Donald Trump and his supporters claim they can rewrite the Electoral College map by taking Rust Belt states away from Democrats. By flipping Michigan, Pennsylvania, Ohio, and Wisconsin from the 2012 election results, Republicans can get to exactly 270 Electoral College votes and win the White House. Trump will do that, advocates argue, by claiming the working- and middle-class voters left behind by the economy over the last 30 years, especially by making trade a key issue in the campaign.

That issue has certainly been at the forefront of the primary debates in both parties for almost a year now. And yet, a new Bloomberg poll of middle-income voters in these four states shows Hillary Clinton with a seven-point lead, and Trump not quite to 40%:

Donald Trump trails Hillary Clinton by 7 percentage points among middle-income voters in the Rust Belt, a key demographic he almost certainly needs to become president.

Likely voters with annual family incomes of $30,000 to $75,000 in Michigan, Ohio, Pennsylvania, and Wisconsin back Clinton over Trump, 46 percent to 39 percent, the latest Purple Slice online poll for Bloomberg Politics shows.

The findings should sound an alarm for Trump because they show he’s failing—at least so far—to dominate among the sort of voters thought to be more sympathetic to him. The poll also splashes cold water on suggestions that the real-estate developer and TV personality is well positioned to win in the Rust Belt.

The news isn’t all good for Hillary in this poll. Despite the natural Democratic advantage in these states — the last time any of them except Ohio went for a Republican presidential nominee was in 1988 — Hillary’s favorables are still negative at 42/56. Trump scores worse at 34/64, while Obama — who won all four of these states twice — only gets a 50/49 despite an uptick in his national job approval ratings. Bernie Sanders has the best favorability rating of anyone mentioned at 55/41.

Also, if Trump is considering Newt Gingrich as his running mate for this Rust Belt-o-rama, forget it — he has a worse favorability rating than Trump at 23/58.

The key to this result comes in the personal-qualities measures. Trump scores well on changing the way Washington does business (50/22), but middle-income voters think Hillary will fight harder for the middle class (43/29), cares more about people like themselves (39/26), and understands their challenges (38/27) than Trump by substantial margins. On presidential temperament, Trump loses badly (22/49), as well as on foreign-policy skills (22/54).

To win in these states, the Trump campaign has to make itself relevant to voters at the local level. As I wrote in my book Going Red, voters in swing states don’t make emotional connections to 30,000-foot messaging and big rallies. That comes through localized, peer-to-peer politics that build emotional connections to the national campaign by contextualizing it into the lives of the voters and their communities. Mitt Romney and John McCain tried the 30,000-foot model and ended up on the wrong end of the “cares about you” question by wide margins, and the elections as a result.

In order to take states and demographics away from Democrats — either on the Rust Belt path or on the more traditional swing-state model in Going Red — the Republican candidate has to outfight Democrats at that level where Democrats have succeeded for at least two straight presidential cycles, and in three of these four states for a lot longer than that. Until Trump runs that kind of campaign, then the Rust Belt model won’t work for him any better than the 2012 swing-state model will.

Finally, in case people believe this to be an outlier, check out the RCP averages for head-to-head polling in Michigan, Ohio, Wisconsin, and Pennsylvania. The Rust Belt and middle-income models are hypothetically valid for Trump, but there is still zero evidence of any success on either path at this stage. Hillary Clinton might well be a terrible candidate, but at least thus far there’s no evidence that she’s actively blowing the election either. On top of that, a GOP campaign that aims at those models at the expense of expanding the reach of the GOP in key demos — youth, African-American, and Hispanic voters — will almost certainly end up losing North Carolina‘s 15 EC votes in November, and possibly Arizona‘s 11 as well. Suddenly we go from 270 to 244 even if Trump’s Rust Belt dreams all come true, which may be better than Romney managed … but still puts Hillary in the White House.