The Problem with Polls: How can you predict behaviour in today’s fragmented society?

Share on facebook
Share on twitter
Share on linkedin

In recent years polls have become increasingly ineffective at predicting the views of the populace. YouGov admitted that “We were, like most people, surprised at the outcome of the 2016 U.S. elections.”

Part of the problem are the antiquated methodologies used; they suffer from biases in sample selection, over-generalisations, and in an age of big data, all have a limited sample size.

We live in a digital world where a significant proportion of people publish their opinions openly on social media. While the world reeled from the results of the EU referendum and the US general election and traditional polling fell short, analysis of social data predicted both outcomes.

Could history be repeating itself with the by-elections in Copeland and Stoke-on-Trent today? Journalists are putting their money on a Labour victory in Copeland, Labour sources we have spoken to are convinced of a victory in Stoke.

However, analysis across social media conversations in recent days has seen a huge upswing in pro-UKIP sentiment concerning the Stoke by-election both country-wide and when the West-midlands are isolated.


Surge in pro-UKIP sentiment since February 18th

Meanwhile, in Copeland, conversations that favour a Conservative win have overtaken those that are positive towards Labour, with a similar right-wing surge over the last few days of the campaign:

This late surge in online conversations against the established political narrative is exactly what we saw with the Brexit vote in the UK and Trump’s victory in America, as voters go to the polls today it will be interesting to see if the voters in Stoke and Copeland have more surprises in store.

If they do, it will be another blow to the polling industry and another challenge to politicians, governments and brands on how to try and understand an increasingly fragmented public.

Why do Polls no longer work?

A magnitude of reasons explain the growing inefficacy of polls. The primary reason being sample selection bias, the people questioned are no longer representative of the population. According to political scientist and former president of the American Association for Public Opinion Research, Cliff Zukin, this is a result of two factors: “the growth of cell phones and the decline in people willing to answer surveys”. Surveys rely on automated calls to landlines, in the US, in 2014 around 50% of households did not have a working landline. Moreover, those with a landline are older and more affluent, thus not representative of the wider population. Polls use weighting to get around this, often involving generalizing minority views, and not accounting for the possibility of diverse views within a minority group.

Not only is there sample selection bias through the use of landlines, but self-selection bias occurs because those most likely to answer the phone and agree to take part tend to be more liberal.

Exit polls in the states and the UK are notoriously inaccurate; often predicting a more liberal outcome than an actual election or referendum result. This may be due to liberals being more open and willing to talk about their feelings. Furthermore, even when you have someone on the phone who has agreed to be interviewed, people are easily influenced, the style of questioning can sway responses and people rarely give answers that reflect their true beliefs when put on the spot.

Some further reading:


To find out more about how onefourzero’s digital analysis and insights can help you identify opportunities for growth and potential risks, contact