For those who have been following the BES report that concludes that there was no “youthquake” at GE17 the former President of YouGov, Peter Kellner, makes some controversial observations in Prospect casting doubt on the core conclusion that has made the headlines.
“…their (BES) latest pronouncement goes way beyond what their data can support. They base their analysis on two post-election, face-to-face surveys after the 2015 and 2017 general elections. Their sample size in 2015 was 2,987; in 2017 it was 2,194. These are larger samples than in most individual polls conducted for the media—though some research reported by the media involved far more people (such as the 50,000 polled weekly by YouGov, which formed the basis of their prediction of a hung parliament, and their indication that the Conservatives were in trouble in Canterbury and Kensington).
Where the BES team skate on thin ice is when they seek to draw precise conclusions from small sub-groups. They derive their main conclusion from the 1,400 respondents that they have crossed-checked against the electoral register, to confirm whether those who say they voted actually did so. This is a valuable exercise which, by definition, campaign polls cannot do, because people have not yet voted (or abstained). Even doing so after the election is expensive and time-consuming. So, congratulations BES, for doing this.
It is the figures for the under 25s that have caused such a stir. The figures for all the other age groups are broadly in line with what pollsters reported months ago.
Here’s the problem. BES interviewers questioned, and confirmed the turnout answers, of only 157 under 25s in 2015 and 109 in 2017.
Data obtained from such small subsamples are subject to large margins of error. The normal formula indicates that the reported turnout for this group could have been eight points adrift of reality in 2015 and almost ten points adrift in 2017. Applying those figures to the BES data, we may deduce that the correct figure for the turnout of the under 25s was 41-57 per cent in 2015 and 34-53 per cent in 2017. (Technically, we would expect the true figure in each election to be within those wide ranges 19 times of out 20; but one time in twenty, a perfectly well conducted survey would be beyond even these limits.)
On those figures, we can say nothing sure about the change in turnout among under 25s in these two elections. (By the by, I am amazed that the BES report data with a near-10 point margin of error to a decimal place. This purported precision is utterly spurious.)
In fact, the true margins of error are greater than that—though how much greater is impossible to calculate. The formula used above assumes a perfectly designed sample with a 100 per cent response rate. BES’s sample design was fine; but its response rate was below 50 per cent. Nobody can be sure whether the voters they did not reach behaved like the voters they did manage to interview…”
The full Kellner article is well worth reading. I think he is right to raise these questions before the “no GE17 youthquake” narrative takes hold.