Monday, November 7, 2016

Final Model Prediction

This is my final forecast.

If you haven't followed this model over the years, it's done a pretty good job of calling the election. It has called every election correctly since 2000, one year even getting the electoral count precisely right as well. The model uses prediction (betting) markets state-by-state data, multiplying the odds of a candidate winning a state by the number of electoral votes available in that state, and then adding them all up.

So, for instance, if Trump has a 74% chance of winning Arizona, and Arizona has 11 electoral votes, Trump is allotted 8.14 electoral votes (11 x .74 = 8.14). This is a probability weighted outcome.

Right now, that model has Hillary winning 302 to 236.

However, candidates cannot win partial states (with the exception of Maine and Nebraska). As such, in the final forecast only, I force the model to allot votes in each state entirely to one candidate or the other.

Here is how the states break down:



So, the final prediction is 307 to 231 for Clinton. 

God help us.

What needs to happen for Trump to win? First, hold North Carolina, which I think he will. Then, he needs Florida, which has been trending away for a few days, but is also doable. That's 260. Then he could win both Nevada and New Hampshire, which is a bit of a stretch. Or Michigan and Colorado. Pennsylvania would do the trick, or Michigan plus New Hampshire. (Interestingly, several scenarios get you to a tie, which throws the election to the House, which means Trump wins.) Bottom line: while none of these paths is out of reach, they are still difficult. I wish I could say otherwise.

Here's where I feel obliged to note that while prediction markets tend to be the best way to predict political outcomes - better than polls - they are not perfect. Sometimes the wisdom of crowds blows it. Brexit is a great example, and there may be undercurrents in this election that are similar. Specifically, voting for Trump/Brexit is seen by elites as unsophisticated or rube-like. Poll respondents may be reluctant to admit to pollsters their views. If this is a two or three point effect, Trump wins. Prediction markets may be preferable to polls, but they are undoubtedly influenced by them.

The other thing that could cause polls to be off are turnout models. Republicans should be careful about this, because many thought that pollsters in 2012 were intentionally using biased turnout models. Well, they weren't, or they were and got lucky. But while pollsters are very good at predicting how people will vote, they are not that good at predicting how they will actually vote. Right now, most pollsters are assuming about 6% more self-identified Democrats will turn up than self-identified Republicans. Given the highly visible enthusiasm gap - we've all seen the contrasting images of Trump and Clinton rallies - this does seem like a reach. One wonders if all the disenfranchised, poorly educated, mostly white voters, who don't typically vote in high percentages, will turn out in surprising numbers. Anecdotally, it sure seems that way, and just like Brexit, pollsters might be missing it. On the other hand, I might be wishcasting (my new favorite word).

Go vote tomorrow.