Poll of Polls

Feb 05 2016

We have had about 25 polls published each year since the last election and as we now build up to the next election we could come close to reaching that figure just in the next two months. There will be a lot of information but how can we make sense of it?

By Michael Marsh, Emeritus Professor, Trinity College Dublin

All poll findings should of course be treated with some caution. There is the expectation of sampling error in any random sample, which is often said to be plus/minus 3 per cent, but is actually smaller than that for any party getting much under 50 per cent of the vote, and is very small - more like plus/minus 1 per cent - when estimates are much less than 10 per cent. Against this, it is likely that the real margin of error is higher because of the nature of most sampling. There is also variation between polls in how the data are weighted to conform to a representative sample, not least because many survey respondents do not vote at all when Election Day comes around. This makes it unwise to conclude that there has been a change when polls from different companies in successive weeks give different results. Polls can also be influenced strongly by events, which can have a dramatic, but ultimately very short-lived impact on opinion. One solution to all of this variation is to conduct a poll of polls, a calculation based on a set of polls, to provide us with a safer estimate of where public opinion is most likely to be at any given point. This is what I do here. It must be admitted that this sort of exercise will not help when all or even most polls suffer not just from random error but also from some kind of systematic error, as was the case in the last UK election. I will write more about this in a later article, exploring the evidence for explanations of why the polls got it wrong in the UK and assessing the implication of this for our own election. For now, I will just assume polling error is random.

Poll of Polls: Mid-December 2015

Taking all the published polls carried out by Behaviour and Attitudes, IPSOS mrbi, Millward Brown and Red C since the last election I have essentially calculated a moving average, weighting each poll according to how far it was from each time point. This gives us a fairly smooth trend for each party across the series of polls. I have done this calculation for the four main parties and for the general group of Others. (To date there is little information on the various elements of Others, but most companies are now trying to measure support for these elements and it will be possible later to get some better estimates of each of them.) The graphs below show changing support for FG, FF, SF and Labour as well as Others. However, as of mid-December, when the most recent poll was carried out, the poll of polls suggests party support as in Table 1. FG leads at over 30 per cent, FF and SF share second place 19 per cent and Labour is on 8 per cent; 23 per cent opt for Independents or other parties.

Poll of Polls.

Looking at the profile of each party in Figures 1-5 we can see that for some this is an improvement on the recent past while for others it is part of a downward trend. Brian Hayes, FG Director of Elections made a remark recently along the lines of 'the trend is your friend'. In the case of FG, this seems to be so. The party has been rising in popularity from the start of 2015, when it stood at just 24 per cent, and should this trend continue for another couple of months FG could come much closer to the 36 per cent it won in 2011 (indicated by the horizontal dotted line) than seemed possible this time last year. However, FG's coalition partner, the Labour Party seems to be showing little sign of movement, as it has been stuck at 8 per cent for the last 9 months. FF and SF also show relatively little change in recent times. Although FF's support went up to 25 per cent in 2013 it has not been above 20 in the last year. SF also seems to have passed its peak (22 per cent in mid 2015) and has been at 19 per cent since August 2015. The real contrast to FG's rise is the decline in support for Others, which was at 27 per cent for a long time, but has declined steadily since last summer.

Combining polls in this way clarifies the real trends that the natural variation between polls often obscures. Combining polls is also helpful when we want to look at the subnational variations in party support: where is each party strong, or weak, and who tends to support them. This analysis is facilitated by the cooperation of the pollsters mentioned above and the companies who sponsor those studies (The Sunday Times, The Irish Times, Independent Newspapers, Sunday Business Post and Paddy Power) and we are most grateful to them for making their data readily available. The tables below examine these subnational variations, in terms of region, social class and age. In any one poll the number of people in each subgroup is very small, but by combining polls the sample size is much larger and so data provided are more reliable. Essentially, these tables show averages across polls, but following a further correction to make the overall distribution of party support in each poll conform to the latest Poll of Polls figures in Table 1. I have combined here only the most recent poll from each company. There is little space in this brief article to comment on these tables, but some of the more notable features seem to me to be: the difference between Dublin and elsewhere in party support, particularly for FF and Labour; the contrast in the class profiles of FG and SF; and the stark contrast between the strength of the 'old' party system among those over 65 and its weakness among those under 25.

Regional support

Using most recent poll from each company but weighted to most recent Poll of Polls party averages.

FF 14 19 21 26
FG 30 30 28 35
LAB 12 7 9 2
SF 18 22 17 20
OTH 26 21 26 17


Class support

Using most recent poll from each company but weighted to most recent Poll of Polls party averages.

Class: AB C1 C2 DE Farmers
FF 15 17 20 21 30
FG 39 32 28 22 47
LAB 11 10 5 8 3
SF 9 17 27 25 7
OTH 26 24 21 24 13

Support by age

Using most recent poll from each company but weighted to most recent Poll of Polls party averages.

Age:  Vote16 18-24 25-34 35-44 45-54 55-64 65+
FF 17 19 15 17 24 24
FG 24 26 31 32 27 38
LAB 5 7 8 8 8 9
SF 34 24 21 16 19 10
OTH 20 24 25 27 22 19


Note: 35-44, 45-54 and 55-64 categories exclude IPSOS mrbi data that has different age divisions.

Finally what does this pattern of party support mean for the composition of the next Dáil? In any Irish election there is a strong correspondence between votes won and seats obtained, but in many elections there are surprises: FG won a lot more seats than votes in 2011, and FF won a lot less. FG also did badly in 2002. To predict seats this time I have estimated the proportion of seats that has been won on average by a party winning a particular proportion of votes since 1981. There are almost no systematic differences between parties here, but Labour has tended to exceed its due share and I have made allowance for that. On the basis of the estimate for mid-December from the Poll of Polls, expected seat shares are shown in Table 5, and graphed with confidence intervals in Figure 6. These estimates do not allow for FG getting the sort of bonus it won in 2011, nor does it allow for Independents and others failing to translate votes into seats if support transfers out of the group. Either, or both, of these outcomes are more than possible in the particular conditions of 2016.

Seat estimates from Poll of Polls


Prof Michael Marsh of the Department of Political Science at Trinity College Dublin talks to Brian Dowling about his Poll of Polls analysis and the forthcoming election in a special edition of RTÉ's Road to the Election.