So, for the last couple of weeks the Pundits of MSNBC have been HIGHsterical. It goes like this: "we blew it in 2016 by misjudging the polls"!! Well, no. As stated many times, and many places (here, certainly), the pollsters got the national vote dead on (you should read the whole thing).
But here's a stubborn and surprising fact — and one to keep in mind as midterm polls really start rolling in: Over the past two years — meaning in the 2016 general election and then in the various gubernatorial elections and special elections that have taken place in 2017 and 2018 — the accuracy of polls has been pretty much average by historical standards.
As also stated here many times, stratified random sampling is the toughest thing in all of quant. At the national level, it's quite a bit easier to say how the vote totals will go, but when you have something as biased as the electoral college getting in the way, making a presidential prediction gets to be problematic.
The number that matters to the 2016 experience is 79,000. That's the high end of various numbers that have been reported for PA, MI, WI margins for Trump. Turns out, while vote totals by state are readily available, number of precincts by state aren't so easy to find. I was able to find both numbers for MI. Here they are:
Trump margin: 10,704
So, the margin is about 2 votes per precinct!!! There is no way on God's Green Earth that a national poll would find so few aberrant voters. The midterms aren't biased by the electoral college. Even so, MSNBC is showing as I type, and they've put up numbers for some interesting races, and the margins are running in the 4 to 6 point range. But... if you look at the fine print in the graphics, the MOE (that's what it says, rather than spelling out margin of error) runs to 3.5+ points! There's a reason that the House as been held by Democrats so much of the time: the mass of the voting public isn't Trumpian and never has been. It's essentially Liberal, even though many won't admit it.