There are several possible reasons for the surveys to fail:
The turnout this year was unusually high, the highest it has been in more than 100 years. In the past, high turnout has mostly resulted in good Democratic results. This is due to the fact that in previous elections the “blue” candidates — that is, the Democrats — were able to attract disproportionately large numbers of members of minorities to the ballot box in these cases. In the election this year, however, Trump in particular was able to mobilize his clientele. The proportion of white voters without a college degree was unexpectedly high. The US political scientist Salvatore Babones describes another reason in a guest article in the “Sydney Morning Herald” as the “dirty secret of public opinion polls”. Accordingly, the response rates to telephone surveys in the USA have downright collapsed. According to the polling institute Pew, 20 years ago 36 percent of those called took part in surveys by telephone. This figure is now six percent. Babones also writes that there are rumors in the industry that only around three percent will participate. His guess as to what could be behind this is: In an age in which almost everyone is equipped with a smartphone, hardly anyone takes calls with a hidden number. However, what speaks against this thesis is that many online surveys were also not more precise.
An effect that also exists in other countries and that is also known in Germany could also have played a role. It is about the concept of social desirability developed in 1972 by sociologists Derek Philipps and Kevin Clancy. According to this, some people refuse to take surveys or knowingly give wrong answers because they fear that providing truthful information violates social norms. As a result, predictions about opinions outside of the social mainstream are much less reliable. In elections in Germany, this effect could be observed in the surprising success of the Republicans in the state elections in Baden-Württemberg in 1992 and in the state elections in Saxony-Anhalt in 1998, where the right-wing extremist DVU got 12.9 percent of the votes from a standstill — last polls saw the party at only six percent. And the election results of the AfD are traditionally characterized by this effect — in some cases considerably. In the last state elections in Baden-Württemberg, Rhineland-Palatinate and Berlin, the polls on election day (exit polls) showed a deviation of 25 percent. In the 2013 federal election it was almost 50 percent.
In the US, this effect is also explained with “Shy Trump Voters”, with “shy” Trump voters. How big the effect really is cannot yet be said. The Republican camp has also tried in the past to develop a political narrative out of it — that of the “silent majority”, the silent majority that supposedly stands behind Trump. This is supported by the fact that a poll in Florida, which was conducted by computer votes, was much closer to the actual result. Because in front of a computer nobody has to be ashamed of their preferences. Against this, as mentioned above, the fact that even anonymous online surveys were sometimes unable to provide exact results. It is possible that Trump was able to convince many Americans in the last few meters in the election campaign. In contrast to his competitor, the president held mass events in the important swing states despite the corona risk. The incumbent may have been able to mobilize many voters in the last few days before the election — and it was already too late for this development to be reflected in the polls.