@matt_blackwell He should just post a graph showing polling outcome variance declining as we get closer to the election. He didn’t need to get into the part about the binomial distribution.
@arindube@NateSilver538 Agreed. Seems easy to say that someone is herding (for some sets of outcomes) and basically impossible to say that any individual pollster is herding.
@gelliottmorris I agree with all of these criticisms of pooled battleground state polls but national polls are even worse (for the purpose of predicting an election).
@gelliottmorris In your screenshot the point estimate for PA is Harris +0.9. In the current 538 polling average it’s Harris +0.1. That’s a big difference!
@matt_blackwell My understanding is polls are weighted so the topline is representative, but subgroups are not. So if for some reason black men are more likely polled than black women, that would show up as low black support for Harris. Is that right?
@binarybits Is there a rule taught in journalism school that for every good thing you’re required to find worst aspect of it and pretend the downside is important? This is like people complaining that wind turbines don’t get recycled.
@arindube@MayankSeksaria I agree that when Harris’s number was around 38 it was very predictable that it was going to move towards 50, but it’s asking a lot for his model to handle the DNC well without an ad hoc adjustment given the candidate switch.