選挙前の世論調査は当てにならないか？

Contrary to popular belief, the recent performance of election polls has not been outside of the ordinary; if anything, average polling errors are getting smaller, not bigger, reports a study published online this week in Nature Human Behaviour.

Pollsters came under fire in the wakes of both the 2015 UK general election and the 2016 US presidential election, amid widespread concern that the industry is in crisis and that polls are becoming increasingly unreliable.

This concern is not justified, report Will Jennings and Christopher Wlezien, after analysing over 30,000 national polls from 351 general elections in 45 countries between 1942 and 2017. They found that polling errors have remained more or less the same over the past several decades: they miss the mark by about 2% on average. In fact, the authors found that there has been a small - but statistically significant - increase in the accuracy of polls over time.

The authors do not argue that election polls are without error and they acknowledge that there have been polls that missed the mark badly. Their results do show, however, that there is no evidence to support claims that there is a crisis in polling accuracy. Rather, the researchers suggest that the polling industry has adapted well to the challenges it has faced (such as the dramatic decline in survey response rates across the last twenty years).