Everyone harks on about the 2016 polls being wrong. At the national level, the "polls of polls" were broadly right - Hilary won the national vote by 2%.
So the tracking polls had an aggregate bias to the Democrats of about 1 - 2%.
What the polls failed to pick up was the shift in Wisconsin, Pa. and Michigan. That was a mixture of some white working class voters switching from Democrat to Republican, some Democrats staying at home, some Independents voting for third party candidates instead of for the main parties, and some Republicans coming out to vote for Trump who had stayed home the last two elections.
Pollsters have a terrible habit of bolting the stable door after the horse has fled. Whatever the polling error was last time, they make ad hoc adjustments to correct for that, thereby creating new errors because each election is different.
In 2012, the polls under-estimated Obama's support, suggesting that his lead over Romney was 1% or under, when in fact he achieved 3%, i.e. there was an aggregate bias to the Republicans of about 1 - 2%
It is difficult to escape the conclusion that a part of the 2016 polling debacle was crude ad hoc adjustments to the polls designed to correct for the 2012 bias: adjustments that understated the Republican share by 1-2% would largely produce the 2016 polling error which overstate the Democrat share.
Why this time is different:
1. The national polls.
One thing that both the 2016 and 2012 national polls had however was that there were various national level polls showing the Republican candidate winning in the run up to the election. Which is what you would expect if the result at the national level was going to be a 2-3% win for the Democratic candidate, due to the margin of error in polling.
What is completely and starkly different this time is that there is not a single national poll this year showing Trump beating Biden. Not one. In fact, the national "poll of polls" is not even close and never has been:
While electoral college maths means that Trump could lose the national vote by more than the 2% he managed last time, there is a limit to how far that can be taken because national polls are polls which include battleground states, and while individual battleground states have specific cultural, economic or demographic factors, those factors are found elsewhere in other states.
Even if you assume that the polls are infected with exactly the same bias as they were last time and that no attempts have been made to correct the overstatement of Democratic support, Biden's lead in the national poll of polls is 6-7%. Even if the poll bias were at the upper end of the spectrum, and was overstating his share of the national vote by 2%, that would give him a 4-5% national lead. Any analysis in which Biden wins the national vote by 4% (2% more than Hilary) but fails to win 3 of: Penn, Wisconsin, Michigan, Florida, North Carolina, Ohio, Arizona, Georgia is going to involve some contorted thinking.
Trump's core support is older white men. They continue to shrink as a percentage of the demographic.
3. Turnout & Early Voting
Contrary to perceived wisdom, higher turnout does not always favour the Democrats: turnout in Florida and Ohio was higher in 2016 than 2012 which translated into a Trump win.
More important is early voting: that bakes in votes before last minute tightening of the polls, and in 2016 Hilary did suffer a wobble in the run up to election day associated with the Comey emails.
This time round, Biden's lead is solid, and votes are being banked in huge numbers (circa 53 million to date), and the Democrat lead in early voting is marked.
This leaves the Republicans having to have a massive effort to get their vote out between now and election day itself, while the Democrats, having already banked much of their vote, have more time resources available to chase up every last vote.