One of the suggestions that a number of users of this site have made is that we should supply a poll of polls of referendum vote intention. Our blogs often contain a reference to the average vote share for Yes and No in recent polls, but until now we have eschewed running such a statistic as a regular highlighted feature.
There is one simple reason for this. As regular readers will be aware, one of the features of polling in the referendum is that different companies consistently produce different results. This raises some difficult questions about how best to calculate any poll of polls.
Given the diversity of poll results, one option would be take the average of the last poll conducted by each company that is polling regularly in the referendum. But some companies are polling more often than others. Ipsos MORI, for example, have not polled since the end of February, nearly three months ago. By now that poll might be thought to be out of date – and the whole point of a poll of polls is try to summarise where public opinion seems to stand now. Moreover, if a new pollster were to enter the fray we would have to change how the figure was being calculated in mid-stream.
An alternative approach is to take, irrespective of who conducted them, the average of the last N polls. Here N is some number that ideally is big enough to produce a reasonably robust figure based on the findings of a variety of pollsters, but not so big that we include polls that were conducted quite a while ago. Exactly what N should be is inevitably a matter of judgement.
There is, though, a downside to this approach. If those pollsters that tend to produce a relatively high Yes vote conduct more polls than those who produce a relatively low one, the poll of polls figure for Yes could well be pushed up. Last month, for example, two of the pollsters that tend to produce relatively high Yes votes, Panelbase and Survation, both conducted not one but two polls. No less than five of April’s six polls came from pollsters who tend to produce a high Yes vote.
Despite that risk, we have decided to adopt the second approach. Since Christmas six pollsters have been conducting polls, and so we have set ‘N’ to six. Our poll of polls will simply show the average Yes and No vote – once the Don’t Knows have been excluded from the denominator – in the six most recent polls. Every time a new poll comes out it will be included in the calculation, and the oldest of the polls in the previous calculation will drop out.
As well as showing the latest poll of polls we will also show how it has evolved over time. To that end, we have calculated the poll of polls all the way back to last autumn, and the results are shown in a line graph underneath the latest poll of polls. The date assigned to each poll of polls is simply the date on which the most recent poll finished interviewing.
Our first poll of polls, which is based on polls conducted up to 28 April, is No 55%, Yes 45%. The No lead is narrower than at any time since last autumn. There has been a clear and consistent trend in recent months towards a narrower No lead.
But here comes the health warning. As we have already noted, those pollsters that tend to produce a relatively high Yes vote have been more active recently – and this has helped push up the Yes figure in our current poll of polls. It may well fall back again if and when some of the less optimistic pollsters for the Yes side publish more polls.
The poll of polls is best regarded as a way of quickly getting a broad impression of the direction in which public opinion has moved over the long term. But only get excited about a shift when it proves to be sustained over a number of readings – and certainly only after checking which companies have provided them.