Istanbuller wrote:No. Most polls are conducted at university campuses or urban areas. It is very likely that sampling is very poorly done. That is why almost all polls in 2016 was wrong.
They weren't all that wrong. They did
call the popular vote out, Trump was elected due to the electoral college after all.
But more importantly, they were massively misinterpreted. When people were saying Hillary had 70% probability of winning, they were saying the probability she would get more votes than Trump was 70%, right? But this is misleading, because it conveys a more certain result than there really was. One way to notice this is to look at the percentage of the vote for each. Here Hillary had something like 48% and Trump had something like 46%... And most polls had a margin of error greater than 2%. This can still then be translated as saying that the polls estimate that the probability Hillary would get more votes was 70% using the typical normal distribution and a reasonable estimate of the variances, but
this 70% figure has its own associated confidence interval (since it's also an estimate and thus random), which was not reported. Even worse, this was only a statement about popular rather than electoral votes, since it's very expensive to field polls that are representative at both the national and State/Territory levels (let alone County level). A good estimate of the electoral votes for each candidate is possible but would need a large and expensive survey, and even then it would require a lot of adjustments since many people don't like to answer polls.
The truth is that the polls weren't strongly calling the election for anyone in 2016. And they aren't doing so now either, that 8% or so difference between Trump and Biden may probably be quite imprecisely estimated. They are an OK guide but they are far from a definitive statement.
Even worse, with the pandemic raging, I also find it extremely hard to predict voter participation this time around. This adds even more uncertainty than usual.