Johnson has definitely declined from his peak, but his peak was something like 9% in the polling averages, and the polling averages now have him at about 6%. So sure, that's a decline, but not the kind of "crash" that some were predicting six months ago when some were saying "Oh, his ceiling is going to be something like 2%."
Most polls still have him north of 5, yet when we get polls like this that have him below 5, everyone starts jumping on it like "See, I told you Johnson was going to crash and burn!" In the threads on polls that still have him in high single digits, there isn't a corresponding chorus of "I guess Johnson isn't going to drop like people were saying."
Third parties typically decline even more on election day itself than in the polling leading up to election day. Nader polled at 2.3% in 2008 and he got 0.6%. Barr polled at 1.5% and got 0.4%.
Neither of them were actually included in many polls that year. Should be noted that Perot '92, Perot '96, and Nader '00 all did about as well on election day as they were polling a few weeks before the election. I guess what I'd say is that polls are really bad at distinguishing between a 3rd party candidate whose true support is 0.5% and a 3rd party candidate whose true support is 5%. It's not clear to me that we can figure out which of those categories Johnson is in in advance of the election.
According to PPP, among early voters in North Carolina Johnson got so few votes that his number rounded down to 0. His and Stein's support is extremely soft, I wouldn't be surprised if most of the people who say that support them don't even bother to vote at all.
Do we have any evidence that "early vote polling" is predictive of anything? I am skeptical. We really don't know how people who vote early might compare to those who vote on election day, or how polling of self-identified early voters stacks up against final results.
Well to be fair we do have a very limited amount of data that some counties in California report that break down numbers by absentee/VbM and same day in person.
I don't know how many other parts of the country even count or break down numbers in that manner, but there were some interesting data points looking at some of these counties in California during the 2016 Primaries.
But yeah.... not sure how much we can really discern about early voting polling in general.
