Predictions can replace actual elections

From the polling station to the election resultHow forecasts and projections are made

It is May 14th, the evening of the state elections in North Rhine-Westphalia. At 6 p.m., moderator Jörg Schönenborn announced the forecast in the first:

"In North Rhine-Westphalia, too, voter turnout will have risen a good bit. And what effects this will have, who will benefit from it, can be guessed if you now see Infratest Dimap's forecast for the state elections in North Rhine-Westphalia. The SPD drops dramatically, minus eight and a half points, 30.5 percent, that would be the worst result in NRW history ... "

Polls in selected polling stations

At this time, the employees of the survey companies who provided the data for this forecast are a long working day behind them - long before the first actually cast vote is counted. Infratest Dimap collects the figures for ARD and the polls research group for ZDF. Whether state or federal elections - the principle of data collection is the same. Infratest will have around 1,200 employees on duty next Sunday. One of them goes to one of 624 representative polling stations. One of them is Renate Perseke, who was in the state elections in North Rhine-Westphalia in May. In front of a polling station in Cologne, she addressed voters who had just cast their vote.

"Hello, may I ask you briefly? You have just voted. Would you be prepared to simply inform us of your voting decision anonymously and voluntarily? That is for the ARD forecast at 6 p.m."

Communicating means: The voters will again receive a kind of ballot paper on which, just as in the booth, they can tick their two crosses. You should also enter your age and gender. On every sixth piece of paper, additional personal data is requested, such as educational level, occupational group and the reason for the voting decision. This results in data from which, for example, the voter migration, which is important in the election analysis, can be represented.

Intermediate results processed in call centers

"We have certain periods of time in which we conduct surveys. A thumbs-up for an hour. And then we pass on the data so that those at Infratest Dimap always have the current data during the course of the day Art, we are in a position to have results by 6 p.m. "

In the Infratest call center in Berlin, around 200 employees receive the interim results. In the course of the day, around 100,000 data records come together, which after evaluation and weighting come relatively close to the actual election result, says Infratest Managing Director Michael Kunert:

"The forecasts at 6 p.m. have a ... These are actually forecasts, where we say we like to be measured by the differences. And the differences are extremely low." The meaningfulness of the forecast at 6 p.m. has become more and more reliable in recent decades - unlike the Sunday question, which in the weeks leading up to the election only conveys current moods and must by no means be misunderstood as a forecast of the election result. But: Both in the Sunday question and in the survey in front of the polling station, not all of the voters polled take part.

Forecasts become extrapolations

Certain groups are more represented than others. That means that the demoscopes have to deduce from them all voters - and that is getting better and better, said WDR television director Jörg Schönenborn before the NRW state election on the Facebook page of his station:

"And these forecasts on election evening, the 6 pm forecasts, are so accurate because there are x empirical values ​​from previous elections, and the computers do nothing other than take into account the old experiences."

With new parties like the AfD, however, that could be difficult, said Schönenborn, because the AfD has a different electorate in almost every federal state, so that their behavior is more difficult to predict.

"And to that extent the institutes ... just needed a bit ... it's like focusing the camera. In the past, manually by hand, that is, without autofocus. The extrapolations are actually autofocus, what is done arithmetically is autofocus. And it works stop with the AfD, with a new party, not. "

The forecast from 6 p.m. then becomes extrapolations. Little by little, numbers on the votes actually cast are also available from polling stations, which are continuously incorporated into new calculations at Infratest. Where there are already results in the representative selected polling stations, these replace the post-election survey. If results from other polling stations are added, they are weighted into the extrapolation. Infratest can create such a new projection at any time - when it is published, the broadcasters decide on the basis of editorial considerations.

"But now at 6:12 p.m. the first projection for the state parliament of North Rhine-Westphalia from Infratest Dimap with slight changes compared to the forecast."

This results in extrapolation after extrapolation - until the election workers in all 88,000 polling stations and postal voting districts nationwide have reported the results to the Federal Returning Officer.