class: center, middle, inverse, title-slide # Populism among German politicians ## A Twitter Mining approach<br> ### Sebastian Sauer, Sandra Sülzenbrück, Yvonne Ferreira, Rüdiger Buchkremer ### FOM<br><br>DGPs 2018 --- class: center, middle, inverse # Introduction --- class: left, top # Research question - How populist are (chatter) present German politicians? - Do tweets provide a useful measure for populism? --- # Popper's theory of the tribal community <br> <br> <br> <br> *tribal community* <i class="fas fa-arrow-right "></i> *person centered community* .footnote[Source: Popper, Karl. 1972. Die Offene Gesellschaft und ihre Feinde. Band 1. Bern: Francke UTB.] --- # Populism as the desire to go back to the tribal community - Losing (having lost) the security of the tribal society hurts - The modern person centered (individualistic) society is marked by.. - insecurity - ambiguity - abundance of possibility - accountability - Feeling of "being lost" and "drifting" emerges - Populist politicians as the advocates of the regression to the tribal community - Populism is fueled by ... - problems and crises of modern live - change, progress, and instability of and in the society --- class: center, middle, inverse # Methods --- class: top, left # Data source: Tweets - Communication styles of politicians (and parties) gathered via tweets - Tweets as data source helps... - processing large data volumes - quite objective analysis criteria - reproducibility - replicability --- # Populism score Unweighted median (z-values) of 8 indicators <table class="table" style="margin-left: auto; margin-right: auto;"> <thead> <tr> <th style="text-align:left;"> Indicator </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;"> word shortness </td> </tr> <tr> <td style="text-align:left;"> ratio of negative/positive words </td> </tr> <tr> <td style="text-align:left;"> proportion of negative words </td> </tr> <tr> <td style="text-align:left;"> proportion of emotional words </td> </tr> <tr> <td style="text-align:left;"> negative emotion score </td> </tr> <tr> <td style="text-align:left;"> overall emotion score </td> </tr> <tr> <td style="text-align:left;"> proportion of CAPITAL LETTERS </td> </tr> <tr> <td style="text-align:left;"> ratio of adjectives to adverbs </td> </tr> </tbody> </table> --- # Hypothesis <br> <br> <br> <br> - Order of populism scores per party: - Trump > AfD > else .footnote[preregistered: <https://osf.io/bwszu>] --- # Data - `\(d \approx 400k\)` tweets from German politicians - `\(t \approx 32k\)` Tweets by Donald Trump - `\(w \approx 6.3 Mio\)` words - `\(k = 200\)` Twitter accounts - `\(p = 8\)` political parties - AfD, CDU, CSU, FDP, Grüne, Linke, SPD, (fraktionslos), Trump/GOP - Data were collected in August 2017 --- # data aggregation  --- # Analysis 1. Compute all 8 populism indicators per politician 2. Recode (if necessary) and compute z-values per indicator and per politician 2. Compute overall populism score (median) per politician 3. Computer overall populism score (median) per party --- # Most (many) Twitter accounts of politicians are included - [List](https://twitter.com/dw_politics/lists/german-politicians/members) "German Politicians" of [Deutsche Welle](http://www.dw.com/de/unternehmen/profil/s-30626) - Twitter accounts of [Bundesvorstand der AfD](https://www.afd.de/partei/bundesvorstand/) - Twitter accounts of [Präsidiuums der FDP](https://www.fdp.de/seite/praesidium) - Markus Söder .footnote[as of August 2017] --- # Tweets timeline <img src="img/p_tweets_timeline2.png" width="70%" style="display: block; margin: auto;" /> --- class: center, middle, inverse # Results --- class: top, left # SPD: largest share of accounts <img src="img/p_accounts_per_party.png" width="60%" style="display: block; margin: auto;" /> --- # SPD: largest share of tweets <img src="img/p_tweets_per_party.png" width="70%" style="display: block; margin: auto;" /> --- # Trump tweets the most So great, great genes <img src="img/p_tweets_day_party.png" width="70%" style="display: block; margin: auto;" /> --- # Grüne: A lot of tweets, CSU: relatively silent <img src="img/p_tweets_day_party_md_no_trump.png" width="50%" style="display: block; margin: auto;" /> --- # AfD and Linke: highest overall populism score; Trump's still great <img src="img/p_party_pop_scores_boxplot.png" width="90%" style="display: block; margin: auto;" /> --- # Union/SPD are *least* populist <img src="img/p_pop_geom_boxplot_wo_trump.png" width="90%" style="display: block; margin: auto;" /> .footnote[Trump excluded from this and subsequent analyses] --- # AfD and Linke: similar populism pattern <img src="img/p_party_pop_scores_details_no_trump_point_limited_grey.png" width="100%" style="display: block; margin: auto;" /> --- # "Top 10 populists" - many from the Linke cum *large* grano salis <img src="img/p_pop_screenName.png" width="70%" style="display: block; margin: auto;" /> --- # Closeup view of "top-1 populist" cum *large* grano salis <img src="img/p_most_pop_screenName.png" width="70%" style="display: block; margin: auto;" /> --- # Sample tweets of "top-1 populist" ----------------------------------------------------------------------------- Nr Tweet ---- ------------------------------------------------------------------------ 1 'Genau dies unkonkrete BlaBla nervt. Was ist Eure Alternative zu Sozialdemokratie? Demokratie ist nicht Kapitalismus https://t.co/KkVbghNMSv\' 2 '#Alternativmedizin Wunderheiler bei Krebs die Not von Patienten ausnutzen müssen bestraft werden. https://t.co/tin8rVy3zG\' 3 'Wir brauchen mehr Abstimmungen als Gewissensentscheidung: Historische Entscheidung Ehe für alle https://t.co/o4X2SLg1lt via @SPIEGELONLINE\' 4 'Krankenhäuser zerfallen, kaum Ganztag, Kita-Mangel, Dauerstau: SPD: Sparfetisch gefährdet Zukunft https://t.co/GW2mshLVeQ via @SPIEGELONLINE\' ----------------------------------------------------------------------------- --- # "Horde words" ``` #> [1] "Stamm" "Bund" "Deutsch" #> [4] "deutsch" "Deutschland" "Volk" #> [7] "Gemeinschaft" "Schöpfung" "Ehre" #> [10] "Respekt" "Held" "Heroe" #> [13] "heroisch" "heldenhaft" "ehrenvoll" #> [16] "völkisch" "Volksgemeinschaft" "Kampf" #> [19] "Sieg" "Vernichtung" "Blut" #> [22] "Boden" "ausländisch" "Ausländer" #> [25] "Krieger" "Krieg" "Tradition" #> [28] "Überlieferung" "heilig" "heilige" #> [31] "Gnade" "ewig" "Ahnung" #> [34] "Erlösung" "erlöst" "erlösen" #> [37] "Verdammung" "Fluch" "verdammt" #> [40] "verflucht" "verdammen" "verfluchen" #> [43] "Verheißung" "Anmaßung" "anmaßend" #> [46] "anmaßen" "Böse" "Verachtung" #> [49] "Verlangen" "Heimat" ``` .footnote[source: https://osf.io/b6ge7/] --- # "Horde score"  --- # AfD/Union use words such as "deutsch*" relatively often  --- class: center, middle, inverse # Discussion --- class: top, left # Strengths - Interesting, thought provoking results - Relatively little used method - quite objective - automated processing --- # Room for improvement - Indicators were chosen with some degree of subjectivity - More and different analyses are warranted - Modeling - Effect sizes - statistical uncertainty - Was the sample representative for German politician chatter? - Association with external criteria is still unknown --- class: middle, center, inverse # Thank you Sebastian Sauer #### <i class="fab fa-github "></i> [sebastiansauer](https://github.com/sebastiansauer) #### <i class="fas fa-link "></i> <https://data-se.netlify.com/> #### <i class="fas fa-envelope "></i> ssauer@posteo.de #### <i class="fab fa-linkedin-in "></i> [Sebastian Sauer](https://www.linkedin.com/in/dr-sebastian-sauer-4791762) #### <i class="fas fa-file "></i> Get slides here: <http://data-se.netlify.com/slides/populist-twitter/populist-twitter-dgps2018.html#1> #### <i class="fas fa-code "></i> Get code and data here: <https://github.com/sebastiansauer/polits_tweet_mining> CC-BY