Changes to polls could predict a more accurate result

Market research company YouGov has developed a new method forecasting how the public will vote this week.

As one of the UK’s leading pollsters, it has already changed their methodology after the general election in 2015. 

The changes included: polling more people who were previously under-represented in their own community but not nationally taking into account the views of those who have little interest in politics and removing a variable that was based on newspaper readership.

Previous methods employed by YouGov largely only polled those who had a keen interest in politics.

For the general election in 2017 YouGov has developed a new technique called Multilevel Regression and Post-stratification (MRP).

MRP uses polling data from YouGov’s polls and surveys from the previous seven days and considers the different variables such as constituency, voter demographics, past voting behaviour and other profile variables.


The pollster interviews 7,000 panelists a day. Over the course of the week, data is collected from about 50,000 panelists across the 650 constituencies. The sample size is much larger than other polling groups, such as Ipsos Mori, ComRes and ICM, who sample fewer than 5,000.

Next, the MRP estimates the probability of how respondents will vote in conjunction with statistics from the Office of National Statistics.

YouGov says that this is only to forecast, not completely predict the results, although it has 95 per cent confidence that its data is reliable. Constituencies are labelled as ‘safe’, ‘likely’, ‘leaning’ or ‘toss-up’.

YouGov is a member of the British Polling Council, an association of polling organisations that publishes polls.

It published a report at the end of May detailing recommendations on how the polls can improve after the perception that they got it wrong in the general election in 2015.

The report is intended so that people can rely upon to provide a more accurate estimate of Conservative and Labour support.

Their recommendations include:

  • determining whether voters had already completed a postal vote by the time they are answering the poll,
  • obtaining more representative samples from groups such as those who are not likely to vote,
  • interviewing people who have not voted first,
  • taking variables (such as education background) into more consideration when weighing respondents’ answers. 

In 2015 the final polls before the result estimated that Conservative and Labour were tied at 34 per cent, when it was actually 38 and 31 per cent respectively. This led to a widespread perception that the polls were inaccurate, according to the BPC.

John Curtice, the president of the BPC, said that the role of the BPC is to ensure transparency about how polls are conducted because people are still asking for what the polls have learned from 2015.

“The report is intended so that people can rely upon to provide a more accurate estimate of Conservative and Labour support in the 2017 election,” he said.

Currently YouGov forecasts a Conservative lead of 42 per cent and Labour trailing behind at 38 per cent.

Leave a Reply