From the beginning of the campaign, we’ve been hearing that Pennsylvania is the key state (you might even say the “keystone” state) to Trump’s strategy for victory, because his unorthodox positions on trade and immigration could pull in disaffected blue collar white voters who feel both parties have abandoned them. And, also from the beginning of the campaign, Republicans have fretted about Trump’s racially-divisive rhetoric being a potential problem for the GOP in states like Florida with large and growing non-white and Hispanic population.

So it’s probably worth noting that Trump is consistently underperforming his national numbers in polls of Pennsylvania, and that he’s consistently outperforming his national numbers in Florida.

Trump is also outperforming his national numbers in heavily-Hispanic swing state Nevada, while underperforming in largely white and frequently cranky New Hampshire. States like Georgia, which should be very safe for a Republican, or North Carolina, that should only be competitive in a 2008-style blow-out, are actually looking about as good for Clinton as Florida is. Meanwhile, states like Arizona remain close even as the national numbers have moved strongly in Clinton’s favor.

What do I conclude from the above?

Tentatively, I conclude the following.

First, polarization cuts both ways. Alabama and Mississippi have very large minority populations. They also have very racially-polarized voting. If you’ve got a white majority, even if it’s a relatively narrow one, and you can mobilize that majority to vote as a bloc, then you can win even if the other bloc votes in a similarly solidaristic manner. Arizona, Nevada and Florida all have narrowing white, non-Hispanic majorities — but they are still majorities, at least for now.

Second, achieving that kind of polarization is more plausible when there is a real divergence of interests between the groups. Arizona, Nevada and Florida are all states with large numbers of recent Hispanic immigrants — but also with large white retiree populations. There’s a generational divide that lines up with an ethnic/racial divide, which may drive economic competition between groups that are relative strangers to each other.

That may explain why Trump is doing relatively better with these particular swing states. But why is he doing worse in places like Pennsylvania?

Well, one possibility is that for all his rhetoric, Trump actually has very little to offer on the economic front. His Detroit speech, for example, was mostly a rehash of very standard and longstanding Republican boilerplate. Relatedly, his emphasis on racial and cultural issues may reinforce the impression that he doesn’t have any actual answers to manufacturing decline. But most important, Trump may be losing white-collar whites at a rate that more than offsets any gains he’s making among blue-collar voters.

This may also explain Clinton’s relatively strong performance in states like North Carolina and Georgia. Prosperous, Republican-leaning suburbanites in these states, a contingent that includes many internal migrants from states like New Jersey, Michigan and Ohio, may not be looking to overthrow the establishment, because the establishment is working for them.

Trump is following a version of the “Sailer strategy,” and what he may be proving is that the strategy only works when white voters view their situation as highly precarious and see racial and ethnic solidarity as a compelling response. And while that may be true in certain states, it just isn’t true on a national basis. Instead, a strategy of mobilizing blue collar whites who feel left behind threatens to undermine the position of more successful communities, driving them to the other party.

That’s probably a good thing for the country, overall. But it’s a bad thing for the constituency Trump is speaking to, who need a tribune who could actually get them a seat at the policymaking table, and not just drive them to further marginalization.