After waiting more than four days to declare the winner and with the result of several states settling by a few thousand votes, the general feeling is that the polls, which predicted a comfortable lead for Biden, have failed again. Something similar to what already happened in 2016 –when Trump won against the forecast– the model of Nate Silver prediction gave Hilary Clinton a 71% chance of winning the day before the election
After the 2016 elections, several investigations were launched on the performance of the surveys, the most important being the sponsored by AAPOR, the association that brings together companies and researchers dedicated to the study of public opinion. The result of this investigation showed that the surveys carried out at the national level had been within the expected error margins. So much so that the mean absolute error was lower than in the 2012 elections (2.2 versus 2.9 points). However, polls conducted in some states, which were used to predict the winner and estimate the distribution of the delegates, showed much larger deviations. In Tuesday’s elections, in the absence of completing the recount in some states, this pattern has been reproduced:
Both in the national total and in most states in dispute, the polls have underestimated the result of Donald Trump. This mismatch comes after the pollsters made modifications in the collection and processing of data to avoid the problems registered in 2016. Some hypotheses about this new generalized ruling point to the behavior of voters, who increasingly decide to vote later, to possible errors of measurement or composition of the sample.
A hypothesis that was partially supported by research post mortem 2016 is that of beats swing. According to this theory, the voters who decided their vote in the last days and hours did so disproportionately in favor of the Republican candidate. This hypothesis makes sense in some states, where surveys are conducted less regularly, something that does not happen at the national level, where data collection ends on the eve of elections. On this occasion, the poll aggregators have detected an upward trend in the Republican candidate during the last two weeks, which could explain part of the mismatch in the polls.
In a recent interview, several managers of polling companies spoke about the problems they could face in the last elections. One of them pointed to the possibility of not detecting a part of Republican voters, the so-called hidden vote. This theory claims that Trump voters are less likely to disclose their preferences when polled, the effect shy trump. This possibility was already considered after the problems of 2016, although at that time it was found little evidence that would support it.
It is possible that, as in 2016, the composition of the sample could explain part of the deviations. So low-educated people were underrepresented, a group that votes overwhelmingly for Trump. Despite the methodological changes implemented in the last four years, the polls seem to continue to have problems representing a part of the Republican electorate, and this deficiency could be related to the polarization climate and the loss of confidence in the institutions that the country lives.
According Pew ResearchConfidence in the media has declined in the last five years, especially among Republican voters. In the same way, asked by YouGov About whether they were confident that their vote would be counted correctly in this election, 25% of Trump supporters were skeptical, compared to 7% of Democrats. And polling companies are no exception: 52% of registered voters distrust polls published in the media, a statement that finds more echo among Republican voters.
For the interviewers, one of the keys to guaranteeing the quality of the data collected resides in the ability to generate a relationship of trust with the respondents. One way to do this is to be seen as independent entities. Undoubtedly, the distrust of the Republican base towards the polls is not a new phenomenon, but the current situation of tension has been able to contribute to transform this attitude of rejection into a refusal to complete the interview. When pollsters are seen by part of the population as part of the political problem, their ability to measure public opinion is compromised. We await the analysis that will help us understand the bias in the voting estimates.