Our goal is to generate a single recommendation for each region which results in the minimal number of votes needed to increase the overall number of Remain seats by at least one according to the D’Hondt method. Example: “In the North West, Change UK need 60,000 votes to increase the number of Remain seats.

Detail of the data sources used can be found here

If you are interested in finding out more about how we provide recommendations, our full source code is available on GitHub. You can download it and run the model with Python.

The approach is to simulate the results of recommending each party in a region until we find the smallest number of votes that will swing a seat from Leave to Remain.

We assume our voters are representative of the general population, to avoid making assumptions about those we can influence.

For a single region, start by calculating how many seats go to which parties according to D’Hondt. We also calculate the expected number of votes per party in a region (eg. Brexit party may have 1.1 million votes).

We hypothetically give 10,000 votes to a single Remain party. Those 10,000 votes can’t come from nowhere, so we move 10,000 votes from the other parties. We choose how to move the 10,000 votes based on survey data of voting intentions - eg. we know that more Labour and Liberal Democrat voters are likely to move to small Remain parties than Conservative parties.

We run D’Hondt on these new votes and see if the total number of Remain seats has increased. If it doesn’t, reset the votes and give 10,000 to each of the other Remain parties in turn. If none are successful, increase it to 20,000 and try giving it to each Remain party again.

Stop when one trial (eg. Liberal Democrat with 70,000 votes) results in a Leave party losing a seat, and a Remain party gaining a seat. This is our recommendation.

Example for a region eg. North West

From current polls, calculate the allocated seats (2 Conservative, 2 Brexit, 2 Labour, 1 Change UK, 1 Lib Dem) and absolute number of votes according to expected turn out (2304923 for Conservative, 2343243 for Labour, etc.)

Take 10,000 voters and back Change UK, removing 10,000 votes from the other parties according to voting preferences. With this new allocation of voters, D’Hondt still gives [2 Conservative, 2 Brexit, 2 Labour, 1 Change UK, 1 Lib Dem]. So moving 10,000 voters to Change UK does not win a seat for Remain.

Reset the 10,000 votes and try backing Green to see if it swings a seat to Remain, then each of the other Remain parties in turn.

If none of the parties gives a successful outcome, go back to step 2 and try giving each Remain party 20,000 votes, then 30,000, etc.

Stop when one trial (eg. Liberal Democrat with 70,000 votes) results in a successful outcome (eg. one fewer Conservative seat, one more Liberal Democrat seat). That is the recommendation in the North West.

Approach Q&As

Why one recommendation per region, instead of multiple (“First Green, Second Liberal Democrat”) or conditional (“If you would have voted Green, vote CUK, else vote as you intended”)?

A single recommendation is:

  • Easier to understand, communicate and share.
  • Requires a larger number of people to vote tactically.
  • Assumes our audience is similar to the overall population.
  • Sometimes counter-intuitive, ie they may recommend voting for the party not currently leading in the polls.

Multiple/conditional recommendations are:

  • Harder to understand, communicate and share.
  • Requires a lower number of people to vote tactically.
  • Reliant on very precise targeting (only the exact target group will switch, and all others will match current polling).
  • More inclusive. Fewer parties are not recommended and therefore fewer people will be upset.

Precise targeting is impossible. Instead, single recommendations are easy to communicate and we aim to drive more tactical votes through awareness.

What happens if a Remain seat takes another Remain seat?

The criteria for a successful recommendation is where the number of Remain seats increase and Leave seats decrease. If a seat switches between two Remain parties, it will not result in a recommendation.

What about uncommitted voters or additional turnout?

Additional turn out is a positive for Remain. We will incorporate this into the model as polls becomes clearer.

What if you lose a seat for a Remain party?

We take a do no harm approach. Voting intentions show that around 50% of voters will stick with the party they voted for in the 2017 general elections. Our recommendations focus on the 50% who are likely to be swayed. We are particularly interested in the 20% of the population who are reported to have voted tactically in the last general election.

What about other Remain tactical voting groups?

We are working on coordinating with other tactical voting groups, since we all have the same ultimate aim, to get more Remain MEPs. Ultimately it’s a tool which we hope people will be persuaded by.

What about Northern Ireland?

Northern Ireland doesn’t use d’Hondt, so it’s much harder to do tactical voting. Nonetheless, we will make a recommendation based on polling.

More information on the implementation details to come later.