opinion poll

Attempt to measure public opinion by taking a survey of the views of a representative sample of the electorate; the science of opinion sampling is called psephology. Most standard polls take random samples of around a thousand voters, which give results that should be accurate to within three percentage points, 95% of the time. The first accurately sampled opinion poll was carried out by George Gallup during the US presidential election in 1936.

Opinion polls have encountered criticism on the grounds that their publication may influence the outcome of an election. Rather than simply predicting how people will vote, poll results may alter voters' intentions – for example, by establishing one party as likely to win and making the voters wish to join the winning side, or by making the lead of one party seem so great that its supporters feel they need not bother to vote. In France, opinion polls cannot be published during the final week of a presidential election campaign.

From unscientific beginnings in the United States in the early 19th century, public opinion polls began using statistical sampling techniques in the early 20th century and were established independently of their sponsor newspapers in the late 1930s in both the USA and in Britain. Since World War II and particularly during the 1960s the popularity of opinion polls has grown.

In Britain two methods of sampling opinion are practised by the polls; random and quota sampling. The former consists in selecting persons to be interviewed at random from lists of names that cover the whole of Britain, such as the electoral registers. Only the selected person is interviewed. When carefully carried out it is more accurate and reliable, but more expensive, than quota sampling, because it covers people from all walks of life and it does not allow the interviewer to choose whom to interview.

Quota sampling involves selecting a sample of people which is representative of the population in terms of certain criteria such as age, occupational group, sex, and region. The technique allows interviewers latitude in finding individuals to question, and is therefore biased towards those who are more approachable. Both techniques are sufficiently reliable to indicate broad trends, within sampling error. But both methods are fallible if the questions asked are either so loaded or obviously biased as to suggest a particular answer to them.

When forecasting general election results the same provisos apply, although it is vital to test opinion until the last possible moment in order to catch any changes that may occur just before election day itself. In recent years it has been suggested that the publication of polls, both during elections and at other times, can influence both the way people vote and the views of their political representatives. There is no evidence, however, to suggest either of these contentions is true.

Exit polls, taken on polling day, are increasingly used, but the accuracy of their results varies. During the 1992 UK general election they were misleading, but during the 1997 general election they gave a good forecast of the eventual result.

The techniques used in polling are most commonly used in the sphere of commercial market research and all the polling companies do more market research than opinion polling.