The declaration of the result of the European Parliament election in Scotland is of interest for two reasons so far as the referendum campaign is concerned.
First, inevitably the parties and campaigns will try to impose their own interpretation of what has happened and why it matters for the referendum debate. Second, the declaration gives us an opportunity to assess the accuracy of some of the referendum polling by seeing how close the final polls were to the actual result.
This is not the place to get too deeply involved in assessing the claims and counterclaims about the substantive implications of the result. One point though is obvious. The SNP had hoped that UKIP success in England & Wales would be accompanied by the party’s failure to win a seat in Scotland. Symbolically there seemed no better way of demonstrating that the attitudes and values of people in Scotland are very different from those of their neighbours south of the border. Alas the plan went awry when UKIP managed to win a seat north of the border – ironically, because the SNP themselves were two or three points short of what was needed to keep UKIP out. However, the No side will probably want to keep its distance from Nigel Farage and his colleagues so both campaigns may well decide that the outcome of the Euroelections is best soon forgotten.
The same, however, might not be true of its lessons for the polls. They do not emerge from the elections unscathed. Altogether seven polls of European election poling intention in Scotland were published between January and polling day. None was claimed to be a final poll designed to predict the eventual outcome, but two polls (one from ICM and one from Survation) conducted in May and published in the final week certainly set expectations of what would happen. In any event, apart from registering an increase in UKIP support during April, together with some easing of SNP support, the polls painted much the same picture throughout. There is no particular reason why we should presume that there was a ‘late swing’ after the last two polls were conducted that would explain any discrepancy between the polls and the result (and no evidence of there having been one in the more numerous British polls).
The following table compares the estimate produced those last two polls and the actual outcome:
ICM Research 12-15.5.14
Ave. Error: Average Error – the sum of the absolute value of the errors for each individual party divided by the number of parties (6)
Both polls (both of which were conducted via online panels) told much the same story, and both are characterised by much the same discrepancy – SNP support was apparently overestimated while that for the Conservatives was underestimated. The latter is a problem with which the polling industry in the UK has been grappling ever since the disaster of the 1992 UK general election in which the final polls pointed on average to a one point Labour lead when in reality the Conservatives were eight points ahead – and (ICM apart) indeed it is also evident in the UK-wide polls of Euro vote intention, which on average pointed to a 22% Conservative vote when in practice the party won 24%. It is evidently a danger against which polls have to be ever vigilant.
Overestimation of SNP strength is, however, not a persistent feature of the polls in Scotland. It was not apparent at either the 2010 UK election or the 2011 Scottish one (though contrary to popular myth the polls did not systematically underestimate SNP strength on the latter occasion either; what they underestimated was the party’s performance on the list vote, a vote that the polls always struggle to get right because of an apparent tendency for respondents to state their second preference). So we might dismiss the apparent overestimation on this occasion as an accidental misfortune.
Still, bearing in mind that many referendum polls are being conducted by different players and/or different methods than in 2010 and 2011, doubtless some will ask whether the discrepancy is a sign that some polls at least might be overestimating Yes support too. Of course Yes support is not the same as that for the SNP, and indeed the ICM poll conducted before the European election recorded a sharp drop in Yes support at the same time as little change in that for the SNP in the Euro-elections. But there is a link between the two and thus a poll that overestimates SNP support would certainly seem to be at risk of overestimating that for Yes.
On the other hand one problem that faces any pollster estimating party support in the Euro-elections is that most people do not vote, a problem that is not thought will bedevil the referendum. The polls certainly struggled to estimate turnout in European elections. According to ICM 46% were certain to vote, while the equivalent figure in Survation’s poll was 55%; the actual turnout was just 34%. Perhaps in the event potential SNP supporters were more likely to have stayed at home – or did indeed have a last minute change of mind. Even so, we have all been given a timely reminder of the potential limitations of polls.