One question that we get a lot about our forecast model (and that wasn’t answered as part of our model Q&A that we had a few weeks ago) tends to come from people who look at our map first and start counting electoral votes from there. Take the map above, which you can find on our presidential forecast homepage: It has 273 electoral votes in the “Lean Democratic" category or better banked for Hillary Clinton, in the states that are various shades of blue. That's a win, but it’s sort of the bare minimum, only narrowly clearing 270. That map hasn’t changed much over the last month (if you look back to when we unveiled it on Sept. 9, the only difference among the key swing states was that Nevada was shaded gray).
What causes confusion is that each day we also show a projected electoral vote margin. You can see it in the ticker at the top of the page: On Thursday, it gave 288 EVs to Clinton and just 250 to Donald Trump. The same problem also pops up a lot in our Senate forecast (though not today, as we’ll discuss shortly … which is good news!). Many days, the number of races that show the Democrat with better than 50-50 odds would add up to only 49 seats, and yet we'd be showing a projected result of 50 seats for Team Blue, which makes all the difference in terms of a majority. So why the difference?
What’s happening is that we’re looking at the full range of possible results, and the number we show at the top of the page is the median result of all the thousands of simulations we run. In other words, 50 percent of all of our simulations will end up with more electoral votes than, say, 288, while 50 percent of all of our simulations even up with fewer than 288. It’s not simply what you get when you look at each state individually, and then total up which ones have better than 50 percent odds and which ones have worse.
Take Florida, for instance. It's currently at a position where Clinton wins it in 48 percent of simulations and Donald Trump wins it in 52 percent of them. That means that you’re going to have a lot of simulations where you end up with 273 electoral votes (where Clinton wins the big four of Colorado, New Hampshire, Pennsylvania, and Virginia, along with Michigan and Wisconsin, but that’s all), but then you’re also going to have almost as many simulations where she ends up with 302 electoral votes. So it stands to reason that the median result is going to be somewhere between those two points, in either the high 280s or low 290s.
In fact, if you look very closely at the histogram of all the simulation results, you can see that it’s mostly a normal distribution, but you have some little blue spikes where the most common permutations are. You’ll notice that, in fact, 273 and 302 are the most common results, followed by 308 (which is what you get when you win Nevada as well as Florida). That looks like it’s followed by 279, which is what you get when Clinton wins Nevada but doesn't win in Florida. And when you look at the overall bell curve and where it crests (in other words, the median), it's somewhere in between those points, in the 280s.
The same disparity has happened many days in the Senate projection: the Democratic candidates had odds above 50 percent of flipping the seats in Illinois, Indiana, Pennsylvania, and Wisconsin, while below 50 percent odds of holding the open seat in Nevada … which, if you add them up individually, would leave the Democrats with only 49 seats. However, the median result of all the thousands of simulations we’d run would often be 50, because there were more permutations where we’d also pick up one of the other seats with sub-50 percent odds, like also flipping New Hampshire or North Carolina, or maybe holding Nevada.
Think of it this way: each state in the electoral college, or each state’s Senate race, is a badly lopsided coin. If you have a lopsided coin that ends up with “heads” only one out of three times you flip it, and you have only that one coin, chances aren’t very good you’re going to get “heads.” But if you have four similarly lopsided coins, chances are better than 50 percent that out of those combined four attempts, you’ll get at least one “heads.”
Notice that I said “most days,” though, because today’s not one of those days. We’re actually at a point where there are a net four Senate races where the Democrat has 50 percent chances or better. There are Illinois, Indiana, Pennsylvania, and Wisconsin, while still losing Nevada … but now for the first time, we’re showing better-than-average odds for Deborah Ross in North Carolina! She now has a whopping 51 percent chance of defeating incumbent Republican Senator Richard Burr.
This race didn’t start high on anyone’s radar, compared with opportunities with more high-profile Dem candidates in New Hampshire, Florida, and Ohio. But a strangely desultory effort by Burr has helped Ross gain a lot of traction. There were four different polls of the North Carolina race released on Monday, and two of them showed Ross in the lead, which is enough for her to pull into a near-tie in both the polling aggregate and projected odds. And the improved odds in North Carolina are enough to almost single-handedly push the Democrats’ overall odds of 50 Senate seats back into positive territory, currently at 56 percent overall. (Improving odds for Katie McGinty in Pennsylvania also help.)
The number that probably first caught your eye, though, was the 68 percent overall odds for Hillary Clinton in the presidential race, up from 64 before. You might be thinking, “duh, that’s because of the debate bounce,” but, no, it isn’t. There haven’t been any state-level polls with post-debate field periods released yet (although by the time you’re reading this on Thursday, several PPP state-level post-debate polls will probably be out … but they will not have been added to the model yet).
Instead, it seems to mostly have to do with improving Clinton odds in Florida, and in North Carolina (where all four of those polls released on Monday, even the one from Gravis, had small Clinton leads). There were some subtle improvements for Clinton’s national numbers late last week (she’s currently back up to a 5-point lead nationally, according to Huffington Post Pollster’s methods), and the many polls we saw on Monday (notwithstanding that CNN poll giving Trump a 1-point lead in Colorado, which was the one real bummer in that bunch) may reflect some of that increase trickling down into the state polls, which, as we often point out, tend to lag the national polls.
That’s good news, because if there is a broader bump coming for Clinton as a result of her strong debate performance on Monday, that will just get layered in on top of what’s already looking like some subtle improvement from before Monday (which may simply be a result of the “differential response” by Democratic leaners in the wake of Pneumonia-ghazi two weeks ago starting to fade into the distance).