This year YouGov developed a first-of-its-kind Looped Stratified Simulation (LSS) election model to cover the 2026 general election in Denmark, with its debut outing showing a very strong performance.

Incumbent Danish Prime Minister Mette Frederiksen called the early election after her government received a boost in public perception after her approach to handling the Trump-Greenland conflict. Whilst her party, the Social Democrats, have experienced losses, she has likely succeeded in securing a third term in office, with the Social Democrats as the largest party in parliament for the fourth successive term.

Our final call the day before the election gave a central estimate of 20.4% for the Social Democrats, 13.9% for Green Left, 9.9% for the Danish People’s Party, 9.6% for the Liberal Alliance, 9.3% for Venstre, 8.3% for the Conservatives, 7.2% for the Red-Green Alliance, 6.3% for the Denmark Democrats, 4.6% for the Social Liberals, 2.4% for The Alternative and 2% for the new Citizen’s Party.

Our LSS model is specifically designed to handle more proportional electoral system elections and allowed us to correctly predict 11 of the 12 parties who entered parliament within the ranges of our final call.

In aggregate, we predicted the red bloc parties would collectively receive 48.5% of the vote (85 seats) and the blue bloc would receive 45.4% (79 seats). The final shares had the red bloc on 48.2% (84 seats) and the blue bloc on 44.1% (77 seats), so our model was particularly accurate here.

In terms of vote share, our average error across all parties was just 0.88 percentage points, and the YouGov model successfully predicted the extent of the Danish People’s Party’s vote share recovery to within 0.8pts, a much greater degree of accuracy than almost all of our competitors.

When it comes to seats, we had an average error of just 1.67 seats across all parties projected.

All of the main pollsters performed well in this election, with only one of six leading pollsters having average errors above 1 percentage point. Modelling across the industry showed a similar story – the incumbent Social Democrats were going to suffer losses since their victory in 2022, to the benefit of Green Left, but they were going to remain the most prominent party in the parliament. The Denmark Democrats were going to suffer losses, and the Danish People’s Party were going to benefit from this.

Where we will look to improve

Whilst the Green Left did climb substantially to 20 seats, coming second, and winning over many former Social Democrat voters, our model did predict a higher vote share. This was our largest inaccuracy, and the only error to fall outside of our LSS-calculated ranges, at +2.3pts from the actual result. Understanding where the model failed to correctly capture Green Left support will be one of our core focuses for the coming months.

One source of error which may have impacted our predictions could be our extended fieldwork period diluting any late swing movement. The LSS model requires a lot of data and given the consistency in voter behaviour and relatively short campaign period, we made the decision to maximise data points and extend the date period as long as possible. Post-hoc analysis of data collected up until the date of the election, and shortening the date span to just a week (17-24 March, n=2056) found that we may have better captured some potential late swing away from Green Left (F) and reducing our error well into the acceptable range (from 2.3 to 0.8).

Separately, our model also slightly underpredicted the Social Democrat, Moderate and Citizens’ Party vote shares. Potential sources of development to minimise this error going forward that we will investigate include improving the sample and weighting methods and refinements to the LSS model itself.

About our new LSS methodology

LSS leverages one of our greatest strengths - the size of our YouGov panels, and the fact we can talk to hundreds, thousands of people every day. Using this approach, we gathered the voting intention of our panellists on a rolling basis for over a fortnight leading up to each call, with a final fieldwork span of 5th to 23rd March and 4,444 responses contributing to our final call released the day before the election.

The unique innovation LSS provides is it uses iterative, randomised stratified sampling to test thousands of different models of electorates and voter behaviour, before pooling these estimates into probability distributions. This allows us to produce central, lower bound, and upper bound projections for each party.

Subscribe to the YouGov newsletter