Big Data Shows “Le Pen And Mélenchon Will Qualify For Presidential Second Round Run-Off,” French Academic Says.
Antoine Bevort is an Emeritus Professor in Sociology at the CNAM (National Conservatory of Arts and Crafts, a famous and prestigious doctoral degree-granting elite school in France). He is a regular contributor to The Conversation and is the author of four books on work and social issues.
For months now, he has been studying the links between the big data from social media during recent election campaigns and their impressive accuracy when it comes to predicting election results, and how we should learn from it. Professor Bevort’s big data method indeed accurately predicted the results of the Austrian Presidential election last year, as well as the surprising results in the French primaries of the Centre-Right party Les Républicains and the Left-Wing Parti Socialiste.
In a few days’ time, eleven candidates will become only two. For months now, Far-Right candidate Marine Le Pen has been tipped by most opinion polls to be one of the two finalists in the second round of the election. After Centre-Right François Fillon’s endless list of affairs for which he has since been placed under official investigation, it’s the Independent Centrist Emmanuel Macron who was meant to fill the second vacancy. But, for the last week or so, it is Far-Left Jean-Luc Mélenchon who seems to be on the rise in the polls — something that had been spotted by Professor Bevort and his method long before the media even started to get all excited about it.
ON OPINION POLLS VS SOCIAL MEDIA DATA.
Last month, an article titled“Is Emmanuel Macron The Wrong Candidate To Beat Marine Le Pen?”published in the weekly politics magazine PoliticsMeansPolitics.com, raised the possibility that opinion polls may have had it all wrong again, alike for the Brexit vote and the Donald Trump win. The article claimed that Mr Macron might actually fail to qualify for the second round of the Presidential election, according to the social media engagement data analysis provided by Professor Bevort.
Do you think the methods used by pollsters nowadays are still accurate enough to predict election results?
“I think pollsters pain more and more to understand the development of public opinion. In fact, the difficulties the pollsters have to predict the results are of the same order as the difficulties the representative institutions (political parties, political institutions, trade unions, media, etc.) have to represent the society. The underlying assumptions and patterns of analysis are inadequate.”
What is the difference of approach between opinion polls and social media data?
“Studies about social media have neither the same object nor the same methodology that opinion polls have. First, they do not work with representative samples, and second, they do not measure voting intentions. Social media data are rather exhaustive and comprise both the politicians’ communication and the way internet users receive it. Indirectly, they measure the political preferences of voters in a less biased way than pollsters do. Even if all voters do not have a social network account, a majority of French people have one, and half of the French check Facebook daily, for example.”
Should we read polls and online data together, then?
“The two approaches should be combined. They (opinion polls and social media monitoring) are two complementary tools. The survey methodology asks at least as many questions as the analysis of social media data. But the polls are well established and continue to serve as the almost exclusive reference in the monitoring of political opinion, despite some major examples of inability to read that very opinion. There is also, most generally, an assumptive criticism against social media data studies. It is, nevertheless, an approach that needs to be examined and improved. It is a promising method.”
“It is an approach that needs to be examined and improved. It is a promising method.”
ON THE MEDIA’S SCEPTICISM OVER THE BIG DATA.
In a post published on his blog last week, Professor Bevort explained that the French Media outlets have decided to dismiss the online data because of the opinion of one social media expert, Nicolas Vanderbiest, who claimed that “the social web is not a good indicator for evaluating the power of a candidate at the polls.”
In an article he wrote for RTBF.be, Mr Vanderbiest thus ironically explained: “The new craze is to see what the future holds through the social networks. Where the true people talk. Far from the ivory tower (of the media) or pollsters, incapable of predicting Brexit or the glorious win of Trump in the United States.” According to the social media specialist, “we end up with a discourse where social networks would be more reliable than opinion polls, whereas the indicator is attesting it is directly influenced by the polls. A beautiful snake biting its tail.” His conclusion was that social networks can indeed provide us with clues about the strength of a candidate online, but in his opinion, the big data “will never be a way to determine for sure the results of an election.”
You don’t agree with Nicolas Vanderbiest. Why do you think the media don’t include big data analysis to their opinion polls analysis?
“I have noticed that the media and some social sciences researchers have concerns about the use of social media data to analyse the changes in political opinion or even treat it with scepticism. Newspapers and pollsters are confused by these new communication media, of which they had not anticipated the importance and which doesn’t fit in their usual methods of analysis. These concerns are also fuelled by a distrust of the social media that supposedly convey fake news, as if propaganda and fake news were born with the social media.”
“Newspapers and pollsters are confused by these new communication media, of which they had not anticipated the importance and which doesn’t fit in their usual methods of analysis.”
ON THE DATA COLLECTION METHOD.
Professor Bevort explained that during the campaign for the French Centre-Right primaries that saw François Fillon win the nomination for the Presidential election, the social media data “confirmed Mr Fillon’s breakthrough and ultimately his victory in the first round”. Several newspapers took an opposing view, such as Le Parisien. The daily newspaper judged that the data from social media were “bad thermometers of public opinion.” Another paper, La Croix, “developed the same analysis,” the Professor in Sociology added.
How and where do you collect the social media data?
“To analyse the social media data, I use the online data provided by a programme called Semrush. The company updates, on a daily basis, the data relating to Facebook, Twitter, Instagram, Google + and YouTube, which is collected straight from the social networks themselves.”
How does it all work? What data are you looking for?
“Communication on social media is measured in three ways: audience, activity and engagement. Audience measures the total number of followers, which are subscribers acquired on a given date, and it depends mainly on the past. Political power generates many ‘followers’. The limitation of audience data is its enormous stock effect and static nature. In contrast, activity reports the number of posts, tweets and videos. It measures the communication activity of a politician over a given period. Finally, engagement evaluates the number of ‘likes’, ‘shares’, ‘retweets’ and comments for the same period. Because it measures the impact of political communication, engagement is the most significant indicator.”
So, in your opinion, engagement data is the most accurate data out there?
“Based on the audience alone, many thought — among which the quoted newspapers — that social media had not anticipated Mr Fillon’s victory. But looking at the engagement allowed to foresee Mr Fillon’s success, and later Mr Hamon’s success(in the Socialist Party’s own primaries).”
Which social networks provide you with the best data?
“The originality of my method is that I mostly use the data from Facebook and Twitter, and secondarily the data related to YouTube and Instagram. Thanks to their high numbers of daily users, the combination of data from Facebook and Twitter allows a good understanding of the changes in political opinion. Most of the predictions, however, rely on the data from Twitter, which I know generates a lot of criticism.”
Overall, how accurate is your social media data method?
“My approach allowed me to predict the winners of the Right and Left primaries in France. It also predicted the outcome of the Austrian presidential election. However, it failed to predict the result of the Dutch parliamentary elections.”
Would you have been able to predict the Brexit result or Donald Trump’s election?
“I would need to study the data to be able to answer the question, but I believe there has already been some work produced on this subject.”
Would you need anything to make your method even more accurate?
“It would be necessary to devote more time and resources to data analysis. There is so much data available that there is scope for multiple analysis.”
“There is so much data available that there is scope for multiple analysis.”
ON THE UPCOMING FRENCH PRESIDENTIAL ELECTION.
Using the most recent data available to him, Professor Bevort was most recently predicting the possibility of a surprising second round opposing Far-Right Marine Le Pen to Far-Left Jean-Luc Mélenchon.
In a recent blog post, he indeed explained that “Jean-Luc Mélenchon has accentuated his dominance over the social networks in the final days before the election. In the first two weeks of April, Mélenchon’s number of followers increased the most, his followers engaged the most and Mélenchon’s team was the most active.” If in February and March, Marine Le Pen was the strongest candidate on Facebook and Twitter, leaving only YouTube to Mr Mélenchon, the professor notes that“since the beginning of the month, Jean-Luc Mélenchon has taken over the lead and even widened the gap with all the other candidates on the social networks.”
Do you think it is possible that Emmanuel Macron, who is currently leading in most — if not all — opinion polls, could be the greatest upset on election day?
“The impact of Emmanuel Macron on the social networks is surprisingly low for a candidate that opinion polls continue to give as a favourite for the French Presidential election. According to the big data from the social networks, Marine Le Pen and Jean-Luc Mélenchon will, in fact, qualify for the Presidential second round run-off.”
“According to the big data from the social networks, Marine Le Pen and Jean-Luc Mélenchon will, in fact, qualify for the Presidential second round run-off.”
To what extent do online audience and social media data variations reflect a change in the candidates’ political influence?
“The candidates themselves are also starting to point at favourable online trends as a sign of increased popularity. We have so far interpreted things in the same way. However, such a reading favours the ‘contagion’ effect: a politician’s growing influence as an explanatory hypothesis. Experts often refer to it as the complex ‘contagion phenomenon’ or influence mechanisms.”
Why do you think some candidates get such good engagement results online (Mélenchon, Le Pen), while others so bad (Macron, Fillon, Hamon)?
“There seems to be another process at work that could explain variations in data. As they become aware of the importance of those figures for their propaganda, candidates start to use more sophisticated tools like NationBuilder or Typeform to generate audience and engagement for their websites and social media activity.”
Could the candidates’ community managers manipulate the data using these tools to artificially inflate the audience of their websites and social media accounts?
“Fake (paid) traffic or bots are known to distort the data, but these tools, which generate inflated audience and engagement numbers, may be a much bigger issue. The significant variations in data could be in part due to a ‘boiling’ effect produced by the tools and activists: an over-increased activity of many followers without a real increase in follower numbers, for instance. The ability to distinguish between the two effects ‘contagion’ and ‘boiling’ appears as a prerequisite to avoid being misled by data variation. This distinction applies to the audience of websites as well as to the social networks’ activity.”
If election results could be predicted from online data, does it mean that any candidate (even the smallest ones, like Philippe Poutou or Jacques Cheminade, usually credited with less than 2% of votes by opinion polls) could actually win an election? Not because of the money they use during their campaign or because of the size of their political party, but simply because they would have a great social media presence and a super social media manager, alike Jean-Luc Mélenchon this year.
“No, I don’t think so, because one of the conditions for the development of their social networks presence is that they must use sophisticated and costly NationBuilder-like tools. The same goes for having super social media staff which happens to be very expensive too. But the social media definitely offer new ways to enter the political arena because of their greater openness. In France, the political system (Presidential election, simple majority system in the legislative elections, weak political democracy) accumulates the hurdles that prevent new political forces.”
“Social media definitely offer new ways to enter the political arena because of their greater openness.”
Should a small candidate with a significant online presence still be considered as a minor candidate?
“No, in my opinion, a small candidate with a big presence on the social media would not be a small candidate.”
ON THE FUTURE OF THE BIG DATA ANALYSIS.
If your social media data analysis correctly predicts the two finalists of the Presidential election next week, will you expect the media to pay more attention to your work and start including online data to their election analysis from then on?
“I think that the media and researchers have a vested interest in cooperating to develop new methods of analysis of the political opinion and the society that better understand the gap between political institutions and the voters. There can be no solution to the crisis of representation without involving pollsters, the media and the academic world. If the social media data get it right at the upcoming election, there will surely be positive consequences. However, if the data gets it wrong, this miscalculation will face more criticisms than any of the pollsters’ past blunder. It would be a mistake, however, not to consider the big data as a new, preferred method to understand in a fine, detailed and quick manner the changes in the political opinion.”
“There can be no solution to the crisis of representation without involving pollsters, the media and the academic world.”
Disclaimer: Political predictions in this article are based on the social media data analysis of Professor Bevort and don’t reflect the views of the journalist, the magazine or its owners.
- “Politics and Big Data: Nowcasting and Forecasting Elections with Social Media”, Routledge, 2017.
- “A Meta-Analysis of State-of-the-Art Electoral Prediction From Twitter Data”, Social Science Computer Review, 2013.
- “Twitter a Digital Socioscope”, Cambridge University Press, 2015.
(This article was first published on the Politics Magazine PoliticsMeansPolitics.com)
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
J.N. PAQUET is an author, journalist and political writer. He is the editor of PoliticsMeansPolitics.com. His book series on populism and nationalism “The Tip of The Populist Iceberg?” is available in print, eBook and audiobook worldwide.
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *