Back to Programme

Adding 2 and 2 to get 5? How Far has Recent Reporting of Election Polls Conveyed Conclusions that Could be Justified by the Data?

Roger Mortimore (Ipsos MORI / King's College London)
Sir Robert Worcester (Ipsos MORI)
Mark Gill (MORI Caribbean / King's College London)

Keywords: Media, polling and electoral behavior

Abstract

After several recent prominent elections and referendums around the world, the opinion polls have been attacked for inaccuracy and for giving a misleading impression of the state of play to politicians and to voters. Yet the impression that the polls give depends very largely on the way their findings are reported by the media, and in some of these cases the fault has been much more in the reporting than in the polling. In the 2016 US Presidential election the final polls were broadly accurate, showing correctly Clinton’s narrow lead in the national popular vote and mostly suggesting that key swing states were too close to call, so that a Trump victory in the electoral college was perfectly possible. In the British referendum on leaving the European Union (2016), some of the polls were roughly right and some probably wrong, but more polls published during the campaign put the campaign to Leave ahead than put it behind, yet the impression seems to have been given of a consensus that a Leave vote was impossible. In the British general election of 2015, although the polls were probably wrong throughout the campaign on the relative vote shares of the biggest parties, and the reporting of the polls’ findings gave a misleading impression of the likely eventual outcome in parliamentary seats, the two errors did not directly coincide: over-confident translation of vote share standings into seat projections would probably still have led to the polls being attacked for inaccurately predicting the election outcome even had their vote share measurements been considerably more accurate than they were. In this case, the standing of the parties as read from the polls became in itself a substantive political issue, since the Conservative Party made the risk of the Scottish National Party exerting influence in a ‘hung’ parliament a central plank of their election campaign. We explore the accuracy of the final and campaign polls in these elections and of the impressions of their findings conveyed by media and other reporting, considering in particular the extent to which the conclusions being conveyed to the public could have been properly drawn from the polling evidence on which they were apparently based.