March 18, 2018 2:00pm by David Farrar

Are opinion polls doomed?

Scott Keeter writes in Nature:

Is election polling — and  by extension, all of survey research — really

facing a crisis? Fortunately, the answer is no. Despite  the serious challenges facing survey  research, a study by Jennings and Wlezien in Nature Human Behaviour finds that  there is no evidence that election polling is any less accurate now than it was in the past

If anything, polling may be becoming more accurate. Their conclusions rest on

a detailed analysis of a large collection of  election polls conducted across 45 countries over the past 75 years.

The elections where polls are out get lots of attention. The ones where the polls are almost spot on (like the last NZ election) get far less publicity.

Readers who follow election campaigns may find this conclusion difficult to accept.

And so it is important to note that the authors do not claim that election polling

is without error, though the mean size of the errors tends to be small (on the order

of about two percentage points, averaged across all of the public polling in a given

election).

A 2% average error isn’t too bad with a 3% margin of error.

The authors employ a database of nearly 31,000 election polls conducted in 45

countries between 1942 and 2017.

Pretty huge.