Those who follow me on Facebook know that I am not a fan of Nate Silver. Over the last few election cycles, many in the media have falsely claimed that Silver is a “political scientist” because he can “accurately forecast elections”. However, both of those statements are 100% incorrect, and this is why.
First, political scientists use theory to create their forecasting models. Basically, they say that A causes B, with B being vote choice. For example, most of the election forecasting models posit that if the economy is good (which is A), then the voters will reward the incumbent party (which is B). With Nate Silver, he does not have any theory at all. He does not suggest any causation as to why voters decided to vote R or D. Instead, he just looks at averages. Therefore, he just does math and should not be considered a political scientist.
Second, forecasting based on theory requires a few things. According the Michael Lewis-Beck, forecasting requires four major components: accuracy, lead, parsimony and reproductibility. In the case of Nate Silver, he violates two of these, lead and parsimony. The idea of “lead” is that you can predict an election result way before Election Day. Therefore, predicting a result just a day before the election is not really a forecast. It would be similar to saying that you are forecasting the weather one hour before, and basing that weather forecast on the big storm cloud approaching. Therefore, it really has no utility…especially since FiveThirtyEight updated their Florida prediction numerous times throughout the day before Election Day.
Additionally, his models aren’t parsimonious. Each election, he adds additional variables which make it harder to follow. The least amount of variables that you need, and the sooner you can make a prediction, makes for a better forecasting model.
So, with that being said, what about the political scientists? How did their models do? Well, they were a bit all over as well. But unlike the people who guess hours before Election Day, some models by political scientists did predict a Trump win months ago. Instead of going explaining every model, I am going to list the models and the results, with a big shout out to the website PollyVote for listing the various forecasting methods used.
Hibbs’s Model: 53.9% for Clinton
Fair’s Model: 45.7% for Clinton
Cuzan, Heggen, Bundrick Model: 48.2% for Clinton
Holbrook’s Model: 52.5% for Clinton
Lewis-Beck and Tien’s Model: 48.9% for Clinton
Campbell “convention bump” model: 51.2% for Clinton
Campbell’s “Trial-heat” model: 50.7% for Clinton
Abramowitz’s “Time for Change” model: 48.6% for Clinton
Norpoth’s Primary Model: 47.5% for Clinton
Of these nine different models, five of them predicted Trump would win. Still, even though they predicted he would win, these models actually predict Democratic vote totals among the two major parties, not Electoral College results. Therefore, with Clinton currently having 50.7% of the vote, the Campbell “Trial-heat” model is spot on. Still, the fact that five of the nine models predicted a Trump victory shows you that there was not agreement when it comes to Clinton winning the election outright.