The Oracle of Diffley
Published
Researcher Will Scheffler introduces Diffley Partnership’s modelling for the 2026 Scottish Parliament election. Our modelling suggests the SNP remain the most likely party to lead the next Scottish Parliament, but even relatively small shifts in public opinion between now and May 7th could dramatically reshape the chamber.
We are delighted this week to unveil our in-house modelling for the 2026 Holyrood election, using polling conducted by both Diffley Partnership and other research agencies to estimate what the Scottish Parliament could plausibly look like in May.
Our modelling approach utilises several variables to make seat projections, namely results from the 2021 election, current polling, known tactical voting patterns, and controlled randomness, which allows for local variation that polling alone cannot capture.
The ‘spine’ of our modelling is rooted in the results from both the constituency and list ballots in the 2021 Scottish Parliament election. These results provide the baseline from which we estimate how support may have shifted since the last election, capturing the historic strength of parties in different areas before applying updated estimates of support from polling.
We then adjust these baseline results using current polling. National and regional polling averages are used to estimate how party support may have shifted since 2021, with more recent polls receiving greater weight than older ones.
Next, we incorporate commonly observed patterns of tactical voting in close constituencies. Depending on the likely top two parties in a seat, we allow for different behavioural responses. For example, when the SNP and Liberal Democrats are the main contenders, some Labour supporters may lend their vote to the Liberal Democrats. In contrast, when Reform is the challenger, that same voter may be more likely to support the SNP.
Finally, because even the best models cannot capture every local factor, such as a particularly strong constituency candidate or a popular MSP retiring, we introduce a small amount of random variation into each simulation. This allows different constituencies to vary slightly from the national picture in ways that resemble real elections.
Where the strength of this approach comes in is that it allows us to run a number of polling day simulations. For every seat in the Scottish Parliament, we simulate the election 1000 times and observe how often each party wins. Below is our baseline prediction, based on all of the above factors:
This approach importantly allows us to tell the constituencies where the race is the closest. Across 1000 simulations, we can identify the most frequent winner in each constituency and the number of times (out of 1000) in which that party wins.
This simulation-based approach also lets us to tweak polling parameters to estimate counterfactual scenarios. For example, if between now and election day, polling were to shift by three percentage points from the SNP to Labour across Scotland, we estimate that the chamber would look something like the below:
You may be asking where the tipping point in this scenario lies. The percentage swing from the SNP to Labour we would have to see to make Labour the largest party is roughly in the region of 6-7% from our current baseline scenario. For Reform to be the largest party, the Swing from the SNP would need to be closer to 8-9%.
Now, we recognise that any significant electoral realignment would likely have wider knock-on effects beyond what we include in a simple SNP to Labour swing scenario. But we show these largely to illustrate a point: that while polls suggest the SNP is on the margin of being able to form a majority government, even a small change before May 7th could shatter their chances of this, potentially putting their seat count at levels not seen in nearly twenty years
Finally, it is important to include a health warning about election modelling in general. All seat projections rely on a number of assumptions and are only as reliable as the polling data that feeds into them. As illustrated above, small changes in vote share can sometimes translate into large changes in seats. For that reason, these projections should be read as an estimate of what could happen based on current polling and potential shifts in opinion, rather than a prediction of what will happen on polling day. Tracking marginal seats allows us to illustrate this clearly. In many of our simulations, a handful of constituencies change hands, meaning the final seat totals could easily move by several seats in either direction.
As new polls continue to be released over the next two months, we will be continuously updating our seat projection, so do keep an eye out for our up-to-date commentary as the race for Bute House continues to heat up.