BPC Comments on PRCA enquiry into political polling and predictions

The British Polling Council welcomes the interest of the Public Relations and Communications Association (PRCA) in the work of the polling industry, with its announcement that it is to hold an enquiry that, according to PR Week, ‘will address trust and confidence in political predictions following the May 2015 General Election, the UK referendum result in June, and the US Presidential Election result earlier this month’.

The BPC would like to highlight the extensive investigations that have and continue to be made into the performance of the polls. The evidence and conclusions of these reviews will, the BPC believes, provide a valuable source of material to assist the PRCA in its enquiry. These reviews set out the challenges facing the opinion polling industry, covering both what they are getting right as well as the premise as to whether (and where) the ‘pollsters got it wrong’ in the Brexit referendum and the US Presidential election.

Following the undoubted underestimation of the Conservative vote and over-estimate of the Labour vote at the 2015 general election, the BPC, along with the Market Research Society (MRS), sponsored a wholly independent inquiry under the chairmanship of Prof. Patrick Sturgis. That inquiry published an extensive 115 page report and made a series of recommendations that the BPC is currently implementing. Meanwhile, the BPC and MRS are sponsoring a public seminar on the performance of the polls in the EU referendum that will be held at the Royal Statistical Society on 8th December.

On the other side of the Atlantic, the American Association of Public Opinion Research (AAPOR) announced last spring that, in line with its practice after previous presidential elections, a panel under the chairmanship of Courtney Kennedy will analyse and report on the performance of the polls in the US Presidential election. The Panel is due to report by May 2017.

Pollsters in the UK are far from complacent about their performance in recent ballots. However, it is to be hoped that the PRCA’s enquiry will add to rather than duplicate the existing attempts that are being made to assess and improve the methodology of the polling industry. The BPC would certainly welcome an opportunity to advise the enquiry of the efforts that have been made to date.

It is, of course, the case that most of the final polls conducted by BPC members pointed to a victory for Remain rather than for Leave. But in most cases they also suggested that the outcome would be very close. Meanwhile, of the 34 polls of voting intentions conducted during the four week official campaign period between 27 May and 22 June, 17 put Leave ahead, 14 Remain while three suggested the two sides were tied. The evidence from the polls did not point unambiguously to a Remain victory or the simple assertion that the pollsters ‘got it wrong’.

Meanwhile in the United States it is all too often forgotten that, as the nationwide polls anticipated, Hillary Clinton won more votes than Donald Trump. At present she leads in the nationwide vote by around two points. This compares with an average lead for Ms Clinton in the final polls as calculated by Real Clear Politics of 3.2 points. If, when the final tally is available, the nationwide polls are shown to have overestimated the Democrat candidate’s lead by just over one point, this will be lower than the average error in polls of presidential elections between 1968 and 2012 of two points.

While the record of the polls in both the EU referendum and the US Presidential election was certainly not perfect there appears to be a widespread misperception that in both cases the polls pointed to a clear winner. They did not.

We therefore hope that the findings of these reviews into the opinion polls will provide a firm evidence base for the PCRA as it starts its enquiry and the BPC welcomes the opportunity for more dialogue as the review takes shape – including the question of how the findings of opinion polls can be better communicated to the public at large.