Handyman
Hug someone you care about...
Same way gerrymandering works.
Harris wins California, New York, etc. by high enough margins, and loses Texas and Florida by far lower margins, that the popular vote isn't in question.
But if all 7 swing states go for Trump by a couple thousand votes, he wins the electoral college big.
The mistake in 2016 was that all the aggregators (minus 538) assumed all state polling was independent, and didn't account for the fact that a miss in PA might also indicate a miss in Wisconsin and Michigan in the same direction. Or that a miss in AZ might also mean Nevada was off similarly. So they assumed there was no way that all 7 could miss in the same direction.
Not wanting to make the same mistake twice, they've probably overcorrected and now assume that a miss in one will mean a miss in all in the same direction.
There is definitely some of that going on...though the weird thing is they made that overcorrection also in 2022 and screwed the whole thing up so I wonder if it is as pronounced as we think. I mean all state polling is, in fact, independent. Just because Harris might win WI big doesn't mean she will win Michigan let alone big. (I mean the odds are in that favor but other factors are at play) Pennsylvania being tied doesn't mean the other "Blue Wall" States are going to be super close. There is an obvious amount independence at play. If they are trending the same way then you can link them but I mean Michigan has a bigger say, Gaza problem than WI does so that needs to be taken into account. This is where analysis of the data plays in, something aggregators tend to struggle with.
To me the mistake in 2016 wasn't that per se, it was that people believed that in the end the Democrats who were undecided or non-plussed over Hillary would eventually come home on Election Day in certain states. You could tell pollsters and pundits definitely believed this and I think some of the people like Silver believed that as well because that was conventional wisdom. I don't think they expected a depressed turnout on par with 2012 (2012 had 125 million votes and 2016 had 128 million roughly) they were expecting a 2008-like number closer to 130 million. That might not seem like a huge number in total difference but she really only lost by like 70k votes in 3 states so every vote mattered. Plus you had Jill Stein getting roughly that amount of votes in those states which really screwed up the calculus.
Based on the map shared I agree they seem to be assuming that a miss in 1 is a miss in all...and that is just a ridiculous overcorrection. I could see maybe doing it by region (Rust Belt, Sun Belt...etc.) but issues in the states are way too different to assume any sort of homogenized outcome. Arizona and Nevada have abortion on the ballot, NC has a crazy loonbat polling in the mid 30s as the GOP Gubernatorial Candidate, Michigan has Gaza, WI finally has fair maps and a Supreme Court which protects voting rights...just way too many variables. They are set up to fail.