#but the poll unskewers had a different methodology AIUI

[tanadrin]

do polls of the whole country tell us anything? don't you need to look at swing states?

National polls aren't useless, because movement in polls tends to correlate. If a national poll moves a couple points in one direction or another, that will tend to correlate with some degree of movement in several states. As I understand it, though, certain subgroups of states (e.g., ones with similar demographics) have much closer correlations in how their polls move, so that a shift in the polls in one Sun Belt state should correspond to a similar shift in the polls in another. This means you can make predictions like "If Donald Trump wins Virginia, he's probably winning a crushing victory nationally," because his performance in Virginia should correlate to his performance in many other states.

Swing state polls are very valuable, but keeping those correlations in mind helps to understand whether a swing state poll is an outlier or not. If a poll shows a shift in a certain direction, but that's not correlated with similar movement in similar states, it's worth questioning of that poll is accurate. Ditto if the poll shows unusual breakdown of results in demographic subgroups: if Trump is winning (say) 30% of young black voters, given the way demographics and party alignment usually break down, he should be winning a massive margin with other groups.

One reason I am not so bearish on Biden is that my understanding is that a lot of polls have had these demographic anomalies, with Trump's lead coming largely from support among younger, politically disengaged voters of color, and Biden, apparently, doing well with demographics like older whites. It is not a coincidence, in this view, that Trump seems to be performing unusually well with demographics that are particularly hard to poll in the modern polling landscape--response rates to telephone polls are very low among millennials and gen Z--and while there are various ways you can try to compensate for non-response bias, those depend on your model of the electorate.

Now, I am not extremely confident about this, because I am the furthest thing in the world from a polling expert, but as I understand it, there are two possible situations here:

One: the polls are broadly correct, and Trump is ahead. The election in November, if current trends continue, will feature a historic realignment of voters along demographic lines like age and race of the likes not seen since the 1960s (called "depolarization" by some commentators), perhaps driven by the rise in far-right internet media and social media.

Two: the polls are broadly incorrect, and we should be more agnostic about the state of the race, or even assume Biden is a little ahead, because such a massive realignment is extremely unlikely to have occurred in only two years since the 2022 midterms (where no such realignment was in evidence, and Democrats broadly overperformed polls), and polling right now is plagued by historically low response rates in the same key demographics that give Trump his lead.

Some commentators, including commentators whose field is polling, seem to want to have it both ways: the demographic crosstabs are wrong, but the top-line polling numbers are right. I'm not sure how this can be true. On top of that, big political realignments usually take time (i.e., we should have at minimum seen some evidence of this coming in 2022), and are unlikely to occur in a race where both candidates have been president before.

So on balance I think the second scenario is more likely. Now, I am not a stats person, nor particularly knowledgeable about polls; all of this opinion is second-hand from other commentators. As such, I am not going to claim any kind if ironclad certainty about this, and you're perfectly entitled to rub it in my face if I turn out to be totally wrong. And if I do stumble across someone who does know the polls really well with an explanation of why I'm wrong (even just at the level of "you are factually wrong, here's why the crosstabs are actually perfectly normal") I may well revise my opinion.