This is a second report by the HEROIC project team on the use of Twitter by official accounts responding to events in Boston during the week of April 15-19, 2013. In our previous report, we provided background on Twitter use during the bombing event, as well as initial findings related to allocation of attention by members of the public to official accounts. In this report, we discuss the use of relational and conversational features that may have been included as part of each posted tweet – we call these features “microstructure” elements. By looking at tweet microstructure, we can learn about when users share information with each other, and with whom they choose to share it.
Conversational “microstructure” elements are aspects of tweet content and style that have been developed as conventions of Twitter use. These include features such as hashtags, directing messages to other specific users, and retweeting others’ messages. These features also signify varying dimensions of online engagement. For instance, the use of a hashtag allows an individual to identify and follow a stream of tweets that are classified under a particular event; directing content to other users means direct engagement between users, such as an official organization and a follower; and retweeting content helps to amplify a message across additional networks. Aspects of conversational microstructure use are of interest because they provide insights into which Twitter messages are amplified and why.
We start by looking at the mean proportion of messages containing each microstructure element, based upon the level of government in which they operate – the organizational sphere. We group our 29 targeted accounts by local, state, or federal level, and average the number of messages that contain each microstructure feature of interest. We also show variation in the use of these elements across the different organizational sectors.
Figure 4. The mean proportion of messages containing conversational microstructure elements at the level of local, state, or federal organizational operations. We also show the 95% confidence intervals for these proportions in order to demonstrate variation in the proportion across organizations.
In Figure 4 we see that the most consistently included microstructure element is a URL, providing a link to additional information. This ranges from nearly 80% of messages posted by federal accounts to 60% of messages posted by local and state accounts. The second most consistently included microstructure element is the use of a hashtag. Local accounts include this element more consistently than state or federal accounts. The least used microstructure feature is the directed message, indicating a message is targeted to a specific user, most likely in reply to a specific question or comment. While local accounts did so on a most consistent basis, it was infrequent. Importantly, we see that most of the tweets with “original content” (non-retweets) come from local and state organizations. Nearly 40% of federal tweets are retweeted messages, signifying a role of reinforcing or echoing local messages in a disaster response.
Previous research (e.g. see Spiro et al. 2012 and Spiro et al. 2012) has shown that the patterns identified in the use of microstructure elements are consistent. We expect that local organizations will have more directed communications to individual followers and produce more original content than organizations that are outside of the geographical area of impact. Furthermore, we expect to see that state and federal organizations will serve to amplify messages from local organizations by retweeting original content to a broader public following. An unexpected pattern, however, is the use of the hashtag for this week of events. While there were a series of events throughout the week, including the detonation of improvised explosive devices at the beginning of the week, the killing of a police officer at MIT, and the lockdowns of Boston and Watertown, there was no indication that a consistent hashtag emerged or trended among official organizations to organize their content into a traceable stream. Hashtags that were utilized varied by sector, such as #tweetfromthebeat, #WANTED, and #CommunityAlert by law enforcement, and #oneboston from local government, indicating different aspects of the response. However, a single hashtag, related to the weeklong investigation and subsequent manhunt and capture, did not emerge.
We are also interested in those messages that received the most retweets by the public. This retweeting activity amplifies messages across broader networks and extends the reach of a single message to additional populations. In Figure 5 we show the percentage of tweets that received more than 450 retweets after they were posted by one of our targeted accounts in each of seven response sectors (crisis response, health, environment, law enforcement, government/elected official, transportation, and other).
Figure 5. Percentage of tweets with more than 450 retweets according to organizational sector.
Notably, the sector that produced the most messages in this high retweet category is law enforcement, followed by government or elected officials. A single tweet posted by the Boston_Police on the evening of April 19, 2013 had more than 137,000 retweets:
Other highly retweeted tweets include the following:
“Suspect in custody. Officers sweeping the area. Stand by for further info.” (Boston_Police, 77,376 retweets)
“In our time of rejoicing, let us not forget the families of Martin Richard, Lingzi Lu, Krystle Campbell and Officer Sean Collier” (Boston_Police, 59,165 retweets)
“#MediaAlert:WARNING: Do Not Compromise Officer Safety by Broadcasting Tactical Positions of Homes Being Searched” (Boston_Police, 20,150 retweets).
In contrast, the tweet that received the most retweets at the beginning of the week, on April 15, provides an update of the hazard impact in terms of deaths and injuries as the event was unfolding. It was posted by Boston_Police and mentions Boston Police PIO Cheryl Fiandaca. It also includes a hashtag consistently used by policing agencies across the U.S.
“Update 23 injuries 2 dead #tweetfromthebeat via @cherylfiandaca” (Boston_Police, 7900 retweets)
The difference between the retweet volume at the beginning of the week and the end of the week may be an artifact of the fact that most of the city was on lockdown at the end of the week, and local members of that population had more time on their hands. It may also indicate that there was an increased level of attention directed towards the activities in Boston and greater levels of activity across the twittersphere as the manhunt unfolded over the last days.
In Figure 6 below, we show dynamics of posting activity for the organizational accounts that had the highest activity levels (most tweets) throughout the weeklong series of events. The size the the dot indicates the number of retweets for each individual tweet. The greatest volume of tweets appears at the end of the week, April 19th, followed by the beginning of the week, April 15th. We note that law enforcement again has the highest volume of tweets as well as retweet counts, with a noticeable increase on April 18, at the time of the shootout at MIT. We also find posting behavior shows time of day seasonality effects. In Figure 6 this is seen in the alternating pattern of activity followed by no activity (blank intervals where no tweets are posted).
Figure 6. Top 10 accounts with the greatest amount of tweeting activity over the week.
Features of conversational microstructure allow us to learn about the differences between organizational roles in communicating about a disaster and its effects. The sphere (local, state, or federal level) and the sector (law enforcement, response, government, etc) appear to function differently from one another in this particular series of events. This is consistent with previous research conducted by HEROIC, but differs upon the basis of the hazard type.
The findings in this second research report re-emphasize that “all disaster is local,” and supports the idea that hazard type will influence who leads public information sharing efforts. With this in mind, it becomes even more important that interorganizational network connections be made prior to an event so that organizations in a supporting role can amplify the messages of a local organization. In addition, organizations should be prepared to communicate across the Twittersphere, knowing that online attention will increase as events intensify and there is greater opportunity for people to search for and confirm information.
This material is based on research supported by the National Science Foundation under awards CMMI-1031853 and CMMI-1031779 and by the Office of Naval Research under award N00014-08-1-1015.
The above material is based on analysis completed by HEROIC team members. Please cite as follows.
Sutton, J., Spiro, E., Johnson, B., Fitzhugh, S., and Butts, C. (2013). “Tweeting Boston: The Influence of Microstructure in Broadcasting Messages through Twitter.” Online Research Highlight. http://heroicproject.org