The three assumptions that explain our voting intention

  • Insight
  • 30 May 2024

As we get closer to the General Election in July, there is an unusually high range in the voting intention numbers between pollsters. More in Common’s latest poll had Labour leading the Conservatives by 19 points, but polls at the same time found Labour leads of anywhere between 12 points and 27 points. 

This difference is driven by varying assumptions that pollsters make about the population, and how they will behave on voting day. So in this blog, we’ve set out the three biggest assumptions that underpin our methodology. And we’ve also made a tool so you can see how our results would change if these assumptions were different.

Assumption 1) Weighting and sample frame

Polls work by choosing roughly 1,000 to 2,000 people and using their opinions to represent the broader UK public. But this is only accurate if your 2,000 people sample is representative of the country.

To do this, pollsters set quotas on the number of people from different age groups, genders, regions etc. to ensure that their sample roughly matches the population. Then, if any groups are under-represented or over-represented, they are weighted up or down to make sure the final results are accurate.

But we can't put a weight and quota on everything - doing so decreases the “weighting efficiency”, which indicates that the weights are introducing more bias to the results. So tradeoffs have to be made. Some pollsters weight on 2016 EU referendum vote, others weight on ethnicity, others weight on social class.

More in Common allocates quotas for age/sex interlocked, education level, and region. We also weight on 2019 General Election vote and ethnicity. We think that these demographic factors are likely to be the most predictive of voting intention at this election, and avoid recall issues from asking people to remember their 2016 vote.

Here’s an example of why sample frames matter. More in Common has always set specific quotas for 75+ year old voters, whereas other pollsters opt for a broader category of 65+ year olds. Given that 75+ year olds are more likely to vote, and more likely to vote Conservative, this has the effect of giving us a slightly higher Conservative vote share than we would have otherwise.

(There are good reasons for why other pollsters don’t recruit for 75+ year olds. For example, you could argue that lower internet uptake among this group means that 75+ year olds on online panels could not represent the wider 75+ age group).

When we ran this experiment in March, the result was a difference in Labour lead of 23 when we did not allocate for 75+ year olds, versus 17 points when we did, keeping all other methodology exactly the same.

Assumption 2) Where will undecided voters go? 

Another dynamic that makes this General Election particularly unique is the current political skew of undecided voters.

Currently, undecided voters are much more Conservative than the general population. 37 per cent voted Conservative in 2019, and only 9 per cent voted Labour. 

Currently, 19 per cent of them say they would Conservative if forced to choose, and 21 per cent would vote Labour - a significantly smaller Labour lead than is seen nationally.

To account for this - we ask undecided voters a “squeeze” question - who they would vote for if they were forced to make up their mind. Undecided voters still tend to say they don’t know when squeezed, but they also skew more Conservative than the general public. If they give a second choice and they say they are very likely to vote, their responses get fed into our final VI. 

Pollsters with higher Labour leads tend not to make this reallocation and exclude the Don’t Knows - this essentially has the effect of assuming they’d ultimately vote in the same proportions as the rest of the sample, and pollsters with lower Labour leads tend to make a more involved approach to reallocating undecided voters - predicting how they will end up voting based on their demographics and preferences rather than using a squeeze question.

Assumption 3) Who will turn out to vote?

A large proportion of the country do not vote in General Elections - something that is unlikely to change at this election, a challenge then for pollsters is figuring out which of our panelists are likely to vote.

To decide which of our respondents will and will not vote, we ask them to rank how likely they are to vote on a scale of zero to ten, where zero means they definitely will not vote, and 10 means they are certain to vote. 

We currently have the threshold set at seven, so anyone who ranks their likelihood to vote at seven or above is included in our headline figure. You can see how changing the likelihood to vote threshold impacts implied turnout at a General Election in the graph below.

As you can see, our implied turnout is higher than you would expect at the moment. This is because at the start of the election campaign we changed the question wording from the more generalIf a general election was called tomorrow, how likely would you be to vote? to the more specific An election has been called for July 4th this year. How likely are you to vote in this election?”. This had the effect of increasing predicted turnout - you can expect us to change the threshold to account for this in the next week.

Likelihood to vote thresholds also change Labour’s lead. This is because Tory voters tend to be older and more likely to vote, whereas Labour supporters are younger and less likely to vote. So a higher threshold means a lower Labour lead. You can see the effects of that below.

What is the combined effect of all of these assumptions? Build your own voting intention.

These assumptions have big implications for headline voting intention numbers. So we’ve built a tool below where you can explore how different combinations of assumptions change our headline voting intention.