Results from The Sunday Times GB Poll
19 May 2017 - 23 May 2017

Updated methodology:

This poll appears at first sight to buck the trend of a narrowing Conservative lead, showing a slight increase since the last one we released, but that is a product of changes to our approach. Like for like, the gap would actually have narrowed by 6%. We will be releasing another poll in the next day or two and early indications are that it will show a further narrowing of the Con-Lab gap.

In common with other polling companies, we have been reviewing our approach to sampling and weighting since the 2015 election miss. For example, we quickly introduced a separate 75+ age group for quotas and weights which had the expected effect of slightly boosting the Conservative share over Labour (75+ being even more likely than 65-74 to vote Conservative). Although this was to some extent cancelled out by the consequent lower effect of weighting by previous vote, a more representative raw sample is clearly a preferable starting point.

The relative accuracy of our internal polling before the EU referendum led us to conclude that weighting respondents based on their 2015 vote should be somewhat more effective in 2017 than previous voting history had been in 2015, with a five year gap since that previous election. However, again in common with other agencies, we still had ongoing concerns about the apparently substantial difference between claimed likelihood to vote across different demographic groups in surveys and the reality of turnout based on sources like the post 2015 election British Election Study – particularly on age.

We have been working with data sets from before the 2015 election through to the first 4 weeks of our 2017 campaign polls, in which we have included several attitudinal and demographic questions from the BES to check for any significant disparities. As a result we are now amending our analysis model for our remaining GB polls, to provide a better balance across age groups and locations as well as reducing the impact on the results of people who were eligible to vote in 2015 but didn’t do so. In summary:

– Age groups for the voting intention questions have been weighted to match the profile of voters in the 2015 BES
– Regions have been weighted to match the turnout of voters based on stats from the Electoral Commission
– Our 10-point likelihood to vote question (previously reported on 8-10) has been tightened to include only people who say they are certain to vote (10/10)

We don’t anticipate making any further changes in our remaining polls other than increasing the sample size as polling day approaches and pressing undecided voters for an answer, as we did in 2015.