How MRP modelling works – and what it means for the general election

Data model producing constituency-level predictions 'getting more accurate all the time'

Deconstructed infographic showing UK regions and polling station signs
The latest YouGov/MRP poll for Sky News predicted a bigger Labour landslide than 1997
(Image credit: Illustration by Stephen Kelly / Getty Images)

YouGov's first MRP projection for the 2024 general election spelled bad news for the Tories. It found that Labour is on course for a bigger majority than its 1997 landslide.

The MRP, which stands for "Multi-level Regression and Post-stratification", suggested that "the Conservatives could be set for near wipe-out across many areas of the country", said the polling firm.

The Week

Escape your echo chamber. Get the facts behind the news, plus analysis from multiple perspectives.

SUBSCRIBE & SAVE
https://cdn.mos.cms.futurecdn.net/flexiimages/jacafc5zvs1692883516.jpg

Sign up for The Week's Free Newsletters

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

Sign up

How does it work?

MRP is a "type of poll that gets pundits excited because it draws from large amounts of data, including a large sample size and additional information like locations", reported Sky News, which commissioned the YouGov analysis.

It uses a large sample of survey results on how people intend to vote to "discover relationships between people's characteristics and their answers to the voting intention question", said YouGov. "It then combines these relationships/patterns with information about the characteristics of people living in different constituencies to construct estimates of how vote intention would look in each constituency."

This means that an MRP is not technically a poll, said the survey firm, but rather a model that "uses polling data".

The YouGov MRP predicted that Labour would win 422 seats, up 220 from 2019, while the Conservatives would lose 225, leaving them with just 140.

A separate MRP from More in Common released on Monday predicted a smaller win for Labour, of 382 seats, and 180 seats for the Conservatives.

How is it different from regular polling?

Traditionally pollsters conduct research on a sample of people who are representative of the whole population, which makes it hard to tell how the national vote share will translate into seats in Parliament.

In "simple terms", said The Guardian, an MRP uses polling data and "adds other details about the respondents, such as their age, qualification level, income, previous voting patterns and where they live". This allows it to then make granular predictions based on demographics.

What MRP "can't do is account for very specific local factors", said Sky News, "such as a hospital or large employer closing down in a constituency, or a scandal relating to a particular candidate".

Is it more accurate than other polls?

MRP is a "valuable resource" in an electoral system like Britain's, "where each general election is in fact hundreds of local contests taking place simultaneously", said the UK in a Changing Europe think tank. It can model "the many variations between one local contest and the next, something which isn't really possible with traditional polling".

YouGov's experimental MRP model correctly predicted the 2017 UK general election result, including shock results in seats such as Kensington and Canterbury, when traditional polling methods were wide of the mark.

MRP has been used to successfully predict recent votes such as Spain's general election last July, and the recent Catalan parliament election. The final YouGov MRP poll for the last UK general election underestimated the Conservative majority eventually secured by Boris Johnson, but it did accurately predict big gains in the so-called Red Wall.

As with all polling, any MRP model involves a "lot of assumptions and estimates, and different models can produce different results even with the same starting data", said UK in a Changing Europe.

It is also important to remember that MRP models "can only tell us about the state of opinion when the polling data is collected – voters can, and often do, change their minds as the campaign progresses".