More in Common’s first Northern Ireland Voting Intention
Introduction
As we begin to publish our Stormont Assembly election voting intention polling regularly, we wanted to share some of the methodological decisions that we have made and some analysis of how Northern Ireland’s voting intention has changed since the last Assembly election in 2022.
Although pollsters all use similar methods and standards in their work, some decisions have multiple valid choices and can affect our results, and so we felt it important to explain what these choices are and how we made them.
Our approach to weighting
Pollsters usually aim to get 1,000–2,000 people to complete each survey they publish. Because Northern Ireland has a smaller population to draw on, we aim for 1,000 responses in our surveys. Unfortunately, it is impossible to ensure that these 1,000 people perfectly represent the demographics of Northern Ireland as a whole. Pollsters deal with this by assigning individual responses larger or smaller weights, depending on whether groups are under- or over-represented in the sample compared with the true population of Northern Ireland. For example, according to the 2021 Census, 45 per cent of Northern Ireland’s population comes from a Catholic community background. If 43 per cent of our respondents came from a Catholic background, their responses would be weighted up to ensure that our sample was representative.
Our weighting scheme uses a range of weights which are standard across the polling industry, with the addition of some in-house demographic choices which other pollsters do not use. Like many other pollsters, we weight responses to our surveys by age and gender, local authority, whether respondents voted in the 2022 Assembly election and which party they voted for, and ethnicity (white versus non-white).
We also weight by community background combined with age and with whether respondents have a university degree. Because community background is so central to attitudes and voting preferences in Northern Ireland, we want our weighted sample from each community background to reflect the age and education profile of that group.
Dealing with turnout
Another issue that all pollsters have to deal with is turnout. People who take part in online polls are often more politically engaged than the average person, and sometimes overstate how likely they are to vote because they see voting as socially desirable. Meanwhile, academic research on past elections has shown that changes in electoral support can often be driven by previous voters staying at home, or by new voters choosing to vote. Our recruitment methods help reduce the risk of an overly engaged panel, but we still have to think carefully about how we ask respondents about their likelihood to vote, and how we identify those we expect to actually turn out to vote on election day.
We ask respondents to rate how likely they are to vote on a scale of 0 to 10, where 10 means they are certain to vote. This gives respondents plenty of scope to say how likely they are to vote, without concerns that we may bias voter responses by using worded or categorical scaled responses such as “fairly likely”. In some elections it can be helpful to use secondary criteria too, such as asking respondents who are old enough whether they have voted before, and how much attention they have been paying to the election so far. Previous experiments we did to study effective turnout filters in four English mayoral elections in 2025 suggested that a simple cut-off point on the 0 to 10 scale was usually best, with cut-offs at 8 and 9/10 minimising errors and coming closest to the actual result in those elections. As a result, we use a simple filter which only includes respondents who rate their likelihood of voting at 8 or above. In our most recent survey, this implies that turnout would be around 67 per cent in a Stormont election held today.
Here you can see how different turnout filters compare with turnout at other recent elections in Northern Ireland, and how each filter would affect our headline voting intention.
Dealing with undecided voters
One other decision all pollsters have to make is how to handle respondents who tell us that they don’t know or are unsure who they will vote for. This matters more the further away an election is. The next Stormont election is expected in May 2027, and for much of the electoral cycle an election will be even further away than that – so a considerable share of people simply won’t have made up their minds yet.
Many voters are fairly unsure who they would vote for so far in advance of an election. If we forced all respondents to choose a party, it would likely cause many to disproportionately pick “other” or independent candidates, or we would overstate support for parties whose voters are less certain.
In order to account for this, we let respondents choose the party they intend to vote for, or say that they would not vote or don’t know who they would vote for. We then ask those who say they don’t know which way they would lean if they had to choose.
The question we then have is how to treat these voters when reporting our headline voting intention (i.e. what we actually think an election held today would look like). Pollsters can report undecided voters as a separate share, exclude them and report only the remaining share for each party, or model how they think undecided voters would vote. All three are viable options.
We use a combination of the last two of these approaches. Because we model turnout alongside voting intention, we have already filtered out respondents who say they are not very likely to vote (as above). This means that we are essentially figuring out how to interpret responses from people who say they are very likely to vote, but are unsure who they will vote for. We choose not to ignore these voters, and trust that they are voters who sincerely intend to vote but have yet to decide their vote. Accordingly, if someone says they are unsure how they would vote in a Stormont election, but names a party when asked which way they lean when prompted again, we count them for that party. We exclude only those who say they are still unsure who they would vote for, or prefer not to say. In our most recent poll, only 1 per cent of respondents said they were likely to vote, but answered “don’t know” or “prefer not to say” to both questions.
How voting intention has changed since 2022 and since July
Our latest round of polling reveals a few notable changes since 2022 and since July this year.
Sinn Féin, who came first with 29 per cent of the vote at the 2022 Assembly election, remained the largest party in both our July and October polls this year, and appear to have recovered from a dip to 25 per cent in our July poll and are now at 30 per cent.
Though the DUP remains broadly stable since July (up one point), its October performance of 18 per cent represents a three point decline in support since 2022. This may be explained by rising support for the other unionist parties since 2022. The UUP and TUV, on 13 and 11 per cent respectively, are up 2 points and 3 points since the 2022 election, and both are stable since July.
The Alliance Party has seen the largest drop in support, from their 2022 high of 13.5 per cent (its best ever Assembly result) to 9 per cent in our latest poll. They are also down from 11 per cent in July. Meanwhile support for the SDLP is unchanged at 9 per cent, the same as at the 2022 election and in July.
More in Common will continue to refine our voting intention model in the lead-up to the next Assembly election. If you have any feedback on our model or approach, or would like to discuss it further, please contact us at: northernireland@moreincommon.com