Swing State Polls

I do love Nate Silver. I love that he took his love for statistics (It was one of the first patches I earned in the Nerd Scouts, BTW) and turned into a work of pure righteousness in the arena of Political Pollsters. Before Nate and his "538" crew, polling was considered sacred and was in reality a den of iniquity! He exposed their lies and biases against facts and then created a sophisticated and surprising accurate way to rate those who dare to report on the political blood pressure of our democracy.

Through his work, we all learned about the practice of push polling--which still goes on, I was treated to it back in March just prior to our most recent primary here in Maryland. He created algorithms to differentiate between the accuracy of pollee categories like "Voter," "Registered Voter," and "Likely Voter". On his own website he uses a range of factors to review polls as they come in and then offer an adjusted result based on the likely accuracy of the poll. This is accomplished through a wide range of factors from the history of the company doing to polling, the sample size, the pool of those polled, the methodology used to conduct the poll, and even an analysis of the quality of the actual questions asked rating them for bias and clarity. It's all fascinating stuff to me. But what's best about it, he has the results in the end to back-up his process. That's called integrity, and I like that most of all.

With this in mind, I spent some time today using his data to create a set of graphics illustrating the polling from 6 swing states. On top are all the polls since June 1 for the states.  I used Silver's system for ranking these polls and then eliminated the one's falling bellow 40% accuracy--which I thought was a rather generous way of weeding out some of the "bad" players.  Oddly enough that criteria nixed nearly all of the polls that showed any advantage for Herr Donald Trump....go figure? 

One thing that might not be as clear in the presentation is the lighter ban running through middle of each graph--that represents the margin or error inherent in polling.