Last year I wrote an article on How Not To Report Opinion Polls. It included advice on ignoring small cross breaks, margins of error and not cherry-picking. That is, if there is a long data series with lots of noise and random error, don’t pick out the one random outlier than supports your case and ignore the rest.
There is a classic example in the Guardian today. John Harris writes about polling of young people and says they are voting Tory. He writes: “One recent YouGov poll put support for the Tories among the 18-24s at 31%, with Labour trailing at 27%. By way of a contrast, Tory support among those aged 40-59 was at 29%, with Labour on 40%. In other words, the time-worn wisdom about politics and the young may be in the process of being turned on its head.”
Well, yes, one recent YouGov poll showed that. This one. However, other YouGov poll this month have tended to show Labour leads amongst young people and the Conservatives doing better amongst older people, a far more normal pattern. The poll the Guardian linked to was not typical of recent polling. On average YouGov’s daily sample contains around about 150 people under 25, about a third of which say don’t know or that they wouldn’t vote. This means the daily voting break for under 25s is based on about a hundred people, and has a margin of error of plus or minus 10 points. In other words, if a party actually had a lead of around about 8 points amongst young people, then random error alone will spit out polls showing leads of between plus 28 and minus 12. You can’t just take one out of context that happens to show figures you like.
Taking an average across the whole of June so far YouGov’s crossbreak for under 25s has the Conservatives on 31%, Labour on 38% – a significantly higher level of Labour support. Even that needs some caveating though. Opinion polls are weighted to be representative of the country as a whole, they are not necessary weighted so that the crossbreaks are internally representative. For example, overall there will be the correct number of people with a C2 social class, but there may be too few old people who are C2 and too few young people, or whatever. In theory this should even out over time, but there are no guarantees.
If you really want to know about the views of a particular sub-sample of the British population you need polling specificially aimed at them. Luckily enough, the Sun commissioned a specific YouGov poll of young people earlier this month, which was specificially weighted on things like education and employment status and level of educational qualification. It didn’t ask voting intention, but it did ask young people which party they thought best reflected their views – the results were 23% Labour, 12% Conservative, 7% Lib Dem, 7% Green, 6% UKIP, 39% none or don’t know.
In short, all the other findings that John Harris writes about on social and economic issues are fine (and are largely drawn from MORI’s generational data based on very large aggregate samples), but the idea that the Conservatives are suddenly the leading party amongst young people is really not true.