Call for Evidence

More opinion polls than ever have been seen in recent elections, but their greater frequency has not been matched by greater accuracy. In the last seven general elections pollsters have got the result wrong three times. The committee sought to understand the impact of polls on voters and politicians, and their influence on politics and how we are governed. The role of media coverage of polls and polls carried out for interest groups, was also of interest. A Call for Evidence was made on 20 July 2017.

Summary of the Turing’s submission

The submission suggests that a possible way of improving polling methods is to encourage open access to raw poll results and population sample sizes, as well as third-party analytics. With regards to media, the suggestion is made that media analysis of polls would be more useful if it focused more strongly on quality of information and intensity of preference.

Social media naturally also has an impact on public engagement with political opinion polling, the accuracy of polling, and how seriously and ‘rationally’ we assess choices. Social media platforms such as Facebook can affect the moods of users and patterns of voter turnout, which could alter election outcomes.

However, using social media to predict elections is fraught with difficulty. The impacts of vulnerabilities in digital and social media-based polling are not well understood, and are difficult to detect and correct for. Any predictive models need to be tested through multiple elections in different countries and languages, to test their performance and generalisability. Models also can’t tell policy makers why a political shift occurred, only that it did. However, it is suggested that future advances in machine learning and natural language processing may help remedy this.

Additional information

Adrian Weller, Turing Fellow, University of Cambridge

Rob Procter, Turing Fellow, University of Warwick
Maria Liakata, Turing Fellow, University of Warwick
Jonathan Cave, Department of Economics, University of Warwick
Adam Tsakalidis, PhD candidate, University of Warwick

Turing affiliated authors