Why is a computer model with 100% accuracy for predicting presidential elections since 1980, with margin error of at most 30 electoral votes, being ignored?
And why is a baseball mid-level statistician Daily Kos poster, granted a New York Times columnist position and given a 250k book deal, who bases his work on ‘analyzing’ other people’s polls while not informing people he was using internal campaign polling, elevated to the status of an expert in political science?
Critics of the baseball mid-level statistician, Nate Silver, suspect he is just a hack from Chicago that Axelrod is using to steer media coverage over poll coverage in an attempt to control the narrative. As an example of this, the critics point to Nate Silver writing a column called ‘Gallup Versus the World’ where he dismisses the most historical polling outfit in the US because its polling was trending Romney.
The response goes something like this: “It is because Nate Silver’s 538 Model uses a statistical approach. Other approaches use ‘savvy’.” This is an utter distortion. There are many models out there which all use a statistical approach. Political science courses in colleges do require classes in statistics.
The alternative to the 538 Model is not ‘savvy’ or ‘punditry’ but other models. Omitting the fact that Silver used internal campaign data for his ‘analysis’ in 2008 (and probably 2010 as well), the 538 model relies heavily on polling data especially state polling. Polling can be flawed. (Famous polling errors of the incorrect 1980 polls or the infamous Ohio 2004 polls are just some examples.)
A political scientist has many tools at their disposal to get at the nature of a race. Aside from polls, they have historical data, economic data, demographic data, and many other forms of data than just ‘polls’.
The issue isn’t why the 538 Model is being shoved in everyone’s faces, but why the Bickers and Berry model, which has far more history at predictions and 100% accuracy, isn’t. Wouldn’t the good scientist want to observe ALL the models out there? What is occurring is a type of anti-science. Every statistical model not based on polling data is dismissed. People who go to the 538 website are being told there are no other statistical models out there. The only alternative to the 538 model is ‘savvy’ and ‘punditry’ which is just not true.
What is the Bickers and Berry model? While the model does take in some elements like the home states of the candidates and past voting behavior, the model relies heavily on economic data. Economic data, just like demographic data, historical data, and polling data, are all useful for the political scientist. Since the Bickers and Berry model hasn’t yet been wrong, this certainly highlights how important economic data can be in determining elections especially since 1980 polling data was wrong.
But Nate Silver dismisses economic indicators!
Shockingly, he says favoribility rating (such as whether you want to have a beer with Obama) matters more than economic factors such as unemployment. History of presidential elections show that the economy certainly does matter.
I looked for an example across the web of how people respond when introduced to both the 538 model and Bickers and Berry model. On this doping message forum (of all places), the thread has people in it dismissing there is any other viable statistical model than one based on polling. The only alternative to the 538 model are other models based on polling. One person replied to the Bickers and Berry model that:
There are others who make wildly different predictions based on different assumptions about different numbers, most notably a pair of University of Colorado professors whose software has Romney winning with 330 electoral votes.
I think the Bickers and Berry prediction is weak. They based their prediction on economic indicators not poll results. They tracked indicators like unemployment and income and then correlated those with the outcomes of past elections.
Now I’m not saying that’s valueless but I think that approach has its limits. People do make voting choices for economic reasons but there are also dozens of other reasons. And more critically, I think it’s wrong to make a prediction based on what you think people should believe and ignore contradicting data on what people say they believe.
The reply sounds like it came from a NLP seminar. He says it is weak. Why is it weak? It is because “it uses economic indicators instead of polling results”. Apparently, polling results are the only statistical models that are allowed. There is much history of polling data being wrong and for the history of economic data being correct… well… the Bickers and Berry model has never been wrong. And then comes the money line:
And more critically, I think it’s wrong to make a prediction based on what you think people should believe and ignore contradicting data on what people say they believe.
This is projection. A thirty two year old computer model doesn’t make predictions based on what people believe. And Bickers and Berry are not ignoring contradicting data unlike Nate Silver and others who say polling data is supreme and economic indicators do not matter. As I understand it, the University of Colorado is a very leftist place, and I’m getting the impression that Bickers and Berry wants their model to fail this election.
The Bickers and Berry model exists in peer reviewed science journals. The 538 model exists as a blog site. Which sounds more like science?
I’m using that thread’s example to showcase what is going on everywhere. The Bickers and Berry model is being dismissed because economic models are being dismissed. This is odd because economic models were not dismissed four years ago.
Bickers and Berry model is obtaining more notoriety this cycle because it is not just an outlier to all the polling data, it is an outlier with a perfect accuracy rating. The Bickers and Berry model got lost in the noise of 2008 where, everyone agreed, economic indicators were extremely important. The famous model has been touted in previous election cycles. However, this is the first time I’ve ever heard anyone say that economic indicators are irrelevant. (If this is true, then why won’t the government release the monthly jobs report?)
Political Science became its own science by splitting off of Economics. Much of the foundation of political science is economic indicators. Or as James Carville famously coined: “It’s the economy, stupid.”
In the upcoming posts, we’re going to go through the Bickers and Berry model, see what its margin of error fully is, its past history, what type of economic data it uses, and the limits of this model.