SOURCES
The FT’s main UK poll tracker uses data from every new national survey of voting intentions conducted by a member of the British Polling Council. Voting intentions in demographic subsamples use data from Deltapoll, Ipsos, Omnisis, Opinium, PeoplePolling, Redfield and Wilton, Savanta, Survation, Techne and YouGov.
METHODOLOGY
We use general election polls covering the UK or Great Britain to calculate both an average voting intention and a range of probable values for each party.
To calculate the average at a particular point in time, we take each poll published in the last 21 days and assign it a weight based on how long ago it was published, the sample size, and how often the pollster publishes polls. The moving average is the weighted average of these polls.
Polls are first weighted according to an exponential decay function, so that a poll published today is weighted fully, while a poll published three weeks ago is not weighted at all. Surveys with larger sample sizes also carry greater weight than those with smaller sample sizes. If a pollster has released more than one poll in the last 21 days, each poll from that pollster is weighted less to ensure that each pollster accounts for the average equally.
For demographic survey tracking, we use a longer time window and more smoothing to account for fewer surveys and smaller sample sizes.
To obtain a range of probable values that each party could win if the election took place tomorrow, we add two distinct sources of error: sampling error and polling industry error.
Sampling error represents the risk that the opinions of a randomly chosen subset of the population do not match the opinions of the entire population. For each party, we estimate the range of values that our moving average could have taken given the sampling error of each poll.
Polling industry error, or nonsampling error, represents the risk that all polls are systematically biased in one direction or another. Sources of error in the polling industry include the use of biased samples, voters who are undecided until Election Day, or voters who do not tell pollsters their true intentions. We estimate non-sampling error by considering how much previous election results differed from pre-election poll averages. The values we have chosen are based on a combination of academic and original research.
Please note that most surveys of UK voting intentions do not include Northern Ireland, where the main political parties are not the same as those in England, Scotland and Wales. As a result, voting intentions in Northern Ireland are mostly excluded from our poll monitoring.
We thank Jack Bailey at the University of Manchester for his help with survey aggregation methods.