The methodology behind political opinion polls requires increasingly greater attention
February 27, 2026
In Epinion’s Polls and Social Science department, we have been looking forward to the election campaign. Of course, because we ourselves will be voting as citizens. But especially because we take great pride in measuring how the public intends to vote.
Publicly available opinion polls hold a range of democratic potentials. They can spark important political debates. They can signal to politicians how voters feel between elections, rather than only every four years. And finally, the public availability of these polls helps reduce some of the information asymmetry that exists between citizens and parliament.
However, opinion polls also carry the seeds of democratic challenges. Polls themselves can influence voting behaviour. Most clearly, when a party gains momentum in the polls, even more voters may feel inclined to support it.
This is not in itself a democratic problem. But since we know these effects occur, it becomes absolutely crucial that the polls being published are conducted as rigorously as possible. At Epinion, we believe this requires as many qualified eyes as possible to scrutinise our work. And those eyes should be sharp. Because producing accurate political opinion polls is difficult – and perhaps more difficult today than in the past.
Three key challenges
First, a poll measuring party support ahead of the 2026 Danish general election is not one statistical test. It is twelve different tests – one for each eligible party. And the more parties there are, the more statistical tests are performed, increasing the likelihood that at least one estimate is wrong. The logic is similar to the fruit aisle: the more cherry tomatoes in the box, the greater the chance that one of them is spoiled. And just like moldy cherry tomatoes in a cardboard box, one error in an opinion poll rarely comes alone. The statistical tests are interdependent. If we overestimate support for one party, we must underestimate another. This means that even in the ideal scenario – where everyone answers our survey – there would still be roughly a 40% probability that at least one party estimate falls outside the reported margin of error. That is simply the nature of statistics.
Second, recent shifts in voter behaviour are challenging the precision potential of political polling. A majority of voters (53%) changed party between the 2019 and 2022 Danish general elections. In fact, more voters than ever are switching parties between elections. The democratic implications of this development depend on one’s perspective. But one thing is certain: it makes accurate polling more difficult. Political polls in Denmark are often “weighted” based on respondents’ answers to a question about how they voted in the previous election. In polls ahead of the 2026 election, we therefore ask respondents not only how they intend to vote now, but also how they voted in 2022. We then use this information to inspect and adjust our sample if it differs from the actual election result. This process is known as “political weighting”.
The challenge is that when more voters switch parties between elections, it becomes harder for people to recall how they voted previously. And memory bias does not affect all parties equally. For instance, in our ongoing surveys we can see that fewer and fewer respondents say they voted for Nye Borgerlige in the 2022 election. If respondents’ answers about their previous vote become systematically inaccurate, political weighting may end up introducing greater errors than it corrects.
Third, Large Language Models represent a potential new threat to survey research – including political opinion polling. Language models are becoming increasingly capable of answering questions posed by humans. Survey questions are no exception. This makes it harder than ever to distinguish genuine human responses from those generated by AI systems. As a result, there is a growing risk that someone – whether for financial or perhaps political motives – might attempt to undermine surveys through mass-produced fake responses.
How we are addressing these challenges
At Epinion, we are currently working on how to best tackle these challenges.
First, we are focusing on how to communicate the statistical uncertainty associated with our polls as clearly as possible. The goal is for as many people as possible to understand what our data can – and cannot – say.
Second, we are examining how to refine our political weighting procedures so that systematic errors in voters’ recall of their previous vote affect the results as little as possible. This includes exploring whether survey techniques can reduce recall bias and whether repeated measurements can help us better understand these errors.
And not least, we are working on how to best safeguard our surveys against responses submitted by entities other than Danish voters.
The latest updates on this work can be found in our transparency reports (in Danish), which we encourage you to explore if you are interested in learning more about how our polls are conducted. After the election, the data collected during the campaign will also become part of our offer to researchers and students who wish to access our datasets. At Epinion, we believe that the best results are achieved when as many sharp minds as possible examine our work.
The inherent statistical uncertainty cannot be eliminated – but we will do everything we can to minimise the other sources of error.
