Do Retweets indicate Interest, Trust, Agreement? (Extended Abstract)
Panagiotis Takis Metaxas, Eni Mustafaraj, Kily Wong, Laura Zeng, Megan O'Keefe, Samantha Finn
DDo Retweets indicate Interest, Trust, Agreement?(Extended Abstract)
P. T. Metaxas, E. Mustafaraj, K. Wong, L. Zeng, M. O’Keefe, S. Finn
Department of Computer ScienceWellesley College, Wellesley, MA, USA
ABSTRACT
Arguably one of the most important features of Twitter isthe support for “retweets” or messages re-posted verbatim bya user that were originated by someone else. (This does notinclude modified tweets that sometimes are referred to asretweets.) Despite the fact that retweets are routinely stud-ied and reported, many important questions remain aboutuser motivation for their use and their significance. In thispaper we answer the question of what users indicate whenthey retweet. We do so in a comprehensive fashion, by em-ploying a user survey, a study of user profiles, and a meta-analysis of over 100 research publications from three relatedmajor conferences. Our findings indicate that retweeting in-dicates not only interest in a message, but also trust in themessage and the originator, and agreement with the messagecontents. However, the findings are significantly weaker forjournalists, some of whom beg to differ declaring so in theirown user profiles. On the other hand, the inclusion of hash-tags strengthens the signal of agreement, especially whenthe hashtags are related to politics. While in the past therehave been additional claims in the literature about possiblereasons for retweeting, many of them are not supported, es-pecially given the technical changes introduced recently byTwitter.
1. INTRODUCTION
Twitter is a real-time information networkthat allows itsusers to write short messages (“tweets”) up to 140 charac-ters in length. We chose to focus on Twitter in particu-lar because, unlike other popular social media platforms, ithas grown into a real-time news source created by everydayusers. Twitter is credited for its role in political events suchas monitoring elections, aiding the so-called “Arab Spring”[2], and for drawing attention to news stories that were ini-tially largely ignored by traditional news media such as theWendy Davis filibuster and the Michael Brown shooting in Texas filibuster on abortion bill rivets online, by HeatherKelly. CNN.com, June 26, 2013. Ferguson, MO.Over the years Twitter has developed its own syntactic com-ponents. This paper focuses primarily on retweeting: Users retweet (RT) to forward a message from another sourceto their own followers. Retweeting was not part of the orig-inal design of Twitter, but was created through user initia-tive. Because of its later adoption date, tweet forwardingcan be done in one of two ways (one that is supported bythe Twitter API and another that is not): either by click-ing a “Retweet” button provided by the Twitter client (andsome other clients), or by modifying the original tweet andmanually typing “RT @[user]” or adding “via @[user]” in anew tweet. We call the tweets produced by typing “mod-ified retweets” (MRTs) to distinguish them from the firstkind. This distinction is important because an unmodifiedretweet is treated differently by the Twitter platform and isguaranteed to point to its originating source.The retweet button’s utilization initially was slow. In 2010,[5] counted 2.99M retweets generated by the retweet buttonon Twitter, which was 36.34% of the total of 8.24M retweetscollected in a corpus of 74M tweets. In a collection of 39Mtweets in mid-summer of 2012, about one fifth of retweetsare created manually by users.As a platform, Twitter has several characteristics that makeit convenient for research. Yet, it is a valid question to askwhether it is interesting to study any specific social mediaplatform. If Twitter does not exist in a few years, will thispaper’s findings matter? We argue that the answer is yes:Twitter is recording human communication that requires rel-atively little effort to produce and consume. This communi-cation is what we want to study. While any particular socialmedia platform may cease to exist or lose popularity in thefuture, the importance of human interaction through socialmedia is unlikely to change. Humans are social animals andtheir desire to communicate with each other and commenton their social environments is one of their universal andunique characteristics. We have chosen to study the interac-tion of humans through social media in an abstract way, nota way specific to the particular social media platform. Weare simply looking at the behavior as revealed through theirinteractions. Further, its service makes it easy for people tosay something. The effort in contributing to the general so-cial dialog is far less than that of writing a comment a blogon a web site or a newspaper op-ed, and it has wider impactthan talking person-to-person or via email. In addition, the a r X i v : . [ c s . S I] N ov ffort to propagate a message sent by someone else is also re-markably small – giving rise to degrading characterizationsof online social participation such as “slactivism” [4].So, why do people retweet? The published literature doesnot give a conclusive answer to this question, though it wasstudied as early as 2010 [1]. Some of the claims made in thepast can be characterized as straightforward (in the senseof being self-evident, ie. that people retweet to broadcastinformation) and others as outdated (in the sense that tech-nical changes in Twitter do not support them anymore, ie.that people retweet to appropriate information). Howeverthere is no agreement in the literature whether retweetingindicates trust, agreement or even endorsement of the orig-inator or the message. Moreover, there is no clear under-standing how the retweeting rates are affected by emotion,political intention, or propaganda, if they are affected at all.Our paper tries to answer conclusively these questions.The remainder of this extended abstract is organized as fol-lows: In the section 2 we present the results of the user sur-vey that answers the question of what Twitter users thinkabout their practice of retweeting. In subection 2.3 we studyan important user subgroup, reporters, that includes mem-bers whose retweeting practice appears to differ significantlyfrom that of the general user. We also present an analy-sis of user profiles that include a disclaimer about retweet-ing not indicating endorsement. Section 3, truncated inthis short paper, presents a meta-analysis of research pa-pers from three major conferences related to Twitter data.Finally, section 4 has our conclusions.
2. USER SURVEY
In mid-July 2014, we conducted a survey asking Twitterusers about various aspects of their retweeting behavior.The survey responses indicated that there is general consen-sus on several important factors that the majority of Twit-ter users consider when deciding whether or not to retweeta given message. However, we also found that there werealso significant variations across subcategories of users, in-cluding differences across the responses from reporters andnon-reporters, individuals interested in politics, users of dif-ferent ages, and users who use Twitter with different levelsof frequency.
We designed a comprehensive twenty-one item questionnairethat asked participants about their retweeting behavior. Ini-tial survey questions inquired about whether the participanthad a Twitter account, and if so, for how long. Other ques-tions relating to general Twitter usage included how oftenthe participant uses Twitter, how often they tweet, whythey use Twitter, how many followers they have, and howmany accounts they follow. The survey also asked partici-pants about what factors they find important when decidingwhether or not to follow an account on a scale from 1 (notat all) to 5 (extremely).After answering these questions, we asked participants toindicate how often they retweet, if at all. Only partici-pants who responded that they are familiar with retweet-ing were led to a page of questions asking them about whatfactors, both about the message and the account, they con- sider important when choosing to retweet a post. In termsof the message, options were whether they find the mes-sage interesting, emotionally resonant, entertaining, some-thing they endorse, something that they agree with, some-thing their followers might find interesting, something for-mative or something they find trustworthy. In terms of theaccount, options were whether they find the account cred-ible, a celebrity, someone they know personally, someonewho shares their opinion, someone they support, or some-one they trust. Both cases were multiple-choice ending witha write-in (“other”) option. These questions were posed asstatements on a likert-scale from 1 (strongly disagree) to 5(strongly agree). They were also asked about what is moreimportant when they decide to retweet: the message, theaccount or both.Additionally, participants who had reported retweeting inthe past week were asked to recall certain features aboutone particular retweet of their choice. These additionalquestions were designed to demonstrate whether users’ ac-tual behavior is consistent with their overall impressions oftheir own behavior. The final questions captured basic de-mographic questions about the participants including age,self-described ethnicity, gender and educational level.
We launched the study by utilizing our social links andreached directly audiences, primarily in academic and ethniccommunities that we had social and professional links (Twit-ter, Facebook, Linkedin and email). Reflecting on the modeof launching our survey, the responders’ sample ( n =316)was highly educated, with 25.3% possessing a Bachelor’s de-gree and an additional 43% a Master’s Degree or higher. Ofthe responders, thirty-three (11.4%) identified themselves asbloggers, journalists, or reporters.Our sample was representative of both males (43%) and fe-males (54.4%), with 2.6% chosing not to disclose gender. Interms of ethnicity, we left it as an open-ended question andrecoded participants responses as Caucasian/White (46%),Asian or Pacific Islander (11%), Black (3.8%), other (22%),and no-repyling (17%).The age demographics of our data is consistent with Pew’sreport of Twitter users on several respects. Reflecting thesimilar trend that younger individuals tend to be on socialmedia platforms more than older ones, including Twitter,38.6% of the participants report being under the age of 24,25.9% between 25-34, 13.9% between 35-44, 12.3% between45-54, and 9.2% over the age of 54. Additionally, our datacaptured users who use Twitter on average as 42.2% of par-ticipants use Twitter daily, 23.8% use it weekly, and 34%use it monthly or less. 55.7% have less than one hundredfollowers while 8% have over one thousand. 40% of partic-ipants provided their Twitter handle. Figure 1 summarizessome of our results about participants primary motivationsfor using Twitter and their retweeting behavior.When it came to retweeting behavior, the majority of par-ticipants responded that they believed that the message was Pew Research Internet Project: Social Networking FactSheet. Data from 2014. igure 1: Summary of the results of our surveywhy people use Twitter (top), and how often do theyretweet (bottom). more important than the account when choosing to retwteeta post. That is, when chosing whether or not to retweet atweet, they evaluated and put more significance on the tex-tual content of the tweet than on the user who had writtenthe tweeted. In fact, some users even reported to not payattention to who tweeted the post in the first place.Participants provided consistent responses when respondingto survey questions that asked about various factors theyconsidered important about a given message when choos-ing whether or not to retweet it. The majority of partic-ipants indicated that factors such as how interesting themessage is to themselves as well as their followers are impor-tant. They also cited trustworthiness, informativeness, andwhether they agree with the given message as other factorsthat were important (see Fig. 2, top).Some of the initial insights that we found helped guide ourfurther exploration of the data that was collected. First, wefound that the majority of participants tended to retweetposts from accounts that they follow (as opposed to accountsthat they do not follow) and therefore it is important tounderstand why they choose to follow certain accounts inthe first place. Consistent with the hypothesis that users care about credi-bility and personal interest when navigating through socialmedia platforms for information, responses to the surveyindicated that specific factors including “shares my inter-est” and “trustworthiness” were most commonly cited as im-portant factors to consider when choosing whether or notto retweet a given account (see Fig. 2, bottom). Mean-while, factors such as “celebrity” and “entertainment” werenot rated as important in Twitter users’ decision-makingprocess. Additionally, there were other factors that yieldedresponses similar to that of “shares my opinion” where therewas no strong opinion. The participants’ responses wererecoded from a likert-scale from 1 (strongly disagree) to 5(strongly agree) to “agree,”“neither”, and “disagree.”Reflecting the amount of emphasis the participants reportedplacing on the account as opposed to the message, resultsregarding what qualities about an account are important toconsider when choosing to retweet a message indicated lessagreement and certainty. The criterion that seemed mostimportant overall was that the account is “credible.” Oursurvey did not ask people about how they judge whetheror not an account is credible and whether or not that istruly the case. However, this finding suggests that Twitterusers do care about whether or not a given account appearstrustworthy.
In light of the previous findings, of particular interest is thestatement encountered in certain Twitter users’ profile sum-mary that retweeting does not mean endorsement or agree-ment. It seems that this statement contradicts the findingspresented in this section, so we examined it in greater depth.There are relatively few users who include such statement.Using the Twitter Search interface for the terms “RT, en-dorsement” we found 384 accounts containing both of theseterms. An exhaustive categorization of these account showedthat 197 (52.7%) of them belong to people in the Media (re-porters, journalists, media producers) including 11 bloggers,99 (26.5%) to politicians, 43 (11.5%) to non-media relatedcompanies, 24 (6.4%) to people who list politics in their in-terests, and 9 (2.4%) to academicians. The remaining 0.5%of accounts were from people indicating that their RT is, infact, endorsement.We then tried to collect a comprehensive sample of suchprofile statements. Using the Twitter API we collected allthe user profiles that make a statement along these lines. Infact we found that the terms “retweet”, “RT” co-occur withthe terms “agreement” or “endorsement” in 2,585 profiles.We randomly sampled uniformly 10% of them and foundthat media-related accounts comprised 52.65% and 12.65%to politicians.There is an over-representation of journalists, reporters andmedia producers who include this disclaimer in their profiles.Other journalists have challenged this practice since its in- E.g., see Commentaries at the Washington Posthttp://wapo.st/19OIcfy and NPR’s “Retweets are endorse-ments” http://bit.ly/YJBQgQ and http://bit.ly/1v9lBY3lusion in one’s profile is not an effective disclaimer (manypeople will likely miss it).Finally, one has to consider the reason for people placingsuch a clarification statement in their profile as addressinga concern that one’s retweeting practices may be mistakenby others. It is reasonable to argue that the mere fact ofincluding this disclaimer is an implicit admission by thoseusing it that others may mistake their intentions,becausefor most people retweeting is endorsement.Journalists andpoliticians may worry about the effect of their retweets ontheir curated public images. As we mentioned, several journalists on Twitter include intheir Twitter profile description a disclaimer stating that“retweeting does not equal endorsement.” It’s almost as ifthis subgroup of individuals feel a need to distinguish them-selves from the attitudes and behavior of other general userson Twitter. To see if participants in the study responded inways that reflected this distinction, we ran five Welch TwoSample t-tests, as well as the Wilcoxon rank sum tests, toconfirm the t-tests. While the t-test’s assumption of nor-mality is fairly robust, we wanted to confirm that we wouldgenerate similar results even if that assumption is not made.There was a significant difference in the degree to which re-porters (M= 3.4, SD=1.1) versus non-reporters (M= 4.1,SD=0.9) agreed with the statement that “When retweeting,it is important that the message is something that I agreewith,” t(32.3)=-3, p=0.006. We also found that 57% ofreporters agreed and 17.9% disagreed with the statement,compared to 75% of non-reporters agreeing and 3.8% dis-agreeing with the statement.There was also a significant difference in the degree to whichreporters (M= 3.3, SD=1.1) versus non-reporters (M=3.9,SD=0.9) agreed with the statement that “When retweeting,it is important that the message is something that I endorse,” t(32.7)=-2.9, p=0.007. While 57.1% of reporters agreed and21.4% disagreed with the statement, 69.5% of non-reportersagreed with the statement and 6.6% disagreed.Finally, there was some difference in the degree to whichreporters (M= 3.7, SD=1.0) versus non-reporters (M= 4.0,SD=0.9) agreed with the statement that “When retweeting,it is important that the account is one that I find credible,” t(35)=-1.88, p=0.07. While 72.4% of reporters agreed and13.8% disagreed with the statement, 76.2% of non-reportersagreed with the statement and 5.7% disagreed.
There was a significant difference in the degree to whichpolitical users (M= 3.6, SD=1.1) versus non-political users(M= 3.1, SD=1.1) on Twitter find it important to follow ac-counts that share their opinion, t(116)=2.7, p=0.008. While76.1% of political users were motivated and 4.5% believedthat when choosing to follow certain accounts it was im-portant that the accounts share their opinions, 71.6% ofnon-political users believed it important and 5.9% did not.The test for the difference in the amount by which politicalusers versus non-political users on Twitter tend to retweetaccounts that they don’t follow was not significant.
3. A META-ANALYSIS OF RESEARCH PA-PERS
To focus our research question a bit further, we did a meta-analysis of the research corpus that has been published inthe last several years on Twitter. Starting with abstractinspection, we collected over 100 relevant papers publishedsince 2008 and 2013 in three major conference venues re-lated to the subject: AAAI ICWSM, IEEE SocialCom, andWWW. From the initial core we expanded the coverage toreferences found in the papers from these three venues. Thepage limit of this conference does not allow us to expandthis section here. We simply state that our findings indicatesupport in the survey results, and provide insight on how theemotions associated with hashtags can affect the retweetingrates and visibility of a topic being discussed on Twitter.
4. CONCLUSIONS
While in the past there have been several claims in the lit-erature about possible reasons for retweeting, no conclusivestudy of why people retweet appears. This paper aims toclose this gap. Employing a user survey, a study of user pro-files, and a meta-analysis of over 100 research publicationsfrom three relevant major conferences, we answer the ques-tion of what users mean when they retweet. Our findingsshow that retweeting indicates, not only interest in a mes-sage, but also trust in the message and the originator, andagreement with the message contents. However, the find-ings are significantly different for journalists, who are morelikely to include a disclaimer on their user profiles that theirretweeting does not mean agreement or endorsement. Theinclusion of hashtags strengthens the signal of agreement,especially when the hashtags are related to politics.We believe that our findings are an important contributionto social media research and to understanding of how somejournalists treat Twitter. In particular, it can help makesense of the crowd’s belief of the validity of a rumor spread-ing [3] (see http://bit.ly/twittertrails ).
5. ACKNOWLEDGEMENTS
This research was partially supported by NSF grant CNS-1117693 and by the Wellesley Science Trustees Fund. Theauthors would like to thank Prof. Jonathan Cheek and Prof.Robin Akert for advise designing the survey.
6. REFERENCES [1] D. Boyd, S. Golder, and G. Lotan. Tweet, tweet,retweet: Conversational aspects of retweeting ontwitter. In
HICSS , pages 1–10. IEEE, 2010.[2] A. Bruns, T. Highfield, and J. Burgess. The arab springand social media audiences: English and arabic twitterusers and their networks.
American BehavioralScientist , 57(7):871–898, 2013.[3] S. Finn et al. TRAILS: A system for monitoring thepropagation of rumors on twitter. Computation andJournalism Symposium, NYC, NY, 2014.[4] M. Gladwell. Small change: Why the revolution willnot be tweeted.
The New Yorker , October 4, 2010.[5] B. Suh, L. Hong, P. Pirolli, and E. H. Chi. Want to beretweeted? large scale analytics on factors impactingretweet in twitter network. In
SocialCom , pages177–184. IEEE, 2010., pages177–184. IEEE, 2010.