With the reliability of traditional opinion polls called into question, the search is on for a new way of tracking the public mood and predicting election results. Our successful prediction of the outcome of the French presidential election earlier this month shows that big data can disrupt traditional polling — if it is done right.
Earlier this year, we launched the French Election Tracker, a measure of how much interest each candidate received that was based on highly granular and comprehensive live data. Using billions of data points spanning the entire election campaign, we also issued a series of accurate predictions for the outcomes of both rounds of the election.
Ours was the only big data prediction to correctly predict Macron as the winner in both rounds. We also predicted Mélenchon’s surge long before it became apparent in the polls. Moreover, the FET predicted the correct outcome an hour before official results were published in both rounds of the election.
Prediction: 23.7%. Result: 24.0%.
Prediction: 22.9%. Result: 21.3%.
Prediction: 21.0%. Result: 20.0%.
Prediction: 17.3%. Result: 19.6%.
Prediction: 15.1%. Result: 6.4%.
Prediction: 64.7%. Result: 65.8%.
Prediction: 35.3%. Result: 34.2%.
As people from all demographic groups spend more time online, the data created as clicks and likes accumulate may one day render obsolete any polling based on small, representative samples.
Yet the French presidential election showed clearly the potential risks of relying on new methods. Most big data predictions were far off the mark, counting Macron out and variously predicting Presidents Le Pen, Fillon and even Mélenchon.
Our correct predictions were based not only on uniquely comprehensive and high-quality data, but also on a thorough understanding of French politics, the country’s history and its unusual electoral system. Moreover, we were transparent about our methods, our data and the limitations of our model.
This focus on context, quality and transparency is what turns a lot of data into big data. We think it constitutes the gold standard for big data predictions, a standard which we will continue to uphold as we move on to building our German Election Tracker.