Tweeting Insights: Université de Montréal Researchers Analyze Public Response to Health Measures via Social Media
Did the extent to which people used social networks – especially Twitter – really reflect how much they supported – or not – the public-health measures put in place during the COVID-19 pandemic?
Quebec health experts Hélène Carabin and José Denis-Robichaud, helped by Erin Rees, who specializes in risk assessment at the Public Health Agency of Canada, went looking for answers to that question.
They analyzed more than 40,000 responses to 27 polls that the Angus Reid agency conducted in Canada between September 2020 and March 2022.
“We wanted to find out how much people used Twitter [now X] and what impact it had on buy-in to the health measures,” said Carabin, a professor in the pathology and microbiology department of UdeM’s Faculty of Veterinary Medicine, as well as in the Department of Social and Preventive Medicine in UdeM’s School of Public Health.
“We did that by looking at whether they talked about the pandemic on Twitter and, if so, what were the tone and substance of their comments about the public-health measures. At the same time, we wanted to analyze their behaviour and whether there was a relationship between their conduct and their use of social media.”
Only 1 in 5 used Twitter
Twitter users are a minority in Canada: only 20.6 per cent of Canadians used the platform during the period under study, and of these, fewer than one third (29.9 per cent) tweeted about COVID-19.
When it comes to behaviour, only 11 per cent of respondents to the Angus Reid polls said they were not following the mask rules and 10.8 per cent were not getting vaccinated. Twitter users were less likely to wear masks and get vaccinated than non-users, and those who criticized the health measures in their tweets had the lowest compliance rates of all.
The study also found regional and linguistic differences.
“There were fewer Twitter users in Quebec than elsewhere,” said Denis-Robichaud, who is an independent consultant in veterinary preventative medicine and epidemiology. “But when francophones posted angry tweets about health measures, it was more likely to be reflected in their behaviour than was the case for anglophones.”
Among francophones, Twitter users who denounced the health mandates were seven times more likely than non-Twitter users not to wear a mask, whereas among anglophones the ratio was four times.
Carabin, Denis-Robichaud and their research team are the first to show a correlation at the individual level between people’s compliance and non-compliance with health measures and their tweets.
Despite their findings, the researchers say it’s too early to accurately interpret their results.
“The sociodemographic characteristics of social media users are not representative of the general population,” warned Carabin. “So we have to be cautious about using tweets to assess behaviour at the population level.”
New lines of investigation
The study opens up new lines of investigation for the researchers, who are already planning the next phase.
They want to see if, in the event of a new pandemic, it would be possible to use Twitter or other social networks to estimate the rate of compliance with public-health measures based on whether and how people talk about them.”
They are now working on improving their tweet analysis tools. In July, at the ADELF-EPITER international conference for French-speaking epidemiologists in Limoges, France, Denis-Robichaud presented the preliminary results of a research project he’s doing.
It aims to evaluate the ability of an algorithm to read and classify messages posted on Twitter according to the tone of the tweets, and to compare the results with a classification by non-expert humans.
“The algorithm was good at classifying the types of public-health measures discussed in the tweets but not at reading the tone or the sentiment expressed,” Denis-Robichaud said. “But recently, we used ChatGPT 4.0, which proved to be almost as good as humans at classifying messages by tone.”
The next step is to use these improved tools to analyze the impact of the Twittersphere on the actual effectiveness of public-health measures.
“If health mandates are strict and people don’t want to follow them, we could see if that affects the disease transmission rate,” said Carabin.
“This could be important for public health because it would help predict the transmission rate, and it could be used to develop new strategies for improving public acceptance and reducing transmission.”