Ipsos political polling methodology

As in 2019, Ipsos’ political polling design for GB voting intentions is based upon a telephone quota sample using both landline telephone numbers (generated by random digit dialling) and mobile numbers (a mix of random digit dialling and targeted sample for some hard to reach groups such as young people, this sample based on opt-ins from a range of sources including online surveys and petitions, warranty cards and competitions), weighted to demographic population totals, aimed at achieving a representative sample of all GB residents aged 18+.  Quotas are set on gender, age, region, whether working full time, ethnicity and educational qualifications.

A two-part voting intention question is used with prompting on main parties and a follow-up “squeeze” question of those who were undecided or refused to the first question.  The Conservatives, Labour, Liberal Democrats, SNP in Scotland and Plaid Cymru in Wales are prompted, others are unprompted (currently this includes the Greens and Reform UK; in 2019 a split-sampling approach was used for the Brexit Party whereby half were and half were not prompted).  A turnout filter is used based on two questions: current stated likelihood to vote (at least 9 out of 10 certain to vote) and previous voting behaviour (if they say they “always” or “usually” vote in general elections or that “it depends”).  Later in the campaign this is supplemented by including those who say they have already voted by post, and excluding those who say they are not registered to vote.  Again this is the same as in 2019.  In 2019, an additional question was asked in order to reallocate supporters of the Brexit Party/Liberal Democrats/Green Party/Plaid Cymru to their second party preference if their first preference was not standing in their constituency; in 2024 we have just done this for the relatively small number of seats where Reform UK and the Greens are not standing.

Data are weighted to match demographic population targets from the 2021 Census, ONS population estimates, and other large-scale surveys. These weights are applied simultaneously by rim-weighting. As in 2019, data are weighted by age, gender, work status, ethnicity, car in household, social grade, work sector (public sector/other), Government Office Region, housing tenure, newspaper readership, and educational qualifications, some of which are interlocked.  Since 2019 we have updated our weighting targets in order to use the most updated figures from different sources (depending on which sources are available and most up-to-date, we have also needed to update which combinations of weights are interlocked, but the factors used remain the same). 

We continually review our methods, and may make further refinements and changes to our methodology if we think it is necessary for accuracy.

Article updated June 2024