Programme comparisons: More variance = more significance?

Voting Advice Applications (VAA) help voters to compare the political offers of the parties more easily. They make it possible to work out differences on the basis of election programmes and, increasingly, past votes. In Germany, one tool has established itself as the de facto standard: the Wahl-O-Mat. However, other tools can provide significantly better differentiation, particularly with regard to the second objective. For example, the Wahltest identifies differences between the programmes with a significance that is around two to three times higher.

Since 2019, wegewerk has been offering the ‘Wahltest’, an election recommendation tool that was first used in 2001 on behalf of Sender Freies Berlin, on its own initiative for Bundestag and European elections. Users frequently gave us feedback that the election test is characterised by a much greater differentiation in the results compared to other election recommendation tools. Either the order of the parties differs significantly, or the overall result is clearer. In order to systematically investigate this phenomenon, we compared the results for prototypical voters in various election recommendation tools. The aim was to find out how large the gap is between the first-placed party and the second and third-placed parties and which parties occupy these places.

In a first step, we compared the results of the election test for prototypical voters with those of the Wahl-O-Mat. A ‘prototypical voter’ here means that we analysed a scenario in which there is 100% agreement with a party. All parties that will be elected to the Bundestag in 2025 were analysed.

Our hypothesis was that the different answer formats contribute significantly to the differentiation performance. The Wahl-O-Mat has adopted the principle of presenting political issues in the form of questions that can be answered with ‘Yes’ or ‘No’ from its Dutch model StemWijzer. In addition, there is the ‘Neutral’ position, which corresponds to ‘Neither yes nor no’ and is categorised as a match - although it remains unclear whether the party and voter actually represent comparable positions. In essence, this is a binary response scheme that simplifies the complexity of political positions, presumably to make the tool appear more user-friendly.

In the election test, on the other hand, a multivariable answer scheme is also possible in addition to ‘yes’ and ‘no’. In the 2025 federal election, for example, there were seven different possible answers to one question, as the parties analysed held fundamentally different positions on the question of how pensions should be secured in the future. To substantiate our hypothesis, we also compared the results with the Voteswiper, another voting recommendation tool that - similar to the Wahl-O-Mat - uses a strictly binary answer scheme, but without the ‘Neutral’ option. According to our hypothesis, the differentiation performance of the Voteswiper should therefore be either just as high or even lower (depending on whether the ‘Neutral’ option in the Wahl-O-Mat actually provides differentiation).