Thread by @Redistrict

I applaud Marist for indeed being one of the most transparent pollsters out there when it comes to its sampling. It also allows us, like Marist, to study what might have led to a chronic pro-D bias in Oct/Nov 2018 polls... 1/

https://twitter.com/LeeMiringoff/status/1286683161213140993

First, we should note that Marist's Oct/Nov 2018 polls were fairly precise even if there was a chronic pro-D *inaccuracy:*

MO: McCaskill +3 (actual result -6)
IN: Donnelly +3 (-6)
AZ: Sinema +6 (+2)
FL: Nelson +4 (-0.2)
TN: Bredesen -5 (-11)
MN: Walz +17 (+12)
WI: Evers +10 (+1)

What's one thing these surveys had in common? In all seven cases, non-college whites were seriously under-sampled relative to their share of the actual electorate based on estimates using Census/CPS data compiled after the election.

Here's Marist's breakdown of college/non-college whites in '18 likely voter pools (vs. Census/CPS-based estimates):

AZ: 34/32 (32/41)
FL: 30/33 (27/40)
IN: 37/47 (31/55)
MN: 41/45 (39/50)
MO: 38/47 (32/52)
TN: 32/45 (32/50)
WI: 39/48 (34/57)

Now, I should also note that Marist also under-sampled non-college whites (per CPS-based estimates) in #GAGOV, where it was pretty much on target, and in #NVSEN, where it showed Rosen (D) trailing 44-46 in early Oct. (she ultimately won by 5).

But on balance, weighting whites by education might have gotten Marist closer to the true result in more states. Would it explain the entire gap? No. But would it strongly suggest weighting for education in the future? I'd say so.