Polls, Damn Polls and Statistics

6 May 2015

They are such a predominant feature of the media landscape every time an election rolls around, that it can be easy to mistake opinion polls for an accurate forecast of upcoming results.

In reality, opinion polls are an attempt to find out what the general population thinks at a particular point in time, but without actually asking everyone.

Because of the methods used to achieve this, the accuracy of poll results can vary, and sometimes different polls asking the same question can have completely different results. What's more, opinion polls themselves can influence the way people vote, so the poll itself is an influence on the very result it's trying to predict. And tactical voting (£) in certain marginal seats may make the results more difficult to predict.

While they can be a very useful tool in measuring the mood of the nation as a whole, there are several factors you should take into account when interpreting the results.

After all, history reminds us just how spectacularly the polls can get it wrong.

Margins of error

For obvious practical reasons, no poll can ask the opinion of every person in the country. So instead, pollsters survey a small cross-section of society to simulate the electorate as a whole. This method, however, comes at a price, and that price is a poll's margin of error. The fewer people surveyed, the greater the margin of error.

A poll that surveys 1,000 people typically comes with a margin of error of plus or minus three per cent. Imagine a poll which estimates support for party A at 35% and party B at 33%. The margin for error allows support for party A to be as low as 32%, and as high as 36% for party B. So with information like this it's not possible to say with certainty that party A is ahead of party B.

To achieve a margin error of just plus or minus one per cent, a poll would need to survey 9,604 people, which is rarely considered feasible.

Not all polls are created equal

There are many different polling companies employing potentially varying standards. Many companies are members of the British Polling Council (BPC), a body which requires its members to adhere to certain quality standards to ensure the most accurate results possible. But not all companies are members, and even though a company that's not a member may still employ scientific methods, results from such companies should be treated with caution.

Polls conducted on public websites, or those relying on a non-representative user base, such as an online newspaper poll, cannot be considered a scientifically-accurate representation of public opinion.

Question the questions

It is important to know what questions are being asked. Most BPC member poll companies will publish the data they use to produce a poll, so you can see the questions for yourself. It's worth asking whether a question asked may be leading people to give a particular answer, and what impact that might have on results.

Methods and trends

For practical purposes, most polls are conducted by phone or online. Yet while both these methods have their drawbacks, historically online polls have proven less accurate.

It's also important to remember that the results of one poll may be tempered by a wider trend. If after a period of polls showing the fortunes of one candidate in steady decline, one poll that shows an improvement in popularity doesn't necessarily mean that candidate has turned things around.

Case in point

Ahead of the 2015 elections UKIP has polled as high as 18% and as low as 8%. Without knowing the final result of the election it's difficult to explain the difference, but nonetheless it's there. There are several possible reasons for this.

Firstly, polls conducted online appear to favour UKIP more than those conducted by phone. Some poll questions also appear to have 'prompted' a UKIP response by explicitly including their name in the poll rather than letting the interviewee state 'Other' when asking which party they intended to vote for. There is some evidence this boosts UKIP's share, but other pollsters have found that prompting small parties over-inflates their numbers.

Getting the clearest picture

With such varying results between polls, one popular measure is what is known as a 'Poll of Polls'. These attempt to compensate for the various biases among the different results and offer an 'average' result from across selected polls.

When reading polls for yourself, be sure to look at as much of the data as possible, including consulting the polls of polls and checking the results of other polls for trends, not just between polling companies but also over time.

But perhaps most important of all, observe Twyman's law: "if a poll is interesting, it is probably wrong".

Full Fact fights bad information

Bad information ruins lives. It promotes hate, damages people’s health, and hurts democracy. You deserve better.