A recent NY Times article flogged a recent poll showing Biden trailing Trump in five battleground states (data available here). While there is something useful to be drawn from the poll (we’ll get to that), there are serious problems with the poll.
The key thing is battleground states are close (hence the name), so the difference often comes down to subtle shifts in small demographic groups–and in these polls, the actual counts of these groups are quite small: for example, the PA data for all Black voters are based on 56 total respondents. Once that gets subdivided by age, gender, propensity to vote and so on, any oddities are going to skew the results. This can be compounded by assumptions about who turns out to vote, adding another source of skew.
Speaking of Black voters, all of the state level polls have Trump getting 20-21% of the Black vote. Not 20-21% of a subset of the Black vote, but the entire vote. Remember: Trump has run twice (‘the only poll that matters’), and never come close to that, so either Black people have radically reconsidered Trump, or something is very wrong with how the poll is weighting small samples. Consider that the range of counts Black respondents of Black voters is between 19 – 173 (even n = 173 could lead you astray). And nn Wisconsin, which according to the poll, is the one state Biden *wins*, Trump gets 23% of the Black vote.
All that said, across the six states, it does appear that the bullshit claim of Biden being doddering has caught hold (the numbers appear to be high enough compared to Trump that a few percentage points of error in the point estimate don’t matter). Professional Democrats need to do something about that–and they could start by doing some polling that includes Trump’s age in the poll (a 77 year old man who, at times, slurs his speech, is not ‘young’), and then promoting that polling.
So these will be close races (battleground states!), but this particular poll isn’t telling us much, probably because it’s overspecified and has small sample sizes.