Labels

Tuesday, August 28, 2012

Digital Politics: FiveThirtyEight

It's time to round out our coverage of election prediction sites with FiveThirtyEight.com. If you are looking at just one site to understand the current state of the race and electoral math, this, probably, is the one. Why is that?

It isn't because Nate Silver, the guy who runs it, is, necessarily, "the most accurate." Everyone claims extreme accuracy and, by the end of the race, everyone pretty much has been. The 3 weeks before election seem to really "tighten up" the polling. But, even though this isn't "the primary reason" to go to FiveThirtyEight, it's a very good reason: He's very accurate.His track record is good.

It also isn't because Silver's model is "so much better than anyone else's." I suspect it is better than most everyone else's--but I'm biased: I've been a 538 fan for a long time and, to be frank, I lack the skill to determine if the results forecast is better at PEC or 538 or whatever. You can read their track records and decide for yourself.

No, the reason I'm recommending 538 if you "just read one" is because of the quality of the blog--not because I'm willing to bet the farm on Silver's 70-30 Obama-Romney split for Nov 6th. In short: the quality of Silver's writing about electoral politics is at the top of the blogo-pyramid. There are other people who are, in my opinion, kind of in his space--but almost no others who are, frankly, in his class. Jay Cost is, to my mind, insightful--but thoroughly partisan. Sean Trende makes a lot of sense but doesn't publish as often (or, if he is, I'm missing the feed).

Silver's articles (and his co-blogger's) are deep-dives into numerical realities. He doesn't say things that the numbers don't say and the whole concept of "big data" is that "the math" is too hard for humans to do in their heads. There are counter-intuitive realities that are at play all around us and using moderately advanced statistical methods as the lens through which to view them brings reality into stark focus.

Does this mean he's always right? No. No one is always right--and it doesn't mean he's not partisan (Mr. Silver leans Democrat)--but many, many voices on both the left and the right respect his analysis and that ... that is almost if not entirely unheard of. The right's general willingness to give 538 the time of day is, in this hyper-partisan environment, not the gold standard. It's platinum ... or iridium ... or something.

I'll say that again: Nate Silver's credibility (and to be sure, it's not 100%--Sean Trende asserts Silver is taking the data and "making the best case for Obama that's possible"--but that's a LOT better than, say, what CNN generally gets credit for) on the right is more convincing to me than anything he does with numbers. That is because while I lack the depth of statistical insight to know if what I'm reading is the best-practice or not, I do trust the blogosphere in general to poke holes in everything they can.

Silver's analysis currently has very few holes.

Let's Look
When you go to 538, down the right is the "state of the race." Here's a snap-shot:
Those Trend Lines Spell Out BORING RACE in Sanskrit
This shows where his model--which includes economic forecasts, state fundamentals, and polling--thinks the race is right now projected out to Nov 6th.You can click the Now-Cast to see where he thinks it is today. So far as I know, he is the only guy really doing this (PEC does something like it--but not in the same way).

Then we get his next view:
Mt. Blue is over there at around 330 ... But Look: Florida is RED!
This shows where each state stands, what his probabilities of electoral vote look like, and for the bottom map, which states are likely to "make the difference." It's Ohio. Yeah? Didn't we know that?

And then you get to the articles. His articles are, as I said, a wealth of insight into electoral science. He discusses poll biasing, the political geography of individual states, and many, many other things that are almost always fascinating and always well written.

This white-paper from PredictWise compares the 538 analysis with the InTrade polling predictions. InTrade was, in fact, very very accurate for the 2008 election. Their conclusion was that taking the raw numbers for both polling (which 538 uses) and betting markets (InTrade) and "debiasing"--that is "removing bias" gives the edge to InTrade--however, the de-biased polling (538) is better than the raw InTrade. Considering that 2008 was the best we've ever seen for betting markets one would conclude that, if nothing else, 538's analysis is not adding "more noise" (a real concern for prediction modeling).
It'd Be Prettier And More Convincing In Color
 In any event, no matter how convincing you find FiveThirtyEight's election-porn math and charts, like Playboy, you should read it for the articles.

2 comments:

  1. Just found your site. It looks good! Please keep it up!

    ReplyDelete
    Replies
    1. Thanks wsisaac--Like us on Facebook--and check us each week day! Tell your friends :)

      Delete