dotCommonweal

A blog by the magazine's editors and contributors

.

538 and Nate Silver

Over the past few weeks, I have become addicted to Nate Silver's New York Times Blog, "538," which analyzes the polling on the presidential election. A University of Chicago math prodigy, Silver made his name analyzing baseball statistics--and then transferred his skills from batting to the ballot box.What I find tremendously impressive about him is his natural ability as a teacher. He is a great writer, and explains things clearly, concisely, and non-polemically. The mysterious world of polling has become a lot clearer thanks to him.I suspect that a whole course in statistics could be structured around his blog. Has anybody done anything like that?

Comments

Commenting Guidelines

Thanks, Kathleen. "538" is a great site on which any political junkie worth his or her salt can while away yet more hours until the election. The detailed breakdown of the tipping point states was especially interesting.

Thanks, Kathy. I am recent fan of Nate Silver. Discovering him and "538" has given me new hope for Election 2012. As of September 2, Obama had a 73.1 percent chance of reelection per Nate Silver. For the 2008 election Silver predicted the popular vote within one percentage point, predicted 49 of 50 states results correctly, and predicted all of the resolved Senate races correctly. His only miss was Indiana. The graphs are a bit hard for me to follow, but I'm not really a graph person.

Indiana--he needs to visit us to make up for his lapse! I really would like to get him to ND to speak.

Great site. Almost philosophical! At least about probabilities.

I've followed Nate Silver since the 2008 campaign, when he set the bar for analysis of polls. I do have a slight problem, though, with his 'probability' tracks. While I don't claim to have Nate's high degree of statistical sophistication, I do have some knowledge, and a lot of experience in reporting.I'd contend that probabilities are best used to describe events of a recurring nature. Take a fish bowl, and fill it with black and white marbles. Let's say there are sixty black marbles and forty white ones. The probability of drawing a black marble is .6 or 60%. But you wouldn't be drop-dead surprised if you drew a white one, even though the probability is 30% less than drawing a black one, or .4. It's only over time that you would expect your draws to 'even out' to the actual statistical probability of six black marbles to every four white ones. (We are assuming replacement of the drawn marble.)Looking at it another way, if you were playing a game of chance, say a card game, and the odds of you winning were .6, your best strategy would be to play many hands, and walk away rich, NOT to bet all your money on the first hand knowing your odds were 6-4 in your favor.But a Presidential election is a single event, not a recurring one (the conditions and candidates are different each time.)So, for me, Nate's 'probability,' even if correct (and it is all based on imperfect data) overstates the certainty of a specific outcome in a single draw or hand.That said, I love Nate's analysis and reporting.

JBrun: Aren't probabilities all about "probable." Maybe recurring events offer a better measure of the chances of one candidate being elected over another. BUT..While candidate's and issues may shift from election to election, I am taken with the recurrence of similarities or commonalities among campaigns, e.g., years when a candidate is the incumbent (and years when there is no incumbent); years when there is a presidential race and years when there are only congressional races; years when the economy is strong and years when it's not. Silver's parsing of these may not always predict the winner, but it adds to the thickness of the discussion.I also think his take on different polling organizations and his weighing of their results is pretty fascinating.Did I say I've never taken a statistics course in my life, and I really don't know what coefficients are, and likewise probably wouldn't know a probability if it walked up and asked if I'd like to run for president? But that aside....

jbruns =As I understand it, the probability of a particular outcome is never certain == the only certainty is the ration between the likely and the unlikely. And though presidential elections are different in many ways in many ways they are the same. The trick, apparently, is to figure out the relative influence of both the like factors and the different ones.Silver has a book coming out at the end of this month, "The Signal and the Noise: Why so Many Predictions Fail - But Some Don't". It looks philosophical enough that non-mathematical me is getting it. http://www.amazon.com/s/ref=nb_sb_noss_1?url=search-alias%3Daps&field-ke...

Oops -- the ratio, that is.

Maggie and Ann: Yes and Yes. I'm a big fan of Nate. I'm just not a fan of his use of probabilities in this particular context.BTW, statistician or not, if you are not familiar with The Monty Hall problem, it is a great diversion and a good example of how probability defies intuitive sense.http://www.theproblemsite.com/games/monty_hall_game.asporhttp://math.ucs... for a full explanationhttp://en.wikipedia.org/wiki/Monty_Hall_problem

Thanks, jbruns. That certainly defies my way of thinking.If each time I open a door I have a fifty/fifty chance of being right (and wrong), why would it make any difference whether I chose a different door or not?

Ann,I hope you read the explanation. But, very simply put, the first time you have a 1/3 chance of being correct, which stays the same if you don't change your choice. If you do change your choice it is 1/2. And if you do the simulation, you'll see.

(P.S. this only works if the 'host' knows where the prize is).

jbruns --It seems to me that what this "paradox" does is to conflate two different questions:1) what are the odds of either A or B if they are the only choices2) what are the odds of either A or B if there is also a COnce you remove the C you are talking about a different problem.

Ann: Right!

j --Probability theory is big in philosophy right now. Has been for a good while. It seems that a lot of people have trouble with it because the way we talk about what is "probable" suggests that it somehow measures/establishes causality, which would introduce a certain necessity into the problem. But some theorists think that expressions of probabilities really just express measures of our expectations based on numbers of past events. Hmm. I can't buy that either. Too simple.Whenever you get into questions of possibilities (which probability does do) you run into some doozies of philosophical problems. EVen Arisotle had major trouble with possible future events.