I have seen many social scientists arguing that the poor predictive power of the American polls were due to the “shy voter” effect. The shy voters are people who conceal their preferences from polls, thus distorting the relevance of the statistical measures.
They are right about the phenomenon that perhaps made the Brexit polls erroneously predict a Remain victory. However, it is way more complex than this in the case the American elections. First, the media minimized the strength of Trump’s support, saying often that they are marginal electoral groups that emerge during this campaign. This might not be the case and we cannot make any assumptions until we have electoral data to compare it with 2012. I have a hunch that the electoral support came from both Republican traditional vote base, as well as some marginal groups propelled by the Tea Party. But only the voting data will confirm this.
Second, why were the polls so wrong? They couldn’t capture well the voter. True! But not necessarily because of the shy voter phenomenon, but because the sampling methods weren’t updated to capture changes in the electoral base structure that emerge due to the non-conventional campaign. Trump mobilised certain rural areas more than other Republican candidates (early data confirms this) that wasn’t reflected in the sampling method. This happened because polling is not a precise instrument and it mostly uses predictions from the past events to make assumptions about the future. As we know, we didn’t have an average campaign, we had a new political player that was an outsider and was prone to mobilising groups of voters in a different way.
Third, there wasn’t any spiral of silence either. Trump’s voters were the most vocal out of all the voting groups, they were the loudest on all the rallies, very active on social media. They weren’t hiding in the closet and waiting to pop-up discreetly during the election day.
Fourth, there was this nonsensical assumption that the Republican voters would not vote for Trump. We don’t know yet if this was the case, but I have a hunch that American politics is based on long-standing partisanship that transcends the person. What would be truly amazing (hypothesis to test when we will have the election dataset) : Democrats swinging massively towards Trump (nobody predicted); people who do not vote usually went massively to vote.
Fifth, yes, the Democrats’ enthusiasm has fizzled because: 1) Bernie Sanders was not the candidate; 2) Clinton had problems with the law, but I am almost sure it wasn’t because they were overconfident she would win. This was just the portrayal of the media (same as in the Brexit case).
What is to be taken from this election is that we shouldn’t dump predictive models, but we should be more careful with the data we input. Also, before publishing articles in famous newspapers, social scientists should wait for the dataset and test their theories.
P.S. The democratic system (electoral colleges) in the US is outdated and simply bad for so many reasons I am not going to write today. And, yes, there would be a few people who would argue against what I said because US is the “greatest” democracy in the world. It is not! Let’s get over it.
UPDATE: Yes, the voting base for Democrats has shrunk from 2008 to now, while the Republicans’ base has remained constant. This invalidates the theory of a “shy voter” or spiral of silence. It also confirms that Clinton didn’t have the same power to mobilise her party supporters.