Can we trust general election polls? Bristol expert tells us how to make sense of the figures


There is still one week to go until the nation heads to the ballot box and already the news is a veritable sea of polling statistics, figures and predictions.

From YouGov to independent surveys, it appears that every app and newspaper is publishing a new set of results on a daily basis – each with wildly different forecasts of the outcome on June 8.

Over the past six days various polls have predicted that Labour and the Conservatives will both increase and decrease their majority and that there could even be the possibility of a hung Parliament.

And it not just the current general election which has left people stumped, in fact pollsters en-mass have gained a reputation for unreliability in recent years. Hardly any predicted the result of the 2015 election, and yet more analytical models were thrown off the scent ahead of the European Referendum.

So it is not surprising that many people feel confused and frustrated by the variety of polling data around.

But there are ways in which polling can be a useful resource to track the progress of the election, if you can understand what you are looking at.

Why do we conduct election polls?

Polls are used by statisticians, parties, pressure groups and journalists as a method of ascertaining the views off the whole or a certain group of the electorate. The information gained from polls is put to different uses by different groups including as a way to measure the overall popularity of a party or to inform changes to individual policy.

How do they work?

In very basic terms the purpose of a poll is to ask a random sample of people the same questions and then use the results as a representation of the population as a whole.

The BBC has a wonderful analogy of “you don’t have to eat an entire bowl of soup to know if it tastes good – if properly stirred, one spoonful is enough.”

How do polling companies get their samples?

Most polling companies contact people by phone or over the internet. One of the largest polling firms, YouGov, has collected a panel of more than 360,000 people and picks a sample of people from the group when it needs to conduct a particular poll.

Read More

General Election 2017

Why do polls asking the same question give different results?

Differing results from the same standard polling test come down to differences in the samples.

Political sociology lecturer at Bristol University, Paula Surridge, said: “One of the main issues we have will polling is sample problems.

“Sometimes the sample of population questioned in the poll are not representative of the population as a whole. For instance in the 2015 election a lot of polls used the views of very politically engaged young people and these did not reflect the attitudes of the larger youth population, so pollsters unintentionally built in a bias.”

The disastrous predictions made two years ago prompted the British Polling Council (BPC) to carry out an independent investigation in to polling practice.

The body found in general companies spoke with more Labour voters than Conservative ones and did not weight the results to compensate for this disparity.

It also found polling companies were “herding” their questions to make sure their overall results were close to their rivals.

The BPC made a number of recommendations to polling companies about sample size and distribution. Some of these measures have been put in place by the pollsters, but the speed of this snap election has meant not all of the proposals have been rolled out.

Another issue which can change results is the weighting placed on certain elements of the questions or answers given.

Some companies will simply ask people how they plan to vote on Thursday, while others will weigh this information against the individuals past voting habits, socio-economic status, age and where they live.

Ms Surridge added: “Pollsters can also change the weighting which they put on to elements such as turnout figures which can affect the results. For example, if I go out and collect the distribution of attitudes towards the death penalty I am not then going to try and predict what the result of a referendum on the death penalty because not everyone who I have spoken to will vote. You therefore have to use some form of equation to work out which percentage of the sample you have spoken to will turn out to vote.”

And while most polling companies are very open about the sources of their data, polling is a business and the exact algorithms used to calculate the results are commercially sensitive information.

Which polls should we trust?

Although not all polls are created equal, it is unfair to say that pollsters set out to deceive.

Ms Surridge said: “Polls in general are reliable, I think what people need to be aware of is what the data can tell you and how it is interpreted.

“Take the YouGov model which made it on to the front of The Times. The paper said that the results showed that the general election will result in a hung parliament. But this was only one of a range of potential outcomes. The Times took the median of the results, but by doing this they are suggesting that the model has given a very black and white answer, which is incorrect.

“Journalists and the media have a great responsibility in their interpretation of the results as they stand on a platform which reaches so many people. One thing I would say in defence of the YouGov statistical model is that a similar model was used to correctly predict the vote share of the US election and the Brexit result last year – so that is a good indicator of reliability.”

However, the lecturer says people should take a methodical approach to deciphering polling.

She said: “My advice is to look at a number of different polls to see the way they are moving. It is clear the Conservative lead is narrowing but not clear by how much.”

Read More

General Election constituencies

What is the future of polling?

It is hoped the full range of BPC will be rolled out by the time the next general election or national referendum takes place.

Ms Surridge said: “Polling is only going to get more accurate, hopefully after this election we should have a pretty good idea of what models work best.

“There are some people which say that pre-election polling should be banned like it is in France and Italy, but I don’t think this is the right way to go. I do however think that we have too many polls now.

“And the more polls you have the greater the chances of producing outlying results. For instance for every 100 polls I would expect a certain percentage, say five, to return slightly unusual results for a number of different reasons.

“So it stands to reason that as the number of polls published in a campaign increases then so too do the number of outlyers, and unfortunately this outlying results are the ones which make the front pages.”

Source