Thursday, July 05, 2018

What People Fail To Understand About Opinion Polls

This summer I have been reading Factfulness by Hans Rosling. In this best selling book Rosling sets out ten reasons why we are wrong about the world and pursues his mission to educate us about facts and statistics. This resonated with me when I read an article on the limitations of opinion polls. People simply fail to understand the margins of error on opinion polls and the potential impact.

Polls in the UK general often have a sample size of 1,200 to 2,200 people. For polls with 1,200 people the margin of error is roughly 3 percent, for polls with 2,200 people the margin of error is closer to 2 per cent. This margin applies equally to each party's vote share.

Thus a poll of 1,200 people which gave both Labour and the Conservatives 35 per cent would still be statistically accurate for either of the following results:

Labour 38%
Conservatives 32%

Labour 32%
Conservatives 38%

Reporting on a study by Will Jennings and Christopher Wlezien, The Guardian says polls are as accurate as they have ever been.’ This is true but the study still found an average polling error of 2.6 percentage points for polls a week before election day. For presidential polls 6 months before an election the same researchers found an average margin of error of 5.4%.

Given the closeness of party polling currently, and on issues such as Brexit, people need to be very cognisant of margins of error.