David Wasserman has an interesting piece up on the NBC News site on where Trump has bled support from his former coalition. The graphic below is informative though I do think there are some problems with Wasserman's analysis.
His basic take:
"In the wake of a pandemic and the protests following George Floyd's death, voters' support for President Donald Trump has tanked.
His average deficit against Joe Biden in national polls has ballooned from 6 percentage points in March to 9 points in July. In the latest NBC News/Wall Street Journal poll, published last week, the former vice president leads Trump 51 percent to 40 percent — larger than his 7-point lead in the NBC News/Wall Street Journal poll from June.
But a closer look at the cross-tabs of recent surveys yields some surprising findings: Trump is actually performing a bit better with nonwhites than he did against Hillary Clinton in 2016. And the group fueling Biden's polling surge is seniors and white voters with college degrees."
So it's seniors and white college grads, he says. White I agree that both shifts are very important, I have some slight disagreements. First, I think his analysis shorts the importance of the white noncollege vote. The chart below shows white college grads moving Democratic by 12 margin points (+9 to +21) but also shows white noncollege voters Republican margin being compressed from -24 to -18. That's 6 margin points which is not nothing. Moreover, since the white noncollege group is roughly 50 percent larger than the white college group the impact of that 6 point shift is magnified so that the impact on the overall Democratic margin is fairly close between the two groups (white noncollege impact is about 3/4 of the white college impact).
The second point is that his analysis is sensitive to the baseline chosen for 2016. Here he uses the Cooperative Congressional Election Survey (CCES) to compare to his basket of 2020 polls. The CCES is a fine survey but it is not, by itself, the best choice for a comparison. Better, for example are the States of Change estimates which actually combine the CCES data with Census data on vote demographics and actual election results down to the county level. These data when compared to the 2020 polls show a 13 points pro-Democratic shift among white noncollege voters compared to a 14 point shift among white college voters--so roughly the same. But since white noncollege is so much larger a group, the impact of the white noncollege shift is significantly larger.