Take YouGov’s MRP super-poll with a pinch of salt when it comes to individual seats in Wales

Ifan Morgan Jones

YouGov’s 2017 MRP poll has gained something close to mythological status after it bucked the trend and projected a hung parliament in that year’s election, when many others had foreseen a Tory majority.

YouGov has now published its first MRP poll of the 2019 General Election, which suggests that the Conservatives are on course for a comfortable majority.

The full projected results are:

• Conservatives: 359 (+42)
• Labour: 211 (-51)
• SNP: 43 (+8)
• Lib Dem: 13 (+1)
• Plaid Cymru: 4 (-)
• Green: 1 (-)
• Other: 1 (-)

In Wales, the poll suggests five seats will change hands, which are the main Tory target seats in the country: Ynys Môn, the Vale of Clwyd, Clwyd South, Brecon and Radnorshire and Wrexham.

There are a few however that they had their eyes on but will fail to gain, according to this model, such as Cardiff North and Gower.

The Liberal Democrats will also, according to this projection, fail to take Ceredigion from Plaid Cymru despite only needing a 0.13% swing to do it.

I would caution against reading too much into this poll, however. The first perhaps obvious point is that this MRP poll has been published almost a week earlier than the 2017 iteration and there are two weeks of campaigning to go.

The polls are still very fluid. This is particularly the case in Wales where Labour jumped 9% in the last Welsh Barometer Poll published on Monday. The current direction of travel seems to be that the gap between Labour and the Conservatives is narrowing slightly.

Data for the YouGov MRP poll was collected over the last week, and so it may have missed some of these changes during that time.

The second note of caution is that while the 2017 MRP is remembered for getting the Tories’ number of seats broadly right, it didn’t actually do as a good job of predicting the results of individual seats.

In Wales, the 2017 poll was wrong when it came to three seats – Arfon, Ceredigion and Preseli Pembrokeshire. It had Labour winning Arfon from Plaid Cymru’s Hywel Williams, Mark Williams of the Liberal Democrats retaining Ceredigion (Plaid Cymru took it), and Stephen Crabb of Preseli Pembrokeshire losing his seat to Labour.

These were all very tight contests with fewer than 400 votes in each one, so the miss wasn’t that bad.

However, in Ceredigion, the MRP’s numbers were all over the place. It had Plaid Cymru’s Ben Lake on 19.7% of the vote, in third place behind the Conservatives and a full 10% behind his eventual 29% result tally.

In fact, the model seemed to have a bit of difficulty predicting Plaid Cymru’s numbers across the board. In Ynys Môn, for example, their predicted vote was 5.1% lower than what the party eventually won.

With the current MRP showing a very narrow 3-way race on Ynys Môn, such an error could be enough to change the result on the day.

And it’s worth keeping in mind that while this poll can tell us, for instance, whether Ynys Mon fits the bill of an average Conservative voting constituancy, it doesn’t know how the actual Conservative campaign is going on the island.

Misses

The MRP itself is based not just on 100,000 interviews but also on data about age, gender, education, past vote and other factors.

This model is then applied to the demographic make-up of individual constituencies to provide projected vote shares for each seat.

While this potentially makes it much more accurate than other polling models, if any of the assumptions underpinning it are wrong it can lead to a big miss across the board.

According to Nate Silver of the website Five Thirty Eight, YouGov’s MRP poll actually made two big errors in the 2017 election:

• They overestimated the Conservatives’ vote share (+4 rather than +2.4)
• They underestimated how many seats they would win on the basis of that vote share

However, luckily for YouGov both these errors cancelled each other out, they got the results roughly right, and their model was considered a great success as a result.

Opinion polls are important tools that can give us a good idea which way the wind is blowing nationally. But as for the result itself, when was the last time we had an election night that didn’t surprise us?

At the 2015 and 2017 General Elections, and the 2016 US Presidential election, they were gasps of breath as the result came in on election night.

If you’re campaigning in an individual seat, there is no room for defeatism or complacency. Anything can happen on the night, and in some constituencies it no doubt will.