The conventional wisdom on mid-term elections is that the party in the White House loses seats. How many they lose is supposed to be highly correlated to the incumbent President’s poll numbers.
Yet, Democrats suffered only very modest losses–albeit with large consequences since their majorities were so thin–given that President Biden’s approval number was in the lower 40s, a number that historically suggested greater losses.
I think I can explain at least a big part of that: the polls missed something important. The classic question asks what the voter thinks of President Biden’s job performance. The choices offered are Very Satisfied, Somewhat Satisfied, Somewhat Unsatisfied, and Very Unsatisfied. They’ll usually take a “no opinion/undecided” if you offer it. Critically, the polls don’t ask why voters are unsatisfied.
Commentators assume, based no doubt on experience, that voters unsatisfied with the President are more likely to take it out on the party at election time. The more unsatisfied they are, the worse for the party. That logic fails in the case of Biden because a significant fraction of the unsatisfied voters think he is too moderate or insufficiently aggressive against Republican tactics. Those voters are never going to vote for Republicans. They may stay home, contributing to the extent that turnout matters, but if they do vote, they are going to vote for Democrats, and all the more so if the Democrat on the ticket seems like a real progressive rather than a triangulator.
This group probably shows up in all the polling groups: If you think Biden is not progressive enough, you might nonetheless say say you are “Very Satisfied” on the grounds that at least he’s not the other guy. Or, like me, you might say you are “Somewhat Satisfied” on the grounds that while Biden did some good things, I would have liked much more. I can easily see how some progressives might also say they are Somewhat or Very Dissatisfied to have their high hopes disappointed. Until we poll better, we’ll never know how this breaks down, but I’m certain this group exists, and that while not enormous it’s well sizeable enough to move the needle.
TL/DR: Democrats did better than expected because the way Biden’s approval number is measured ignores the existence of Progressives who wish Biden were different–but are never going to vote for Republicans.