Predicting the Electoral Map
[Note: The analysis presented in this post relied upon poll and model data available as of August 1st. The probabilities stated and used herein were derived directly from that data. Polls and model results will inevitably fluctuate over time, however the major results and conclusions of this piece have not been significantly affected.]
Dependent Probabilities
There are many great models out there for predicting the electoral outcome of the 2016 presidential election. Fivethirtyeight presents its model in three different forms, and The Upshot tracks several others. Many of these models rely heavily on individual probabilities of victory that have been estimated on a per state basis. The modeler then runs thousands of simulations (20,000 in the case of fivethirtyeight), based on all those state-by-state probabilities. The problem here is that, as far as I can tell, these models assume that the outcomes of each state are independent.
For example, they might say there's a 75% chance Clinton will win Virginia, and separately, a 50% chance she will win North Carolina. If we treat these as independent probabilities, however, it would suggest that there's a 13% chance she will win North Carolina, but not Virginia. That sounds like a small number, but I would suggest it should actually be much smaller. Demographically and politically, VA is quite similar to NC, except that VA moderately favors Clinton over Trump (Clinton polls about 5% points ahead of Trump in VA, while they are essentially even in NC). The same relationship between these states has been observed in previous elections. NC consistently is about 5 points more Republican than VA, and that's a pretty sizable margin. Therefore, it is exceedingly unlikely that Clinton would lose Virginia but win North Carolina, because that's not how they're linked.
Noncompetitive States
This is not just a problem for this one pair of states. There are similarities between many, many states that suggest that they are likely to as a group vote similarly. What happens in North Dakota is correlated with what happens in South Dakota. This brings me to a second criticism of the models I've seen: does South Dakota really have a 20% chance of voting for Hillary Clinton? Any reasonably politically-attuned person would put the chance at 5% or lower -- wouldn't they?
And this problem doesn't just appear for South Dakota. All over the place, modelers are hedging their bets on states which we know -- through several sorts of information -- are extremely likely to vote a particular way in the 2016 presidential election. Minnesota has been the most consistently Democrat state in the modern political era. Reagan didn't win it in either of his landslides. You have to go back to George McGovern's failed 1972 bid -- in which McGovern won only Massachusetts and the District of Columbia -- to unearth the traces of a Republican winning MN.
Unhelpful Hesitation
Now of course historical patterns aren't indefinite guarantees. At some point, a Republican presidential nominee will perhaps win Minnesota. Just not this time. In 2016, it turns out, MN looks much like it has in past cycles. This habit of suggesting that a state is competitive when it really isn't appears to be fairly pervasive, from what I've seen. The obvious explanation is that the political media would like their audience to think the race is closer than it is -- there's so much up in the air and anything could happen! Also, I suppose they may just want to avoid making any prediction that could possibly turn out wrong. At the end of the day, if they're going to say that California and Oklahoma are the only sure things, then there's very little reason to tune in. There's very little reason to look at a map like the one below.
Georgia is not a swing state. Trump has a huge advantage there. And Clinton will win Oregon. And very likely Virginia. And why is Missouri not red? And Michigan not blue? How does Mississippi only lean Republican? The truth is, we know pretty well how this election will turn out, especially now that we are post-conventions. Together, the history of our electoral politics and piles of fresh polling data are pretty powerful predictors.
Being Realistic
A more honest map -- colored just by what we know, and not a financial need to have you check back on it daily -- would look more like this.
Now, I don't mean to remove all gradation and suggest that all states are either A) obviously Republican, B) obviously Democrat, or C) a 50-50 toss-up. Obviously, it's more likely that Trump will win NC than NV. I just want to be more realistic about the vast majority of states. It's kind of silly, I think, to estimate a probability of the winner on a state-by-state basis, run tens of thousands of replications of a Monte Carlo simulation, produce electoral scenarios that are fairly often absurd, and then count those all up to estimate an overall probability of one candidate winning. There's only going to be one election, not 20,000. And Clinton isn't going to lose Virginia but win North Carolina. I would argue that the vast majority of "potential" outcomes are simply statistical imaginings.
Now, if you know your electoral map, you probably realize that the version above looks quite good for Hillary Clinton. She already has 268 electoral votes and needs just 2 more to reach the magic number: 270. Winning any of the six toss-up states on the board will put her over the top. Note: winning Maine's second congressional district would put her at a 269-269 tie, in which Trump would need the Republican House to decide it for him. That detail aside, it's very easy for Clinton to win at least 1 of 6 toss-up states.
So the verdict is...barring any major changes you don't really need to bother to follow this election.
A Very Simple Analysis
Well, okay, maybe not quite. There's one important assumption built into the electoral map above: that Pennsylvania is safely a Clinton win. The history and polling indicate that it is, by my reading. But, let's for a moment imagine what Donald Trump's plan would have to look like. It's been widely suggested that Trump would need to win all the top prizes: Pennsylvania, Ohio, and Florida (also known as the "Big Three"). They are, unsurprisingly, the three states that both A) have lots of electoral votes, and B) appear reasonably close. Ohio and Florida are in fact veryclose. And that should come as no surprise. So then, the real problem standing in the way of a Trump victory is Pennsylvania. The million dollar question is: what really are Trump's chances to win PA?
Looking across about half a dozen models, the current estimates fall between 15 and 30%. Let's take the most optimistic case for Trump: his real odds of winning PA are 30%. So, what would that mean? How might he win the election? What are the realistic scenarios? I'm going to use an extremely simple model to answer that question, because I think it's all that's needed.
Let's assume there are five toss-up states: Nevada, New Hampshire, Florida, Ohio, and Iowa. We'll assume they're all 50-50 (quite plausible).
We'll give North Carolina to Trump because he's a bit stronger there and it'll offset any low-probability weakness of Clinton in some state already given to her.
We'll also give Trump Maine's second congressional district, just to make things simpler.
Assuming Trump has a 30% chance of winning PA, we will evaluate what might happen if he wins or loses, and come up with an overall estimate for the likelihood of a Trump electoral win by looking at just a few states.
There's Just One State to Watch -- Or Maybe Three
A pretty simple model, right!? And it just needs 30 seconds of arithmetic, not 20,000 simulations. Looking at the results, one fact jumps out: there's a 70% probability Trump loses PA, and he has virtually no hope in that case. There's something like a 3% chance that he could still sweep all five toss-up states to miraculously pull together a win. Trump would be much better served, however, to look for a victory that depends on that 30% chance in PA. A PA win is an excellent way to start. It doesn't guarantee electoral victory, but it looks good. If Trump wins PA, I assume that Ohio and Iowa also go to Trump, because they are easier to win with the same voters that would give him PA. Additionally, I assume that a win in PA would mean that Virginia is then also in play, so I make it a 50-50 toss-up as well.
So, given a PA win, that leaves 4 toss-up states: FL, NV, NH, and VA. At this point, Clinton has 235 electoral votes, and Trump has 251. So, out of those 4 toss-up states, Trump needs to find 18 more electoral votes. We'll assume that these four states will in fact decide independently (because they are in this case quite different), and that gives us 16 (2^4) equally likely outcomes for the toss-up states. In 10 of those 16 outcomes, Trump gets enough electoral votes to win. In the other 6, Clinton holds on despite the PA and OH losses. That means, given a PA win, Trump would have about a 63% chance to become president. Accounting for the unlikelihood of the initial PA win, Trump has about a 21% chance to become president.
Here's an interesting fact about those 6 Hillary-holds-on scenarios: each one involves her winning Florida. It's a huge prize (29 electoral votes -- that's more than the other three toss-ups combined), and it's really her only way to make up for losing both PA and OH. However, it's not quite a golden ticket. Trump can still overcome a FL loss if he picks up VA and NV (regardless of the result in NH). So he doesn't absolutely need the Big Three. But realistically, he does.
So, here's my prediction for how election night will go:
Trump probably loses Pennsylvania, and it's an early night. Cable news executives cry themselves to sleep.
There's a chance Trump runs close in PA, which means he probably wins OH, IA, and NC, and is also close in VA.
Assuming he can actually pull out wins in those states (PA, OH, IA, NC), then it all comes down to FL.
If Trump can win Florida (probably not too hard by this point -- the most difficult wins are behind him) then he wins with the Big Three.
No, Colorado or Wisconsin or Georgia or Arizona is not going to be the key. I'm sorry. Electoral politics are just not that interesting.
Then, strap in for a rather similar 2020! (Looking at you, Ted Cruz.)