An $834 billion cut to Medicaid, the government’s health insurance for the poor, is the biggest single change in the health-care bill that the US House of Representatives passed in May and the US Senate is currently considering. The Congressional Budget Office estimated this cut would leave 14 million more low-income Americans without coverage by 2026 than would otherwise be the case.

The map above shows that many of the places hardest hit by these cuts are also places that disproportionately supported Donald Trump during the 2016 election.

The map compares, on a county level, the share of votes Trump obtained with the average share of the population enrolled in Medicaid and other low-income health programs between 2011 and 2015, which likely undercounts Medicaid recipients since the program has expanded more in recent years.1

To read the map, use the color key. The dark blue in the top right of the key represents counties that were in the top third of both Trump votes and Medicaid-enrolled residents. Conversely, the grey in the bottom left represents counties in the bottom third for Trump votes and Medicaid participation.

Where you see dark blue and purple counties, then, you see places that heavily backed Trump but whose poorer citizens are now likely to see cuts to their health care.

Many of these counties are in swing states that played a major role in Trump’s shock election. Note the darker patches in Pennsylvania (which Trump won by  0.7 percentage points), Florida (1.2 pp), Wisconsin (0.8 pp), Michigan (0.2 pp), and North Carolina (3.7 pp). Now these voters may reap the bitter rewards.

For example, in Somerset County, Pennsylvania, 76% of voters backed Trump; 17.8% of the population was using Medicaid and like programs. In Levy County, Florida, 70.6% of voters endorsed Trump and over 26% were also participants in Medicaid and its sibling programs. And in Barron County, Wisconsin, 60% of voters checked off Trump, but 20.5% of them and their neighbors were enrolled in low-income health care program.

It has been argued that the Trump voters in these districts supported him because they resented all the Medicaid users around them as welfare scroungers. But it’s hard to sustain that argument. For one thing, the magenta splashes across the map show there are plenty of places with large Medicaid populations that went for Trump’s rival. And anyway, Trump himself repeatedly vowed not to cut Medicaid spending—a promise that, if the bill passes, will turn out to be a bait-and-switch.

Even voters who backed Trump and aren’t Medicaid users themselves are still likely to feel the effects of these cuts. In rural areas, hospitals depend on Medicaid revenue to keep their doors open; without that backing, even well-off people in those areas will see access care to reduced. And Medicaid is a big funder of programs that fight opioid addiction and other public-health risks that devastate entire communities.

The political implications of this map stretch beyond the president himself. Republican senators from states with high levels of Medicaid enrollment—like Ohio’s Rob Portman, Pennsylvania’s Pat Toomey, Missouri’s Roy Blunt or, though it’s not shown on this map, Alaska’s Lisa Murkowski—all have political incentives to protect the low-income health program, so they may vote against the bill.

Trump is a visual learner. If he has seen a map like this, it might explain why, having thrown a party to celebrate the passage of these cuts before he understood them, he has now told lawmakers that the bill is  ”mean, mean, mean.” The real problem, for Trump, is that it is especially mean to his most important supporters.

Correction: An earlier version of this story misrepresented the data by showing the percentage of Medicare beneficiaries who are eligible for Medicaid, rather than the percentage of all US residents. The map has been updated with new data reflecting estimated enrollment in Medicaid and other low-income health programs for the 2011-2015 period.