Statistics, probability, and Nate Silver

Honestly? It's cool to see a statistician getting media love. I imagine for most people he's like a real life character from Big Bang Theory. I've seen him on tv several times and he comes off as incredibly nerdy, but he's skilled at talking about statistics and making it relatable to the average person.<p>In a way, he's a spokesman for statistical reasoning to the world. It's odd to see how much pushback he gets for saying "A candidate with a slight lead in October wins 75% of the time". This is a fact, easy to look up. But because the other side and the media don't like it, Nate is put into the position of defending common sense to America. And he does an admirable job.<p>Hopefully he can find a way to use this for good outside of politics. Maybe he can educate people about statistics and probability in other ways. Starting a blog that correctly predicts the behavior of reality tv stars might be the best thing he could do for the cause of reason in this country.

One thing that really bothers me: most of Nate Silver's supporters and detractors don't understand jack shit about his models.<p>On the Democrat side of the table, many treated 538 like a source of religious prophecy and source of comfort - when truth be told Nate is just a smart statistician who developed a model that's accurately predicted the outcomes of three elections (the first being the Hillary / Obama primary.) On the Republican side of the table, major trends in national and state-level polling data were ignored because pundits believed that the pollsters were oversampling Democrats when the truth is that the Democratic base has expanded since 2008.<p>Other legitimate statisticians, like the University of Colorado, predicted a Romney win and obviously their models were wrong.<p>Scientists, statisticians, economists, and engineers shouldn't be worshiped as sources of incontrovertible truth - they're just humans trying to do the best objective, accurate job that they can!

&#62;Predicting an election based on polls is an entirely different matter. The election will turn out one way or another. If the same people voted for President 100 times without an external factor interfering differently across samples, the outcome would be the same every time.<p>I'm glad someone pointed this out, since I think it gets obscured a bit in most discussion of 538's statistics.

I understand what the post is trying to convey, yet it is easy for someone to cast the election &#38; football in an equivalent framework: increasing the number of polls one of the statistician uses in order to narrow the confidence interval is equivalent to increasing the number of measurements you make on the football players &#38; their environment before the game in order to narrow the confidence interval on the match outcome..<p>in other words you can interpret the confidence in the case of the election just as in the case of the football game, by stating that:<p>1. if your mathematical model of the population is accurate (eg. for instance your population really is voting using a binomial law of a certain probability p_democrat of voting for the democrat candidate) 2. and if the measurements (polls or player states) you have made give you that at least k voters out of N polled would have voted democrat<p>then you can consider all the (hypothetical) instances of a population which would have resulted in a same or higher number of democrat voters for this poll, and you can compute the confidence interval that gives you the probability that such a population would indeed elect a democrat president<p>So you don't necessarily have to view it in terms of "not changing the population vote, but changing the precision of the data", you can also cast it in just the framework of the football prediction

&#62; There is internal chaos in the game that forces the probabilistic distribution. Predicting an election based on polls is an entirely different matter. The election will turn out one way or another. If the same people voted for President 100 times without an external factor interfering differently across samples, the outcome would be the same every time.<p>This is a distinction without a difference. There is "internal chaos" in the game because we don't know its initial conditions exactly, just like in the elections. If we had precise knowledge about the particles in the ball, the field, and the players, we could theoretically predict the outcome with high certainty. Except for quantum mechanical effects, statistics isn't about true randomness, it's about modeling lack of knowledge.

One additional point that doesn't get much play in the debate about Nate Silver's numbers: political science, like economics, is a soft-science. You <i>cannot</i> 100% accurately predict the election based on polls, just as you cannot accurately predict market movements based on economic indicators. This is because the system is made up of self-aware human actors that respond to meta-information like Nate Silver's numbers. Even if the polls were 100% accurate in measuring voter sentiments, as soon as they are published they are picked up by political operatives and become part of the spin machine which changes those sentiments.<p>But if you want to use polling data, Nate Silver is demonstrating the least-bad way to do it.