Is this your prediction for the election?
No – this is not a prediction. It is our estimate of what the range of possible results would be if the election took place today and the voting intention was showing what it currently shows, based on our current data. It is not a prediction of where we will be on 12 December. We don’t know how things will change between now and polling day, but it would be surprising if things didn’t alter – one way or the other – before then.
Is this a poll?
No – it is a Multi-level Regression and Post-stratification (MRP) model that uses polling data. Polls aim to ask a representative sample of people a question and then show the results to that question, broken down by demographics. Our model uses data from a poll to discover relationships between people’s characteristics and their answers to the voting intention question. 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, if we were able to do very large polls in every constituency.
What is the model and how does it work?
It works by modelling vote intention based on analysis of key demographics as well as voting behaviour in the 2016 EU referendum, the 2017 general election, and the 2019 European Parliament elections. Each day, YouGov conducted approximately 14,000 interviews with registered voters from our panel, who are shown just the parties and candidates running in their particular seat. This data is used to assess how voters are making choices across the many different types of constituency in Britain, using information about the voters in each constituency as well as the sets of candidates who are standing in different constituencies. From this, the model calculates voting intention and seat estimates.
Have you assessed every local variable?
No, but we have used a very wide variety of variables that might capture different voting patterns. These include demographic characteristics of individuals and constituencies, past vote choices of individuals and shares of constituencies, information about the candidates standing for different parties, as well as a range of other features of constituencies. We cannot reliably capture factors specific to single seats, but we can capture patterns that affect even a small number of seats, if we have identified them as potential patterns to look for.
Who made the model?
The model was developed primarily by Professor Ben Lauderdale of University College London in conjunction with Jack Blumenau (University College London), YouGov’s UK political team, and YouGov's Data Science team headed by Doug Rivers of Stanford University. The data are streamed directly from YouGov's survey system to its Crunch analytic database. From there, the models are fit using Hamiltonian Monte Carlo with the open source software Stan. Stan was developed at Columbia University by Andrew Gelman and his colleagues, with support from YouGov and other organisations.
Are you saying these numbers are exactly right?
Both the seat and voting intention estimates have uncertainty. For each party in each seat, as well as the seat total overall, we report ranges that we believe have a 95% chance of including the outcome of the election if it were held today. The number presented is the most likely value, which is typically very close to the middle of that range. So when assessing the data it is best to remember that both the seat and voting estimates for the two main parties could very well be higher, or lower, than this midpoint. As the vote and seat shares for the smaller parties are lower, the ranges on those figures are also narrower.
How have you accounted for turnout?
Turnout is assessed on voters’ demographics and is also based on analysis from 2015 and 2017 British Election Study data.
How many people have you polled in my constituency?
It varies from seat-to-seat, but about 150 per week in each constituency. Naturally, we can’t ask everyone in a constituency how they would vote and a model cannot produce as accurate a result as a full scale poll in each seat. However, the sample size in each constituency is not all that important given that the demographics of each constituency – based on census data – are what is actually used to map voter data into each constituency.
Why are you doing this?
Elections are becoming more difficult to assess and so, as well as running traditional polls, we are looking at new ways to meet the challenge. We know we run a risk publishing so much data in the heat of a campaign but as data scientists we are committed to innovating, to increase both accuracy and specificity. We hope that voters, campaigners and the media are able to form an accurate picture of the current state of the election as a result, and make the decisions that each of those groups make in a campaign based on more accurate information rather than less accurate information.
Why should I trust you over other research companies?
There are other seat-level estimates being published and reported on during this campaign. YouGov has the strongest track record of applying these methods successfully in US and UK elections. We have published the general strategy in an academic journal article, so you can read about it in more detail. Nonetheless, we recognise that these methods are relatively new, and that each election can present new challenges. Methods which have worked well in the past may no longer do so. We are committed to constantly improving our methods and we make every effort to ensure that our work represents our best estimation of what the world thinks.
