I think we’re finally getting to a place where I feel comfortable about relying on the available data to make a forecast. As a first stab at this, I began with the current Cook Political Report Scorecard.
I then assigned the following probability distribution:
- Solid D = 0.99 chance of a Democratic victory in this state (with its electoral votes, of course)
- Likely D = 0.90 chance of a Democratic victory in this state
- Lean D = 0.75 chance of a Democratic victory in this state
- Tossup = 0.50 chance of a Democratic victory in this state
- Lean R = 0.25 chance of a Democratic victory in this state
- Likely R = 0.10 chance of a Democratic victory in this state
- Solid R = 0.01 chance of a Democratic victory in this state
I’m open to other quantitative interpretations of these qualitative terms (e.g., “Lean D”). I’d especially like to know how often states that we ranked as, for instance, Lean D went for the Democrats in the past. But I haven’t found that information, and the sample size might prove to be too small anyway. I’m also open to corrections about the likelihood that any given state will go one way or another. CPR is a good guide, but it’s not omniscient.
Anyway, the result is that the expected Democratic yield of electoral votes is 279, just 9 more than is necessary to win the presidency. That’s pretty damn close. In contrast, President Obama won 322 electoral votes to Gov. Romney’s 206 in 2012. (Strange, it sure seemed close at the time! And in a way it was; President Obama received only 3% more votes that Gov. Romney.) If the Democratic candidate – let’s be frank – if HRC gets 279 electoral votes, while the Republican candidate – probably Mr. Trump – gets the remaining 259, the 2016 election will be even closer than the 2004 election, in which President Bush beat Sen. Carry 286 to 251.
But the expected number of electoral votes isn’t very helpful without a sense of the variety of ways that things might play out on election day. So I used the data from the CPR to run a 1,000 round Monte Carlo simulation. (It would have been better to have a higher number of rounds, but it was a pain-in-the-neck because each round had 50 independent randomized numbers, which is hard to manage on a spreadsheet, so for now I’ll stick with this for now.) The results were pretty interesting: Democrats received more than 269 electoral votes 63.3% of the time, averaging slightly less than their expected total at 277.9 electoral votes.
However, I’m not ready to say that the Democrats have a 63% chance of winning in November. First, Monte Carlo simulations like the one I ran treat the results of each state as if they were independent of one another. They’re not. A poor debate performance or an email scandal is likely to hurt a candidate in (nearly) every state simultaneously. “Caute,” as Spinoza liked to remind himself. Second, as mentioned above I have reservations about both the qualitative-to-quantitative translation and the qualitative assignments from CPR. Third, it’s still damn early. But this is an improvement, I think, over my prior. For the moment I’m blending my 50-50 prior with the results above: The Democrats have a 56.5% chance of winning the Presidency.