Social science research using Twitter to gather attitudinal and behavioral data must account for bots and trolls but can still render meaningful results, says David Broniatowski, PhD, FPsyS, assistant professor of systems engineering at George Washington University in Washington, D.C.

To help researchers accomplish this, Broniatowski and colleagues published a guide to identify and deal with the assorted “malicious actors,” including bots, spambots, content polluters, fake followers, and human trolls.1,2

In previous research on the safety of vaccines, Broniatowski found that social media has become a battlefield of divisive rhetoric, some of it posted just to create the illusion of a “debate” and keep people divided. To establish this false equivalency, some tweets from the same sources have posted both negative and positive messages about vaccines, he found.2

Given the potential to undermine the data collected, should social science researchers simply avoid Twitter? “No, I don’t think that is necessary,” he tells IRB Advisor. “Twitter is still a rather useful tool to assess attitudes and behaviors — with the caveat that anything that you do on social media is subject to any of the distortions that come with the medium. I think every medium has its own possibilities for distortion.”

Twitter in particular can amplify misinformation within its open platform.

“In a sense, it is like doing social science research in a crowd and someone is holding a megaphone,” he says. “That doesn’t mean you can’t ask everybody else their opinions. You can look at their opinions and get useful information. But the person with the megaphone is going to have an outsized influence, and you want to try to control for that.”

Typical bots that imitate human messages are more likely to attack Twitter, while spambots linked to advertising show up on media like Facebook, he explains.

“Depending on how you are doing your analysis, you may not be able to completely remove the effects of Twitter bots,” he says. “But it’s not as if these bots in and of themselves exist in a vacuum. In many cases, they may be retweeting messages that were generated by human beings or vice versa.”

As these trends continue, more IRBs may become directly involved in providing oversight for research using social media.

“We have thought a lot about that,” he says. “The real question there comes down to who are the at-risk populations? When those populations are at risk, what can be done to mitigate that consistent with IRB laws?”

For example, he cites a recent paper written by a colleague that found that quoting Twitter messages verbatim in research could result in deidentification that could violate the privacy of those quoted.

“It’s often possible to reverse-identify the accounts that generate those tweets,” he says. “That is problematic. One of the things they propose is in research involving Twitter data, researchers should consider paraphrasing tweets rather than using them verbatim.”

In the study3 in question, researchers reviewing a series of research articles using Twitter found that 72% quote at least one participant’s tweet. Searching for the quoted content led to an identified participant 84% of the time. Moreover, 21% of the articles made a participant immediately identifiable by citing their Twitter username.

“Only one article reported obtaining consent to disclose identifying information, and institutional review board involvement was mentioned in only 40% of articles, of which 17% received IRB-approval and 23% were deemed exempt,” the authors concluded.

The authors recommended that researchers aggregate findings to protect participant identities, noting that editors should reject papers that refuse to do so. IRBs should be vigilant for these issues in overseeing research that includes social media.

“It is imperative that we protect participant privacy even in social media studies,” the authors noted. “First, privacy settings are set by the account owner who may post sensitive information and then later delete or make their post private. There are documented cases of people compromising their job, college admission, or relationships when their postings were rebroadcast on other media channels.”

REFERENCES

1. Jamison AM, Broniatowski DA, Quinn SC. Malicious Actors on Twitter: A Guide for Public Health Researchers. Am J Public Health 109, 1 May 2019;doi:10.2105/AJPH.2019.304969.

2. Broniatowski DA, Jamison AM, Qi S, et al. Weaponized Health Communication: Twitter Bots and Russian Trolls Amplify the Vaccine Debate. Am J Public Health. 2018;108(10):1378–1384. doi:10.2105/AJPH.2018.304567.

3. Ayers JW, Caputi TL, Nebeker C, et al. Don’t quote me: reverse identification of research participants in social media studies. npj Digital Medicine 2018;(1):30: doi:10.1038/s41746-018-0036-2