Facebook Doesn't Know What German Voters Think
Judging by how the German election is playing out on social networks, the far right Alternative for Germany party should be well ahead. But as in France earlier this year, Facebook and Twitter engagement is a poor predictor of what's going to happen in Germany -- a phenomenon data enthusiasts should try to understand after their recent forecasting triumphs in the U.S. and the U.K.
Social network engagement -- how active the supporters of a party or candidate are and how enthusiastic they appear to be about their choice -- was more predictive of the results of the Brexit referendum and the U.S. presidential election than traditional polls. It also indicated (along with the better-constructed sociological surveys) that Jeremy Corbyn's Labour Party would do better than expected in last June's U.K. election. These successes have fueled concerns about data mercenaries shaping election outcomes through Facebook targeting, which now appears to be at the center of U.S. investigations into Russian election meddling.
But attempts to use "big data," including social network parameters, to forecast French presidential election results last spring were largely disastrous, while traditional pollsters got it right. Emmanuel Macron, who had few diehard supporters before the first round of the election in April, appeared to lose out to candidates who actually had a voter base -- certainly nationalist Marine Le Pen, conservative Francois Fillon, media-savvy and charismatic leftist Jean-Luc Melenchon. In addition, Le Pen and Melenchon were more fun to watch than Macron; on the social networks, that counts big-time. One "big data" model based on measures of interest in the candidates, devised by social media consultancy Echobox, did perform in line with the polls, but didn't get as much attention.
In Germany, Echobox's election tracker -- which measures the electorate's interest in the parties, expressed as the total traffic generated by media articles about each of them -- has the AfD in first place.
This, of course, is not a prediction that the party will win -- just proof that it generates the most interest, both positive and negative. The party also has the highest Facebook engagement, according to Newswhip, another social media consultancy.
On Twitter, the far-right party has fewer followers than its more established rivals, but they are the most active: The AfD gets 76 retweets per tweet, almost five times as many as Chancellor Angela Merkel's Christian Democratic Union.
There are two reasons, however, why these numbers probably don't spell an AfD victory on Sept. 24. One is that Germany's Facebook penetration isn't as high as in the U.K., the U.S. or even France. That makes it less likely that Facebook activity will translate into real-world voting.
The other reason is that in Germany, like in France, traditional polling methods are better-aligned with the election system. Half the German parliament is elected by voters picking from a nationwide menu of parties; there are no quirks such as the U.S.'s electoral college or the U.K.'s reliance solely on individual constituencies. This makes nationwide representative sampling more useful than in those countries. Polls show the AfD winning some 11 percent of the vote, much lower than the expected performance of the two leading political forces -- the CDU and the center-left SPD. An upset is far more unlikely, based on these data, than it was in the U.S. on in both recent U.K. votes.
That doesn't mean the polls can't be wrong at all. AfD's internet activity may well pay off with a better performance than predicted thanks to voting for specific candidates in constituencies, which shapes the other half of the parliament. The far-right party may end up with more seats than expected, raising its prominence and even its power in the unlikely case Merkel cannot form a ruling coalition and is forced to make do with a minority cabinet. But the polls are still highly likely to call the election outcome more accurately than "big data"-based alternative methods.
Once that happens, there will be conclusions to draw both for the forecasters and for politicians. One striking possibility is that the less a country relies on Facebook, the less it tends to go for the political extremes. In normal, everyday life few people are radical: They are held back by old-school social norms that exist among co-workers and neighbors. On the social networks, these are less important. Instead, filter bubbles and echo chambers encourage behavior that would be perceived as antisocial offline. The propaganda swirling in those bubbles, whether foreign or home-brewed, reinforces extreme beliefs. A healthier society resists Facebook the way a healthy body can resist infection better.
There's probably more to it than that, though. There are still 31 million regular Facebook users in Germany, so it's not clear what explains Facebook's relatively small influence. Current research shows people everywhere, including the U.S. and the U.K., have little trust in what they read on social media. And yet social media engagement metrics work better as predictors of democratic outcomes in the English-speaking countries. Better-designed research is needed to understand this: Perhaps asking people about trust levels is the wrong approach and something else, such as emotional affinity, needs to be measured in a subtler way than the trust surveys use. The difference in how German, French, British and U.S. voters react to what the social networks push to them is worthy of more attention than it's getting.
This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
To contact the editor responsible for this story:
Therese Raphael at firstname.lastname@example.org