On 22 September 2017, our model projected the following vote estimates for the upcoming German federal elections on 24 September 2017:
CDU/CSU 35.7%, SPD 22.4%, The Left 9.5%, Alliance 90/The Greens 7.8%, FDP 9.3% and AfD 10.8%.
However, as you can see in the election forecast chart above, these values are by no means certain. The bars for each party indicate in what range we expect the actual value to be with an 83% probability. This means that the probability that the actual election outcome will not lie in this range is the same as that of rolling a six with one die—not very likely, but still possible.
What is more decisive is which coalition governments would have a majority in the new Bundestag in absolute terms. Owing to the uncertainty of our forecast, we can only indicate a probability of a certain coalition having an arithmetical majority. It is almost certain that a grand coalition between the CDU/CSU and SPD would have the majority of seats. The probability of a so-called "Jamaica coalition" (black/green/yellow) between the parties of CDU/CSU, Greens and FDP is currently 88%. The corresponding probabilities of other coalition options based on our prediction are indicated above in a graph.
We are a team of electoral researchers from the universities of Mannheim, Zurich and the HU Berlin. Our model combines historical information about German federal elections with current empirical data. Zweitstimme.org will accompany the campaign for the 2017 Bundestag election with scientifically sound forecasts. For feedback on our forecast, please use our contact form or write to firstname.lastname@example.org.
The structural component of our model is based on factors that have already proved to be relevant for projecting election results in the past (since 1949). These include, for example, the performance of parties in past elections, historical survey data, and information on whether a party provided the Chancellor. In other words, the structural component learns from the regularities of all past federal elections. The early availability of this information (already 200 days ahead of the election) allows an early prediction of the election outcome.
However, the structural component alone is often not sufficient to reflect short-term adjustments in the party system or fluctuations in the political climate. We therefore use published values of the so-called Sunday question to take account of an election’s dynamics. In simple terms, to forecast the actual outcomes we combine the information on the regularities of past elections with what we are currently observing in the polls. While the structural component of the model remains stable, our forecast is continuously updated with each newly published poll.
Our model uses a so-called MCMC algorithm, which—metaphorically speaking—repeatedly simulates the election outcome; in our case 100,000 times. Based on these simulations, probabilities can then be calculated for all events that are directly related to a party’s predicted vote shares. For example, if the CDU/CSU is ahead of the SPD in approximately 80,000 of the simulations, this corresponds to an estimated probability of 80% that the CDU/CSU will actually perform better than the SPD.