I'm not a fan of this style of analysis because these polling margins do take into consider undecided voters and therefore you create a systemic bias.
It would be much better to compare Trump polling average to Trump results and Clinton polling average to Clinton results.
1) Southern states could really be at play this year. Statewide polling in GA & TX were near perfect in 2016 and 2018. Let's see if the trend continues.
These states should be easier to poll given how monolithic various voting demographics are, any good likely voter screen should do well. It probably shouldn't surprise anyone that the least accurate polls are in states with the most (brain damaged?) Obama-Trump voters.