French Election – neck and neck, but we’ll stick our neck out..

French Election Final

Well, we had hoped that things would become a bit clearer after the debate, but our DataSwarm Analytic Engine is still showing that Macron and Le Pen are essentially neck and neck, but with Macron losing his definite nose in front from  earlier this week. As with last time, we thought we would make our prediction on Friday using French custom , ie no polling on the day before the election.

Most of the French polls are showing something like a 60/40 split for Macron and growing. We are not seeing that.

What we are seeing, for the first time with our system, is a “split vote”, or more accurately a straddled vote. In essence we have a number of metrics which in the past have all lined up but this time are split. It works as follows:

  • If we measure the “influence” of the social media commentariat then Macron is ahead by a good 6+ %
  • There is more conversation overall (just) for Macron
  • But if we measure the “buzz” on social media then Le Pen is ahead by several % points
  • If we measure the “velocity” of support, ie what is happening at the time, we saw a switch over the day from Macron to Le Pen, there is a surge for her.  It may be bots, but our system is good at controlling for those- we will look at it on the weekend. (Update – It started before the Macron hacked emails were dumped, so that was not the cause.)
  • Our composite “Zeitgeist” measure over the day switched from a small 1-2% lead for Macron to a similar one for Le Pen. But it is very sensitive, the slightest change in weightings shifts from one to the other.

If we interpret what the graphic display is signalling, it is essentially that there is a large rump of Macron conversation (big orange blob) but it’s not particular influential – general conversation. There is a smaller very influential grouping (green blob in front). The size of the gap between the two is very unusual. The conversation for Marine & Le Pen (slightly different constituencies) are in the middle, and though smaller and less influential in total to date, en masse are more energised.

On balance, the majority of the signals are still for Macron but it is a weak signal overall. Sailors out there will know how confused the various telltales on a boat are when the wind changes, and it feels a bit like that.

We’ll admit freely that the system isn’t perfect, and we started to track politics to try and calibrate our system for this very reason – to help us in situations like this, as you get a known outcome and can then look at how your metrics performed. But this is the first time our system has “sat on the fence” – or straddled it, more accurately – while tracking elections. Every time before its been clear on who is winning.

There are two possible causes which could explain these variances in our system:

Firstly, we are using just Twitter, which we have found in the past gives us a good enough overview as a predictor.  Caveats are:

  • France has a lower penetration than the UK and US so the population on Twitter may not be a representative sample.  When for example we looked at the UK 2010 election on Twitter, the Liberal Democrats were going to win with 60% of the vote. They didn’t get even half of that. We have done a bit of triangulation with this project, but not much*
  • We know that social media in its early days trends liberal, so it may well be that Le Pen is further ahead – except that Macron is not a liberal either, and social media also tends to favour passionate extremes, and in essence France rejected the mainstream candidates, both are “outsiders” – which brings us to…

Secondly, the conversations are driven by how the potential voter groups are built up:

  • Both Le Pen and Macron have their support from round 1, but that was , 21.7% vs 23.7% respectively, a c 9% gap in Macron’s favour. But that is only c 45% of those who voted in the first round.
  • The question is where will the other candidates’ voters go? The DataSwarm Analytics Engine is showing a significantly stronger memetic bond between Macron and Fillon than Le Pen and Fillon, but with Melenchon its about 50/50 to either. Fillon and Melenchon had c 19.5% each and the remaining c 6% does not seem to be massively for one or the other (to be honest we haven’t looked at this tail in great detail).
  • But regardless, in essence this election will depend on the 55% of voters who voted for neither of the 2 remaining candidates who are both right-of-centre, whereas the “55%” are all left of centre.
  • In past elections there has been a concerted “banding together” of the non Le Pen camps to keep the Front Nationale out, but to make it more “interesting” this time around, other fissures are appearing – apart from Left/Right issue, Melenchon’s supporters are more nationalist in flavour, like Le Pen, whereas Fillon’s are more Globalist.

French commentators say that the mainstream media is heavily behind Macron and that is possibly driving our smaller leading green blob, but most Macron conversation is by people who aren’t his natural supporters so are less energised so this could explain the trailing big orange blob. Le Pen’s support is smaller but more committed, and it seems is more high energy.

We noted in our EU hypothesis model that so far (in UK, Holland) the Centre-Right candidate has made a grab for some of the Far Right turf to limit their support, and the centre-left has fragmented more left and more right, but in France Macron has not really made a grab for the right,  and the further left is more nationalist like the further right so this hypothesis will be tested in detail here.

In other words, France, as per usual, is different.

So, it may well come down to the weather, and whether the 55% who don’t have a dog in this fight can be bothered to vote, and if so who comes out in bigger numbers – the left-right or the left-nationalists.

And the forecast is for bad weather, which is said to favour Le Pen.

So there you have it….

Further Excuses Footnotes

*In an ideal (well paid assignment) world we’d triangulate this with other datasets in detail but we do this election prediction work on our own meagre dime. We didn’t need to triangulate much with Brexit or Trump, but this is the tightest election picture we’ve seen, We think it has something to do with the short time for the field to move from many to two candidates. and it’s possible that we would see a slightly clearer picture if we went into our system data in more detail, but we didn’t have the time and its better to wait for the outcomes.  We may have to factor this sort of straddling in future, but we’ll know better once we see the results and calibrate the systems.