This is it. The last week of the campaign has passed, and we are about to enter the media embargo. These elections, the first of Macron’s five-year term, will determine the current balance of power at the national and European levels - a strong performance in France gives you more weight in your group at the European level.
They also mark the beginning of an electoral marathon, since there will be at least one election per year until 2022 - municipal 2020, departmental and regional 2021, presidential and legislative 2022. This Sunday will therefore signal the strengths and weaknesses of each party for the upcoming elections – although in a noisy way, as each election has its idiosyncrasies.
A few hours before Election Day then, what does the model tell us about this balance of power? What are the most likely outcomes? Where are the uncertainties? Let's embark for one last ride in the bowels of the model.
What can each party expect?
I know it's not very sexy, but I’ll still dare: more or less the same as since the beginning of March. In the end, the balance of power has not changed that much over the last three months: centrist LREM and far-right RN are still the big favorites; right LR is still alone in his corridor, clearly behind LREM but clearly ahead of far-left LFI and the Greens (EELV); the latter are still neck and neck, while the Socialists (PS) and nationalist DLF are fighting to get at least a few seats.
There were indeed some movements, the most important being for RN (+3,5 points of median support in 3 months, from 21% to 24,5%) and DLF (-2 points, from 6% to 4%). But the other parties have hardly moved - nothing that can really upset the hierarchy.
5-in-6-chance interval of the popular vote
Solid lines represent the median share of the popular vote of each party.
Shaded areas show the range in which the true popular vote is, with a 5 in 6 chance
(83%). So a hypothetical range from 20% to 25% with a 22.5% median means
the party has a 5-in-6 chance of getting 20% to 25% of the popular vote,
a median share of 22.5%.
Why take a 5 in 6 chance as benchmark? Look at it as the probability of getting any number but the 6 when throwing a fair die. Hover over the chart to see the details. You can hide/display a party by clicking on its name in the legend.
What about all that excitement around polls that put RN in the lead? As an attentive reader, you had not been fooled: as we reported last time , the two parties have been neck and neck for months, so it is not surprising to see the hierarchy fluctuate in the polls - hence the interest in aggregating these polls .
Another way of saying this is that suggesting "RN is winning in the polls" is at best an abuse of language. A more relevant description will argue that "RN is winning in some polls (about 2/3 since the model gives RN a 2 in 3 chance of finishing first)". The others give either LREM 1st or a tie.
When the polls are so close, basically anything can happen. Of course, RN is the number one favorite for the first time since December, but the model clearly indicates that it is a very slight chance - only a 2 in 3 chance of finishing first, while LREM has a 1 in 5 chance. Similarly, for a long time, LREM had about a 3 in 5 chance - a very weak favorite indeed.
Probability of getting the most seats
At this point, the final position of the two favorites will not be very informative. The more interesting will be the gap between them. According to the model and the available data, there is a two-point gap between RN and LREM (24,5% vs. 22,5% of the popular vote). It is difficult to decide between them on the number of seats as well: the median scenario is 22 seats for the centrist majority and 24 for RN, with a 5 in 6 chance of getting between 19 and 25 seats for the former and 21 to 27 for the latter.
The most likely result is that RN ends up with 2 more seats than LREM - as seen in the distribution below, this scenario has a 1 in 7 chance (13%) of occurring. But the uncertainty is high, since there is a 5 in 6 chance that the result will fall between +2 for LREM and +5 for RN.
Distribution of the gap between RN and LREM
What are the main points of attention?
1. How big will the polling error be?
Obviously, we are going be interested in the magnitude of the polling error. French institutes are quite accurate and their errors hardly depend on the election’s type:
Mean absolute error of all polling institutes, by type of election
|Type of election||Far left||Left||Greens||Center||Right||Far right||All parties|
The absolute error measures the distance between the weighted average of the pollsters and the election results. PollsPosition calculations on nearly 800 polls in 15 elections. See our method for more details.
However, they tend to misjudge leftist parties during European elections. In this case, this could help the PS: since its level is already very low, a polling error is more likely to underestimate it than overestimate. It could also be good news for the rest of the left if this announces an electorate further to the left than expected by polls. But the opposite is also true, as our simulations suggested three weeks ago: according to the model, it would be surprising to see the left-wing parties outperform if one of them underperforms. Such underperformance would probably signal an electorate leaning to the right and center, or a demobilized left-wing electorate.
The table also shows that the far left is among the parties best assessed by pollsters, regardless of the type of election. This is yet another way of challenging the so-called intentional underestimation that pollsters would subject LFI to. In general, if you are constantly finding a way to criticize polls that favors your party, it is probably because the political hooligan inside you has taken over. Again, until proven otherwise, pollsters are not systematically biased against any party.
Except against the far right. Out of the 800 polls and 15 elections in our database, they overestimated the former National Front (FN) by an average of 2 points - this trend being strongly influenced by the last five elections. This is probably due to a combination of factors: an electorate that may be harder to poll than others; herding behaviors more pronounced at the end of campaigns; poor incentives given by the media and the public - institutes are criticized without nuance when they undervalue the extreme right, but they are unhelpfully ignored when they overvalue it.
We have to keep that in mind when, over the last two weeks, we see polls showing that RN is increasing despite a very stable campaign overall. To speak like a Bayesian, I would be more surprised if polls undervalued the far-right party than if they overvalued it.
2. Turnout and 0-seat parties
We dealt with turnout in detail in our previous analysis , so we will only do a brief reminder here. Turnout should be low - around 40%. One of the added values of our model is that it takes into account historical uncertainties - and therefore, in particular, turnout. This is very useful, but it is not magic: if this election’s uncertainties dwarf previous elections’ uncertainties, the model will not be able to do much about it.
On the other hand, I am skeptical that this black swan could come from turnout: historically, it is quite similar from one election to another - it is much more likely to decrease from 40 to 35 than from 40 to 10 for example. As a result, I tend to think that conventional wisdom focuses too much on turnout, thereby forgetting other variables of interest.
Finally, European elections have a threshold effect: you need at least 5% of the popular vote to get seats. As the election is proportional, 5% of the votes represent 4 seats. This means that you get either at least 4 seats or 0 seat.
With an average of 5.2% and 4% of the votes respectively in our model , PS and DLF are the most likely not to get any seat – a 4 in 9 chance (46%) for PS, and a 2 in 3 chance (66%) for DLF.
3. Macron’s popularity
As we pointed out in our first analysis , this variable is highly correlated with En Marche’s polling results - when one rises, the other does too. So where is Emmanuel Macron's popularity? According to our tracker , his approval has been stable since the beginning of April (28.5%), while his disapproval ratings have declined very slightly over the same period (68.5%).
This probably does not surprise you, since we wrote above that polls for LREM were stable over this period. The question is, of course, whether popularity follows or precedes voting intentions. But in this case, neither indicator is moving, so this question is less pressing.
Both predictable and uncertain, this election seems paradoxical. It is predictable insofar as polls have been stable for months and draw a clear balance of power: a duel of favorites, a chaser, a duel below 10%, and a tail fighting to have at least a few seats. These four strata did not move during the campaign, looking a lot like silos and an election that is played out in advance. And yet there is uncertainty. In the third stratum, LFI and EELV are very close, while everything is possible in the first stratum between RN and LREM.
In this respect, this election is a great opportunity to practice probabilistic thinking: when the data do not contain enough information to help models discriminate between different scenarios, we are left with integrating this uncertainty into our reasoning.
There will undoubtedly be plenty of deterministic interpretations of the results on Monday morning - "it was obvious that", "everyone knew that". If you want to avoid this trap - and be honest with yourself - I advise you to bet with yourself: what is the most likely scenario in your opinion? What would surprise you the most? Where will each party end?
Basically, the statistical processing of data reminds us of what political life often makes us forget: reality is always more complicated than mottos. In this way, it nudges from ignorant certainty to reflecting uncertainty.