How poll tracker works

  • Published

Polling a sample of the population has often been likened to tasting soup: if it is well stirred then you need to have only one spoonful to tell what the whole bowl is like.

In the same way, a well conducted poll of 1,000 people can, most of the time, give us an idea of what the country as a whole is thinking.

However, there are several problems that pollsters need to overcome to have a chance of accurately reflecting the whole electorate.

Margin of error

The first caveat is that no poll can be 100% correct 100% of the time. Polling companies generally claim that 95% of the time, a poll of 1,000 people will be accurate within a margin of error of +/-3%.

This means that a figure in the poll could be up to three percentage points higher or lower than that shown.

So if the Tories are on 32% and Labour is on 38%, there is a chance they could both be on 35%.

It is, however, more likely that the figures will be 1% out rather than 3%.

The question

How the question is framed can have an effect on the results. Companies have to decide whether to simply ask people who they would vote for or to remind them of the choices (as they would have on a ballot paper).

Weighting

Another issue is how to ensure the sample is representative of the general population. To achieve this, polling companies "weight" their data to match the demographic profile of the UK.

At its most basic level, this means that if a poll of 1,000 people is made up of 550 men and 450 women, it is unrepresentative because it does not reflect the profile of the UK population (51% female).

So the answers of female respondents will be given slightly more weight (in this case they will each count as 1.133 people) to give them a representative impact on the final findings.

Conversely, the men will be weighted to each count as 0.891 people.

The same procedure is routinely carried out for age group, social class and region.

Image source, Getty Images
Image caption,

How can a poll reflect real life intentions?

Past vote

For voting intention polls, further adjusting is required. Some pollsters weight by past vote - they ask respondents who they voted for last time and weight the sample so that it more closely matches the political make-up of the general population.

One problem with doing this is that a certain number of people will incorrectly recall who they voted for last time - and a few will even lie about it.

So polling companies often use a variety of further methods to improve the accuracy of their weighting.

These include using information from a range of previous voting intention surveys.

Likelihood to vote

Most companies then weight or filter by likelihood to vote so that the answers of people who are most likely to vote are given the most prominence in the results.

This does have the effect of reducing the number of people on whose answers the final voting intention figures are based - which in turn raises the effective margin of error.

'Shy' Voters

Finally, several pollsters reallocate a percentage of "don't knows" to the party they voted for last time.

This is to get around the problem that emerged following the 1979 election with the phenomenon of "Shy Tories" or the "spiral of silence" - people who do not like to admit they support a certain party but who vote for them nonetheless.

Timing of publication

The date for each poll is the final date on which fieldwork was conducted.

Previously, date of publication was used, but many polls are now disclosed by blogs or via social media before they appear on newspaper websites or in print.

The BBC poll of polls recalculates every time a new included poll appears.

If there is any delay between fieldwork and a poll being made public, we may recalculate the poll of polls to take account of this.

Rounding

Polls are often rounded, so occasionally a result may not add up to exactly 100% and sometimes change will not sum to zero.

Methodologies in detail

BBC poll of polls

The BBC poll of polls is a rolling average of all polls included in the poll tracker.

Every time a new poll is published, the poll of polls is recalculated to factor in the results of that poll.

YouGov and Populus poll more frequently than other companies. To prevent over-representation, only one weekly poll from each is currently included in each poll of polls.

This is usually the YouGov poll published in the Sunday Times, and the Populus poll published on Friday.

ComRes, external

Question: If there were a General Election / UK General Election (for Scotland) tomorrow, would you vote Conservative, Labour, Liberal Democrat, UKIP or for some other party?

The names of the major parties are rotated in the question at random. UKIP has been included in the prompted parties since November 2014.

If respondents decline to name a party in response to the voting intention question, they are asked how they would probably vote if it were a legal requirement to do so and allocated this party.

All who continue to remain undecided or refuse to say for whom they would vote are then allocated a party, according to the party with which they most closely identify. In both cases, the data is weighted by reported likelihood of voting.

Interview method: Telephone and online

Sample size: Telephone: Approximately 1,000 adults age 18+; Online: Approximately 2,000 adults age 18+

Sample method: Within each region of the UK a random sample of telephone numbers is drawn from BT's domestic database. The last digit is randomised so that unlisted numbers are also hit.

For online polling, part of the sample is selected from an online panel of more than 250,000 GB adults. A second group is selected through website invitations or pop-ups. Respondents are profiled using a series of demographic data questions.

Within each government office quotas are set for the demographic profile of the sample in each region.

Weighting: Data is demographically weighted to reflect the profile of all adults in the UK aged 18+.

ComRes weights the data by likelihood to vote. This is done by asking respondents how likely they are to vote (on a scale of 1-10 - with 10 being absolutely certain to vote).

The answers of respondents who answer four or below are excluded. The rest are weighted proportionately so that someone who answers 10 gets the most weight and someone who answers five gets the least.

Respondents who have answered "don't know" to the voting intention question or refuse are asked a further "squeeze" question: if it were a legal requirement for you to vote, which party do you think you probably would vote for?

This extra data is then weighted and added to the original voting intention results.

ComRes weight by how respondents recall having voted at the last general election. This weighting is based in part on comparing the distribution of this recall vote (for those who declare for whom they voted) with the actual result of the 2010 General Election and in part with the average past vote obtained in the previous 12 polls.

ICM, external

Question: The Conservatives, Labour, the Liberal Democrats and other parties would fight a new election in your area. If there were a general election tomorrow which party do you think you would vote for?

Interview method: Telephone

Sample size: Approximately 1,000 adults age 18+

Sample method: Within each region of the UK a random sample of landline telephone numbers is drawn from BT's domestic database. The last digit is randomised so that unlisted numbers are also hit.

A random sample of mobile telephone numbers is also generated and 150 of the interviews are conducted over mobile phone.

Weighting: Data is demographically weighted to reflect the profile of all adults in the UK aged 18+, including those in households that do not own telephones.

ICM then weights the data by past vote.

The answers are also weighted by likelihood to vote. This is done using a combination of how likely respondents say they would be to vote in a new election (on a scale of 1-10) and how consistent they have been at voting in the past.

Respondents are then asked whether they voted in 2010 and which party they voted for in that election. The vote figures shown in the table are calculated after ICM has excluded those who say they will not vote, refuse to answer the question or don't know who they would vote for.

The figures are adjusted for turnout calculated accordingly to their stated intentions to vote.

Ipsos MORI, external

Question:"How would you vote if there were a general election tomorrow?" All 'undecided' and 'refused' responses are then asked a 'squeeze' question: "Which party are you most inclined to support?" and responses from the two are combined. (Conservative, Labour, Liberal Democrat, [SNP/Plaid Cymru in Scotland/Wales respectively], Other. If the respondent says other they are prompted to choose from a list including Green, UKIP and BNP)

Interview method: Telephone, including mobile telephone

Sample size: Approximately 1,000 British adults aged 18+

Sample method: Random digit dialling of landlines (so as to include households with ex-directory numbers), with sample stratified by Government Office Region (GOR). Respondents are selected on a quota basis to reflect the demographic make-up of the population. 800 respondents are selected this way and a further 200 by randomly generated mobile telephone numbers.

Weighting: Data is weighted to reflect the profile of the population, using the latest census data and recent updates from the Office for National Statistics and other surveys. Data is weighted by age, gender, social grade, work status, public vs. private sector worker, region and car ownership. Ipsos MORI does not weight by past vote.

Certainty of voting filter: Ipsos MORI asks respondents to rate their likelihood to vote on a scale of 1-10 where 10 means absolutely certain to vote and 1 means absolutely certain not to vote. The data is then filtered so that only the responses of those who answer 10 out of 10 (absolutely certain to vote) make up the final figures.

Those who are undecided or refuse to answer are excluded.

Opinium, external

Question: "If there was a general election tomorrow, which party would you vote for?"

Interview method: Online using 'computer aided web interviewing'.

Sample size: Approximately 1,900 GB adults aged 18+

Sample method: Defined from pre-collected registration data containing gender, age group, region , working status and social grade to match the latest published ONS figures.

Weighting: Results are weighted to nationally representative criteria

Populus, external

Question: "If there was a general election tomorrow, which party would you vote for? Would it be [rotate order] Conservative, Labour, Liberal Democrat, SNP [Scotland only], Plaid Cymru [Wales only] or another party?"

If 'another party':

"Would that be [rotate order] the UK Independence Party (UKIP), the Green Party, the British National Party (BNP), or some other party - or do you not know how you would vote?"

Sample: Approximately 1,500 adults 18+

A more detailed explanation of the methodology involved can be found on the Populus website., external

Sampling method: Within each government office region a random sample of telephone numbers was drawn from the entire BT database of domestic telephone numbers. Each number so selected had its last digit randomised so as to provide a sample including both listed and unlisted numbers.

Weighting: Data is weighted to the profile of all adults aged 18+ (including non telephone owning households).

Figures are adjusted for turnout on the basis of respondents' declared likelihood of voting.

In a further step Populus weights the whole sample on the basis of its 'past vote' - adding the most recent poll data to its previous 20 most recent voting intention polls (so as to avoid the random volatility that can appear in comparing any two individual samples) and calculates the past vote weighting from the average recalled past vote in this data, giving a weight of 50% to the actual result of the last election and 30% to the average recalled past vote from the most recent polls.

An additional final step is then taken to address the tendency for 'spirals of silence' among supporters of unpopular parties causing an inadvertent bias in voting polls.

This is done by taking those respondents who will disclose which party they voted for at the last election, but refuse to answer the question of how they would vote in an election now, or say they don't know. They are then reallocated to the party they voted for at the last election on the basis of values derived from a 5,000 sample callback poll immediately after the last election, when Populus re-interviewed people polled in the run up to the election.

Their responses are compared in terms of what they actually did with what they said they would do when first interviewed.

TNS-BMRB, external

Question:"If a general election were to be held tomorrow which party, if any, would you vote for?"

The following party names appear in a randomised order: Would not vote for a party, Labour, Conservatives, Liberal Democrats, SNP - Scottish National Party, Plaid Cymru, Other, Prefer not to say.

If respondents answer 'Other' they are asked: "Which party is that?"

The following party names appear in a randomised order: Green Party, UKIP - United Kingdom Independence Party, BNP - British National Party, Other, Prefer not to answer.

Interview method: Online

Sample size: Approximately 1,200 British adults aged 18+

Sample method: The poll is conducted through the TNS Online Omnibus. Invitations to complete the questionnaire are emailed out to a sample of GB online panelists who have agreed to participate in market research. The survey is open for a limited time period and then closed off a few days later.

Weighting: Data is weighted to the profile of all adults aged 18+ in Great Britain for working status within gender, age within gender, presence of children and Government Office Region. Data is also weighted to the 2010 General Election results. The voting intention figures were also weighted by 2014 voting patterns and a Likely Voter Model based on the British Election Study is applied.

YouGov, external

Question: If there were a general election held tomorrow, which party would you vote for?

Conservative, Labour, Liberal Democrat, Scottish National (SNP) / Plaid Cymru, Some other party, Would not vote, Don't know

Interview method: Internet panel

Sample size: Approximately 2,000 adults aged 18+

Sample method: YouGov has recruited an online panel of nearly 300,000 subscribers. From this panel YouGov selects a sub-sample to match the Great Britain electorate by age, gender, social class, newspaper readership and party ID.

Only this sub-sample has access to the questionnaire. Respondents are paid for taking part.

Weighting: Data is weighted to the profile of all GB adults aged 18+ including people without internet access.

As well as weighting by gender, age, social class and region, YouGov weights by readership of individual newspapers and party ID.

On 7 April 2015, YouGov announced some changes to its methodology, external, introducing weighting according to how likely respondents say they are to vote on 7 May.

YouGov also stated that it would draw daily polling samples from people previously contacted in January and February 2015, and weight the data according to how those people said they would vote at that time (a period when the polls were broadly stable and Labour were, on average, slightly less than a point ahead).

This, the company said, ensured that it could be confident that any material change in the polls would reflect a genuine shift in public opinion since January and February.

Survation, external

Question: If there was a general election taking place tomorrow, and there was a candidate from all political parties standing in your constituency, which party do you think you would vote for?

Sample size: Over 1,000 adults aged 18+ from Great Britain.

Sample method: Survation's poll is conducted through an online panel.

Weighting: Data is weighted to the profile of all British adults aged over 18. It is also weighted by age, sex, region, household income, education and past vote.

Related internet links

The BBC is not responsible for the content of external sites.