I’ve always been very clear about my position in 2018. The polls were correct even though there was a gigantic turnout gap, with many aspects of it (Nevada for example) being unrecognized by pollsters. The polls should have underestimated democrats more given how far they exceeded expectations.
That’s also one case. The average GCB miss is 1.1 in the direction of democrats according to Nate Silver’s graphic, and there’s a logical (though not conclusive) case to be made that the low-propensity vs high-propensity coalitions are becoming wider than ever.
This is a very important point which makes the "but the polls were (largely) accurate in 2018!" argument appear in an altogether different light. It’s possible that the polls were (largely) accurate only because pollsters didn’t account for the turnout gap, i.e. the drop-off in turnout among Trump supporters and non-college-educated voters offset what otherwise would have been a Democratic bias.
The point here is — and we’ve seen this in special elections — that polls are most likely to miss in high-turnout elections.
Would you consider the GA runoff and VA NJ 2021 elections high turnout? The polls were quite accurate for those three.