Antivaccine activity on Twitter: It’s not entirely what you think

Writing about the antivaccine demonstration at the California State Capitol in Sacramento on Wednesday in opposition to #SB 276, the bill co-sponsored by Sen. Richard Pan to tighten up state regulation of doctors writing letters supporting medical exemptions to school vaccine mandates, I started thinking about the whole social media apparatus that supports the antivaccine movement. I remember in essence mocking the first fumbling forays of antivaxers into Twitter in the wake of the birth of the “CDC whistleblower” conspiracy theory in 2014. By 2017, fake news and Twitter bots were running rampant with antivaccine misinformation, and, no surprise here, the antivaccine movement had made major inroads into Facebook. By 2018, there were suggestions that Russian bots might have started sowing discord and misinformation on Twitter as well. Obviously, Twitter and Facebook groups, both private and secret, had provided an excellent platform for the “V is for Vaccine” protesters to organize and then publicize their demonstration. That made a recent study forwarded to me by a reader of interest. Basically, it’s a study looking at the changes in antivaccine hashtags on Twitter since 2010.

The study, published in Vaccine two weeks ago, is entitled Temporal trends in anti-vaccine discourse on Twitter is by Keith Gunaratne, Eric A.Coomes, and Hourmazd Haghbayana, comes from the University of Toronto, Western University in London, Ontario, and Université Laval in Quebec.

The authors note:

Previous studies have confirmed that anti-vaccine content figures prominently across social media platforms, with criticisms encompassing themes of safety, personal freedom, and pharmaceutical and medical skepticism [7], [9]. As many as 41% of parents report exposure to anti-vaccination content on social media [2]. Further, individuals exposed to such content may be more likely to spread anti-vaccine messages, cyclically propagating misinformation [10]. In response to the effects of online anti-vaccine messaging, Dr. James Madara, the chief executive officer of the American Medical Association, formally published a public letter urging leading technology companies, including Twitter and Facebook, to ensure access to accurate information on the safety and efficacy of vaccinations [11]. Given this widespread concern regarding social media’s role in propagating vaccine hesitancy, we sought to characterize the temporal trends in pro- and anti-vaccination discussion on Twitter, as reflected by hashtags, and to determine the extent of inter-communication between these communities.

To accomplish this investigation of temporal trends, the authors searched Twitter for all Tweets containing the English-language index term “vaccine” dating from between 01/01/2010 and 01/01/2019. Next, a systematic programmatic search was utilized to interrogate Twitter’s Advanced Search front-end using a modified version of the Python script to identify all existing Tweets containing the search term “vaccine.” For each Tweet, the author, unique identifier, text, timestamp, contained hashtags, and reply/re-tweet status were acquired.

Then:

Hashtags were extracted from all tweets containing the term “vaccine.” Hashtags present in >100 tweets were reviewed by two authors, with a third arbitrating conflicts: after excluding non-vaccine-relevant or ambiguous hashtags, the remaining vaccine-related hashtags were labelled as pro- (score = 1), or anti-vaccination (score = −1) by manual review of associated tweets. Example hashtags by category include: non-relevant (#coffee, #world), pro-vaccination (#vaccineswork, #vaccinessavelives), and anti-vaccination (#cdcwhistleblower, #vaxxed). No hashtags or tweets were specifically excluded because of non-English language.

And finally:

Twitter.com was re-searched on March 3, 2019, via the Python script for all pro- and anti-vaccine hashtags to create a final dataset of vaccine-related tweets. To validate manual classification and identify subcommunities, Louvain community detection via NetworkX2.2 was employed to perform network analysis [12]. Hashtags were represented as nodes on an undirected graph. Each pair of hashtags that appeared in at least one message together was connected by a weighted edge, with the edge weight equal to the number of unique users that tweeted the pair of hashtags together. Hashtag importance was calculated using the Eigenvector centrality module (also via NetworkX2.2), a quantitative estimate of the influence of each hashtag node within the network [13]. In this model, connections to high-scoring nodes have a stronger influence on the score of an individual node than connections to low-scoring nodes.

Using techniques like this, it is possible to produce a visual representation of the Twitter networks discussing vaccines and how they’ve changed over the last nine years. The result of the search was the identification of 10,043,087 tweets with 3,843 unique hashtags, which the authors classified into non-vaccine-relevant (n = 3382), ambiguous (n = 164), pro-vaccine (n = 154) or anti-vaccine (n = 125) categories. The authors also report that a second search for all 279 pro- and anti-vaccine hashtags found 1,637,712 vaccine-related tweets. Mathematical clustering analysis initially identified nine hashtag-based communities. The authors report:

Anti-vaccine hashtags largely coalesced into one community (119 hashtags; central hashtags: #cdcwhistleblower and #vaxxed) with a small remote secondary community regarding the Philippine dengue-vaccine scandal [15]. Pro-vaccine hashtags segregated into one dominant community (113 hashtags; central: #vaccineswork) and several closely-linked secondary communities including opposition to the Australian Vaccination-skeptics Network (six hashtags, central: #stopavn) [16] and multiple others focused on disease-prevention: influenza (24 hashtags; central: #fightflu), polio (four hashtags; central: #endpolio), hepatitis B (two hashtags; central: #nohep) and HIV (three hashtags; central: #hivvaccineawarenessday). The most frequently used anti-vaccine hashtags were: #cdcwhistleblower (280,779 tweets), #vaxxed (123,382 tweets), #hearthiswell (44,426 tweets), #novax (32,424 tweets), and #cdcfraud (21,750 tweets).

Any of you who who are also active on Twitter will recognize all of these antivaccine hashtags, particularly #cdcwhistleblower, #vaxxed, and #hearthiswell. Antivaxers like to festoon their Tweets with these hashtags as though they were Christmas trees and the hashtags are ornaments.

Here’s the visual representation of these networks:

Twitter figure 1

One interesting additional finding of this study is that before 2014 there were relatively few antivaccine Tweets (and pro-vaccine Tweets, truth be told). For example, the authors report that there were 215 pro-vaccine tweets in the first quarter of 2010, which increased by 1,670 Tweets/quarter (95% confidence interval: 1370–1970 Tweets/quarter, r2 = 0.790) reaching a peak of 73,200 Tweets during the last quarter of 2018. In contrast, the authors report that a median of 906 anti-vaccine Tweets/quarter (IQR 583–1108 Tweets, maximum 6,871 tweets) were observed until the third quarter of 2014, during which there was a surge of 57,845 Tweets; subsequently anti-vaccine-tagged tweets decreased by 2,670 Tweets/quarter.

Here’s the graph:

Twitter figure 2

Notice anything? Notice that big peak in antivaccine activity beginning in the latter half of 2014 and continuing to the middle of 2015? I wrote about it before. This is when antivaxers surged onto Twitter in the wake of the creation of the “CDC whistleblower” conspiracy theory in August 2014. It also encompasses the period of time of the Disneyland measles outbreak after Christmas 2014 and the roughly half-year period of time during which SB 277 was being debated in California. Remember, if you will, that SB 277 was the California bill that ultimately became law to eliminate nonmedical “personal belief exemptions” in California. You’ll then notice that, after the bill passed in the summer of 2015, antivaccine Twitter activity fell rapidly and has been continuing to decline slowly ever since around 2016.

It’s also counterintuitive. After all, to anyone who’s active on Twitter, as I am, it sure “feels” as though antivaccine activity on Twitter has been increasing and is at its highest levels ever. If this study is accurate, that’s simply not true. Of course, one problem with this study is that it relies solely on hashtags to categorize Tweets as pro- or anti-vaccine. A lot of Twitter users don’t even use hashtags very often. I know I don’t. My Tweets would definitely be categorized as pro-vaccine, but this study would only capture the small percentage of them in which I actually used hashtags. It’s also true that I’ve encountered a lot of antivaxers who don’t often use hashtags, although my admittedly anecdotal experience is that antivaxers like to use a lot of Twitter hashtags.

So what does it all mean? Let’s see what the authors say:

Conversely, while few anti-vaccination tweets were posted prior to 2014, a significant surge in anti-vaccine discussion occurred between 2015 and 2016. This period coincides with the 2014–2015 measles outbreak [17], the publication of Vaccine Whistleblower [18] – an anti-vaccine book, linked to #cdcwhistleblower, and the release of Andrew Wakefield’s anti-vaccine film Vaxxed, linked to #vaxxed [19]. While the resultant volumes of anti-vaccination tweets were not sustained, the anti-vaccination userbase has doubled since 2015. There is minimal inter-communication between communities, with only 0.2% of users engaging across networks. Previous investigation of Facebook identified similar segregation of users into pro- and anti-vaccine communities [9]. Such isolation of the anti-vaccination minority may limit the penetration of evidence-based knowledge translation campaigns on social media, permitting ideologic echo chambers to perpetuate.

And:

This study characterizes the evolution of both pro- and anti-vaccine discourse on Twitter, revealing significant increases in the volume of pro-vaccination tweets coupled to decreases in anti-vaccine discourse since the latter reached a peak in 2014–2015. Despite the greater volumes of pro-vaccination discourse in recent years, and the userbase contributing anti-vaccination content being smaller, the anti-vaccine community continues to grow in size. This finding coupled with the minimal inter-communication between communities suggests possible ideological isolation. Further studies are needed to investigate how Twitter may be used to effectively disseminate accurate vaccine information and the real impact of such discourse on downstream vaccinations.

And this “feels” largely true, at last to me as someone who’s on Twitter a fair amount and observes antivaccine. There is a much more robust pro-vaccine community on this platform compared to a few years ago, but it is isolated from the antivaccine community. Most of the interaction I observe between the two communities consist of attacks by antivaxers on pro-vaccine advocates (the “pharma shill gambit” is a particular favorite, naturally), ironic subtweeting, and attempts by vaccine advocates to refute the copious antivaccine misinformation being spread by antivaxers.

Of course, as I’ve said many times, you’re not going to change the minds of antivaxers on Twitter, and that’s not what I’m about when I go on Twitter. Besides my own personal entertainment, there, as here on the blog, I’m about deconstructing antivaccine misinformation for the vaccine-hesitant to see. They’re the ones who might be reached, not the antivaxers Tweeting under the #cdcwhistleblower hastag. The good news is that this study implies that Twitter might not be as effective an amplifier of antivaccine messages as antivaxers had hoped that it would be.