Every General Election campaign has a ‘hold your nerve’ moment.
And last night’s YouGov MRP polling announcement is one such moment.
It’s a wake-up call for anyone who doesn’t want to see a Tory Brexit being delivered in two weeks’ time. And we can’t pretend it doesn’t have challenges for our position.
But the situation is always more complicated for Liberal Democrats. Our national seat campaigns are being rolled out in a heavily focused way.
We can see from recent seat polls in places like Finchley & Golders Green and Wimbledon that, when voters in those specific constituencies are asked how they are voting, we are doing much better than this model suggests.
Because our target seat campaigns are so focused in key areas, it makes it hard for data modelling like MRP to pick up our activity. What is clear is that our local seat activity is shifting significantly more votes our way in these seats than across the UK as a whole. And we know from 2017 that the number of doorstep conversations is the greatest indicator of electoral success.
A General Election isn’t a single UK-wide poll. It’s 650 separate races, and modelling like MRP will not necessarily identify the differences in what is going on in communities up and down the country, where people are struggling to decide how best to simultaneously stop Brexit, avoid a Corbyn Government and deny Johnson a working majority.
Voting choices that seem obvious in one seat are anathema in another.
MRP data modelling is very different in character to traditional polling which we tend to be more familiar with.
Multiple Regression and Post-stratification modelling is an extremely clever way of producing estimates of opinion for defined geographic areas by combining information from huge national samples (but very small constituency samples) with authoritative data from sources such as ONS and the Census.
The MRP authors themselves attach a significant caveat to their report stating “Our sample is large enough that we can identify patterns that occur across relatively small numbers of constituencies, but the largest model errors are likely to occur in constituencies with very atypical patterns of voting. Some examples of these are seats where there is a high profile independent candidate (e.g. Beaconsfield) or where there appears to be a new pattern of local competition in this election (e.g. Kensington)”.
In short, to work properly MRP requires a high degree of interpretation by professional analysts. And assumptions at the margins, can make huge differences when extrapolated out across a national position.