In the final stretch of 2016, some analysts were able to detect loud, clanging alarms ringing for Hillary Clinton beneath the surface that most of us could not. They had access to polling conducted at the congressional district level, which showed Donald Trump gaining ground big time in White working-class House districts.

But most of us don’t have access to such data, which is often conducted by campaigns and party committees.

David Wasserman, who analyzes House races for the Cook Political Report, has access to an enormous amount of this data. Wasserman has been tweeting that this granular polling is telling a different story this time, one in which Trump is not matching 2016 totals in White working-class House districts and falling further behind in districts heavy on college-educated Whites.

I contacted Wasserman to ask about all the data he’s been seeing. There’s tons of it, much of it private, and it shapes ad strategies in contested House races. It’s often shared with nonpartisan analysts. Wasserman told me it’s particularly valuable, because it “can give you a rare window into micro-shifts among certain demographic groups that are harder to detect at the state level."

I spoke with Wasserman about what all this data is telling him in this regard. An edited and condensed transcript of our conversation follows.

Greg Sargent: In recent weeks, how many of these polls have you seen?

David Wasserman: Hundreds — in the most competitive band of around 50 districts.

Sargent: Why are they a good window on the presidential race?

Wasserman: You can measure how far ahead or behind 2016 margins Trump is running. In 2016, district-level polling in late October showed flashing red warning signs for Clinton in districts dominated by White non-college voters.

It wasn’t being detected so much in state-level polling, because the state polling chronically under-sampled those voters.

Sargent: What are you seeing now?

Wasserman: In my 13 years of covering House races, this is probably the most consistent cycle I’ve seen: Trump is underperforming his 2016 margins by eight to 10 points in most competitive districts. If Trump won a district by three last time, he’s probably losing it by six this time. It’s a pretty consistent pattern.

There are exceptions. The first is, in districts that are heavy on upscale suburbs, Trump is underperforming his 2016 margins by even more — over 10 points. Examples include Indiana’s 5th district in the Indianapolis suburbs, Missouri’s 2nd in the St. Louis suburbs, or Arizona’s 6th in the Phoenix suburbs.

The second exception is in heavily Hispanic districts, where we’re seeing Trump come close to or exceed his 2016 performance. In South Florida, Trump is clearly doing better. But he’s also doing better in South Texas and California’s Central Valley.

Sargent: That’s interesting, because it goes counter to what some have said — that there’s a difference in political orientation between Florida Latinos and Latinos in the Southwest and West.

Wasserman: Certainly Trump has gained more in South Florida than in South Texas. But we’re still seeing modest gains for Trump in heavily Hispanic districts outside of Florida.

Sargent: How big a bearing do you think this has on the presidential race?

Wasserman: It basically confirms the trend in national polls. Nationally Biden is up about 10 points. That’s an eight-point margin swing from Clinton’s two-point popular vote win. National polls also show Trump doing better with Hispanics than four years ago, and worse with White voters, particularly seniors, college-educated Whites and blue-collar women.

Sargent: What is the polling showing in districts that are heavy on blue-collar Whites?

Wasserman: The picture is also pretty clear in those districts. Biden is running ahead of Hillary Clinton by a good bit, but still not back to where the Obama-Biden ticket was in 2012. Biden is getting halfway back to 2012.

Sargent: Does that apply in places like southwest Pennsylvania and northern Michigan?

Wasserman: Yeah, and that’s exceptionally meaningful. Trump clearly needs to get back to his 2016 numbers in those places to have any chance of winning.

Sargent: What are those polls showing in [Rust Belt working class white] districts?

Wasserman: The same thing: Biden is getting halfway back to 2012 numbers for Democrats. There are exceptions in districts where Trump’s trade policies are popular — in the Iron Range of Minnesota, we’re seeing Trump support approach 2016.

But take southern Minnesota, which is more farming. We’re seeing much better numbers for Biden than for Clinton. So it does depend on the economy and cultural attachment to Trump.

The main point is that Biden is doing better than Clinton across the board in White working-class districts. But the extent varies by region.

Sargent: Biden is essentially rebuilding the “blue wall” that Trump cracked by getting at least halfway back to 2012 levels with non-college whites throughout the region. Then you tack on the gains among college-educated Whites, which seem more substantial. Is that the basic reading?

Wasserman: Biden’s gains with college Whites are probably more durable, because those voters appear to have abandoned Republicans altogether.

[But] even though Biden is polling really well in non-college districts and getting closer to 2012 numbers, those gains are more fragile. We know Trump has won these voters before.

Trump is poised to benefit from a much larger turnout among White working-class voters than 2016. The problem is he’s not winning a high enough share of them. He needs to recover ground in particular among blue-collar White women. That’s where we’ve seen the bulk of Biden’s gains.

Sargent: Doesn’t Trump have to get those levels up even higher than 2016 to compensate both for demographic change and for Biden’s broadening of the gap among college-educated Whites?

Wasserman: Yes. Trump needs to boost turnout of non-college Whites by five points nationally, just to offset their declining share of the population since 2016. But he also needs to increase the share of those voters he’s winning.

Trump’s gains among non-Whites can only get him so far, because there’s really not much of a Hispanic vote in Pennsylvania, Michigan and Wisconsin. So he’s got to solve this riddle with both persuasion and turnout. He needs to persuade more White voters — both college and non-college — to stick with him. And he really needs to boost non-college White turnout.

That’s a lot to ask in 14 days.

Watch Opinions videos:

Read more: