On November 27, 2019, the polling organisation YouGov predicted the results of the December 12 UK general election. According to their predictions, the Conservative Party would be expected to win 359 parliamentary seats, followed by 211 seats for the Labour Party, 43 for the Scottish National Party (SNP), and only 13 for the Liberal Democrats (LD). As a result, the Conservative Party would gain a working majority of 68 seats in the House of Commons.
YouGov’s results were based upon a survey of 100, 319 panellists from across Great Britain, which had been conducted during the preceding seven days. YouGov used a multi-level regression and post-stratification (MRP) model to make the predictions. The statistical technique of regression involves predicting a single dependent or outcome variable from several independent or predictor variables. In this particular instance, voter preference for a given political party is the outcome variable. The MRP model specifically allows predictions at individual constituency level to be made from nationally collected data, comprising many variables. This eliminates the need for surveys to be conducted separately in each and every parliamentary constituency.
The business of polling panders to the need for people, whether politicians, the media, or members of the general public, to be able to predict the future, presumably for planning their approach to the issues of the day. In the UK, the major polling organisations belong to the British Polling Council, which requires members to demonstrate that they use representative sampling methods and statistical weighting procedures to accurately reflect the views of the population under study, which for general election opinion polls includes all eligible voters in the country.
Opinion polls can sometimes turn out to be inaccurate, even though the accuracy of polls has been improving over time. The present YouGov poll has caused a stir, because the organisation accurately predicted the results of the June 8 2017 general election. On May 31 2017, the YouGov election model predicted 311 Conservative, 255 Labour, 51 SNP and 10 LD seats within a total of 632 parliamentary seats in Great Britain. The predictions were, however, made even closer to polling day, thereby adding to their accuracy.
Many polls have got it wrong, sometimes at considerable expense. An often-quoted example is that of a poll conducted by the American magazine Literary Digest in 1936. A mailing list of ten million names was created, all of whom received a questionnaire by post. Of the 2.4 million who responded, 57 per cent predicted a victory for Alfred Landon in the forthcoming US Presidential election. On polling day, 62 per cent of voters cast their votes in favour of Landon’s opponent, Franklin D. Roosevelt. Despite a large sample size, the pollsters got it wrong on many fronts. By choosing the people to be polled from telephone directories, club membership lists and magazine subscribers among others, they had selected a non-representative sample, consisting mostly of upper- and middle-income citizens. Furthermore, the high non-response rate meant that useful information relating to the characteristics and voting intentions of those reluctant to respond was lost to posterity.
While polls are more accurate nowadays, they are by no means perfect. A poll only tells us what those people sampled are thinking of on the day, and does not factor in the possibility that voters can change their opinions in the immediate run-up to elections. That explains why the accuracy of polls increases the closer they are held to polling day. National surveys also frequently fail to consider local issues and candidate profiles that may influence patterns of voting within individual constituencies.
All polling organisations have to make compromises, as it is very expensive and time-consuming to study larger and wider representative samples of the British public with the resources available. YouGov, an online market research and opinion polls company, was founded in May 2000 by two former Conservative parliamentary candidates, Stephan Shakespeare and Nadhim Zahawi. YouGov has a panel of over one million registered users. Surveys are restricted to sub-samples of people selected from this panel. Respondents are chosen by a process of “Active Sampling” to be representative of British adults in terms of age, gender, social class and level of education. Only those chosen by YouGov are able to fill in an online questionnaire, which is accessed by a username and password. This means that only a selected group of people with internet access can take part in the survey.
Ipsos MORI is another major British polling organisation. On October 31 2019, an Ipsos MORI poll described a 17-point Conservative lead over the Labour Party. For this survey, a representative sample of 1,001 adults in Great Britain, aged 18 and over, were interviewed by telephone between October 25 and October 28. In other words, this poll was restricted to telephone users, another selected group, despite steadily increasing numbers of phone users in the UK.
It is clear that no poll can ever be truly representative, given the resources currently available. Pollsters thus make allowances for potential inaccuracy by providing a margin of error for their results. For the Ipsos MORI poll, it was stated that there was a 9 in 10 chance that the “true value of a party’s support” was to be expected to lie within 4 points of the actual estimates made by the poll. The message is that while polls may well predict the trajectory of a nation’s voting intentions, other factors are constantly at play in the background . Seismic shifts in public attitudes and opinions shortly before general elections can provide unpleasant surprises for the unsuspecting. Exit polls and polls in “marginal” constituencies are probably more reliable predictors of voting trends.
Ashis Banerjee (voter; unswayed by opinion polls; have written a statistics book)