Driven by data; ridden with liberty.
On 23rd April, French voters went to polling stations for the first round of this year’s presidential election.
Despite some expectations there would be another major polling miss, it turns out that French polling companies estimated the first round result with remarkable accuracy.
Emmanuel Macron, a centrist who has only recently started his own party, won 24.0% of expressed votes, with Marine Le Pen of Front National coming second with 21.3%.
To understand the accuracy of these results, I calculated a weighted polling average (based on the total sample size) of the final seven polls of different houses (conducted in the fieldwork date ranges): Odoxa (21st April 2017), BVA (20th – 21st April 2017), Ifop-Fiducial (18th – 21st April 2017), Ipsos (19th – 20th April 2017), Elabe (19th – 20th April 2017), OpinionWay (18th – 20th April 2017) and Harris (18th – 20th April 2017).
There was also an Odoxa poll conducted on 20th April, and an Ifop-Fiducial poll conducted between 17th and 20th April. These polls include the 20th April in their fieldwork range, but are not the final pre-election poll from these houses, and so are not included.
Looking at those seven polls, the single largest average error on any of the five top candidates was for Hamon (PS), who was estimated to have a share of 7.5%. When all votes were counted, Hamon received 6.4% — an error of only 1.1 percentage points.
The mean absolute error for two parties (Macron and Le Pen) was 0.5 percentage points. For the main five candidates, it was just 0.6 points.
Now the second round has been determined, Mssr. Macron and Mme. Le Pen can be directly compared. Front National’s Le Pen is currently polling between 35% and 40% in a direct contest against En Marche’s Macron.
In recent polls, Mssr. Macron leads by at least 20 percentage points.
It would take either a monumental polling miss, unprecedented and not belied by the first round’s accuracy, or a huge shift in real voting intentions for Mme. Le Pen to win the French presidency.