BPC/MRS enquiry into election polling 2015: Ipsos MORI response and perspective

This Forum article considers the unsatisfactory results of pre-election opinion polling in the 2015 British general election and the BPC/MRS enquiry report into polling by Sturgis et al., providing a response from Ipsos MORI and associated researchers at King’s College London and Cranfield Universities Whilst Sturgis et al. (2016) consider how to perfect opinion poll forecasting, why the 2015 prediction was inaccurate when the same methodology returned satisfactory results in 2005 and 2010 at Ipsos MORI is considered here instead We agree with Sturgis et al. that the inaccurate results were not due to late swing or the ‘shy Tory’ problem and with Taylor (2016) that the underlying problem is a response rate bias However, Sturgis et al. critique pollsters in their report for systematically under-representing Conservative voters but the Ipsos MORI final poll had too many Conservatives, too many Labour voters and not enough nonvoters The Sturgis et al. conclusion is convincing that the politically disengaged were under-represented due to quotas and weighting mechanisms designed to correct for response bias Nevertheless, for Ipsos MORI, this explanation does not account for why the polling methodology was inaccurate in 2015 when it had performed accurately in 2005 and 2010 For Ipsos MORI, a more likely explanation is that Labour voters in 2015 became more prone to exaggerate their voting likelihood We offer various postulations on why this might have been so, concluding that to account for the inaccuracy requires a two-fold response, to improve: (i) sample representativeness and (ii) the projection of voting behaviour from the data Unfortunately, the BPC/MRS report offers no blueprint for how to solve the problem of sampling the politically disengaged Whilst Ipsos MORI have redesigned their quotas to take account of education levels, to represent those better with no formal educational qualifications and reduce overrepresentation of graduates, polling in the referendum on EU membership suggests that the problem of drawing a representative sample has been solved but difficulties in how best to allow for turnout persist.