Opinion polls are having a hard time.
Due to the horserace-style coverage of polls, one might wonder if we should move political polls to the sports pages of newspapers. Polls come under fire when they fail to predict the outcome of an election or referendum. This week's US Presidential Election has certainly revived the debate about the accuracy of polls.
Evidence from Irish opinion polls since 1987, shown below, suggests that, while polls close to election day have generally reflected election results fairly accurately, earlier polls in the campaign often overestimate or underestimate support for some parties or Independents.
Additionally, past exit polls suggest that nearly half of voters may have not decided on their first-preference vote choice yet.
Therefore, it wouldn't be surprising if some parties or Independents ultimately do better or worse than current polls suggest.
Why policymakers need polls
Although parties and candidates often downplay the importance of polls, they rely heavily on them.
A single poll, though limited in scope, can give a snapshot of voting intentions or reveal public opinion on important issues like housing, immigration or crime. Tracking trends across several polls provides valuable insights into what is changing and staying the same.
Candidates commission constituency polls behind the scenes and governments frequently conduct private polls.
For instance, in Germany, the Chancellor’s Office conducted weekly opinion polls for its own use. Political scientists Anja Durovic and Tinette Schnatterer from Sciences Po Bordeaux examined more than 350 ofthese weekly polling reports.
They found that after an election, the government focused on polling questions related to its own key issues, but as the next election approached, it paid more attention to voters' main concerns. Politicians want to understand public opinion and may adjust their priorities accordingly.
Aggregating opinion polls
I maintain the Irish Polling Indicator website, originally established by Tom Louwerse (Leiden University) ten years ago.
The Polling Indicator uses a Bayesian statistical model to combine results from multiple opinion polls into a single 'poll of polls' estimate, alongside a margin of error.
This margin is influenced by sampling variation in individual polls, the number of polls available, the degree of variation in support across polls, and structural differences between polling companies.
The Polling Indicator only considers political parties that have been included in the 149 opinion polls released since the 2020 election.
New parties, like Independent Ireland, appeared in only a few polls, and small parties are usually grouped together with Independents. As a result, these parties are merged into the 'Independents/Other' category.
The figure below displays the latest Irish Polling Indicator estimates with 95% uncertainty margins, indicating the likely range of current support in terms of first-preference vote intentions.
Fine Gael is currently the most popular party with support estimated between 22% and 25%, followed by Fianna Fáil at 20.5% to 24%.
Sinn Féin's support is estimated at 17.5% to 20%, on par with Independents and other parties at 17% to 20%.
The Social Democrats are estimated to have 4% to 5.5% support, while Labour and the Green Party each range from 3.5% to 5%.
Support for Aontú and People Before Profit–Solidarity is estimated to be 1.5% and 3%.
RedC is the only pollster that explicitly lists support for Independent Ireland. According to their most recent poll, released today, Independent Ireland currently stands at 3%. RedC polls published since June 2024 have consistently placed Independent Ireland's support between 3% and 5%.
We need your consent to load this flourish contentWe use flourish to manage extra content that can set cookies on your device and collect data about your activity. Please review their details and accept them to load the content.Manage Preferences
Assessing the accuracy of polls since 1989
How can we assess how well polls have performed in the past?
Simply put, we can compare polls taken up to election day with the actual vote shares. If the election result matches the polls exactly, the vote-poll error is zero. For example, a vote-poll error of 2 would mean that, on average, the polls were off by two percentage points from the actual election results across all parties.
The figure below shows polling errors for Irish general elections from 1989 to 2020.
On average, polls were off by 1.32 percentage points from actual first-preference vote shares.
Interestingly, the smallest polling error of 0.74 points occurred during the turbulent 2011 election, when Fianna Fáil lost 57 seats, Fine Gael gained 25 and Labour gained 17.
The largest error was in 2016 (1.99 points), while the error in 2020 (1.24) was close to the average vote-poll error since 1989.
We need your consent to load this flourish contentWe use flourish to manage extra content that can set cookies on your device and collect data about your activity. Please review their details and accept them to load the content.Manage PreferencesIrish elections in international comparison
How do these figures compare internationally?
Political scientists Will Jennings (University of Southampton) and Christopher Wlezien (University of Texas at Austin) analysed over 30,000 national polls from 45 countries to assess polling accuracy.
Based on 350 elections, they found an average polling error of 2.03 percentage points. Errors were generally smaller for legislative elections (1.83 points) than for presidential elections (2.70 points).
With an average polling error of 1.32 points in the last eight Irish general elections, the polls performed notably better than the international averages reported by Jennings and Wlezien.
The accuracy of polls at the start of the campaign
The results above reflect polls collected right up to election day. But how accurately do polls conducted earlier in the campaign predict the final outcome?
To explore this, I extracted Polling Indicator estimates three weeks before each election since 1989 and recalculated the vote-poll errors.
The average vote-poll error increases from 1.32 (for polls taken up to election day) to 1.72 (for polls up to three weeks before election day). More importantly, polls at this stage of the campaign often significantly over- or underestimate some parties.
For instance, in 2020, the Polling Indicator three weeks before election day underestimated Sinn Féin’s support by 3.6 percentage points, while overestimating Fianna Fáil’s support by 3.5 points.
In turn, the aggregated polls three weeks before the 2016 elections overestimated support for Sinn Féin by 4.6 points and underestimated support for Fianna Fáil by a similar margin. The polls right now are unlikely to reflect the exact election results.
Why polls may differ from election results
What are the potential sources of polling error?
Political scientists Christopher Prosser (Royal Holloway, University of London) and Jonathan Mellon (West Point) identified several factors, including late swings in support during campaigns, how undecided voters are treated, and predicting which respondents are likely to vote on election day.
Other sources of error involve methods used to draw representative samples and weighting results to correct for coverage and non-response bias, the time remaining until election day and the phrasing of poll questions.
In Ireland, polls may become more accurate as campaigns progress, partly due to the high number of undecided voters and differences in candidates' canvassing effectiveness. Door-to-door campaigning and personal contact are much more prevalent in Ireland than elsewhere.
Studies have also shown that geographical proximity to candidates and canvassing efforts can strongly affect candidates' electoral fortunes.
Currently, many voters may still be unfamiliar with the names and backgrounds of their local candidates, but this is likely to change in the coming days.
How to read and treat polls
Comparative research shows that polls have not become systematically less accurate. What has changed is the increase in the number of polls and the focus on horserace-style coverage.
However, I sense that pollsters and the media, both in Ireland and internationally, are moving away from an obsessive focus on headline numbers. More attention is now being given to polling trends and survey questions beyond party support.
Additionally, in recent weeks, pollsters in Ireland have experimented with alternative measures of support, such as asking about specific candidates or gauging respondents' levels of certainty about their intentions. These are encouraging developments, as we should view polls as valuable indicators of political support, rather than as prophecies.
Further information
The Irish Polling Indicator website contains more interactive graphs and details on the statistical model.
Durovic, A. & Schnatterer, T. (2024). Why governments want to learn about citizens' preferences. Explaining the representational logic behind government polling. European Journal of Political Research, published ahead of print.
Jennings, W. & Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.
Prosser, C. & Mellon, J. (2018). The twilight of the polls? A review of trends in polling accuracy and the causes of polling misses. Government and Opposition, 53(4), 757–790.
Stefan Müller is an Associate Professor at the School of Politics and International Relations in University College Dublin.
During the campaign, he will be looking at insights and trends from the various opinion polls in his Poll of Polls analysis.
