Dr. Adrian Kavanagh:

In general, I agree with everything Mr. Donnelly said. In regard to the British election, looking at the poll of polls taken on the day of the election, the main issue was that the Labour Party's actual share of the vote turned out to be two or three percentage points lower than expected. The results for all the other parties, however, were pretty much bang on. The poll of polls predicted the UKIP, for instance, would be at 13% or 14%, which is where it finished. The Green Party, as predicted, secured about 4% of the vote. The problem was in translating those poll figures into seat numbers. A lot of people did not take account of the fact that the UK uses a first-past-the-post electoral system, which is a factor. In addition, there was the knock-on effect of the large swing to the SNP in Scotland, which was always going to result in a drastic reduction in the number of Labour Party seats. The latter would probably have had to be above the Conservative Party in percentage terms to have had a chance of winning more seats than it.

It is often the case that the main issue in regard to polls is probably the question of how to translate the data they provide into seat numbers. We have a proportional representation electoral system in this country, but that does not mean the seat outcome is exactly proportionate. For instance, until 2009, Fianna Fáil generally won a seat bonus at every local and general election. In 2011, however, the party won 17% of the vote but only 12% of the seats. There is no neat relationship between actual percentage share of the vote and seat numbers. Another issue to note is that there is a margin of error associated with polls. Most polls tend to have approximately 1,000 respondents and the margin of error is around 3%. If, say, Fine Gael is at 25%, it is feasible that one poll might have the party at 22% while another puts it at 28%. That is another factor to take into account.

In the case of the Seanad referendum, probably one of the most important issues was turnout. Where turnout is low, as is often the case in referenda, opinion polls are less reliable than would be the case in a general election where, generally, turnout levels tend to be much higher. The polls before the Seanad referendum were probably not as inaccurate as they seemed to be. The issue may have been that not enough of those who intended to vote "Yes" actually turned out on the day. In other words, it may have been that the "No" voters were simply more likely to turn out.

Mr. Donnelly made a very good point regarding constituency polls. I agree that most of our national polls are generally quite consistent and accurate, but there is a problem with constituency polls. In the case of the latter, the number of respondents is usually much lower. If, for example, the sample is only 200, the reliability of the data is reduced. The smaller the number of respondents, the greater the margin of error. In addition, there is a name recognition factor with constituency polls. I recall that one particular poll in Laois-Offaly, for example, had Brian Cowen at 33%. Candidates with greater name recognition, which is usually the candidates who are sitting Deputies, tend to have higher poll ratings. However, after an election campaign, other candidates may catch up. Constituency polls are perhaps the most problematic of all polls for the reasons we have outlined.