Yup, I'm still doing this.
Let's get some basic facts about the electoral college out of the way. Here's how it works, in very brief form. For each state, the parties create a slate of electors, numbered according to the total number of House members and Senators allocated to the state. If a candidate gets a plurality of the vote in the state, that candidate's electors get to be members of the electoral college. After the election, the electoral college votes, and whoever gets a majority of the electoral college wins. Yes, there's more, but that's the executive summary.
It's a very screwy system, and it has the potential to give a majority of the electoral college to someone who didn't have a plurality in the popular vote. How often does that happen? Not often. It happened in 2000. Sort of. It happened in 2000 because of the official tally in Florida. Here are the officially certified results, according to the FEC. Of course, the butterfly ballot had a lot to do with that. (Truth in advertising: Henry Brady was one of my grad school professors). Regardless, look at how close those numbers were in 2000. The popular vote-electoral college split doesn't happen without the candidates being nearly tied at the national level.
What happens when they aren't that close? Well, take a look at the current RealClearPolitics map, when you just go with the current polls. That puts the electoral college vote at 340-198 Clinton, even though her polling lead is only 6.7 points in the two-way contests and 5.3 points in the four-way contests. What's the deal?
Here's the deal. While the electoral college can create a split in a tied race, the normal effect is to exaggerate an advantage. Clinton has a lead, and the electoral college is exaggerating it, making a 6.7 point lead, or possibly a 5.3 point lead look like an electoral college blowout. Why? Because winning Florida, even if it is only by 2 points, translates into all 29 electoral votes. Winner-take-all systems magnify small advantages.
Clinton is leading, and the electoral college magnifies that lead.
Now, what does any of this have to do with Nate Silver? He is looking at the electoral college to the exclusion of all else. Forest, trees, or something. Why? Well, he's a sports statistics guy. Me? I hate sports. All sports. Every last one of 'em. Election night is fun. I love watching the returns come in, and they do so by state. So, hey, let's run simulations by state! Silver turns it into a combinatorics problem. Combinatorics is the branch of mathematics associated with counting methods because there are multiple methods to count to 270 (538 electoral votes total, and 270 is a majority), so he runs simulations of ways for each candidate to get to 270. This ain't fantasy sports, though. In statistical terms, all of the data are hierarchical. That means everything is subject to some combination of local and national forces, and separating them, along with separating the over-time effects is brutally difficult, making the simulation approach unwieldy at best. But, when the national effects are overwhelmingly large, there is no need.
Remember, the electoral college magnifies national leads. To ignore that effect and focus on state-level simulations is to miss the forest, or, in statistical terms, miss the hierarchical effect that causes winner-take-all systems to magnify national leads. A popular vote-electoral college split can happen. But not when one candidate is as far ahead as Clinton is now. It happened (sort of) in 2000, but that was when Bush and Gore were separated by about half a percentage point in the popular vote at the national level. That's just not the case now.
Yes, this year is crazy, but really, there is a forest here, not just some pretty trees.
Can Trump win? Yes, but something needs to change at the national level, unless all the polls are wrong, in which case Silver's combinatorics game is misguided anyway because no statistical approach can work with data that far off.