United Kingdom 
United States 
Deutschland 
France 
España 
Italia 
International 
Who we ask
YouGov runs GB voting intention once a week for the Times and Sky. We typically use a sample of just over 2,000 respondents, sampled exclusively from our own panel without using any third-party panel providers. To ensure our sample is representative of the British public as a whole, we set quotas on our samples by respondents' age, gender, education, ethnicity, region, socio-economic grade (using NS-SEC) and how people voted at the last general election. We collect data on how people voted at the last election as quickly as possible after the election, or when people first join our panel, to prevent issues with false recall (that is people misremembering how they voted, or claiming they voted the way they wish they had with hindsight).
What we ask
We ask voting intention using two questions. First, we ask people how they would vote if a general election were held tomorrow. Then we ask people to think about their own constituency, the parties standing there, and ask how they would vote there. It is the answers to the second question that we use to drive our headline voting intention figure. We have found this approach is more effective at picking up tactical voting considerations and how people actually cast their vote on the day, in particular in seats where the Liberal Democrats are in contention.
How we prompt
We currently ask people if they will vote Conservative, Labour, Liberal Democrat, Reform UK, Green, SNP or Plaid Cymru. We do not include smaller parties in the question, but offer people the option of saying "other" and collect data on which other parties people choose. The decision on when to include a new or smaller party in the main prompt is a difficult one that is carefully considered. Putting a small party in the main prompt can artificially boost support, while failing to prompt for a party with substantial support could artificially suppress it. In practice we inform this choice using internal polling asking voting intention with an open text box, without any prompting, and only include a party in the main prompt when we are confident it would give an accurate finding.
How we process the data
After collecting people's responses, we use multi-level regression and post-stratification (better known as MRP) modelling to turn our raw data into headline voting intention figures. MRP is a well-established technique, and YouGov's MRPs have a strong record in the UK and around the world.
We use the same MRP model as we use for our large seat-by-seat projections but done on a much smaller scale. Rather than putting tens of thousands of people through the model and projecting it to the demographics of 632 different British constituencies, we use 2,000 people and project it to the demographics of just one geography – Great Britain as a whole.
The MRP modelling also accounts for two other important factors: turnout and people who say “don't know”. The turnout modelling in our MRP is based mostly on respondents’ demographics, and which groups were most likely to vote at previous elections, rather than how likely respondents themselves say they are to vote. This addresses the problem of more people in polls saying they will vote than actually do in reality (a persistent problem that has contributed to industry-wide polling error in the past). In practice, this means it assumes higher turnout amongst older groups.
The MRP model also reallocates people who say “don't know” when asked how they would vote, effectively reallocating people who say they don't know on the assumption they vote in the same way as other people who share their demographics and past politics. This corrects skews caused by people who probably will vote, but don't yet know who for. For example, in reality men and women are equally likely to vote, but women are more likely to say don’t know in polls. Not reallocating don’t knows would result in headline figures based on too few women.
What we publish
Each week, we publish headline figures from our MRP modelling that takes into account turnout and don't knows. We also publish tabulated results on our website that show the results to the questions without the MRP modelling, and with don't know, won't say and would not vote responses included.