrative -0.15 Male -0.75*
(0.10) (0.11)
Positioning 0.19* Republican 1.20*
(0.09) (0.09)
Providing info 0.23* Senate 1.02*
(0.08) (0.09)
Requesting action 0.23
(0.28)
Thanking 0.08
(0.16)
April 12, 2012
Shapiro, Hemphill, and Otterbacher
20. PREDICTING
POLARIZING VOTES
Measure Coefficient
Narrative -0.05
(0.05)
Positioning 0.09*
(0.04)
Providing info 0.07*
(0.04)
Requesting action -0.13
(0.17)
Thanking -0.24*
(0.09)
April 12, 2012
Shapiro, Hemphill, and Otterbacher
21. USING TWITTER
BEHAVIOR FOR
PREDICTIONS
• Positioning and providing info predict size of
audience
• Positioning predicts extreme voting
• Thanking predicts centrist voting
April 12, 2012
Shapiro, Hemphill, and Otterbacher
23. FUTURE WORK
• Who is not connecting and why?
• What’s the nature of the cross-party
mentioning?
• Are there reciprocal patterns?
• What relationships exist between
conversation networks and offline networks?
• What impact does gender have on social
media communication behavior?
April 12, 2012
Shapiro, Hemphill, and Otterbacher
24. CONTACT US
• Matt Shapiro (mshapir2@iit.edu)
• Libby Hemphill (libby.hemphill@iit.edu)
• Jahna Otterbacher (jotterba@iit.edu)
Illinois Institute of Technology
info@casmlab.org
http://www.casmlab.org/projects/publicofficials/
April 12, 2012
Shapiro, Hemphill, and Otterbacher
27. HYPOTHESES
H1. Twitter is a virtual echo chamber in which officials interact
mainly with themselves and create homophilous networks.
H2. A member of Congress’s location in the network is
significantly predicted by both Twitter-based and non-Twitter-
based characteristics.
H3. The degree to which members of Congress are followed and
befriended is a positive function of positioning and pro-social
statements via Twitter and polarizing voting records.
H4. Polarizing voting records are particularly reflected by
positioning and pro-social statements via Twitter.
April 12, 2012
Shapiro, Hemphill, and Otterbacher
28. RESULTS
Hypothesis Supported?
Positioning and pro-social Yes
tweets predict
followers/friends
Positioning and pro-social Yes
tweets predict voting
records
April 12, 2012
Shapiro, Hemphill, and Otterbacher