There has been a bit of relatively polite insulting going on in the election model community. Based upon my observations, Sam Wang started it. But that isn’t to blame him. I think it was basically defensive. All the other big modeling outfits were saying that the Republicans were heavily favored to take over the Senate next year and he just didn’t see it. His argument is a simple one: all the models based upon fundamentals are actually less accurate because they are just throwing in noise. So his approach to models is really the best you can do: look at what the polls are saying.
Today at Vox we see a little pushback, Nate Silver: Sam Wang’s Model Showing Democratic Senate Advantage “Is Wrong.” That’s a provocative headline and distorts what Silver is actually saying. Nate Silver is kind of a punk, but he is still a numbers a guy and is careful about what he says. (He’s kind of like me when I was that age but with huge success that greatly raises his arrogance level.) So we should pay close attention to what he has to say.
His complaint is that Wang’s model doesn’t include the uncertainty that is inherent in polls. In a sense, this doesn’t seem a valid complaint. Wang provides two probabilities. The first is the percent chance the Democrats will hold the Senate if the election were held today. Currently, this is 80%. He also provides the percent chance the Democrats will hold the Senate in November. This number is only 70%. But what I think Silver is getting at is that his model is more like a Monte Carlo simulation, and without getting into the mathematics, they do tend to tamp down strong conclusions. When I was doing Monte Carlo simulations of atmospheric chemistry and climate models, they had a shockingly consistent tendency to produce errors right around 40%, regardless of how complex or simple the model.
Regardless, I don’t see Sam Wang’s model as being all that certain. His current 70% chance of the Democrats keeping control of the Senate is not much different from 65% and 67% probabilities that FiveThirtyEight and The Upshot gave Republicans not long ago, not to mention the 86% that The Monkey Cage produced. Given that all of these models have now come down to roughly 50%, it would seem their greater uncertainty doesn’t mean a whole lot.
I think what is really going on is that Nate Silver is feeling a little underappreciated. And also maybe insecure. In you look at the Vox Senate model roundup, you will see that the FiveThirtyEight model is the outlier. All the models predict the Republicans with 49 or 50 seats, but FiveThirtyEight predicts 52. Now that isn’t quite as bad as it sounds. The model currently gives the Republicans only a 53% of taking control of the Senate. It is just that the most likely scenario is the Republicans getting 52 seats. There is a 14% chance of this. Just the same, there is a 13% chance of them getting 50 seats and another 13% change of them getting only 49 seats.
At this point, the best bet is that the Senate will be evenly divided with the President of Vice to decide. If that turns out to be the case, it will be yet again that Sam Wang looks good, because he’s been predicting the Democrats retaining control of the Senate for a long time. It will be those who claim the importance of political science fundamentals who will have explaining to do.
Afterword
I don’t want there to be any confusion as to where I stand. I think political science can tell us a whole lot about American politics. But in terms of predictions, the only thing that I’ve found that is really important is the economic trend leading up to presidential elections. In off-year elections, it seems to be a mess. I have my eyes peeled for any fundamentals that are important in these cases. But thus far, all I see is a lot of correlation and very little causation.
