There are three arguments that Donald Trump has made over time to explain how he'll win the White House. He began with vague assurances that he'd win, without providing many specifics. Then he evolved into naming states, seemingly at random. (He has, at times, mentioned deep-blue states like his home state of New York, which he will not win.) Of late, he's homed in on a few more-feasible options: winning the Rust Belt states of Pennsylvania, Ohio and Michigan in November.

More feasible, but not entirely feasible. Pennsylvania and Michigan have voted Democratic in every election since 1988. (Ohio is a swing state, of course, so that's a bit more realistic.) Central to Trump's argument is that he'll increase turnout and support from working-class white voters, enough to counteract votes from heavily Democratic (and less-white) parts of each state.

On Thursday, Bloomberg Politics released a poll that cast some doubt on that happening. Pollster Purple Strategies surveyed voters in Wisconsin, Ohio, Pennsylvania and Michigan who earn between $30,000 and $75,000 a year -- what they call "middle income." Their choice for president? Hillary Clinton, by 7 points.

There's an obvious question here, though. That's all voters. What about the white voters that seem central to Trump's argument that he can win? Bloomberg broke that group out, too, writing:

When only the white middle-income participants are considered, the billionaire beats Clinton 44 percent to 40 percent. ... That's behind where Republican Mitt Romney finished with white voters against Obama in the last election, when exit polls showed the former Massachusetts governor winning 59 percent of the white vote.

Not only is that behind where Romney was, it's well behind what national polling shows.

The Post and ABC News released polling this week that included a question looking at how white voters at various income levels looked at the two candidates. These are national numbers, not specific to the region under consideration (the Midwest), but there's a wide gap from the Bloomberg numbers.

This includes a lot of white voters in strongly Republican parts of the country. If we narrow the region encompassed by our numbers to the Midwest, Trump's lead shrinks, but the numbers are too small to allow us to be confident.

In 2012, Mitt Romney won Ohio by 16 points among white voters, according to exit polls. He won with that group by 11 in Michigan and by 3 in Wisconsin. Romney lost those states by 3, 9 and 7 percent  respectively. To win the White House, Trump needs to improve on Romney's performance. We shouldn't read too much into one poll, but the numbers in the Bloomberg survey are not what Trump needs -- by a wide margin -- if he's to sweep the Rust Belt or even pick off a couple of states.

There's another option, though: Trump can always pick other states that he'll promise to win.

The Bloomberg results are based on an online survey conducted May 18-24 among 803 likely voters in each state. Respondents were recruited to take the survey through through an opt-in online panel. Because the survey is not a random sample of the population, no margin of sampling error can be calculated describing how far results might differ from the overall population.