Our model thinks that some of the big swings toward Biden in other models/aggregates is a bit phantom. That's probably bc it relies heavily on state trends (which have been more stable) when data are available and bc of our partisan non-response adjustment projects.economist.com/us-2020-foreca… 
I'd be careful with normal polling averages right now. We know that debates tend to cause temporary swings in the polls, and our model is picking up what we'd expect to see under a phantom swings — big movement in rdd and phone polls with smaller changes in partisan-weighted data
You can see the diff bt state and national trends at play in PA: If Biden really was up 10-11 nationally, our model thinks we should be seeing a lot more polls around +9-10, but instead we have a load of +5-8s and a few outliers around +11.
There's a similar story in Wisconsin, where the post-debate average of polls is roughly +7, which indicates a national environment closer to +9 than +11.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
