Pre-election polls not becoming less reliable: study


Paris (AFP)

Pre-election polls are not becoming less reliable, said a study Monday addressing public distrust stemming from surprise results in Britain's 2015 general election and the 2016 US presidential vote.

To investigate whether suspicion is warranted, a duo from the universities of Southampton and Texas analysed over 30,000 national voter intention surveys conducted for 351 general elections in 45 countries from 1942 to 2017.

"We find that, contrary to conventional wisdom, the recent performance of polls has not been outside the ordinary," Will Jennings and Christopher Wlezien wrote in the journal Nature Human Behaviour.

"Ultimately, although the polling industry faces a range of substantial challenges, we find no evidence to support the claims of a crisis in the accuracy of polling."

The error margin in surveys used to predict election outcomes has remained more or less consistent over the past 75 years, the researchers said.

The magnitude of the poll error in both the US and the UK elections "was not especially unusual," they added.

The work was prompted by questions raised after voters in Britain unexpectedly gave the Conservative Party a majority in 2015 when a hung parliament had been predicted.

In the United States, most pollsters had predicted a win for Hillary Clinton in the 2016 election, which she lost to Donald Trump.

There are good reasons to fear that polling errors may have increased, the duo said.

These include that response rates to surveys using well-established methods such as face-to-face and telephone interviews have declined -- potentially jeopardising representativeness.

At the same time, cheaper, easier and possibly less rigorous methods such as online polling have become more popular.

But the team said pollsters have adapted to these changes, introducing weighting and other techniques to boost accuracy.

Weighting involves an adjustment of poll data to ensure that a sample of participants accurately reflects the population from which they are drawn.

The researchers stressed they were not claiming that traditional polling methods are working just as well as before, or that polling has not changed.

"Rather, we infer, based on our evidence, that survey organisations have adapted to the well-known challenges they face."

It is inevitable that pollsters will "get it wrong" from time to time, and subjected to scrutiny, added the team.

"Indeed, this can lead to methodological reflection and innovation, and in turn improvements in polling."