Statistics? Good and well, if only the numbers wouldn't be.
Erhard Blanck (* 1942), German writer and painter.
Many opinion pollsters and election researchers would certainly agree with the above quote, especially the latter
did not cover themselves with glory in the prediction of election outcomes.
But how are election forecasts actually made?
The starting point is always the unweighted raw data of the voting intention, which is collected via surveys by the known Sunday question
„Which party would you vote for if federal elections were held this Sunday?"
However, the result of the unweighted voting intention is heavily biased, because some subgroups (e.g. older or unemployed people) participate more often
The political mood is created by weighting in terms of national demographic characteristics (gender, age, occupation or town size).
These values are unsuitable for predictions, since the response behavior of eligible voters differs significantly from their electoral behavior due to
effects such as social desirability or tactical votes.
This applies for example to the Left Party in Germany. Several people fear social rejection if they would list the Left in the answer to the Sunday
question due to its predecessor party, the SED, which was the governing political party of the German Democratic Republic.
From this it follows that the value of the Left Party is always lower in the weighted raw data than on the election day.
In the final step, projection functions are used to the political mood to adjust such differences. These functions often contain a variety of
factors, which are not published and top secret. They are the only way to make unbiased election forecasts and have a substantial
share in the final prediction quality.