Menu
Fri, 29 March 2024

Newsletter sign-up

Subscribe now
The House Live All
Health
Health
Press releases

University of Exeter research could lead to more accurate political polling

University of Exeter

2 min read Partner content

In light of pollsters’ failure to predict the 2015 General Election result, the University of Exeter is training computers to correct ‘misreporting’ by cross-examining responses from a larger pool of online surveys.


The work of University of Exeter experts could help to bring about more accurate political polling following the widespread failure of surveys to predict the 2015 General Election result.

Pollsters have long struggled to get accurate results because the public do not answer questions honestly; respondents are more likely to admit to socially-desirable traits and behaviours and under report those which they think are less desirable. This ‘misreporting’ can lead to inaccurate and even misleading conclusions about people’s opinions and, in the case of cross-national research, undermine the validity of comparisons.

Dr Gabriel Katz, from the University’s Politics department, is working to discover if using more data and ‘machine learning’ - training computers to examine information from several different surveys - can correct misreporting and measurement error.

Dr Katz said: “We live in a time of growing public disenchantment from politics, and yet because of online surveys we have more information than ever before about their opinions, behaviour and characteristics.”

Online surveys are becoming increasingly popular because they are easy, fast and cost-effective. Dr Katz wants to find out if this data can be contrasted with the information from other surveys, where respondents are selected, to see if this can help correct errors between what people report thinking, and what they actually think.

“I hope this method will allow me to take full advantage of this paralleled amounts of information,” he added. “Of course internet surveys are usually poor substitutes for surveys where respondents are selected to be represented and chosen at random, but they can be a useful source of information to correct for misreporting in cross-national studies.”

Data from online surveys will be obtained from Kieskompas, a polling company which conducts web-based self-administered surveys. This data will be used to train the machine learning algorithms, and to classify responses in the cross-national probability samples into ‘accurate’ and ‘misreport’ categories, which will then be used to make statistical inferences that take into account and adjust for the possibility of survey misreporting.

PoliticsHome Newsletters

Get the inside track on what MPs and Peers are talking about. Sign up to The House's morning email for the latest insight and reaction from Parliamentarians, policy-makers and organisations.

Categories

Political parties