A tale of two MRPs: can constituency polling really predict the election accurately?

According to conventional wisdom, the Multilevel Regression with Post-stratification (MRP) model is the gold standard of modern-day political soothsaying.

Tapping into a national sample of over twenty thousand participants, researching its findings on a constituency-by-constituency basis and using advanced statistical regression analysis — the MRP poll is there to tell us who is in and who is out. In advance of the returning officer’s declaration, the MRP is there to “call time” on an individual political career before the bell is actually rung.

But does it?

In the last two days, two of the UK’s leading polling companies, IPSOS (18th June) and YouGov (19th June) have released two MRP polls, both with samples of some twenty thousand participants. As such, each poll has received widespread media attention.

Yet strikingly little direct comparison has been made on the findings of these surveys — and doing so highlights the potential fragility of this method of polling.

***Politics.co.uk is the UK’s leading digital-only political website. Subscribe to our daily newsletter for all the latest election news and analysis.***

For while both the Ipsos and YouGov polls both project a Labour landslide — with Keir Starmer’s party set to win 453 seats on the former estimate and 425 seats on the latter, when one delves into the constituency level findings, the projections are often wildly different.

Here are some significant examples:

1) A Reform breakthrough in Barnsley North?

According to the Ipsos poll, Labour is on course to comfortably poll 56 per cent in Barnsley North, easily seeing off Reform UK on 24 per cent on the vote. Yet YouGov has very different findings. This area of South Yorkshire saw some of the best results for the Brexit Party back in 2019, and YouGov has Labour only narrowly holding on in Barnsley North with 37 per cent of the vote — compared to 34 per cent for Reform UK.

2) The Liberal Democrats romping home in Tunbridge Wells?

According to YouGov’s MRP, the Liberal Democrats are set to comfortably gain Tunbridge Wells with 40 per cent of the vote compared to 31 per cent for the Conservatives. Yet the Ipsos poll from a day earlier paints a different picture. Not quite so “disgusted of Tunbridge Wells”, Ipsos has the Conservatives retaining the seat on 35 per cent — with the Liberal Democrats polling just 22 per cent, almost half of the level suggested by YouGov.

3) A Green breakthrough in North Herefordshire?

According to Ipsos, the Green Party is ahead in their target seat of North Herefordshire, polling 36 per cent compared to the incumbent Conservatives who are on 30 per cent. But, hold on: here the picture is almost the reverse of that in Tunbridge Wells. This time, YouGov projects North Herefordshire will be a comfortable hold for the Conservatives who, on 34 per cent, easily see off the Greens. According to YouGov, the party is polling at little more than half Ipsos’ figure of 20 per cent.

4) Liz Truss is under serious pressure — but from whom?

According to Ipsos, Liz Truss may soon cease to harbour such warm feelings toward Reform UK’s leader Nigel Farage, given the knife-edge forecast in her Norfolk South West constituency. The poll places Liz Truss on 31 per cent, compared to 30 per cent for Reform UK. Yet according to YouGov, it is Labour that poses the greatest threat to the former prime minister, with Liz Truss on 33 per cent, Labour on 29 per cent, and Reform UK on 25 per cent.

5) The SNP advancing on the Scottish Borders?

According to Ipsos, the SNP is doing well across the three large constituencies that sit immediately to the north of the English border. The pollster predicts the party will gain Dumfries and Galloway from the Conservatives; in Berwickshire, Roxborough and Selkirk, meanwhile, Ipsos has the SNP polling 33 per cent — just below the 35 per cent of the Conservatives.

But again, the picture between the two MRP polls is markedly different. YouGov has the Conservatives well ahead by 42 per cent to the SNP’s 31 per cent in the Berwickshire seat.

6)  Labour winning the Battle?

According to YouGov, one of Labour’s most remarkable gains could come on the south coast, East Sussex retirement spot of Bexhill and Battle — long one of the Conservatives’ safest seats, and a constituency with one of the lowest proportions of younger voters anywhere in the UK. Yet where YouGov projects a Labour victory by a margin of 3 points, Ipsos reckons the Conservative hegemony will be sustained — with the party winning out by a more comfortable 7 per cent.

7)  And total confusion in Newton Abbot…

One of the widest differences between this week’s two MRP polls is in the Devon Conservative seat of Newton Abbot. According to Ipsos, the Conservatives sit on 36 per cent, seeing off Labour on 29 per cent and with Reform UK in fourth place on 12 per cent.  

But according to YouGov, far from mounting a challenge, Labour actually comes fourth on 16 per cent of the vote. In Newton Abbot, YouGov has the Conservatives on 31 per cent, narrowly seeing off the Liberal Democrats on 25 per cent and Reform UK on 23 per cent.

Can we trust MRP polling?

Over the years, the public has had reason to be sceptical of national opinion polls, the same ones that predicted a Labour victory in 1992, and a Conservative outright majority in 2017.

In this election, the national result appears in little doubt, with all polls projecting a solid Labour landslide as the party polls double that of the Conservatives.

Yet exactly where Labour gains those seats, and in particular how many seats are won by the Liberal Democrats and Reform UK, would appear far less certain.

As the tale of the above two MRP polls shows, drawing on an average of just 35 interviews per seat, constituency-by-constituency polling is throwing up wildly different expectations.

These would appear to go far beyond a basic margin of error.

As such, the findings of an MRP poll may be useful for political parties to display on bar charts in campaign literature — but from the very different findings in the above analysis, you can’t help wonder if these sorts of polls are worth the statistical paper they are written on.

Politics.co.uk is the UK’s leading digital-only political website. Subscribe to our daily newsletter for all the latest election news and analysis.