Tracing Twitter’s Impact on Ebola Response

How well does Twitter represent public perceptions and behavior in a public health crisis?

Researchers at Kent State University are conducting a National Science Foundation (NSF)-funded study of what social media activity on Twitter reveals about perceptions of last October’s Ebola scare on campus and how perceptions influenced behavior.

By comparing Twitter messages with responses from an online and a telephone survey of the Kent State community and people in the surrounding area, the researchers will better understand the role of social media in reflecting and influencing public perception and behavior in a crisis.

What the researchers learn will help inform public health officials in responding to future threats. The study is one of the first of its kind to evaluate the use of social media data in risk communication during public health events as they unfold.

While Twitter data are increasingly being mined by social scientists, the data’s validity as a research tool has not been firmly established. The Kent State study is designed to determine if behavioral responses align with social media comments. If so, it would validate the use of social media to predict behavior. 

Gleaning Data from 40 Million Tweets

The Kent State team is led by principal investigator Eric Shook, assistant professor of geography in the College of Arts and Sciences and director of the High Performance Computing and GIS Laboratory.

Shook, who has a computer science background, is gathering data from 40 million tweets, including those sent during the October 2014 Ebola scare. He will compare this information with 4,300 responses the researchers received in an October 2014 online survey of all Kent State faculty, students and staff.

The research team will conduct a telephone survey of 750 people in the surrounding region, asking them what they knew about Ebola, what risks they perceived and how they changed their behavior in response to news that a Dallas-based nurse who became infected after treating the first Ebola patient in the U.S. had visited with relatives in Ohio who work at Kent State.

The online survey found that people at Kent State sanitized and washed their hands more, avoided crowds and used less public transportation during the scare. 

Twitter usage spiked wildly in the Kent State area on Oct. 15, 2014, Shook said, as news spread about the nurse and Ebola. In Northeast Ohio, tweets about Amber Vinson, the nurse, who later recovered, far outnumbered those about Thomas Eric Duncan, the Ebola patient in Dallas, who later died.

Shook is using algorithms and high-performance computers to sift through the 40 million tweets he purchased from GNIP, a data reseller owned by Twitter. The researchers will examine what was re-tweeted, how often users returned to the subject, and what sources of information, whether valid or inaccurate, they consulted and passed along to others. 

Does a “Real-Time” Response Mirror Surveys?

Twitter data reflect a “real-time” response. Surveys, on the other hand, are by nature historical, occurring after an event, when people have had time to reflect, Shook noted. “Twitter gives us a way to look at the everyday lives of people,” he said.

If Twitter data mirror responses gathered in more traditional surveys, the data could prove to be a valuable research tool and a way to gauge public behavior. Twitter data could be used to build a model to predict how people will respond to an epidemic, for instance.

Shook noted that disease models are built on simplified and often incorrect assumptions. Influenza outbreaks, for example, may prove to be milder than predicted because people change their behavior from what is expected. They might stay home from work when they are not sick in order to avoid contamination, so the outbreak subsides instead of spreading as widely as disease models and health officials expect.

Social media data could provide a window on that type of unanticipated behavior to help health officials respond to a public health epidemic or crisis.

Collaborating on the research are Andrew Curtis, professor of geography and director of the GIS/Health and Hazards Laboratory; Jacqueline Curtis, assistant professor of geography and associate director of the GIS/Health and Hazards Lab; Gregory Gibson, assistant professor of sociology and director of the Survey Research Institute; Anthony Vander Horst, assistant professor of sociology; and Christopher J. Woolverton, professor of Environmental Health Sciences and director of the Center for Public Health Preparedness in the College of Public Health. 

See more (video)