Second-round vote transfers
Experimental
Experimental. This page is experimental. It shows assumptions the model already uses, taken from the 2017 and 2022 elections. They are not 2027 results.
What this is
When the second round pits the RN finalist against another finalist, the model splits the voters of the other candidates between the other finalist, neither of the two and the RN finalist; each bar below shows that average split for one group of voters.
Where the voters of each group go
Experimental
- Other finalist
- Neither
- RN finalist
Voters of La France insoumise and the PCF
Other finalist 42% · Neither 40% · RN finalist 19%
Polls of the voters of Jean-Luc Mélenchon · 36 polls. By election : 2017 (23) — 49% / 36% / 15% · 2022 (13) — 35% / 43% / 22%
Voters of the PS, the Greens and other left parties
Other finalist 67% · Neither 26% · RN finalist 7%
Polls of the voters of Benoît Hamon (2017) and Yannick Jadot (2022) · 33 polls. By election : 2017 (23) — 74% / 20% / 5% · 2022 (10) — 60% / 32% / 8%
Voters of Les Républicains
Other finalist 46% · Neither 28% · RN finalist 26%
Polls of the voters of François Fillon (2017) and Valérie Pécresse (2022) · 36 polls. By election : 2017 (23) — 46% / 26% / 28% · 2022 (13) — 45% / 30% / 25%
Voters of Debout la France
Other finalist 25% · Neither 33% · RN finalist 42%
Polls of the voters of Nicolas Dupont-Aignan (2017) · 14 polls. One election only: how much this group changes from one election to the next is unknown.
Voters of Reconquête
Other finalist 8% · Neither 12% · RN finalist 79%
Polls of the voters of Éric Zemmour (2022) · 13 polls. One election only: how much this group changes from one election to the next is unknown.
The other blocs (for example the centre and Horizons) have no data: for them the model keeps the older uniform synthetic split.
How to read this chart. Each bar stands for all the voters of a group in the first round. It is cut in three: those who would vote for the other finalist, for neither, for the RN finalist. The width of a part is its average in the 2017 and 2022 polls.
Learn more
Under each bar the "By election" line gives the same shares for 2017 and for 2022: the further apart they are, the less sure the average. Shares are rounded, so they may add up to 100 % give or take one point. The model draws values around these averages; the bar shows only the centre.
Where the numbers come from
These are polls taken after the first round of 2017 and of 2022: the voters of a candidate were asked how they would vote in the second round. The rows come from the "Reports de voix" tables of the French Wikipedia polling lists; each row keeps its institute, its date and the page address.
A group takes the mean of the polls of each election, then the mean of the elections; "rolling" polls (overlapping windows) are left out. The code is in presidentielle/transfers.py.
- 2017 : BVA, Elabe, Harris Interactive, Ifop, Ipsos, Kantar Sofres, Odoxa, OpinionWay. Source page 2017
- 2022 : Cluster 17, Elabe, Harris Interactive, Ifop, Ipsos, Opinionway. Source page 2022
132 polls in all. Sample sizes are not in these tables: a row is one poll, not a number of voters.
What this page does not say
This is not an estimate of the 2027 vote. The 2027 transfers are not known: the bars come from 2017 and 2022.
These are assumptions of the model. In both 2017 and 2022 the other finalist was Emmanuel Macron; applying these rates to a different finalist is an assumption that no past election can test.
Two elections are only two points. The gap between them is the only handle on how far off a transfer can be; the model uses it to set the spread of its draws (Dirichlet concentration: 48).
In P(win) these rates enter only when the second round pits the RN finalist against another finalist and no duel poll prices that pair. Everywhere else the model uses duel polls if they exist, otherwise the synthetic split.
How the model is tested on past elections · See also the full method