Constituency polls vs MRP vs 2017 results: where to look for tactical advice?
We’ve had ten constituency opinion polls released, one full set of MRP data (which is a large scale national poll then used to model the results in individual constituencies, as explained here) plus of course the 2017 general election results in all seats.
How do they compare? Let’s look at those seats for which we have all three.
In nine out of ten cases, both the constituency poll and the MRP data used agree on which pair of parties is currently in the top two slots (and hence who should be considered as potential recipients of tactical votes). The one exception is Portsmouth South, where MRP and the constituency poll diverge on who is in the top two.
What’s more, six of those nine cases involve a change in the top two from 2017 and so in those cases both MRP and constituency polling agree with each other but point to a different tactical voting choice than the 2017 results would suggest.
Overall then, MRP and constituency polling are painting a broadly similar picture, including in cases where they both show a major change from the 2017 general election result.
That means that, broadly speaking, based on the public data we have so far, either (a) both this MRP data and the constituency polls are wrong, or (b) the 2017 general election result will not be a great guide to tactical voting decisions.
My money is on (b) as that both would fit with what we know from other data (e.g. the big swings shown in the national polls and the pattern of vote shares in the European elections, both of which suggest significant limitations to using the 2017 results as tactical voting guides), and also as the constituency polls and MRP get to their answers via very different routes. Two different approaches coming up with roughly the same answer gives that answer extra weight.
| Cambridge | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 16% | 52% | 29% | – | 2% |
| Poll (October) | 10% | 30% | 39% | 7% | 12% |
| MRP | 16% | 30% | 38% | 6% | 7% |
| Esher and Walton | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 59% | 19% | 17% | (2% Ukip) | 2% |
| Poll (October) | 45% | 11% | 36% | 4% | 3% |
| MRP | 38% | 9% | 35% | 11% | 5% |
| Finchley and Golders Green | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 47% | 44% | 7% | (1% Ukip) | 2% |
| Poll (October) | 29% | 25% | 41% | 2% | 3% |
| MRP | 36% | 25% | 26% | 5% | 5% |
| Gedling | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 43% | 52% | 2% | (2% Ukip) | 1% |
| Poll (November) | 37% | 42% | 6% | 13% | 1% |
| MRP | 36% | 29% | 16% | 11% | 5% |
| North East Somerset | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 54% | 35% | 8% | – | 2% |
| Poll (October) | 44% | 14% | 28% | 7% | 3% |
| MRP | 41% | 18% | 23% | 8% | 7% |
| Portsmouth South | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 38% | 41% | 17% | (3% Ukip) | – |
| Poll (October) | 27% | 24% | 30% | 14% | – |
| MRP | 30% | 25% | 22% | 12% | 10% |
| South Cambridgeshire | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 52% | 27% | 19% | – | 2% |
| Poll (November) | 36% | 12% | 40% | 7% | 4% |
| MRP | 38% | 14% | 31% | 7% | 8% |
| South East Cambridgeshire | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 53% | 28% | 19% | – | – |
| Poll (October) | 42% | 16% | 31% | 8% | – |
| MRP | 37% | 16% | 29% | 10% | 6% |
| Wokingham | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 57% | 25% | 16% | – | 2% |
| Poll (November) | 42% | 12% | 38% | ? | ? |
| MRP | 40% | 13% | 29% | 8% | 7% |
| Workington | |||||
| Con | Lab | Lib Dem | Brexit | Green | |
| 2017 result | 42% | 51% | 7% | (4% Ukip) | – |
| Poll (October) | 45% | 34% | 5% | 13% | 2% |
| MRP | 33% | 31% | 12% | 18% | 4% |
Is it possible to use a more complex mathematical model that factors in both national change from 2017 to 2019 (ge Lib Dems polling at 2-3 times what they were) and local factors such as how pro-Remain a constituency was ?
Also what is your take on the Labour complaint about using the extrapolation from the ‘Flavible’ website set up by the two data scientists, and them not being a member of the British Polling Council ? Presumably, since they use data that is already acquired, they are not pollsters ?
Apologies ! On reading a bit further – looks like that is what MRP is – although it is not clear what they built in to the constituency estimates.
And it looks like what Flavible did is pretty much what I described.
We have only to consider how good a guide the 2010 results were to what would happen in 2015! This time, given the huge concentration on one UK-wide issue, the results are quite likely to bear even less resemblance to the previous general election.
Really glad someone converted the graphs from Best for Britain; I’d bulk downloaded them all in preparing for a script to analyse the images, but this link has saved me a job I probably wouldn’t have got round to!