Our polling model for VA-Gov is up. It controls for house effects and (attempts to) adjust for partisan nonresponse ("attempts to" bc we detect only modest patterns in the VA polling right now).
It says McAuliffe +2, but the CI includes a loss up to 4 pts
economist.com/united-states/…
It says McAuliffe +2, but the CI includes a loss up to 4 pts
economist.com/united-states/…
One big thing: Polls show ~7% of voters are undecided right now. That really explodes the CI. Luckily, there has been very little correlation so far between the number of undecideds & the gap between McAuliffe and Youngkin. (Versus in CA when we saw undecideds breaking 4 Newsom.)
One thing to keep your eye on, in terms of trying to spot potential polling error, is that multiple polls have come in with Youngkin's vote % with voters of color wayyy higher than Trump's, including among Black respondents. Like ~10pts. Nonresponse or real? We'll know Tuesday...
Here's an inside look at the model. Basically, there's a 1.5-point difference in McAuliffe's margin if you trust polls that weight by party id, past vote, or voter file partisanship scores more than other polls. These charts show trends if all polls had a certain scheme: 
• • •
Missing some Tweet in this thread? You can try to
force a refresh
