THE PSYCHOLOGY BEHIND PEOPLE’S DECISION TO FORWARD DISASTER-RELATED TWEETS

Huaye Li, Howe School of Technology Management, Stevens Institute of Technology, Hoboken, NJ, USA, hli21@stevens.edu

Yasuaki Sakamoto, Howe School of Technology Management, Stevens Institute of Technology, Hoboken, NJ, USA, ysakamot@stevens.edu

Yuko Tanaka, Research Organization of Information and Systems, National Institute of Informatics, Tokyo, Japan, yuko-tanaka@nii.ac.jp

Rongjuan Chen, Howe School of Technology Management, Stevens Institute of Technology, Hoboken, NJ, USA, rchen@stevens.edu

Abstract During a disaster, information spreads through social media. Although information that can save lives exists in social media, it can be difficult to find because it is buried under a sea of unverified information. One challenge in improving the use of social media for disaster management is to facilitate the spread of actionable information that improves people’s well-being and at the same time reduce the spread of misinformation that confuses people and interferes with the discovery of useful information. To this end, we study the psychology of message forwarding in social media. Based on the understanding of the psychology, we develop recommendations for designing tools that augment the forwarding decisions of individuals such that useful information spreads in social media. Here, we present the results from analyzing the forwarding of tweets related to the disasters caused by the Great East Japan Earthquake in 2011. The results from questionnaires suggest that people will be more likely to share a disaster-related message in a social media environment when they perceive the message as more important, accurate, anxiety provoking, familiar, informative, and fluent. The implications of the results on improving the quality of information in social media during disaster response are discussed.

Keywords: Information spread, Forwarding behavior, Social media, Disaster response.

1 INTRODUCTION

Communication during disasters increasingly relies on social media technologies. For example, Twitter played an important role in sharing information and organizing responses during the disasters caused by the Great East Japan Earthquake in March 2011 (Ogiue 2011; Tachiiri 2011). In Twitter, users post brief messages of up to 140 characters. These messages called tweets are broadcast to the users’ followers. Twitter users can also re-tweet others’ tweets through a click of a button, allowing them to share, or forward, information with their followers and contribute to the far-reaching spread of information. People use social media technologies to not only acquire new information but also generate content. Given the growing use and participatory nature of social media, better understanding of people’s behavior in such an environment is essential.

The work reported here contributes to the better understanding of the psychology behind people’s decision to forward disaster-related messages in a social media environment. In particular, it tries to answer the following research question: How does people’s perception of a disaster-related message relate to their decision to share it in a social media environment? Rumor psychologists have identified some perceptions including anxiety and importance that affect rumor transmission in face-to-face environments (e.g. Rosnow 1991). Through questionnaires using actual tweets related to the Great East Japan Earthquake, the work reported here makes theoretical contributions by extending theories of rumor psychology to sharing of disaster-related messages in social media.

Better understanding of how people pass information in social media will be helpful for improving the quality of information in social media, which will in turn improve the use of social media for disaster management. One challenge in using social media during disaster response is that, although social media may allow people to obtain useful information and make sense of the situation (Cataldi et al. 2010; Demirbas et al. 2010; Doan et al. 2012; Jansen et al. 2009; Sakaki et al. 2010; Sankaranarayana et al. 2009; Sriram et al. 2010; Sutton et al. 2008), social media can also facilitate the spread of inaccurate information (Castillo et al. 2011; Starbird et al. 2014; Tanaka et al. 2013). In fact, in the aftermath of the Great East Japan Earthquake, false tweets confused people and interfered with the disaster response coordination (Inose 2011). Realizing this, the Japanese government alerted people to unverified information on social media. Similarly, the Federal Emergency Management Agency (FEMA) created rumor control web sites for the 2012 Hurricane Sandy and the 2013 Oklahoma Tornado (FEMA 2012; 2013). However, these rumor control techniques may not prevent individuals from spreading false information (Tanaka et al. 2013). The results reported in the current paper will provide practical contributions by suggesting ways to reduce the spread of false information and increasing the proportion of useful information in social media during disaster response.

1.1 Why Extend Rumor Research?

Previous work on rumor psychology has revealed that people spread rumors during responses to disasters (e.g., Prasad 1935, 1950; Sinha 1952). A rumor is unverified information in times of uncertainty that is relevant and seems useful to understand the situation and manage risk (Allport & Postman 1947; DiFonzo & Bordia 2007; Rosnow 1991; Shibutani 1966). Similarly, the messages related to the Great East Japan Earthquake used in the work reported in the current paper appeared in uncertain times and were unverified and relevant.

Table 1 shows the four main topics of the current article: the role of (1) perception in the (2) sharing of (3) disaster-related information in (4) social media. Table 1 also shows the topics considered by three areas of research that focuses on information sharing. Most research on online social networks has considered information sharing in social media but has not integrated the role of perception in information sharing and has not covered disaster situations. Research on trend development has considered the role of perception in the sharing of social media messages but tends not to focus on disaster environments. In contrast, most research on rumor psychology has focused on the role perception plays in the sharing of disaster-related rumors but not on social media environments.

Research Sharing Perception Disaster Social media

Social networks online (e.g., Aral & Walker 2011; Huberman et al. 2009) ✔ ✔

Trend development (e.g., Berger & Milkman 2012; Ha & Ahn 2011) ✔ ✔ ✔

Rumor psychology (e.g., DiFonzo & Bordia 2007; Rosnow 1991) ✔ ✔ ✔

Current article ✔ ✔ ✔ ✔

Table 1. The overlap between the topics of relevant past research and the work reported here. A check mark indicates that the specified research focuses on the specified topic.

Research on trend development that considers perception typically focuses on arousal and valence, which are two main measures of feelings (e.g., Berger & Milkman 2012; Heath 1996). In contrast, work on rumor psychology includes a wider range of perceptions that are applicable to disaster environments. For this reason, the work presented here extends research on rumor psychology to a social media environment in an attempt to better understand how people’s perception of disaster-related tweets relates to the sharing decisions. Thus the review will focus on past research on rumor psychology.

1.2 Review of Rumor Psychology

Based on their analysis of the spread of rumors after WWII, Allport and Postman (1947) proposed that rumor transmission was a function of the importance of the rumor to the person multiplied by the ambiguity of the evidence pertaining to the rumor. Importance relates to how relevant, significant, and consequential the rumor is. Ambiguity is how uncertain the information carried by the rumor is due to the surrounding context.

Later, Anthony (1973) introduced anxiety as another key element in rumormongering. In school settings, anxious students were more likely to report that they heard a rumor (Anthony 1973) and that they passed on a rumor (Jaeger et al. 1980). Moreover, how anxious people felt after reading a rumor predicted how likely they were to share the rumor (Rosnow et al. 1988). Following Rosnow et al., in the work reported here, anxiety means how anxious, worried, or concerned a person becomes after encountering a message. Jaeger et al. (1980) and Rosnow et al. (1988) also showed that people were more likely to share a rumor when they believed the rumor to be more accurate or true (see also DiFonzo & Bordia 2007; Li & Sakamoto 2013; Rosnow 2001).

Rosnow (1991) combined these findings and extended the model of rumor by Allport and Postman (1947). He proposed four factors behind rumor transmission: uncertainty of the situation, relevance of the rumor to the person, anxiety associated with the rumor, and trust in the rumor based on its perceived accuracy. Of these four factors, uncertainty and relevance are akin to ambiguity and importance in the Allport and Postman model, and anxiety and trust are new additions that are absent in the model.

In addition to the four factors identified by Rosnow (1991), the perceived familiarity and informativeness of messages can also predict rumor transmission under some conditions (DiFonzo & Bordia 2007). For example, increasing the familiarity of a message can increase its perceived truth (Hawkins & Hoch 1992; Hawkins et al. 2001), and people are more likely to spread rumors that they believe to be true (DiFonzo & Bordia 2007; Li & Sakamoto 2013; Rosnow 2001). People are also more likely to share a message that they believe to be informative, because they want to help others by passing on useful information (DiFonzo & Bordia 2007).

Several studies have tested the proposal from work on rumor psychology in social media environments (Kwon et al. 2014; Oh et al. 2010; Spiro et al. 2012; Tanaka et al. 2012). For example, Spiro et al. (2012) analyzed tweets about the Deepwater Horizon oil spill in 2010 and concluded that perceived importance was a key element of rumormongering based on their observation that an event with high media coverage was more likely to be re-tweeted. Tanaka et al. (2012) asked Japanese subjects how important, accurate, anxiety provoking, and familiar they thought given rumor tweets related to the Great East Japan Earthquake were as well as their intent to share the tweets. They found that familiar tweets were perceived as more accurate, as in the past work (e.g., Hawkins & Hoch 1992), but, inconsistent with the past findings from rumor research, they found that only the perceived importance of a tweet had a significant positive relation to Japanese people’s intent to share the tweet. The current work aims at gaining deeper understanding of the extent to which perceptions that past studies have identified to be related to rumor transmission also apply to information sharing on social media.

Another perceptual factor, which has not been extensively researched in relation to information sharing, is fluency, or the ease of processing information. The feelings-as-information theory (see Schwarz & Clore 2007 for a review) proposes that fluency can influence various judgments, including people’s perceptions of truthfulness (Begg et al. 1992), effortlessness (Song & Schwarz 2008), risk (Song & Schwarz 2009), intelligence (Oppenheimer 2006), and liking (Reber et al. 2004). For example, in the Song and Schwarz (2008) studies, subjects thought that the instructions for exercise routine and Japanese roll recipe written in an easy-to-read font type would require less effort to follow than the same instructions written in a hard-to-read font type. Similarly, people may like the same message better and find it more believable in an easy-to-understand form than in a hard-to-understand form. Fluency might affect people’s perception of messages in social media and their intention to share these messages.

1.3 Hypothesis

Of the factors reviewed previously, importance, anxiety, accuracy, familiarity, informativeness, and fluency are perceptions of disaster-related messages. Uncertainty is a characteristic of the situation or the context. Thus, the work reported here considers the following hypothesis: people are more likely to share a message in a social media environment when they perceive the message as more important, accurate, anxiety-provoking, familiar, informative, and fluent.

2 METHOD

The study involved two questionnaires for obtaining people’s perceptions of actual tweets discussing the disasters caused by the Great East Japan Earthquake and their likelihood of sharing the tweets.

2.1 Subjects

Fifty-one workers of Amazon’s Mechanical Turk (https://www.mturk.com) completed the questionnaire asking their likelihood of sharing a given tweet. There were 30 women and 21 men. Their average age was 36. Forty-three workers who did not participate in the questionnaire on sharing completed the questionnaire asking their perceptions of a given tweet. There were 21 women and 22 men. Their average age was also 36. All subjects were residents of USA. The use of two questionnaires to separate the sharing from perception was to keep subjects from explicitly thinking about perceptions when rating the sharing likelihood. The recruitment of subjects ended when each tweet resulted in the pre-specified number of responses. This procedure, detailed in Procedure, determined the total number of subjects in the two questionnaires.

2.2 Materials

The stimuli were 200 English tweets related to the Great East Japan Earthquake, collected from Twitter on March 5th, 2012. Past work has reported a strong correlation between English and Japanese tweets discussing events related to this earthquake (Doan et al. 2012). The first step was to identify Twitter users who communicated information about the disasters using Twitter. The second step was to collect all of their tweets uploaded after March 11th, 2011 when the Great East Japan Earthquake happened. The next step was to identify relevant tweets by searching the following five keywords: earthquake, tsunami, nuclear, Fukushima, and radiation. The last step was to randomly sample 200 tweets that were unique and were not re-tweets. Figure 1 shows one of the tweets in a text box.

Figure 1. The questionnaire on sharing asked subjects to rate the likelihood that they would

share a given message on social media.

2.3 Procedure

Subjects completed the questionnaires online through Amazon’s Mechanical Turk. A few research groups have shown that researchers can collect high-quality data from Mechanical Turk (e.g.,

!"#$%&'()*+! "#$%! %&'()! $%! *+,'&! -.,-/.0%! 1.%-,2%.! &,! ($%*%&.131./*&.(! $24,15*&$,2! $2!%,6$*/! 5.($*! %'6#! *%! "7$&&.1! *2(! 8*6.+,,9:! $2! -*1&$6'/*1! $24,15*&$,2! 1.;*1($2;! &#.! <=>>!

",#,9'!.*1&#?'*9.:!7#$6#!#$&!@*-*2!,2!A*16#!>>&#B!"#$%!.*1&#?'*9.!&1$;;.1.(!&%'2*5$%:!7#$6#!

/.(! &,!5*%%$C.! /,%%! ,4! /$4.! *2(! (.%&1'6&$,2! ,4! $241*%&1'6&'1.B! "#.! &%'2*5$%! 6*'%.(! *! 2'6/.*1!

*66$(.2&! $2! 8'9'%#$5*:!7#$6#! *44.6&.(!#'2(1.(%! ,4! &#,'%*2(%!,4! 1.%$(.2&%B! D,'1! 1.%-,2%.! $%!

C*/'*+/.E!$&!7$//!#./-!'%!-1.-*1.!4,1!4'&'1.!($%*%&.1%B!

!

,)-.&(#./')-+! F5*;$2.! &#*&! ),'! 1.*(! &#.! 4,//,7$2;! $24,15*&$,2! $2! %,6$*/! 5.($*! %'6#! *%!"7$&&.1!*2(!8*6.+,,9B!G/.*%.!*2%7.1!&#.!?'.%&$,2!+./,7B!

!!

!

0

1(2-./')+! H,7! /$9./)! $%! $&! &#*&! ),'! 7$//! %#*1.! &#$%! $24,15*&$,2! $2! %,6$*/! 5.($*! %'6#! *%!8*6.+,,9!*2(!"7$&&.1I!

! >! <! J! K! L! M! N! !

3'.0".0"440 !! !! !! !! !! !! !! 56.&272480

!

0

999,:08'(0;").0.'0$)';07'&20"<'(.0.=20"<'>20-.".272).?0@42"-2027"/40(-09990

0

0

,:0.=/-0/-08'(&0:/&-.0A,B0:&'70.=/-0<".#=?0@42"-20#'7@42.20.=20:'44';/)%+0

0

F!*5!*! ! ).*1%!,/(! !!5*/.! !!4.5*/.!:!!+,12!$2!&#.!6$&)!,4! ! B!

!

F!6#.69!%,6$*/!5.($*:!%'6#!*%!"7$&&.1!*2(!8*6.+,,9:! !!1*1./)! !!7..9/)! !!(*$/)B!

!

0

B=")$08'(0:'&08'(&0@"&./#/@"./')C0

!

DBEF3GH,0"/*0)'.0&2"#=/)%0>/#./7-0=.I48JK2L'6MN0

O'+5$&!

Buhrmester et al. 2011; Crump et al. 2013; Mason & Suri 2011; Mason & Watts 2009; Paolacci et al. 2010). The procedure in the current work followed their recommendations.

Figure 1 shows an example webpage that subjects saw when completing the questionnaire on sharing. Subjects indicated their responses by clicking a value on a scale of 1 to 7, where 1 indicated not at all likely to share and 7 indicated extremely likely to share. Figure 2 shows the main part of the questionnaire on perceptions, in which subjects rated how familiar, accurate, informative, important, fluent, and anxiety provoking they thought a given tweet was.

Figure 2. The questionnaire on perceptions asked subjects to rate how they felt after reading a

given tweet. The questions were based on the definitions used in the past studies on rumor (see DiFonzo & Bordia 2007) and fluency (see Schwarz & Clore 2007).

After submitting their response, subjects had an option to rate another tweet, which was randomly selected from the available sample. A subject could rate the same tweet only once but could continue to rate different tweets that still needed ratings. Data collection ended when each tweet had ten

!"#$%#&'#$(&)*+%&),##)-&

&!

!

&!"#$%#&$.%,#'&)*#&/0""0,+.1&23#%)+0.%-&

&

45&67#'$""8&*0,&/$9+"+$'&:'#;01.+<$="#8&,#"">?.0,.@&+%&)*+%&),##)&)0&A03B&

&

! "! #! $! %! &! '! (! !

)*+!,+!,--!.,/0-0,1! !! !! !! !! !! !! !! 23+14/4-5!.,/0-0,1!

!

C5& 67#'$""8& *0,& $;;3'$)#& 0'& )'3#& :$%& 0DD0%#(& )0& (+%)0')#(& 0'& /$"%#@& (0& A03& )*+.?& )*+%&

),##)&+%B&

&

! "! #! $! %! &! '! (! !

)*+!,+!,--!,6671,+4! !! !! !! !! !! !! !! 23+14/4-5!,6671,+4!

&

E5&67#'$""8&*0,&+./0'9$)+7#&:3%#/3"8&#(3;$)+0.$"@&+%&)*+%&),##)&)0&A03B&

&

! "! #! $! %! &! '! (! !

)*+!,+!,--!08.*1/,+094! !! !! !! !! !! !! !! 23+14/4-5!08.*1/,+094!

&

F5&67#'$""8&*0,&+9D0')$.)&:%+1.+/+;$.)8&;0.%#23#.)+$"@&+%&)*+%&),##)&)0&A03B&

&

! "! #! $! %! &! '! (! !

)*+!,+!,--!0/:*1+,8+! !! !! !! !! !! !! !! 23+14/4-5!0/:*1+,8+!

&

G5&67#'$""8&*0,&#$%A&:$%&0DD0%#(&)0&*$'(@&+%&+)&/0'&A03&)0&3.(#'%)$.(&)*+%&),##)B&

&

! "! #! $! %! &! '! (! !

)*+!,+!,--!4,;5! !! !! !! !! !! !! !! 23+14/4-5!4,;5!

&

H5&67#'$""8&*0,&$.I+03%&:,0''+#(8&;0.;#'.#(@&(0#%&)*+%&),##)&9$?#&A03B&

&

! "! #! $! %! &! '! (! !

)*+!,+!,--!,830*7;! !! !! !! !! !! !! !! 23+14/4-5!,830*7;!

&

JKLMNOPQ&$+(&.0)&'#$;*+.1&7+;)+9%&*)5"AR4#S0ITU&

ratings. Subjects received one cent for each tweet they rated. This amount was based on the finding that increasing pay would not change the quality of responses (Mason & Watts 2009).

The study reported here took place in April 2012, 13 months after the earthquake hit Japan. All subjects were residents of USA when they participated in the study. The disasters that followed the earthquake were unexpectedly severe and involved worldwide concerns including nuclear plant failures and resulting radiation leaks. Consequently, one year after the earthquake, people in USA were still posting messages related to the earthquake on Twitter, and the event was still fresh in people’s minds, especially those of residents in Japan. Examining the sharing behavior of people in USA is important because messages on social media can have worldwide influence. In fact, one message re-tweeted from London resulted in saving the lives of children isolated at a school during the disasters caused by the Great East Japan earthquake (Inose, 2011).

3 RESULTS

The analyses were based on seven scores associated with each of the 200 tweets: sharing and six perceptions. Each score was the mean of 10 ratings from 10 individuals. All ratings from all subjects were included in the analyses.

3.1 Overall Pattern

Each histogram in Figure 3 shows the frequency distribution of scores for a variable. For example, the histogram on the bottom right corner shows the frequency of tweets that are associated with a specific range of scores for the likelihood of sharing. Each bar is for a range of sharing scores indicated on the horizontal axis. The height of the bar indicates the frequency of tweets, and the sum of the frequencies indicated by all bars is 200 because there are 200 tweets. The histogram for sharing revealed that the likelihood of sharing ranged from about 1.5 to 6. The mean likelihood of sharing the 200 tweets was 3.58, slightly below the middle of the scale.

The histograms for the six perceptions indicated that the messages tended to be relatively high on accuracy (mean = 4.73) and fluency (mean = 5.36) as suggested by the shift of distributions toward the right side, relatively low on anxiety (mean = 2.76), familiarity (mean = 2.72), and importance (mean = 3.11) as suggested by the shift of distributions toward the left side, and relatively average on informativeness (mean = 4.18). The 200 tweets varied in the scores for the six perceptions as well as for sharing, making correlational analyses meaningful.

3.2 Relationship between Perception and Sharing

The main focus of the work presented here was to test the hypothesis that people are more likely to share a message in a social media environment when they perceive the message as more important, accurate, anxiety provoking, familiar, informative, and fluent. Figure 3 shows the Pearson’s correlation coefficient between each of the six perceptions and sharing as well as an associated scatterplot with a regression line. Consistent with the hypothesis, the six perceptions all had significant positive relationships to the likelihood of sharing: messages rated as more important, accurate, anxiety provoking, familiar, informative, and fluent were also rated as more likely to be shared.

Figure 3 also shows the correlation coefficients among the six perceptions and associated scatterplots with a regression lines. The results reveal multicollinearity for the six perceptions. They are highly correlated with one another. Although the presence of correlated predictors reduces the validity of interpreting the estimate for each predictor in a regression model, it does not reduce the predictive power of a regression model as a whole. Thus, the fit of a regression model comes next.

Figure 3. The scatterplots with regression lines and Pearson’s correlation coefficients show

that each of the six perceptions had a positive relationship to sharing. In addition, the six perceptions were correlated with one another. One star, two stars, and three stars indicate significance level at .05, .01, and .001, respectively.

3.3 Predicting Sharing from Perception

It would be useful to be able to predict the spread of new tweets based on people’s perceptions of the tweets. An initial step is to test if a regression model can successfully fit the observed data. Table 2 shows the results of fitting a linear regression model to the likelihood of sharing, with the six perceptions as predictors. As predicted, the regression model was significant, F(6, 193) = 17.29, p < .001, with adjusted R2 of .33, indicating that the model as a whole has a predictive power and that the six perceptions were all related to the likelihood of sharing. Only anxiety and fluency show significant estimates in Table 2 because of the multicollinearity of the six perceptions as mentioned previously. Using people’s perceptions of a new tweet, it would be possible to predict the extent to which the new tweet would spread.

Accuracy

1 2 3 4 5 6 7

0.37*** 0.35***

1 2 3 4 5 6 7

0.52*** 0.37***

1 2 3 4 5 6 7

0.67***

1

2

3

4

5

6

7

0.27***

1

2

3

4

5

6

7

Anxiety

0.41*** 0.36*** 0.63*** 0.59*** 0.52***

Familiarity

0.26*** 0.46*** 0.47***

1

2

3

4

5

6

7

0.30***

1

2

3

4

5

6

7

Fluency

0.42*** 0.62*** 0.43***

Importance

0.65***

1

2

3

4

5

6

7

0.43***

1

2

3

4

5

6

7

Informativeness

0.44***

1 2 3 4 5 6 7 1 2 3 4 5 6 7 1 2 3 4 5 6 7 1 2 3 4 5 6 7

1

2

3

4

5

6

7

Sharing

3.3.1 Estimate Standard Error

Statistical Significance

(Intercept) -0.01 0.59 t = -0.01 p = .991

Accuracy (as opposed to false) -0.13 0.14 t = -0.91 p = .366

Anxiety (worried, concerned) 0.49 0.10 t = 4.74 p < .001

Familiarity (recognizable, well known) 0.11 0.13 t = 0.81 p = .417

Fluency (easy to understand) 0.38 0.11 t = 3.55 p = .001

Importance (significant, consequential) 0.09 0.15 t = 0.59 p = .556

Informativeness (useful, educational) 0.06 0.14 t = 0.44 p = .659

Table 2. The linear regression model of sharing with the six perceptions as predictors was significant, indicating that the model as a whole had a predictive power.

4 DISCUSSION

The results from questionnaires using actual tweets related to the Great East Japan Earthquake suggested that people’s perceptions of disaster-related tweets could relate to their sharing of these tweets. In particular, perceived accuracy, anxiety, familiarity, importance, informativeness, and fluency of disaster-related tweets appear to positively relate to people’s decision to share these tweets, and these perceptions may be able to predict the spread of disaster-related tweets. The study reported here extended past research on rumor psychology, and the findings could provide useful information for improving the use of social media during disaster management. The remainder of this section consists of discussion of theoretical and practical implications and consideration of limitations and future directions.

4.1 Theoretical Implications

The results presented in the current paper suggest that the basic law of rumor transmission and its extension developed in the rumor psychology literature apply to the sharing of disaster-related messages in a new environment, social media. In addition to accuracy, importance, anxiety, familiarity, and informativeness taken from past work on rumor psychology, fluency was a strong predictor of sharing. This finding implies a new extension to the basic law of rumor: perceived fluency, or ease of processing, may play a role in rumormongering. The work reported here also complements research on revealing the relationship between the structure of online social networks and the spread of information and behaviors (e.g., Aral et al. 2009; Aral & Walker 2011, 2012; Cha et al. 2010; Fang et al. 2013; Huberman et al. 2009; Kwak et al. 2010; Leskovec et al. 2007). Moreover, it adds to research on predicting the spread of information based on its content (e.g., Asur & Huberman 2010; Bandari et al. 2012; Ha & Ahn 2011).

4.2 Practical Implications

The results of the work presented here may be useful for the development of recommendations for the design and use of social media technologies during disaster response management. To deal with the spread of unverified information during responses to disasters, one potentially effective solution may be to ask the crowds to report on their perceptions of disaster-related tweets and flag the ones that are

likely to spread. Among the flagged tweets, unsubstantiated ones should be closely tracked to reduce the spread of potentially inaccurate information.

On the flip side, government agencies and media need to spread warnings and instructions during disasters. The current results suggest that people will share warning messages more when they perceive the messages as accurate, easy to understand, important, informative, familiar, and anxiety provoking. Although it is not easy to control all of these perceptions, some may be easier. For example, citing a credible source may increase perceived accuracy, and writing simply will contribute to the ease of processing.

The key message here is that designers and users of social media as well as government officials should be aware that perceptions or feelings like the ones examined here may be able to predict the spread of information on social media. Ideally, users of social media will remind themselves that misinformation abounds during disasters and that they need to turn on their analytic thinking mode and refrain from making quick decisions. However, slowing down and engaging in analytic thinking during stressful situations will not be easy (e.g., Ecker et al. 2010; Lewandowsky et al. 2012; Tanaka et al. 2013). Then, designers of social media should take advantage of people’s reliance on feelings to make snap decision. Designers may consider an option of introducing a disaster mode, in which feelings are primed to either curb or facilitate the spread of information. More work is needed to identify how to prime feelings and what the effect of such priming may be (see Chen & Sakamoto 2013; 2014).

4.3 Limitations and Future Directions

One limitation of the work reported here is the use of residents in the USA only. All of the subjects were far from the disaster center. Examining information-sharing behavior of people who are far from disaster center is important because messages from these people can actually save lives (cf. Inose 2011). However, the current results may or may not generalize to the behavior of people in the disaster center. Persons in a disaster center may perceive and react to information differently from those who are far from the disaster center (cf. Chen & Sakamoto 2013). In addition to the distance to the disaster center, differences in culture and socioeconomic status may play a role in shaping behavior. There are a few studies that have examined the information sharing behavior of Japanese residents (e.g., Tanaka et al. 2012; 2013). Conducting surveys and interviews to obtain more information about the respondents’ experiences and observations will be useful to gain deeper understanding of the psychology behind the sharing of disaster-related tweets.

Another limitation is the artificial nature of the work reported here. Some features of the real social media were missing. One such feature was the structure of the social network. In Twitter, for example, users follow other users, and this structure can no doubt play a role in sharing behavior (Aral & Walker 2011) although its role may be modest (Watts & Dodds 2007). For example, having many followers does not always translate to generating many re-tweets (Cha et al. 2010). The work reported here coupled with research examining the effect of social network structures on sharing behavior (e.g., Aral et al. 2009; Aral & Walker 2011, 2012; Cha et al. 2010; Fang et al. 2013; Huberman et al. 2009; Kwak et al. 2010; Leskovec et al. 2007) will be able to improve the understanding of the spread of information in social media.

Source of information and collective opinion were also features available in Twitter but missing in the study presented in the current paper. Past research suggests that these features may play important roles in information sharing. For example, Oh et al. (2010), who analyzed the source credibility of tweets related to the Haiti Earthquake, concluded that having a credible source could play an important role in reducing the anxiety level of Twitter users, which could in turn curb the spread of rumors. Research on social influence has found that people often use collective opinion, such as the total number of downloads (Salganik & Watts 2008) and the average rating (Sakamoto et al. 2009), when making their own decisions. It would be useful to further examine whether or not citing a credible source and whether or not exposure to re-tweet count could change people’s perceptions of

and decisions to share a disaster-related message in social media (see Li & Sakamoto 2013; Tanaka et al. 2014).

4.4 Concluding Thoughts

One challenge in using social media during disaster response is that the information that can help save lives is buried under a sea of other information and misinformation. Studies like the one reported here, which advance understanding of factors that influence people’s sharing of information on social media, can be fruitful for the development of techniques to prevent the spread of false information on social media and make it easier to find actionable information during disaster response. We believe that continued work in this area will contribute to our society and our future by making social media more effective for disaster management.

Acknowledgments This material is based upon work supported by the National Science Foundation under Grant No. IIS-1138658 and Grant No. BCS-1244742. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. The authors thank the reviewers for their thoughtful comments.

References Allport, G. W. and Postman, L. J. (1947). The psychology of rumor. New York, NY: Holt, Rinehart,

& Winston. Anthony, S. (1973). Anxiety and Rumor. The Journal of Social Psychology, 89, 91-98. Aral, S., Muchnik, L., and Sundararajan, A. (2009). Distinguishing influence-based contagion from

homophily-driven diffusion in dynamic networks. Proceedings of the National Academy of Sciences, 106, 21544-21549.

Aral, S. and Walker, D. (2011). Creating social contagion thought viral product design: A randomized trial of peer influence in networks. Management Science, 57, 1623-1639.

Aral, S. and Walker, D. (2012). Identifying influential and susceptible members of social networks. Science, 337, 337-341.

Asur, S. and Huberman, B. A. (2010). Predicting the future with social media, Proceedings of WIIAT. Bandari, R., Asur, S., and Huberman, B. A. (2012). The pulse of news in social media: Forecasting

popularity. Proceedings of the 6th International Conference on Weblogs and Social Media (ICWSM). AAAI.

Begg, I. M., Anas, A., and Farinacci, S. (1992). Dissociation of processes in belief: Source recollection, statement familiarity, and the illusion of truth. Journal of Experimental Psychology: General, 121, 446-458.

Berger, J., and Milkman, K. 2012. What makes online content viral?, Journal of Marketing Research, 49, 192-205.

Buhrmester, M., Kwang, T., and Gosling, S. D. (2011). Amazon’s Mechanical Turk: A new source of inexpensive, yet high-quality, data? Perspectives on Psychological Science, 6, 3-5.

Castillo, C., Mendoza, M., and Poblete, B. 2011. Information Credibility on Twitter. Proceedings of WWW 2011.

Cataldi, M., Di Caro, L., and Schifanella, C. (2010). Emerging topic detection on Twitter based on temporal and social terms evaluation. Proceedings of the 10th International Workshop on Multimedia Data Mining. pp. 1-10. ACM.

Cha, M., Haddadi, H., Benevenuto, F., and Gummadi, K.P. (2010). Measuring user influence in Twitter: The million follower fallacy. Proceedings of the 4th International Conference on Weblogs and Social Media (ICWSM). AAAI.

Chen, R., & Sakamoto, Y. (2013). Perspective matters: Sharing of crisis information in social media. HICSS-46.

Chen, R., & Sakamoto, Y. (2014). Feelings and Perspective matter: Sharing of crisis information in social media. HICSS-47.

Crump, M. J. C., McDonnell, J. V., and Gureckis, T. M. (2013). Evaluating Amazon’s Mechanical Turk as a tool for experimental behavioral research. PLoS ONE 8(3): e57410. doi:10.1371/journal.pone.0057410

Demirbas, M., Bayir A, M., Akcora G, C., Yilmaz, Y., and Ferbatosmangolu, H. (2010). Crowdsourced Sensing and Collaboration Using Twitter. 11th IEEE International Symposium on a World of Wireless. In WOWMOM Symposium.

DiFonzo, N., and Bordia, P. (2000). How top PR professionals handle hearsay: Corporate rumors, their effects, and strategies to manage them. Public Relations Review, 26, 173-190.

DiFonzo, N., and Bordia, P. (2007). Rumor psychology: Social and organizational approaches. American Psychological Association, Washington.

Doan, S., Vo Ho, B., and Collier, N. (2012). An analysis of Twitter messages in the 2011 Tohoku Earthquake. Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, 91, 58-66.

Ecker, U. K. H., Lewandowsky, S., & Tang, D. T. W. (2010). Explicit warnings reduce but do not eliminate the continued influence of misinformation. Memory & Cognition, 38, 1087-1100.

Fang, X., Hu, P. J.-H., Li, Z., & Tsai, W. (2013). Predicting adoption probabilities in social networks. Information Systems Research, 24, 128-145.

Ha, S. and Ahn, J. (2011). Why are you sharing others’ tweets?: The impact of argument quality and source credibility on information sharing behavior. Proceedings of the International Conference on Information Systems.

Hawkins, S. A. & Hoch, S. J. (1992). Low-involvement learning: Memory without evaluation. Journal of Consumer Research, 19, 212-225.

Hawkins, S. A., Hoch, S. J., & Meyers-Levy, J. (2001). Low-involvement learning: Repetition and coherence in familiarity and belief. Journal of Consumer Psychology, 11, 1-11.

Huberman, B., Romero, D., and Wu, F. (2009). Social networks that matter: Twitter under the microscope. First Monday, 14, 1-5.

Inose, N. (2011). How Tokyo responded to request for help on Twitter Retrieved on April 17th, 2013, from: http://www.nikkeibp.co.jp/article/column/20110314/263638/

Jaeger, M. E., Anthony, S., & Rosnow, R. L. (1980). Who hears what from whom and with what effect: A study of rumor. Personality and Social Psychology Bulletin, 6, 473-478.

Jansen, B. J., Zhang, M., Sobel, K., and Chowdury, A. (2009). Twitter power: Tweets as electronic word of mouth. Journal of the American Society for Information Science and Technology, 60, 2169-2188.

Kwak, H., Lee, C., Park, H., and Moon, S. (2010). What is Twitter, a social network or a news media? Proceedings of the 19th international conference on World Wide Web WWW’ (10), pp. 591.

Kwon et al. 2014 Leskovec, J., Adamic, L. A., & Huberman, B. A. (2007). The dynamics of viral marketing. ACM

Transactions on the Web, 1, 1-39. Lewandowsky, S., Ecker, U. K. H., Seifert, C., Schwarz, N. & Cook, J. (2012). Misinformation and

its correction: Continued influence and successful debiasing. Psychological Science in the Public Interest, 13, 106-131.

Li, H., & Sakamoto, Y. (2013) The influence of collective opinion on true-false judgment and information-sharing decision. Howe School of Technology Management Research Paper Series Number 2013-08 (http://ssrn.com/abstract=2210742).

Mason, W. and Suri, S. (2011). A Guide to Behavioral Experiments on Mechanical Turk. Behavior Research Methods, 44, 1-23.

Mason, W., and Watts, D. (2009). Financial incentives and the ‘performance of crowds.’ HCOMP ’09: Proceedings of the ACM SIGKDD Workshop on Human Computation, 77-85.

Ogiue, K. (2011). An examination of rumors about the Great East Japan Earthquake. Kobunsha, Tokyo.

Oh, O., Kwon, K. H., and Rao, H. R. (2010). An Exploration of Social Media in Extreme Events: Rumor Theory and Twitter during the Haiti Earthquake 2010. International Conference on Information Systems, MI, Saint Louis.

Oppenheimer, D. M. (2006). Consequences of erudite vernacular utilized irrespective of necessity: Problems with using long words needlessly. Applied Cognitive Psychology, 20, 139-156.

Paolacci, G., Chandler, J., and Ipeirotis, P. G. (2010). Running experiments on Amazon Mechanical Turk. Judgment and Decision Making, 5, 411-419.

Prasad, J. (1935). The psychology of rumor: A study relating to the great Indian earthquake of 1934. British Journal of Psychology: General Section, 26, 1-15.

Prasad, J. (1950). A comparative study of rumours and reports in earthquakes. British Journal of Psychology: General Section, 41, 129-144.

Reber, R., and Schwarz, N. (1999). Effects of perceptual fluency on judgments of truth. Consciousness and Cognition, 8, 338-342.

Rosnow, R. L. (1991). Inside rumor: A personal journey. American Psychologist, 46, 484-496. Rosnow, R. L. (2001). Rumor and gossip in interpersonal interaction and beyond: A social exchange

perspective. In R. M. Kowalski (Ed.), Behaving badly: Aversive behaviors in interpersonal relationships (pp. 203-232). Washington, DC: American Psychological Association.

Rosnow, R. L., Esposito, J. L., and Gibney, L. (1988). Factors influencing rumor spreading: Replication and extension. Language & Communication, 8, 29-42.

Sakaki, T., Okazaki, M., and Matsuo, Y. (2010). Earthquake shakes Twitter users: Real-time event detection by social sensors. Proceedings of the 19th International Conference on the World Wide Web. ACM.

Sakamoto, Y., Ma, J., & Nickerson, J. V. (2009). 2377 people like this article: The influence of other’s decisions on yours. In N. Taatgen, H. van Rijn, L. Schomaker, and J. Nerbonne (Eds.), Proceedings of the 31st Annual Meeting of the Cognitive Science Society.

Salganik, M. J., & Watts, D. J. (2008). Leading the herd astray: An experimental study of self-fulfilling prophecies in an artificial cultural market. Social Psychology Quarterly, 71, 338-355.

Sankaranarayanan, J., Samet, H., Teitler, B. E., Lieberman, M. D., and Sperling, J. (2009). Twitterstand: News in tweets. Proceedings of the 17th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems.

Schwarz, N., & Clore, G. L. (2007). Feelings and phenomenal experiences. E. T. Higgins & A. W. Kruglanski (Eds.), Social Psychology: Handbook of basic principles (2nd ed., pp. 385-407). New York: Guilford.

Shibutani, T. (1966). Improvised news: A sociological study of rumor. Indianapolis, IN: Bobbs-Merrill.

Sinha, D. (1952). Behaviour in a catastrophic situation: A psychological study of reports and rumors. British Journal of Psychology, 43, 200-209.

Song, H., and Schwarz, N. (2008). If it's hard to read, it's hard to do. Psychological Science, 19, 986-988.

Song, H., & Schwarz, N. (2009). If it's difficult to pronounce, it must be risky: Fluency, familiarity, and risk perception. Psychological Science, 20, 135-138.

Spiro E. S., Fitzhugh, S., Sutton, J., Pierski, N., Greczek, M., and Butts, C. T. (2012). Rumoring during extreme events: a case study of deepwater horizon 2010. Proceedings of the 3rd Annual ACM Web Science Conference (WebSci '12). ACM, New York, NY, USA.

Sriram, B., Fuhry, D., Demir, E., Ferhatosmanoglu, H., and Demirbas, M. (2010). Short text classification in twitter to improve information filtering. Proceeding of the 33rd international ACM SIGIR conference on Research and development in information retrieval.

Starbird et al. 2014 Sutton, J., Palen, L., & Shklovski. (2008). Backchannels on the front lines: Emergency uses of social

media in the 2007 Southern California Wildfire. Proceedings of the 5th International ISCRAM conference.

Tachiiri, K. An examination of the Great East Japan Earthquake: What did social media report? Discover Twenty One, Tokyo.

Tanaka, Y., Sakamoto, Y., & Matsuka, T. (2012). Transmission of rumor and criticism in Twitter after the Great Japan Earthquake. Annual Meeting of the Cognitive Science Society.

Tanaka, Y., Sakamoto, Y., & Matsuka, T. (2013). Toward a social-technological system that inactivates false rumors through the critical thinking of crowds. HICSS-46.