Software

Issue Salience and Ownership in 2016 US Presidential Election Primaries on Twitter

Description
Issue Salience and Ownership in 2016 US Presidential Election Primaries on Twitter
Categories
Published
of 35
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  Issue Salience and Ownership in 2016 US PresidentialElection Primaries on Twitter  ∗ Veikko Isotalo  University of Helsinki  September 2017This paper inspects political issues and themes associated with the 2016 US Democratic and Re-publicanprimarycandidatesviaTwitterdata. Theanalyzeddatasetcomprisestweetsmentioningindividual candidates between February 12th, 2016 and May 26th, 2016. Tweets were processed by using two distinct methods for large scale text data: pre-determined keywords and topic mod-els. With eight pre-determined issue categories, it was possible to show in which central politicalthemes candidates were most connected to, by themselves or by others, on Twitter. Results showthat on party average the Democratic candidates were more often connected to environmental,economic issues and social programs. Conversely, Republican candidates were more associatedwith immigration. Intra-party differences proved to be pronounced for Democratic candidates.On average candidates from both parties presented the most interest in economic issues, whereasTwitter users mentioned them with social and cultural issues the most. After quantitative analysisof the issue categories, topic models were fitted to the categorized data. Topic models confirmedresults of the lexicon-based approach and provided detail by distinguishing key topics for candi-dates. It remains to be seen whether the signals extracted from the Twitter data can be confirmed by other sources and offline opinions of the voters. Keywords : US election, issue ownership, big data, Twitter, lexicon model, topic model Introduction Politicians’usageofsocialmedia, especiallyTwitter, hasbecomeanessentialwaytocommunicatewith the public along with political campaigning (see e.g., Enli 2017; Morris 2017). The relative importance of Twitter is also well understood by today’s candidates, for instance Hillary ClintonchosetobreakthenewsofrunningforpresidencyonTwitter(Enli2017). TheadvantageofTwitterover other media is that it provides a direct communication channel for candidates to send theirmessages to the feeds of millions of followers (Enli 2017). Thus, Twitter and social media havedecreased the reliance of candidates to reach out through the traditional media (e.g., televisionand newspapers) to get their message out (Enli 2017). Morris (2017) has found that campaign messages of 2016 US presidential election sent via Twitter resonated with the same magnitudewith potential voters as messages sent via traditional media. ∗ Thanks to Pablo Barbera, PhD (University of Southern California), for providing guidance 1  Pew Research Center (2016) reports that 24% of all US online adults (21% of all US adults) were Twitter users. Online adults - less than 30 years old (36%) are three times more representedon the platform than online adults aged 65 and older. Moreover, Twitter is more popular amongthe highly educated than adults with lower education. (Pew Research Center 2016)Twitter, as a social media platform, provides opportunities for candidates to engage in dia-logue with voters, but interaction between candidates/campaigns with voters in the US contexthas been non-existent (Enli 2017). One explanation for the candidates’ lack of engagement indeliberation is the hostility of Twittersphere, where any communicational mistake might lead toharassment and trolling (Theocharis et al. 2016).This paper uses Twitter data, collected at the time of 2016 US presidential primaries, to explorewhich major political issues were linked to the Republican and Democratic candidates. Therefore,the paper focuses on supply side theory of issue ownership and issue salience. Issue ownershipmeans that the issue-owning party is considered to be more capable in handling the issue in ques-tion than other parties (Petrocik 1996). Issue ownership perceptions of voters are relatively fixed,what follows is that candidates are left to influence which issues should become salient for theelection at hand (Petrocik 1996).Methods employed in this article are conventional text mining approaches for large data sets.Lexicons developed by Jackson et al. (2017) provide an essential basis for filtering and structuring the Twitter data into general categories. After lexicon-based analysis, topic modeling, unsuper-vised learning method, was used to identify key topics within the categorized data.The structure of the paper is as follows: the theoretical framework of issue ownership will beintroduced first. Theoretical part is followed by Data and Methods section which first introducesthe research questions and the research process. Then both text analysis methods of lexicon-basedand topic modeling approaches are explained in detail. After that results of both analyses arepresented. The paper is then concluded by discussion. Theory of Issue Ownership Reservations about the explanatory power of social-structural model in determining individuals’voting patterns has increased the demand for other theories that give emphasis on supply sideexplanations, namely actions of politicians and parties (Hosch-Dayican et al. 2013, p. 2; Damore 2004,p.391). Supplysidetheorieshavebeenfocusingonpoliticalissuesasthemostdecisivefactorof individuals’ voting decisions. One of these theories is called issue ownership, which claimsthat parties have a differing reputation in handling political issues. Issue ownerships changeusually slowly and they are formed based on long-term actions of parties (Hosch-Dayican et al.2013, p. 5). One explanation for slow shifts of issue ownership are the links between politicalconflict and social structure (Petrocik 1996, p. 827). Besides reputation, parties can also claim issue ownership via mere association which indicates spontaneous response of the voters to linka party with the issue (Hosch-Dayican et al. 2013, p. 4). Parties have thus established themselvesas experts of taking care of certain issues which gives their candidates credibility on those issues.2  In the late 20th century, in the United States, Democrats were considered to be better at handlingsocial welfare and civil rights issues where as Republicans have more credibility on social issues,foreign policy and government management (Petrocik 1996, p. 831; Damore 2004). In April 2017, the public opinion had shifted so that the Democratic party was perceived doing better job intraditionally Republican topics, including foreign policy, immigration, and government spending(Pew Research Center 2017). Voters’ perceptions of issue ownership are not likely change while campaigning, but candi-dates can try to affect the exposure of issues in which they have relative advantage to other can-didates (Petrocik 1996). All campaigning efforts of candidates thus do not pursue to change the minds of voters, but to “market” problems reflecting the issues on which they have an advantage(Petrocik1996,p.828). Ifthesalientissuesoftheelectionaresuccessfullydeterminedbyoneparty, other candidates are facing an uphill battle to challenge the issue-dominant candidate. Damore(2004) has shown that Democratic presidential candidates, in the time period of 1976-1996, were“trespassing” more frequently on Republican issues than the other way around.It follows from the theory of issue ownership that candidates should prefer avoiding directconfrontation on policy standings with other candidates while running presidential campaigns(Sigelman and Buell 2004). Namely, the objective in running a political campaign has been toincrease issue salience of one’s own issues and avoid taking clear positions on issues that candecrease the support, in order to maximize the electoral base. Conversely, Sigelman and Buell(2004)suggestthatcandidateshavebeeninvolvedinthesameissuesascandidatesoftheoppositeparties, even more than commonly expected. One way to account for candidates converging inthe same issues is that media might pressure candidates to talk about issues, or that candidatesneed to seem informed on the current issues (Sigelman and Buell 2004).Also, it seems plausible that candidates which are trailing in the polls can be expected to bemore prone to challenge their opponents on issues without any claims of issue ownership. Inopposition to the underdogs, candidates which start with strong leading positions could be ex-pected to be vaguer in their communication with voters and avoiding direct confrontation withtheir opponents. Data and Methods In this section of the paper, I will first introduce the research questions, after which research pro-cess is described with detail to enhance replication. Lastly, in this section I will introduce centraltext analysis methods which were selected for this study. Research Questions 1. Which major political issues were each of the presidential candidates linked to on Twitter bya) other Twitter users; b) candidates themselves?2. Regardingthefirstquestion,isthereadifferencebetweenthecandidatesofdifferentparties?3  Are there intra-party differences?3. What topics could be identified?I have no hypothesis for the first and third research question. For the second question, I hy-pothesizecandidatesofthesamepartytoaddresssameorsimilarissues(althoughnotnecessarilyoffersolutionstothoseissues),whiletheyavoidtacklingsametopicsasdifferentpartycandidates. Research Process This paper analyzes a unique dataset containing in total 258,139,170 tweets (93.9 million tweetsmentioning Democratic and 164.2 million Republican candidates) focusing on the 2016 US pres-idential primaries. The data collection ran from February 12th, 2016 to May 27th, 2016. Tweetswere collected by using Twitter Streaming API (Twitter Developer Documentation 2017) by using custom made Java-program which obtained tweets containing specified keywords, see Table 1.Keywords were updated as candidates dropped out the race by removing words linked to can-didates that dropped out. Data collection program was similar than the one used in Isotalo et al.(2016), but it was further modified by tracking candidate profiles thus obtaining all tweet activityof candidate user accounts, also see Table 1 section  Username .The whole of collected Twitter data were stored in csv-files from which candidate specifictweets were sorted to candidate files grouped by date. The criteria for assigning a tweet to acertain candidate was that the tweet had to contain at least one of the candidate associated words.It was sufficient that the word was partially matched in the tweet string. The lists of candidatespecific words are presented in Table 2. These words were selected to be more precise than somewords in Table 1, so that one could be more certain that a tweet really is connected to a candidate,for instance “Cruz” is a too generic word to be linked to the presidential candidate Ted Cruz.After dividing tweets to candidate and day specific files, the tweets were pre-processed (e.g.,setting the tweet text into lowercase) and tokenized to uni-, bi- and trigrams. These n-grams werethen matched with pre-determined keywords of eight categories. Jackson et al. (2017) had createdthe keywords by using tweets, public Facebook posts and debate transcripts of 2016 presidentialcandidates.Exact matches with words from eight dictionaries were checked with an R package Quanteda.For each tweet a number was assigned by the number of matches with dictionary words. Tweettokenswereallowedtomatchwithn-gramsofmorethanonedictionary. Thisphaseoftheanalysiswas computationally heavy.After lexicon-based topic classification the tweets were processed further by removing retweetmarks from the tweet text. Then tweets were classified either being created srcinally by a candi-dateorsomeoneelse(e.g.,potentialsupporter). Inthepaper,Irefertothesedatasetsas candidate and  supporters . In practice the data in candidate data sets contains tweets that might have beencreatedlongbeforebythecandidate,butthosetweetswerenowcirculatingaroundTwitter. There-fore, I considered it to be important not to let those tweets to be mixed with content created by4  Table 1: Most of the key words used to obtain tweets for individual candidates Candidate Clinton Sanders Rubio CruzUsername @HillaryClinton @BernieSanders @marcorubio @TedCruz@SenSanders @teammarco @SenTedCruzOtherSearchTermsHillary Bernie Marco Rubio Ted CruzHillaryClinton Sanders Rubio CruzClinton fellthebern Rubio’s #TedCruz2016Clinton’s BernieSanders Cruz’ssecclinton Bernie2016 #Cruz2016#imwithher Bernie #CruzCrew#trabajocomohillary #Sanders2016 #TedCruz#LibertynotHillary DebateWithBernie#Hillary2016Candidate Trump Carson Bush KasichUsername @realDonaldTrump @RealBenCarson @JebBush @JohnKasichOtherSearchTermsTrump Carson Jeb Bush Kasich#Trump2016 bencarson #jeb2016 John KasichTrump’s #BenCarson2016 #JebBush Kasich4Us#TeamTrump #Carson2016 #Jeb Kasich’s#DonaldTrump #BenCarson #bush NoToKasich#makeAmericaGreatAgain #BC2DC16 Bush’s#TrumpTrain Carson’s#MakeAmericaStupid 5
Search
Similar documents
View more...
Tags
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks