political twitter mpsa slides

Upload: dsparks

Post on 10-Apr-2018

226 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/8/2019 Political Twitter MPSA Slides

    1/36

    Birds of a Feather Tweet Together

    Partisan Structure in

    Online Social Networks

    David B. Sparks

    Duke University

    images from: twitter.com

  • 8/8/2019 Political Twitter MPSA Slides

    2/36

    Congressional Twitterers

    Congressional Research Service (2009)

    Republicans : Democrats :: 2 : 1

    85 Tweets/day collectively Most status updates offer links to press releases,

    etc.

    Only 1.4% of Congressional Tweets are replies

    Members refer more frequently to their districts whenon recess.

  • 8/8/2019 Political Twitter MPSA Slides

    3/36

    Social Network Analyses of Congress

    Co-sponsorship Fowler 2005, 2006

    Community structure and modularity Zhang, et al. 2008, Waugh et al. 2009

    Committee and caucus membership

    Porter et al. 2005, Victor and Ringe 2009 Interest groups, lobbyist contributions, andmailing lists Grossmann and Dominguez 2009, Koger and Victor

    2009, Koger, Masket, and Noel 2009 Ideological scaling

    Poole and Rosenthal, Bafumi and Herron 2007, etc.

  • 8/8/2019 Political Twitter MPSA Slides

    4/36

    Theory/Motivation

    Time and attention are limited, so spending theseis costly.

    Thus, Following an entity on Twitter is a signalthat the entity is something on which it is worthspending resources.

    Thus, there is information about preferencesencoded in Following behavior.

  • 8/8/2019 Political Twitter MPSA Slides

    5/36

  • 8/8/2019 Political Twitter MPSA Slides

    6/36

    Hypotheses

    Hypothesis 1: The online network of politicalelites is distinctly partisan.

    Hypothesis 2: The allocation of attention by non-elites reflects a partisan dimension.

    Hypothesis 3: Attention allocation preferencesreflect political ideology.

  • 8/8/2019 Political Twitter MPSA Slides

    7/36

    H1: Partisan Network

    Following network among 133 Senators andRepresentatives

    Two major partisan clusters In-group and cross-group ties:

    Freemans Segregation Index: 0.59 Log odds ratio: 3.14 Significant at p < 0.001

  • 8/8/2019 Political Twitter MPSA Slides

    8/36

  • 8/8/2019 Political Twitter MPSA Slides

    9/36

    H1: Partisan Network

    Fuzzy clustering (Kaufman and Rosseeuw 1990)on distance matrix from adjacencies

    Probabilities of each observation falling into each oftwo clusters

    Good predictor of Democratic partisanship, less

    so for Republicans Correctly classifies 74.4% of members, a 9%

    improvement over guessing the modal party.

  • 8/8/2019 Political Twitter MPSA Slides

    10/36

  • 8/8/2019 Political Twitter MPSA Slides

    11/36

  • 8/8/2019 Political Twitter MPSA Slides

    12/36

    Hypotheses

    Hypothesis 1: The online network of politicalelites is distinctly partisan.

    Hypothesis 2: The allocation of attention by non-elites reflects a partisan dimension.

    Hypothesis 3: Attention allocation preferencesreflect political ideology.

  • 8/8/2019 Political Twitter MPSA Slides

    13/36

    Hypotheses

    Hypothesis 1: The online network of politicalelites is distinctly partisan.

    Hypothesis 2: The allocation of attention bynon-elites reflects a partisan dimension.

    Hypothesis 3: Attention allocation preferencesreflect political ideology.

  • 8/8/2019 Political Twitter MPSA Slides

    14/36

    H2: Partisan Followers

    Collected IDs of 117,837 users Following anycongressional account

    Mean legislators Followed: 2.2 Mean Followers per legislator: 3,308

  • 8/8/2019 Political Twitter MPSA Slides

    15/36

    H2: Partisan Followers

    If Following behavior is a function ofpartisanship, we would expect to see a tendency

    to Follow MCs of only one party. The most prolific Followers tend toward the

    center

    Generally media outlets and other completists Some prolific Followers do concentrate

    exclusively on one Party

  • 8/8/2019 Political Twitter MPSA Slides

    16/36

  • 8/8/2019 Political Twitter MPSA Slides

    17/36

    H2: Partisan Followers

    Two types of Followers, distinguished byFollowing count

    Major Followers Follow more than 10 Memberaccounts 90th percentile and above

    Predict number of Republicans Followed, as afunction of Democrats Followed and MajorFollower classification

  • 8/8/2019 Political Twitter MPSA Slides

    18/36

    Hypotheses

    Hypothesis 1: The online network of politicalelites is distinctly partisan.

    Hypothesis 2: The allocation of attention by non-elites reflects a partisan dimension.

    Hypothesis 3: Attention allocation preferencesreflect political ideology.

  • 8/8/2019 Political Twitter MPSA Slides

    19/36

    Hypotheses

    Hypothesis 1: The online network of politicalelites is distinctly partisan.

    Hypothesis 2: The allocation of attention bynon-elites reflects a partisan dimension.

    Hypothesis 3: Attention allocationpreferences reflect political ideology.

  • 8/8/2019 Political Twitter MPSA Slides

    20/36

    H3: Ideological Following

    Person-to-group network of Followers andlegislative Twitterers

    Principal Component Analysis of the adjacencymatrix, to reduce dimensionality

    Generates scores for Followers

    and loadings for legislators If ideology explains Following behavior, it should

    be encoded in the first few components

  • 8/8/2019 Political Twitter MPSA Slides

    21/36

    H3: Ideological Following

    0011000politipage

    0000010PJDente

    0000011GR8GLFR

    0111010ElderJustice

    0000000johnny7268

    1111100rawl4senate

    1001100derjensen26

    0011000politikpundit

    cbrangeljohnthunerogerwickerJohnKerryJeffFlakeMarkUdallclairemc

  • 8/8/2019 Political Twitter MPSA Slides

    22/36

    H3: Ideological Following

    First and especially second components appear tohave ideological content

    Followers with scores at the extremes of the scaleappear to be highly ideological/partisan.

    Implying that Followers are well-sorted by the PCA

    Loadings for Member accounts have a clearrelationship with other measures ofideology/partisanship.

  • 8/8/2019 Political Twitter MPSA Slides

    23/36

  • 8/8/2019 Political Twitter MPSA Slides

    24/36

  • 8/8/2019 Political Twitter MPSA Slides

    25/36

  • 8/8/2019 Political Twitter MPSA Slides

    26/36

  • 8/8/2019 Political Twitter MPSA Slides

    27/36

    H3: Ideological Following

    Linear discriminant analysis

    Predicting known partisanship of Members withloadings on first two components

    Correctly identifies partisanship in 94.9% of cases.

    Linear regression analysis

    Predicting roll call-based Optimal Classificationscalings (Lewis and Poole 2000) with secondcomponent loading, controlling for knownpartisanship

    OC scaling correlates with 2nd Component at 0.762

  • 8/8/2019 Political Twitter MPSA Slides

    28/36

  • 8/8/2019 Political Twitter MPSA Slides

    29/36

    H3: Ideological Following

    Regression results indicate a relationship betweenFollower behavior and ideology

    Republican ideology predicted better than that ofDemocrats.

  • 8/8/2019 Political Twitter MPSA Slides

    30/36

  • 8/8/2019 Political Twitter MPSA Slides

    31/36

    Hypotheses

    Hypothesis 1: The online network of politicalelites is distinctly partisan.

    Hypothesis 2: The allocation of attention by non-elites reflects a partisan dimension.

    Hypothesis 3: Attention allocation preferencesreflect political ideology.

  • 8/8/2019 Political Twitter MPSA Slides

    32/36

  • 8/8/2019 Political Twitter MPSA Slides

    33/36

    Conclusions

    Hypothesis 1: The online network of politicalelites is distinctly partisan.

    Hypothesis 2: The allocation of attention by non-elites reflects a partisan dimension.

    Major Followers do not evince partisan leanings Completists?

    Amongst those Following a relatively small number ofMembers, there is a negative relationship betweenRepublican and Democratic accounts Followed.

    Hypothesis 3: Attention allocation preferencesreflect political ideology.

  • 8/8/2019 Political Twitter MPSA Slides

    34/36

    Conclusions

    Hypothesis 1: The online network of politicalelites is distinctly partisan.

    Hypothesis 2: The allocation of attention by non-elites reflects a partisan dimension.

    Hypothesis 3: Attention allocation preferences

    reflect political ideology. Scaling of Followers appears to correctly identify

    partisans/ideologues

    Scaling of Members Successfully classifies partisanship

    Predicts accepted measure of ideology

  • 8/8/2019 Political Twitter MPSA Slides

    35/36

    Future Directions

    Scaling everything

    Challengers?

    Corporations?

    Diffusion of ideas through re-tweets and @

    replies

    Communication with constituents

  • 8/8/2019 Political Twitter MPSA Slides

    36/36

    Thanks.