JavaScript is not available.

The main difference between our polls and recent national polls, to me, stems from two things. 1) The EC-battleground gap, which was 4 points in '16 and could easily grow in '20, esp among RVs. This is basically what we thought the result should look like, back in July

109

2) Unlike national polls, we should Trump holding up with white, working class voters. Take the Fox poll v. Biden yesterday : White, no degree. Fox: 37/50, Upshot: 34/58 White, college. Fox: 50/42, Upshot: 50/41 Hispanic. Fox: 60/28, Upshot: 63/29 Black. Fox: 89/6, Upshot: 82/8

Obviously, white working class voters are a group pollsters have struggled with in recent years, especially in state polls. We also do a lot that can help us here that others don't, like strata on party x region, weighting on education, and response rate adjustment on turnout

I should note that this is methodologically identical to our live polling from last fall But Michigan was a real struggle. We cut our responses there to 500 (v 650 elsewhere), due to bad productivity. High design effect required weight changes. All described in method page.

That said, collective sample here is N=3766 and more than 5000 if you count Iowa. The picture, I think, is very clear over all and it's not really subject to the margin of error. Our sample was quite clear, as well (left column = unweighted; right = weighted)

Another slight difference is that our poll doesn't include people who aren't registered to vote, but say they will. When you take those still unregistered voters out of the Marquette poll, our results (already close) snap into alignment (thanks

for the cut)

And if we had not weighted by party/education, like most of the you've seen, our results would have been: Biden+5 Sanders+3 Warren+1 (I should note that we stratify our sample on party reg x region, which would tend to diminish the importance of weighting on party)

(IE: in FL, our respondents are 37% dem, 35% rep by registration, since that's the state tally; it is possible that if we just collected sample without stratification, like most polls, the underlying sample may have been more D/R in unknown ways)