Why do recent polls fail? Should we fix them?
In the two most important election events of the past years (the Brexit referendum and the U.S. presidency), the polls were wrong. Everybody was certain that we would live in a world with the UK inside the European Union, and with Hillary Clinton as the US president. Yet, here we stand, with the UK bureaucrats preparing the grounds for the Brexit, and Donald Trump ready to enter the Oval Office next January. What has happened?
First, I do believe that the organizations conducting the polls really followed the best standard procedures. However, these procedures lie on a big assumption: people are answering honestly to polls. And why this assumption bedeviled the accuracy of polls only recently? In order to answer this question, let’s take a look at the major candidates for the US presidency over the last few elections.
A big difference of the last pair of candidates (Clinton vs. Trump) is that one of them – Donald Trump – has been more or less demonized. That is not merely being discredited by his opponents (a common practice), but also being discredited by the media, and almost being exiled by his own party. Under these circumstances, it is no surprise that (even in anonymous, faceless polls) some of Trump’s supporters would be reluctant to a priori reveal their political preferences.
A similar situation occurred with the Brexit referendum, where even the always-right bookmakers got it wrong. Brexit was presented (by its opponents, as well as by many national and international media, authorities, etc.) as a terrifying, devilish option. Again, this could have made people supporting it inaccurately stating their choice in the polls that preceded the actual referendum. Having an experience from the troubled Greek politics, the same thing happened there already a couple of times. Most notably, when the extreme-right party (Golden Dawn/ Chrisi Avgi) first entered the Greek parliament, in 2012, polls under-predicted its percentage by more than 50%.
Interestingly, in the next elections (when the party had gained some public acceptance, already being in the parliament), the polls started predicting its percentage a lot more accurately.
Clearly, many other things could have gone wrong in predicting the outcomes of these three elections – predictions are no easy business. The important point is the following. These mis-predictions cannot be attributed that easily to “statistical error”. The latter is random, in the sense that it can either over-predict or under-predict any given result. However, the last mis-predictions had a clear pattern: They always under-predicted the percentage that the previously demonized option would get. This is not random – it is predictable.
If this is indeed the problem, it can be easily fixed: Simply include in a poll a couple of “demonization” measurements, and then adjust the results in favor of highly demonized options. Now from an ethical perspective, the dilemma is, should we do this, or not? Should the results of polls be manipulated in such a heavy-handed fashion? I really have no answer for that. Still, if people do not respond accurately to polls, accurately presenting their inaccurate answers sounds problematic.
By: Antonios Stamatogiannakis