The UK general election campaign has begun, with voters having to decide whether to support the current Conservative government – which wants to proceed with leaving the European Union – or try to replace it.
A pro-EU campaign group, Best for Britain, has launched a designed to help elect politicians who want to change course on Brexit. The tool has been criticised for what some see as odd advice, but the group says their choices are backed by complex statistical modelling.
The Liberal Democrats are the most anti-Brexit UK party, with a policy of notifying the EU that the country no longer wishes to leave, but it is also a minor party, meaning its chances of being elected in many seats are seen as slim.
Advertisement
Despite this, Best for Britain is recommending a Lib Dem vote in 99 seats where the party trailed the incumbent by 25,000 votes or more in the last general election in 2017.
Best for Britain CEO Naomi Smith says this advice isn’t due to partisan bias, but rather the result of how the group powered its calculator. It used a relatively new form of election modelling known as MRP, short for multilevel regression and post-stratification.
“MRP isn’t polling, that’s the key thing that people probably don’t quite understand,” she says. Instead, it is a model that takes a large initial polling sample (more than 46,000 voters in this case) and combines it with numerous other factors (90,000 in this instance) to make much more granular predictions about the behaviour of constituencies – and even individuals.
The method saw a boost in 2017 when polling firm YouGov released an experimental MRP on the eve of the election, correctly identifying the winning party in 93 per cent of seats.
YouGov’s Chris Curtis says the principle behind MRPs is to model types of voters, rather than try to rely on traditional constituency polling.
For example, an MRP will model the behaviour of a 24-year-old woman with a degree in the north-east of England, then work out how many of those there are in different constituencies. It then assesses other factors, such as how marginal a seat is (voters in marginal seats are more likely to switch parties than those in safe seats) and calculates incumbency effects (which are particularly strong for an MP’s first election as the incumbent).
The techniques behind MRPs are decades old, but are largely a new factor in politics. This is partly because existing polling techniques are becoming less effective as it gets harder to reach balanced samples and also because online polling decreases the cost of the large-scale polls needed to power the models.
Both Smith and Curtis are aware of the model’s limitations, however, especially in an election like this one with so many variables. Smith notes there were more than 400 members of parliament in the 2017 election who received more than 50 per cent of the vote in their constituency. Her MRP currently forecasts that fewer than 90 will have such a majority in this complex, multi-party contest.
Smith says just because Best for Britain’s calculator recommends a party in a seat, that doesn’t mean the seat is winnable: the UK’s electoral system means there are many safe seats.
There is also a significant difference in an MRP based on polling in the final days of an election campaign and one weeks before an election is even declared – not because the latter is “wrong”, but because during an election campaign, voters change their minds. Smith says Best for Britain is planning to update its model closer to polling day.
“MRP is still just like all polling, telling us where the public is right now,” says Curtis. “It is not predicting the future.”
