Here's a pattern that I think may have some consequences for polling in 2020, at least for now: response rate by vote history.
In our polls from last month, here's the completion percentage by vote history.
Voted 2016 and 2018: 2.3%
Voted 2016 not 2018: 1.3%
Didn't vote 2016, voted 2018: 2.3%
Didn't vote 2018 or 2016: 1.1%
Voted 2016 and 2018: 2.3%
Voted 2016 not 2018: 1.3%
Didn't vote 2016, voted 2018: 2.3%
Didn't vote 2018 or 2016: 1.1%
To my knowledge, no public pollster adjusts for this--either in terms of sampling or weighting.
Now, here's Trump approval in PA/WI/MI, by vote history:
Voted 2016 and 2018: 47-51
Voted 2016 not 2018: 55-43
Didn't vote 2016, voted 2018: 38-61
Neither: 52-45
Now, here's Trump approval in PA/WI/MI, by vote history:
Voted 2016 and 2018: 47-51
Voted 2016 not 2018: 55-43
Didn't vote 2016, voted 2018: 38-61
Neither: 52-45
Here's Trump approval in AZ/FL by vote history:
Voted 2016 and 2018: 50-49
Voted 2016 not 2018: 44-54
Didn't vote 2016, voted 2018: 44-53
Neither: 40-53
Voted 2016 and 2018: 50-49
Voted 2016 not 2018: 44-54
Didn't vote 2016, voted 2018: 44-53
Neither: 40-53