At the ballot box: Polling and democracy in 2024

x Sponsored content on Research Live and in Impact magazine is editorially independent.
Find out more about advertising and sponsorship.

From Brussels, to London, to Washington DC, to New Delhi, Kyiv, Islamabad, Jakarta and Taipei, 2024 is shaping up to be a huge year for democracy. Almost half of the world’s population are living in countries with elections this year, and around four billion people will be eligible to take part.

For the polling industry, the cataclysms of the poorly predicted results of the 2015 UK general election, 2016 US presidential election and Brexit referendum forced the industry to re-examine, recalibrate and reconvene for future elections. In 2015, the Market Research Society and the British Polling Council launched an inquiry into the UK general election polling deficiencies. The report, which was released the following year, suggested a failure to reach enough Conservative voters was to blame.

Since then, the polling industry has taken steps to address its overcounting of some sections of the population. However, the challenges for this year include not only those historical issues of voter representation, but also current fears about disinformation and future complications stemming from artificial intelligence (AI).

“Polling companies almost always conduct post-mortems and work out whether their weighting schemes or modelling decisions could have been better,” explains Chris Hanretty, professor of politics at Royal Holloway, University of London. “That’s helpful, and can identify improvements, but, sometimes, you end up fighting the last war and worrying about things that went wrong in the last election that won’t be an issue this time around. At the same time, some new issues come up.”

Jane Green, director of the Nuffield Politics Research Centre at the University of Oxford and president of the British Polling Council, says that some challenges cannot simply be addressed by changes in sampling.

“What’s key here is differentiating between election upsets that are down to data and election surprises that are down to interpretation, narrative-building, prediction,” says Green. “It’s all the more reason for great care in how we talk about uncertainty and the complexities of public opinion.”

The key is to make sure any lessons are learned quickly. “We should always bear in mind that, in many ways, people may not know for certain how they are going to vote right up until election day itself,” adds Kelly Beaver, chief executive, UK and Ireland at Ipsos. “One of the key challenges for understanding voting intention is overcoming social desirability bias, where respondents may be tempted to say what they think we want to hear, or what might be ‘trendy’, rather than what they truly believe.”

Risk and reward

How do we better predict elections? “Important insights come when we can use survey data to reveal preferences or corners of the electorate we didn’t know about before,” says Green. “In our British Election Study work, for example, I’m fascinated by the limitations of ‘income’ as a variable, and how we learn so much more in understanding people’s feelings of economic security. In our separate recent intergenerational survey, we identified older voters who have younger family members who are struggling financially, and how that motivates greater support for policies that help younger generations.”

Technology can also help. “Tools that promote representative sample collection and encourage active, attentive respondent participation are some of the fundamental technologies being leveraged to improve accuracy,” says Alexander Podkul, senior director of research science at Morning Consult.

But, he cautions, there needs to be a careful calibration that is “not always fighting  the last battle and not overfitting to recent historical data”. He adds: “Are we contacting and representing respondents from different backgrounds, demographic groups and political persuasions? Being able to understand the population and use that insight in targeting respondents and incentivising responses is necessary for collecting representative, high-quality interviews.”

Social media intelligence, online passive monitoring, biometrics, neuroscience and digital ethnography are all part of the modern polling arsenal. There is also a potential role for AI. The technology is particularly useful for generating automated insights from large data sources, but it could also be used to detect meaningful patterns in tracking data, prepare efficient and effective survey instruments, and prevent survey fraud, says Podkul.

The rise of large-language models (LLMs) represents an interesting turn in how we can measure public opinion, as they can be used to synthesise and distil open-ended text. James Crouch, head of policy and public affairs research at Opinium, says: “This would mean being able to quantify open-text responses much more quickly and without human bias or error. It allows for not just creative new data analysis, but also for new ways to ask questions.

“Pollsters can ask for people’s opinions with much less prompting and this allows for more off-the-cuff responses that may more accurately measure public opinion. It is close to some of the coding that many pollsters already do, but allows for more flexibility on a practical level.”

He sees a positive role for LLMs in freeing up researchers for the types of analysis and work that is most in demand from policy-makers and political parties. “They [LLMs] can speed up efficiency on some of the less glamorous jobs, such as pulling data together over time, meaning that pollsters can get a sense of long-term trends much more quickly,” Crouch adds. “Improved efficiency has a knock-on effect for what it actually means to do polling, as it may free up time from operational tasks for spending more time doing in-depth data analysis.”

Then there is the added challenge of disinformation. “Disinformation is, by its nature, a symptom of a highly connected and fast-thinking society. To see how far it has already spread, we can pinpoint who has probably been targeted by disinformation and quantify the scale of the problem,” explains Crouch. Polling companies can use insight about which groups are most hit by which pieces of fake news to inform strategies to tackle the problem, he adds.

At the time of going to print, a date had not yet been set for the UK general election, and the impact of AI and disinformation remains to be seen. Keeping up with an ever-shifting voter landscape will be the challenge shaping 2024 for pollsters everywhere.

This article was first published in the April 2024 issue of Impact.