Nicht was Du denkst! Was wir wirklich von Wahlvorhersagen lernen können

Abstract

As this forum demonstrates, the competition between election forecasting models along the US model becomes standard also in Germany. This raises the question about the scientific relevance of this undertaking. It is often argued that election forecasts provide a particularly hard and hence good test for the falsification of theories of electoral behavior. In this contribution, I contest this argument and show that the forecast models presented in this issue hardly contribute to the falsification of theories. Instead I refer to an often overlooked relevance of election forecasts consisting in the models' potential for opera-tionalizing important variables. I illustrate my argument by referring to the literature on prediction markets. The Chancellor Model predicts that the 2017 Bundestag election will most likely confirm Angela Merkel for another term as head of the federal government. As of June this year, she leads her Social Democratic challenger Martin Schulz by a wide enough margin in chancellor popularity to overcome any fatigue with her tenure after three terms. Besides continuing with the Grand Coalition, Merkel also has the option of entering a Jamaicacoalition (with the FDP and Greens), which would command a majority in the new Bundestag. A shift in chancellor popularity, however, would change the forecast. In the event that Schulz ultimately surpasses Merkel in chancellor popularity, as he did in late February, the SPD candidate would be more likely to head the next German government.