Early Wednesday morning, The Washington Post and our polling partners at ABC News released polls in two Midwestern states that are critical to the results of the 2020 presidential election: Minnesota and Wisconsin. In each case, Democratic presidential nominee Joe Biden leads — by a wide margin in Minnesota and a narrow one in Wisconsin.

To some extent, those results mirror the 2016 contest, in that Donald Trump lost Minnesota (narrowly) and won Wisconsin (even more narrowly). More broadly, though, the Post-ABC polls appear to reflect some significant shifts in each state that put President Trump at a significant disadvantage.

Comparing these results with 2016 exit polling is tricky for a few reasons. One is that our poll reflects the state of the race at the moment, meaning things will probably change to some extent before Election Day on Nov. 3. Another is both our poll and exit polling include margins of error that introduce more uncertainty than hard numbers can capture.

Nonetheless, the comparison is revealing. Here are demographic comparisons between likely voters in our poll and the 2016 exits, looking at margins of support then and now.

One thing is immediately apparent: The shift has almost uniformly been in Biden’s direction. Republicans in Wisconsin are more heavily supportive of Trump relative to Biden than they were supportive of him four years ago, but other than that Biden’s doing better in nearly every case.

In some cases, Biden’s doing far better. Among women in Minnesota, and White women in particular, Biden’s advantage over Trump is about 30 points larger than what Hillary Clinton saw in 2016. That year, White women preferred Trump by a narrow margin; now, they prefer Biden by a wide one. Independents in Minnesota similarly flipped from a narrow Trump advantage to a big Biden lead.

The changes in Wisconsin are more modest but still important. Most noticeable is the improvement Biden has seen with older voters. A fifth of the electorate in Wisconsin was age 65 or over in 2016, and they were about evenly split (with Trump leading Clinton by a point). Now, Biden leads by 17 points. That helps give him his overall lead in the state.

Keen political observers will quickly point out that polling in 2016 also showed Clinton with wider leads right before the election than she actually enjoyed. They would be right. FiveThirtyEight’s average of state polls at the end of the 2016 race gave Clinton a five-point lead in Wisconsin and an eight-point lead in Minnesota. Those were off by four and six points, respectively.

It is nonetheless the case that Biden’s advantage in FiveThirtyEight’s current averages is still as large or larger in each state than at the same point in each of the past five presidential contests. With the exception of Wisconsin in 2016, the Democratic candidate won both states in each of those elections.

Trump has aimed at taking Minnesota and probably needs to hold Wisconsin to be reelected. If the shifts among women and older voters seen in the Post-ABC poll reflect an actual shift in the electorate — a fair assumption based on a wide range of other state and national polls — that seems very difficult to do.

Particularly given the rate at which the election is approaching. Polling in each state has moved within a fairly narrow range, as has the race overall. Trump needs either a dramatic shift in the race to gain a lead or an even bigger miss on state polls, a hope that would necessarily depend on pollsters making the same mistakes they did four years ago.

Given the stability in the race, such an error might be Trump’s best bet.