Many of you may read Nate Silver's poll analysis over at the New York Times, where he made a name for himself as fivethirtyeight politics.
Silver's a veteran of the Baseball Prospectus crew and a former baseball sabremetrician.
You may know him as an Obama partisan and hack who correctly predicted 49 of the 50 states (he missed Indiana) that went for Obama in 2008 and all the Senate races that year. Of course, we later learned he was privy to internal polling of the Obama campaign, so, well, whatever.
His predictions in the 2010 races were ok, too, but, my problem with Silver isn't really in his expertise as a prognosticator. I find that to be of really little use to me. Any idiot can read polls, apply some weighting to them and extrapolate the results to election day. This is problematic because polls are not elections. Polls do not reflect what would happen exactly if the vote was today. They are an estimation of that. Unlike baseball, where we have real data every day to work with (the actual results), a poll is an estimate. It's as though we asked everyone going to a Yankees series and asked, "What will Jeter do this series?" and then we based his season stats on the fact that 1000 people told us he'd go 4 for 11 with 2 walks, 2 RBI's and a double and a homer. More on this later.
My problem with Silver is that he grew up as a sabremetrician, yet he has evolved and become famous for really just being an accountant. There are literally tens of thousands of smart people working in Finance departments all over Corporate America who can whip up Excel spreadsheets and manipulate numbers to predict the future. They do it every day to predict corporate profit, loss, revenue projections, etc.
What Silver does in election prognostication is nothing special, and it really is completely dependent on the quality of the data input. If Silver does hold some special idiot savant skill, it is in accurately recognizing the correct weighting of the input data. However, like works of Shakespeare, a room full of monkeys could achieve the same results given enough time, or,we could outsource it to a computer. Polls are subject to two major biases. First, they are taken at a given time. As they age, they become stale, and less reflective of the mood of the public on the actual election day. Second, they are subject to selection bias. This is the pool of people that pollsters use to take the actual poll. You can see these biases in every decent poll released. If you don't see them, someone has suggested you are looking at an op-ed. Selection bias is something we rarely see reported in the top-line, horse race number, but it is a key way pollsters can skew their polls. Sometimes you have to dig in to the polls to find this. Pollsters will tell you they try to use pools of people that reflect the expected election day electorate. They would, ostensibly, be hurting their reputation to do otherwise. But, in this day and age of budget cuts at universities and media outlets who pay these pollsters, maybe they can't refine their pools as well as they'd like, and they go with a group that is D+9 (9 percent more Dems than Reps), in a state that has never voted that way in its history. I can't give you the reason this cycle the samples seem to overly skew to Dems. You'll have to figure that one out for yourself, but look for it in any poll. There are websites and analysts who try to remove this bias, but, Silver does not.
So, personally, I think predicting the 2008 election, once you admitted to yourself it was a wave, would have been pretty damn easy. I think this came easier for Silver, since he was already predisposed to favor Obama politically, and he was getting data from their campaign. In other words, I am not that enamored of the guy's skills.
What would I like Silver to do?
I'd like Silver to apply his sabremetric skills to his analysis, as I think that would be useful, and interesting.
As I said, anybody with some excel skills can do what he's doing. It ain't rocket science, as we say.
What makes sabremetrics so compelling in baseball is that it takes all those old traditional ways we have of measuring people's performance, and says that they are not enough, and that, in fact, we largely measure by accounting, rather than using measures that help us predict future performance.
In old baseball world, we used statistics to validate what we saw on the field, but we didn't use statistics to tell us what we would see on the field. Sabremetricians stood this on its head, and wanted to answer 2 questions - Why do we see what we do, and, if we can answer why, we can also predict what will happen (future performance).
When I look at the internals of polls, I see all sorts of data that is largely unbiased data. Sure, it still suffers from the selection bias, but, because that data is in depth, there is much more that it tells us about future performance than just the horse race number. I'd like to see people of Silver's ilk look at that data, and there's a lot of it historically, and let's see if we can use it to predict what will happen, rather than the way we're doing this.
Why do I say this, because when we look at things like independent voter splits, and the horse race numbers for D's or R's, we are starting to unskew the top line results. These have meaning precisely because they are unskewed, and unbiased. Right Track/Wrong Track and other questions that get routinely asked suffer less from this bias.
When I look at these numbers in pretty much every poll I see this cycle, I say to myself, how does Obama win? The country hates him. It's not just my friends, it's evident in the internals of these polls. And clearly his campaign saw this, too, as they have run a campaign based on these internals. Do everything you can to NOT speak about the record, and demonize the opponent. The fact that the needle has moved so little says something to me.
But, I could be wrong. I will admit that. But, I am heartened and validated that on Face the Nation today, David Gergen and the entire panel there basically agreed with me.
Now, they tend to think it will be a close Obama win, but they all agree the Republican enthusiasm gap is real and palpable. Either the Obama ground game really is that great, and they'll pull our Ohio, and win, or it's not and Romney wins.
I happen to think if Romney does win, and the enthusiasm gap is something that is being missed by the polling and the MSM (they're ignoring it), that Romney will win every battleground state but Nevada (taking FL, VA, NC, OH, NH, IA), and he'll add new BG's MI, WI, PA and may take Oregon as well.
If Obama wins give OH and IA to him and the new BG's. It'll be a small victory, just like his campaign, and Pyrrhic.
1 comment:
amazingwork.keepit up.
Post a Comment