Anger and bots on Twitter are not affecting vaccinations in France

Misinformation is not preventing French people from getting COVID-vaccinated

Dear reader,
in this issue, I try to understand the effect of Twitter on the French vaccination campaign. It is true: there is a lot of misinformation on social media. Moreover, there is a chance that there is a misinformation campaign in progress. But, at least in France, anti-COVID jabs are proceeding well.

Is the media overestimating the campaign against the Green Pass? A sentiment analysis, data from the French vaccination campaign, and Twitter data issue a grim verdict for those who think that social media alone can manipulate society. 

The following analysis considers only 18,000 tweets posted between 9.35 and 12.15 of July 23, 2021. The number of tweets is the limit Twitter allows to download from its free API. The level of mistrust in Macron’s proposal about making the Green Pass compulsory for restaurants and museums, among others, is overwhelming reading the following chart. 

The chart sums the sentiment of every tweet in the conversation. I used an algorithm called nrc. This algorithm scores each word in a text by anger, fear, joy, etc. Scores are always integer numbers.


In the chart, we see that more than 50 per cent of the sentiment has to do with anger, fear, or sadness. If one had to trust this chart, one would imagine France being rocked by the protest against Marcon’s idea of forcing people to get vaccinated or renounce going to the restaurant. Yet, this mass of tweets DaNumbers collected proves that the protests failed to gain traction.  

The chart above is slightly more complicated than usual. The problem here is not the chart itself (it is just a scatterplot after all) but the method used to produce it: on the vertical axis, there is the number of retweets (nothing new: everyone knows what a retweet is). Points are single tweets. On the horizontal axis, I performed a slightly more complicated analysis.

Share DaNumbers

Using a statistical technique called Factor analysis, DaNumbers discovered that sadness, anger, disgust, fear, and surprise have a high chance of being together in the same tweets. That allowed DaNumbers to generate a single variable called Anti Green Pass sentiment (here for the code to replicate the results). The next question now was: Can we use the Anti Green Pass sentiment to predict the number of retweets? The flat red line shows we cannot. Besides that, there is little indication the sentiment affected life outside Twitter: After president Macron’s announcement, the speed of vaccinations improved in France.

The campaign peaked four days after Macron announced the extension of the Green Pass. The 7-day rolling average grew substantially, reaching more than 688,000 doses on July 21, the chart shows. People protesting were in the order of 160,000 last Monday, according to some media.

Although policymakers fear that the vaccination campaign will plateau tank to so-called vaccine-hesitant people, the numbers tell another story. France is well beyond 50 per cent of its citizens fully vaccinated. Monday morning, European Commission president Ursula von der Leyen announced that 70 per cent of EU citizens received at least one dose of COVID vaccine. So, what is the fuss?

Maybe, media and policymakers spend too much time on Twitter and give it a role it does not have. For starters, 92.8 per cent of the examined tweets is a retweet. So, the campaign against the Green Pass in France is mostly a sequence of non-original content. Profiles retweet massively and, as far as the sentiment analysis is concerned, the original tweet with the most anger is this

In this tweet, there is an English commentator (a self-proclaimed Marxist who supported Brexit). The guy issues a heated opinion against Macron’s Green Pass extension mixing it with anti-EU rhetoric on Sky News Australia.

How this video ended up on French Twitter would deserve a story on its own. Yet, the important thing now is to use this anger to spot the secrets of these 18,000 tweets. Are people angry, or are we witnessing another bot-led vast misinformation campaign? The following charts will help to answer. 

Here, DaNumbers collected the top-10 angriest profiles among those who contributed to the the18,000 tweets. The numbers are low because of the thousands of profiles that took part in the conversation. A short analysis of this peculiar top-10 shows that they are, at least, anonymous users. How high is the chance that they are not human? The following chart will answer. 

This chart shows how likely it is that those Twitter profiles are, likely, bots. Most of them are above the 70 per cent threshold. These numbers come from the Tweetbotornot R package developed by Missouri University’s Mike Kearney. According to the software, the chance that the Twitter profile of this writer is a bot is 65 per cent, whereas R Tweet bot's chances to be a bot are at 95 per cent. 

The analysis is limited because Twitter’s free API has limits. Tweetbotornot takes 100 tweets from a Twitter profile and assesses them. The limit for this kind of operation is 1,600 requests every 15 minutes, so DaNumbers decided to limit the number of requests to 1,000. Despite limitations, there is a high chance that bots are polluting the French Twittersphere. 

Mainstream media are scrambling to assess the size of the anti-vax movement and of those who do not want restrictions against those who are not COVID-vaccinated. Is there a misinformation campaign targeting Europe? If there is, it is failing. Although there is a lot of attention on the matter, data show that the focus on social media is too much: people are taking their vaccines no matter what they read on Twitter. And that is good news, even if Twitter is failing in keeping its platform bot-free.

Click here for the data and the code.

Thanks for your attention.