Thursday, March 7, 2013

Nate Silver, His Expectation

Nate Silver is an American statistician that is well known for his predictive analysis of the major league baseball and recently political elections. The accuracy of his November 2008 presidential election predictions,  he correctly predicted the winner of 49 of the 50 states , won Silver further attention and commendation. The only state he missed was Indiana, which went for Barack Obama by one percentage point. He correctly predicted the winner of all 35 U.S. Senate races that year.

In an interview conducted with Nate, he gives insight into his view on the future of big data. Silver said "the "big data" era will only be successful if the government is willing to evolve with it."

"In some cases, the government has the best data in the world, but not always the ability to use it," Silver said, "The governments that are willing to evolve with it will benefit, certainly, but there is no end-point to big data."


3 comments:

  1. This comment has been removed by the author.

    ReplyDelete
  2. After reading your post, I got interested in hearing some critique on Nate Silver and what others thought about his process. So, I found an article (link at the bottom) that suggests 4 lessons we can learn from Nate Silver and his approach to predictive analytics.

    1. Prediction Triumphs Prejudice.
    Erik Sherman, the author of this article, suggests that Nate's approach to his analysis of the polling data focuses on hard facts rather than subjective observations. He draws an analogy to Moneyball and baseball. I think there's a great deal of proof in this claim. I think that the underlying question is - "Are humans predictable?". While we naturally would want to say "No", the data suggests otherwise. And we must remember that there is inherent probability in pretty much everything. So, while humans may not be 'predictable' in a Vegas, card-counting fashion, they do 'tend' to act in certain ways.

    2. Probabilities are relative.
    Sherman addresses the critique that polls aren't predictive because they only represent one point in time. That's correct, however if you collect enough polls over time the data will suggest trends that have associated likelihoods. He goes on to say "exact rightness is overrated". I think this is such a profound statement because of how much we clench to things that we KNOW are true. It reminds me of a quote from the Will Smith movie, Men In Black:

    " A person is smart. People are dumb, panicky dangerous animals and you know it. Everything they've ever "known" has been proven to be wrong. A thousand years ago everybody knew as a fact, that the earth was the center of the universe. Five hundred years ago, everybody knew that the Earth was flat, and fifteen minutes ago, you knew that humans were alone on it. Imagine what you'll know tomorrow."

    There's so much truth (outside of the aliens..) to this statement. We really have to understand that every 'fact' will have some value of certainty or uncertainty. So this begs a really challenging question - "what can you be absolutely sure of?"

    3. Patterns can add up in substantial ways.
    The race for president was actually 2 races. The first, the popular vote. The second, the electoral college which was dependent on states voting independently of one another. Many were astounded by Nate's prediction because it held Romney at such a low expected outcome. However, what people failed to see was that they were distracted. The popular vote had absolutely no role in the outcome. Thus, the significance of analyzing the states independently was what yielded the best prediction. We have to remember that not all components of a system effect the outcome. The important thing is to focus on what the "drivers" are.

    4. You're never too small to make data work.
    This is definitely Sherman's weakest point and seems like an after-thought. Not much to say on this other than anyone is capable of making big data work, which seems like an obvious statement.

    Lastly, it's worth mentioning that at the bottom of the article someone commented and posed the criticism that Silver's analytics were based on an assumption that the polling done at poll stations "do not use population-based probability samples, and that makes it IMPOSSIBLE to draw truly valid inferences from them." This is a valid claim. But, as you'll see Sherman points out... Silver predicted 49 out of 50 of the states... so, there's that.

    ReplyDelete
  3. Reference: http://www.huffingtonpost.com/2012/11/06/4-lessons-from-nate-silve_n_2082777.html

    ReplyDelete