A Feeding Frenzy of Forecasts

Researcher Will Scheffler walks through the latest in Scotland’s election forecasting landscape, including our 2026 Holyrood seat projection model, what could shift in the final three weeks of the campaign, and what to watch for on the day of the count.

We have reached the point in the Holyrood campaign where new polls and seat projections are arriving almost daily. You might expect this to bring clarity about the likely shape of the next Parliament. But actually more data is producing less agreement.

Recent modelling illustrates just how wide the range of outcomes has become. Earlier this week, an MRP from JLPartners projected the SNP winning 67 seats, an overall majority. Less than 48 hours later, a model from Election Maps UK, based on the latest More in Common Poll, suggested the party could fall to 48 seats. Both scenarios cannot be true, yet both are grounded in recent data. And each would point to a fundamentally different direction for the next Parliament.

A difference of this magnitude reflects different assumptions about how votes translate into seats: how national polling is distributed geographically, how smaller parties are treated where they are not standing, and how much weight is given to tactical voting and local variation. Small differences in these assumptions can translate into large differences in seat numbers.

Rather than adding another black-box forecast to the pile, we set out to do something different: clearly show how the assumptions in our seat projection model shape outcomes, identify the genuinely competitive seats, and highlight factors that could still move the dial before May 7^th.

I previously wrote in detail about the methodological choices we made when designing and updating our model. Our model starts with the notional results from the 2021 election, based on the updated constituencies and regions following the review of boundaries. It then uses a weighted average of the last five polls to estimate national and regional swings on both ballots, allowing for uneven shifts in support across parties and geography.

A key challenge for forecasters has been how polling treats the Scottish Greens. Polls have generally assumed that the Greens are standing in every constituency across Scotland and allowed respondents to universally select the party when choosing who they will vote for – something we know will not be possible at the ballot box. For that reason, our model initially allows the Green vote to take hold in a constituency, but then redistributes the vote share in constituencies where they will not be fielding a candidate (most of this vote share goes to other parties while some is dropped from the model entirely, reflecting the chance that these voters may simply stay at home).

In constituencies where the projected result is marginal based on these swings, we then allow for tactical voting to shift the result. We recognise that the same voter may exhibit different tactical voting behaviours based on the competitive parties in the seat (i.e. a Labour voter might vote Lib Dem in a LD/SNP marginal seat but vote SNP in a Reform/SNP marginal). This bespoke approach more directly reflects voter behaviour than approaches which apply a uniform national swing towards parties which are expected to benefit from tactical voting.

Finally, we apply a small amount of controlled noise to the result, to simulate local factors that models are unable to quantify (i.e. strong local candidates, uneven ground games across constituencies, differential turnout, etc.).

We then run several thousand simulations of the election, take the most frequent winner in each constituency and apply the D’Hondt formula to produce an overall projection. Based on this approach, our current projection is:

• SNP: 61 seats
• Labour: 17 seats
• Reform: 16 seats
• Greens: 14 seats
• Conservatives: 11 seats
• Liberal Democrats: 10 seats

Many seats across Scotland are on a knife-edge, and small differences between models are often driven by marginal constituencies breaking in different directions (our Director Mark spoke about this in detail on the Ponsonby and Massie podcast this week).

Below are the ten constituencies which appear to be the most competitive across our simulations:

Beyond the headline numbers, our modelling helps frame what could change over the next three weeks. Between now and the election, we will be considering a set of questions that could have major implications for the final outcome:

• Will Labour be able to outperform expectations in the Central Belt and West of Scotland, where we see many constituencies on a knife-edge?
• In the North East, will Reform UK beat the odds and translate pockets of local support into constituency seats, or will their vote share be split with the Conservatives, opening the door for the SNP to sweep the region?
• If other parties are able to beat the SNP in key constituencies, will the governing party be compensated on the regional list, or has their peach ballot support fallen below what would be needed for the party to return regional MSPs?
• What will the result look like in Edinburgh and Lothians East – a region which looks to be hotly contested on the constituency ballot between the SNP, Labour, the Greens and the Liberal Democrats?

Stay tuned for our final voter intention poll as well as our final seat projection, as we set out what these results are likely to mean for the next five years of Scottish politics.