快猫短视频

Being a pollster is an unenviable job. Most polling firms failed to predict the results of the last two UK general elections, the Brexit referendum and the most recent US presidential election, leading many to declare an industry-wide crisis. But despite these high-profile blunders, a new analysis says polling accuracy is actually the same as it has always been.

Polling errors are no worse now than they were 75 years ago. In fact, they have barely changed at all. Across 30,000 national polls from 351 general elections in 45 countries, errors were only 2 per cent on average. 鈥淭here just isn鈥檛 a crisis in polling accuracy,鈥� says co-author at the University of Southampton, UK.

However, one might expect there to have been a massive improvement, given that nowadays, fancy algorithms, huge collections of internet surveys and super-fast computers are all involved in making predictions.

Less to go on

There are at least two potential explanations for this lack of improvement. The first is that although statistical methods have got better, it has become harder to conduct polls. Response rates to telephone surveys have fallen through the floor, electorates seem to be more volatile and demographics just aren鈥檛 as predictive as they used to be. 鈥淧ollsters have become more and more sophisticated, just as it鈥檚 become more challenging,鈥� says Jennings.

In last year鈥檚 US presidential election, there were to draw from than in 2012. Many state newspapers that previously ran their own polls had shut down or no longer had the money for costly surveys. This explains in part why many pollsters correctly predicted the overall popular vote 鈥� Clinton to beat Trump 鈥� but not the result in key states, meaning they missed Trump鈥檚 victory.

The second potential reason is a little more worrying for those in the industry: we may have reached peak polling. The majority of polls are conducted using the same principle, where you ask a representative sample of people how they will vote and then extrapolate from this to guess how a larger group, like a constituency or country, will vote.

Sweet numbers

It鈥檚 like trying to find out the proportion of different coloured sweets in a jar by just looking at just a few, says Marcus Roberts at UK-based polling firm YouGov. Checking more and more sweets, or polling ever larger numbers, gives statistically diminishing returns. 鈥淭here are better sampling techniques and you can ask more people, but fundamentally you鈥檙e still pulling sweets from a jar,鈥� says Roberts.

Despite the new findings, recent high-profile outliers have caused a bit of a shake-up for pollsters. Alongside their regular polling methods, many firms tried experimental techniques for the 2017 UK general election. Some worked and others didn鈥檛, but knowing whether any will stand the test of time is difficult.

Every year there are tens to hundreds of polls, but elections are much rarer. This makes it hard to know if any particular polling technique is a dud or a one hit wonder. People are also unpredictable and often surprising, but perhaps a perfect polling method would be able to capture that too. 鈥淭hat things surprise us doesn鈥檛 mean pollsters shouldn鈥檛 be able to capture people鈥檚 intentions,鈥� says Jennings.