Cookies remember you so we can ensure to give you the best experience possible. By using this website or closing this message, you are agreeing to our cookies and policies

Do not show this message again
UK Insights

How does Kantar Public weight its voting intention data?

Joel Williams

Head of Methods

Politics 13.02.2018 / 09:00


Learn how we ensure that our polling predictions are as accurate as possible.

Like most pollsters, Kantar Public’s samples are drawn from an online panel (operated by Lightspeed) and are broadly balanced with respect to gender, age and locale. Kantar Public then applies a set of more nuanced weights to the data before reporting any results.  This ensures balance with respect to several other dimensions, some demographic – e.g. age crossed by educational level – and some political, including both 2016 Referendum voting and 2017 General Election voting. 

Most pollsters do something like this but Kantar’s weights are distinguished in ensuring that (i) the number of 2016/2017 non-voters in the weighted sample is approximately correct, and (ii) that the implied turnout1 in a hypothetical General Election is within realistic bounds. All polls have too many politically engaged respondents and this is one way (not the only way!) of dealing with the problem.

One consequence of this is that Kantar polls tend to show relatively low levels of political engagement compared to most polls, and this is especially pronounced among young people. Young people in polls tend to be particularly unrepresentative of their part of the population so their data is subject to a greater degree of weighting than average. We are consequently less confident that we know the true level of political engagement among young people but we are nevertheless confident that the weighted version is more accurate than the raw sample version.

What is different about voting intention data when there is no upcoming election?

Voting intention polls between elections have an additional level of uncertainty because the next General Election may be several years away so a question such as ‘how likely are you to vote?’ is more hypothetical than it is when a General Election is only a few weeks away. Because voting tends to be habitual in form, we find that most older people – who have voted many times – will tend to say they expect to vote, even if the next election is several years away. For others – mostly younger people – for whom voting is not (yet) a habit, an element of doubt may be expressed unless the election is right in front of them.  In practical terms, we see that expressed voting probabilities (‘how likely are you to vote?’) tend to drop between elections among younger people but not so much among older people. This can affect the headline voting intention figures if these are ‘turnout adjusted’ and the turnout adjustment at least partly reflects expressed voting probabilities. 

One option is to ignore expressed voting probabilities altogether and simply fix the characteristics of the voting population in advance so it is not at the mercy of these fluctuations in expression. Several pollsters did this for the 2017 General Election but ended up with inaccurate final polls because they assumed that fewer young people and fewer previous non-voters would turn up at voting booths than actually turned up. We did not do that, and we know that this decision helped our final poll to be one of the most accurate2.  In exchange, we accept some additional volatility between elections with respect to likely voter characteristics. At those times, we also quote voting preferences among the whole population, not just those that look to us like voters.

How does Kantar Public estimate voting probabilities?

Kantar Public does not take expressed voting probabilities (‘how likely are you to vote?’) at face value. Firstly, some are not registered to vote so they need to be identified and given a voting probability value of zero once the registration deadline has passed. Secondly, we know from doing post-election re-contact surveys that what people say they are going to do (or not do) is not always a great match for their eventual behaviour. In particular, younger people and previous non-voters are prone to over-estimating their likelihood of voting in the next election.

Our empirically-grounded formula for estimating voting probability takes all this into account and has proven reasonably robust to different election scenarios. For example, this formula led us to believe that the turnout among the Voting Age Population in 20173 would be 64% (it was actually 63%) and that this would vary from 42% among 18-29 year olds to 87% among those aged 60+. Undoubtedly, these figures are not wholly accurate but they are very similar to those obtained from the British Election Study post-election survey that used the highest quality sampling methods4. This gives us confidence that both our weighting and voting probability estimation methods are along the right lines.

Source : Kantar Public

Editor's Notes

1. In our case, ‘turnout’ means the proportion of the Voting Age Population (VAP) that votes.  The VAP includes those that cannot vote because they are not registered as well as those who cannot vote because they are not eligible to do so. We estimate the VAP turnout using a fixed odds ratio to the estimated sample turnout.

2. And before that in the 2016 EU Referendum [See this article for the 2017 General Election final poll and this article for the 2016 EU referendum final poll].

3. The VAP includes those that cannot vote because they are not registered as well as those who cannot vote because they are not eligible to do so.

4. See Prosser et al (2018) for details, although note that the BES covers the Voting Eligible Population (VEP) which excludes those that cannot vote because they are not eligible to do so (a group included within the VAP). We have converted the (validated vote) VEP turnouts into VAP turnouts for this publication, estimating these (unofficially) at 41% for 18-29s and 83% for those aged 60+.

Latest Stories

We ‘go back to basics’ with editor of Marketing Week, Russell Parsons, to cover how marketing can better market itself.

It's another week of prizegiving: time for the BRIT Awards and the Oscars.

As we celebrate World Radio Day (13 Feb), we look at how the medium has evolved and why it still represents a useful channel for brands.

The February Brexit Barometer from Kantar Public also shows how people say they would vote if a new referendum was held.

London Fashion Week begins, it's World Radio Day and of course we celebrate St Valentine's Day.

Related Content