There is a large divide between national and state polls right now, with national poll showing an 11 percentage point advantage for Biden (in two-party %) and state polls hovering around +9.
Our election model is based mostly on the latter — we only really use national polls to fill in the gaps — but I do wonder about what the election would really look like with Biden +11 in state polls too. IA, OH and GA would probably be safely lean D
A 2 point delta nationally is huge in probabilistic terms, esp on election day. It's the difference between FL being 70% Biden and 90%, which might even be enough to push Biden's odds above 95
https://twitter.com/onetruecarlson/status/1315800848740413440?s=20
Yeah, though the error pointed the other direction in 2012
https://twitter.com/sebastian_byron/status/1315801750545207296?s=20
The NYT/Siena polls have also been about 2 or 3 points more favorable to Biden the whole year. Our model subtracts that average house effect, but this is hard for me to accept given that I think they're some of the best polls in the biz
https://twitter.com/skyledavis/status/1315801763753197568?s=20
In terms of house effects, you can do two things. You can base pollster biases on their historical records, which is good but can't control for mode/weighting changes -- or you can treat them as random effects, which risks over-correcting for "biases" w good data (like NYT)
It's a tough call
We’re kinda averse to ad hoc weights, but I think if we do this again in 2024 we might add some measure of quality based on a pollster’s amount of historical bias. I’m really looking forward to writing up our postmortem in a few weeks
https://twitter.com/richardamann/status/1315805382531600384
• • •
Missing some Tweet in this thread? You can try to
force a refresh
