The 2015 election has seen more constituency-level opinion polls than any previous election, most of which have been administered by Lord Ashcroft. In this post, Tom Lubbock examines the differences in reported vote intention when those polled are asked to think about the candidates in their specific constituency, rather than a more general question. The results suggest significant differences from national polls, particularly for the Liberal Democrats.
Certain aspects of this election are quite hard to predict. If you believe the opinion polls carried out in individual constituencies you will also believe that the Liberal Democrats are going to do much better than national polls suggest (national polls show them doing very badly indeed). The Lib Dems are either going to outperform expectations on the 7th of May, or they are going to spectacularly collapse to around 15 MPs. Which of these happens has enormous significance in an election where most forecasts point to Labour and Conservatives a few seats apart.
The best models of this election lean fairly heavily on Lord Ashcroft’s constituency opinion polls. Ashcroft asks two different voting intention questions of those who pick up the phone:
- ‘If there was a general election tomorrow, which party would you vote for?’
- ‘Thinking specifically about your own parliamentary constituency at the next General Election and the candidates who are likely to stand for election to westminster there, which party’s candidate do you think you will vote for in your own constituency?’
It is the responses to this second, constituency-specific, question that are reported in the headline vote share figures. We can see from Graph 1, below, that in a lot of constituencies the vote shares derived from the second question are similar to those from the first. But in 36 of the 146 seats polled by Ashcroft so far the winning party changes between first and second questions (39 of 182 polls).
The Lib Dem vote share doesn’t fit the general pattern. Their vote increases dramatically when the second question is asked — in theory — when voters think about the local contest. There is actually no Lib Dem vote share over 10% that does not increase when the constituency-specific question is asked. On average the Lib Dems jump 10.3 percentage points from one question to the next.
Is this good news for the Ashcroft polls and all the forecasts that rely on them? Are they picking up incumbency effects and tactical voting that the national polls, which shape so much debate, miss out? Maybe, but despite the common sense arguments for asking a question about the specific constituency in which a voter lives we should be sceptical. The first question might be cementing respondents’ answer to the second question or (working in the opposite direction) voters might be switching their choice after being asked a second similar question believing that they are being prompted for a different answer. Both these are examples of priming. These priming effects are the bread and butter of part of the political science literature (or Yes Minister sketches), and switching in particular is a well known feature of criminal evidence.
Graph 2, above, shows how much total change in vote share takes place between question 1 to question 2 in different types of seat. Voters in seats that the Lib Dems are defending, whether against Conservatives or Labour, are much more likely to change their voting intention between the two questions than voters in the other three categories of seat. Within those Lib Dem marginals there is also a good deal more variation than in other types of seat, about 30 percentage points. The striking differences offer initial reassurance about the constituency-specific second question.
The fact that voters in Con/Lab, SNP/Lab and UKIP target seats don’t respond to the constituency-specific question in the same way as voters in Lib Dem seats is a good sign that there is not a blanket priming at work here. If switching or cementing were at work we would expect them to act in all seats. Looking at the same data another way in Graph 3, below, we can see that this is not the case. Even if we suppose that voters who have switched their answer to the first question since 2010 might be more susceptible to priming this also ought to be true of large numbers of UKIP and SNP voters. Again, the absence of this effect is good evidence that the Ashcroft polls are picking up genuinely local voting behaviour.
As we move into the short campaign constituency polls will start to name candidates and should converge towards the actual election result. Come the 8th of May we will have a much better idea of the efficacy of the question ordering in these Ashcroft polls, but we also await experiments conducted by the British Election Study and other pollsters to give us a better idea of the best way to poll individual constituencies.