Sean's chart is critical: polls miss hardest in lopsided states, and they miss the same direction every cycle. Non-response bias is getting worse, not better. Rs in red states and Ds in blue states both beat their polls, in short. Somewhat controversial: in partisan states, polls that aren't regressed against their own past misses shouldn't be used at all at this point, and Senate midterm error bars should be 2-3x'd. (See: NJ governor.) Priors are more powerful now! Model turnout and partisanship off similar past elections, with decay factors back in time, and treat polls as a hint about drift, not as ground truth. The hardest calls become turnout and low-propensity voters, which is exactly where polls fail! This being the problem area twice is a good sign: it tells you where one of the two big issues is. The other is polls getting a valid sample of groups, hard when 0.5% of people even respond. In this election, that means your starting position should be Dems take the House but not the Senate. To make both sides mad: Husted and Hinson win. So does Hochul. Pump the brakes on upsets in party strongholds on both sides! Gun to my head, contested Senate races: Likely D: NH, GA, NC Toss-up: MN, MI, TX Likely R: ME, OH, IA, KS (AK's ranked choice is its own barrel of monkeys, so fuck you in particular, Alaska.) Dems pick up NC but may cough up MI, which leaves almost no path to a Senate majority. 50-54 R is the likely range, with 51 or 52 most likely. Collins alone could be a case study on this entire problem; Sean notes this but the Senator outruns her polls by such wide margins, repeatedly, that it's obvious the error bars are deeply incorrect. Remember: that's error IF the sample is representative, but we know with certainty it's not. You can also see this sort of disparity playing out across the early voting landscape, with folks like @DjsokeSpeaking and @TonerousHyus and @seantrende and @ZacharyDonnini all having a brawl about what is actually going on. Some of this, of course, is partisan cheerleading (nature of the game; you can hate it but it is what it is), yet a lot of this is actually that things are getting both harder to model and harder to predict in advance. Past success does not indicate future success, and in fact may contraindicate as regimes change. My central point thus remains: be cautious of certainty, and be long vol. From a betting standpoint, I think many of the Rs in Senate races are probably underpriced on a volatility-adjusted basis, as the uncertainty is way higher than people think and we're seeing betting markets missing badly (Brazil and Israel) already. That's not a statement they will win (they very well may not), but that things like Paxton in the low 30s is odd to me when in reality, we know so little we can be confident of that your priors should have had that around 50-50, if not slightly leaning R. And before you spin me a bunch of stories about Talarico, remember we saw this with Beto (who was in many ways a stronger candidate), but also, we're hearing the same cope from Rs in NH or NY, where your priors should have you confident Ds will win. Disagree? Check back Nov 4.