The German Süddeutsche Zeitung wrote last, last Saturday on page one that surveys are often wrong. I wouldn’t put it that way. Yes, survey companies face new challenges, but these challenges didn’t annul the basic laws of statistics and probability. The problem is that even in political circles, many people don’t know what public opinion research can and can not do. And in addition to that, some clients are also hesitant to invest into solid enough research designs.
To begin with, one survey is never a prediction of the outcome (and even less of the winner) of an election. If one is interested in a forecat for the upcoming German elections, however, I can recommend a project that five electoral researchers from the universities of Mannheim, Zurich and HU Berlin have launched. Together, they developed a powerful forecast model which they publish on their website www.zweitstimme.org. Their calculations are based on two components: the structural and the polling component. With respect to the structural component, they take into account information such as the performance of parties in past elections. Together with survey data from different sources, they then calculate forecasts for the parties’ performance (with a certain statistical uncertainty). They then update their figures continuously as soon as new survey data are published. Their forecast a few days before the election is as follows: CDU 36,3, SPD 23,0%, AfD 9,9%, Linke 9,5%, FDP 9,2%, Grüne 7,8%.