Never mind Nate Silver, "Yahoo Signal" should get the biggest props

Politics & Elections
41

Whatever you say, boss. Oh, but wait: you’re not my boss and I say it *is *arguable. Guess we’re a’arguinnnit!

This, even as I showed you were completely wrong in your assumption that I was operating with an established animus against Silver? Don’t think no one noticed how you elided that point.

We were comparing 60% and 99%, I thought; but even still, are you saying it doesn’t matter if Romney has a 9% or a 1% chance of winning? That the difference is negligible? The former is 800% more likely to occur! I would probably literally bet my house on an even money bet I only had a one percent chance of losing, whereas a 9% chance would make me nervous at those stakes.

Hardly. I’m no Ph.D. either, but I aced the ACT and SAT in math and have taken (with As and Bs) several math major courses in college, including differential and integral calculus, and most importantly for this subject, discrete mathematics. I have no trouble with “an average” though more precisely we are talking here about an arithmetic mean (a median is really a better representation of the quotidien meaning of “average”).

It’s not just a “layout”, it’s what he chose to label as his “forecast”! You sound like the Republicans parsing Obama’s “act of terror” comment, LOL.

42

Chronos, you make a good point. Come to think of it, it’s a little like the interminable argument about theoretical vs. experimental physicists. And usually, I’m sympathetic to the latter there (not least becaus they are the downtrodden underdogs). But Silver is a little bit less important in his role, it seems to me, because absent his observations of reality (what people intend and plan to do vis-a-vis voting), the election will happen anyway and confirm or not confirm the theoretician’s prediction.

43

Denial of the obvious is not an argument.

Ok, so revise it this way: whether you had something against him before, you are not discussing his work fairly now, and I detailed the ways you were doing that. (And you haven’t responded to those other than going on about the word “forecast.”) That’s why I concluded you disliked him.

I don’t know why you thought that. At no time did I compare those two figures to each other. I said several times that their Election Day forecasts (99% for Wang and 91% for Silver) are pretty similar in the big picture. By both Silver and Wang’s reasoning, Obama’s low point was right after the Denver debate. Silver rated Obama’s chances of winning at about 60% after that debate. I don’t know what percentage Wang might have assigned at that point, but if you look at his graphs, he is in the same ballpark.

I never said it doesn’t matter. I said you’ve been characterizing Silver’s forecast as wishy-washy and Wang’s as definitive and confident, and I don’t think that’s reasonable when they always said the same general things and gave Obama a better than 90 percent chance of winning.

Then why are you having such a difficult time with the 313.0 EV number?

44

I’m not having any trouble with it, you are: alternately trying to argue it wasn’t his prediction (when it was his “forecast”, LOL); that it is an attempt to split the difference on Florida (never mind that you argued more or less simultaneously that he was assigning Florida to Obama’s column, yet 313 is closer to 303 than to 332); that it is just some aesthetic element of his “layout” (which might be the funniest of all).

If Obama had not won Florida, I have to strongly suspect you would be much more interested in the 313 figure, would not be claiming 332 was his prediction, and would argue “he said Florida could go either way”. This is why it’s so suspect for Silver to have these multiple figures lying around to cover his bases depending on what happens.

45

I simply don’t get this argument. The different numbers were based on different potential scenarios, with different probabilities assigned to each. It’s not “covering bases” - it’s what the data spat out. Do you honestly expect him to just stick a single number out there as his prediction of what would happen and ignore less likely but still possible outcomes? That’s not how his approach works and I’d be a lot less interested in his work if it did.

46

It’s what Wang did, and I do have more respect for that. We see ITT how some are trying to argue with a straight face that Silver did predict 332 (even though the number appears nowhere in his forecast!), thanks to all the hedging and multiple scenarios. If you make multiple predictions, you do heavily increase your chance of your partisans playing exactly that cherry-picking game after the election.

47

You are touting your math acumen while mistaking what the 313 figure is. And you’re bickering about the meaning of words like predict and forecast and playing cite the dictionary - ‘this word means this word, so when Silver says this it means that!’ even though it obviously does not - and you’re ignoring a very easily readable chart that shows what Silver expected. This is not that complicated. If you understand the numbers you are looking at, your argument here is difficult to explain. You’re either playing word games to make Silver look worse, or you’re failing to understand a very simple chart.

Just about, although that’s a bit of an oversimplification.

He said Obama had a 50.3% chance of winning Florida. Yes, Wang and Silver and Yahoo all had Florida could basically go either way. That was true. But Silver was the only one who said it was more likely Obama was going to win it than Romney, and that was correct.

I gave a very loose explanation of why that is: Silver had Obama getting 332 EVs, but the state most likely to be wrong was Florida (for Romney), then Virginia (for Romney), and so on - which shaded the average lower overall. I don’t know why I have to explain this over and over.

You’re wrong about that, and you’re now resorting to an extremely lazy argument: “if something else had happened I bet you’d be singing a different tune!” That’s irrelevant, and no, in this case I would not. I would still say that Florida was a virtual tossup, which everybody agreed it was, but I would say that Silver got Florida wrong and that Wang and Yahoo Signal got it right, which would make their EV forecasts more accurate than his. I wouldn’t be pretending that an average of simulation results was an prediction of the exact outcome of the election because that would be wrong and totally ridiculous regardless of the final EV total. And many, many posters (sometimes including me) explained that Silver’s forecast was not a projection of the exact EV count. It was an average of the simulations he ran on his model. It’d be ridiculous to say Silver expected the EV total to be split by a decimal place. That’s impossible.

So you’re very knowledgeable about math, but you think Silver “just has figures lying around?” And again, you’re criticizing Silver for stuff Wang also did. I showed you where he did this.

48

It would be intellectually dishonest for Silver to not “have those figures lying around”. He gave us the best information he had available to give, and he explicitly told us just how good (or, alternately, how bad) that information was. Yeah, he would have appeared more confident if he’d omitted the error bars, but that confidence would be a lie.

49

No, it really isn’t. He put it out there as the most likely outcome from a range, which was never in question HAD YOU ACTUALLY READ WHAT HE WROTE.

Here’s a plot from his blog the day before the election showing that the two most likely results are 303 and 332 EVs with a razor thin edge to the 332.

From the last paragraph of his election eve blog post:

[QUOTE=Sam Wang]
Election Eve prediction is not the most impressive of feats. What we did in August (here, and here) was the interesting part. Those were true predictions, and were centered around Obama 315 EV, Romney 223 EV.
[/quote]

So, he did the same things Nate Silver did (even to producing an average value around 315) but merely presented it differently and for that you have more respect?

50

That’s what I’ve been saying. And I also agree with what you said in your last post; I’ve been trying to express that point but maybe I haven’t succeeded.

51

LOL, by “very easily readable chart” do you mean the bar graph that only marks the x axis in increments of 60 EVs, and tops out at 20%? You’re telling me that a putative first time visitor to 538, the classic “reasonable person”, should be expected to ignore the EV number up top under “Forecast”, scroll down and find that small graph, figure the bar that’s a little taller than the others and touches the 20% probability line must be Nate’s prediction (of course!), and eyeball it as being exactly 1/30th of the way between 230 and 260? If I didn’t “know” you as well as I do, I’d suspect you were putting me on to see how far you could push your gag.

I mean, even back when I was a huge Silver backer–and before the election I was visiting it several times a day–I never took 332 to be his single, top-line “number”. Do you have a link to any of his posts in which he used it? I wouldn’t be shocked if he mentioned it among his forest of text somewhere, but I think the context would be interesting. (Hey, maybe you could even find verbiage that would support your case, although that would still mean he did a poor job with his sidebar.)

Could you at least weigh in on why you think he so prominently featured the number 313? I’m **not **talking about what it *is *mathematically. I’m asking why you think it is sitting up there in pole position. Someone else characterised that position as “unfortunate”; do you agree? If so, why do you suppose the ever-precise Nate Silver would do such an “unfortunate” thing? Honest question.

Antibob, perhaps it got lost somewhere along the way, but I’m only a little more impressed with what Wang did than what Silver did. I’m much more impressed with what Yahoo Signal did than either of them. My intent was never to take, or defend, the “Wang kicks Silver’s ass” position.

As Marley seems quite a fan of Wang, though (he was the one who introduced him to this thread), I would be curious to see him respond to an interesting line you quoted from Wang:

“Election Eve prediction is not the most impressive of feats.”

O RLY?

Marley, you’re pretty slippery when (all) wet, so I’m sure you’ll come up with something entertainingly wriggly in response to that. ;0)

(Or maybe you’ll just pretend not to see it.)

52

The real data, which anyone interested in real data will find, is in the histogram. What he shows in the big bold type for anyone who just wants to glance at it is a single-number summary of the histogram. Now, there are a variety of different numbers one can pick as a single-number summary of a histogram, but the arithmetic mean is probably the most commonly-used. I suppose you’re arguing that he should have used the mode instead? Maybe, but mode is a rather unstable measure of typical value of a distribution.

53

Chronos, I’m not saying he should have used anything different (although Wang used the median and mode and left the mean out altogether). All I’m saying is people should stop trying to claim he predicted 332 EVs for Obama.

54

He predicted that that was the scenario with the single-highest likelihood and as it happens that is how it turned out. The fact that a scenario where Obama lost Florida was almost as likely doesn’t make his prediction less impressive, nor if Obama had barely lost Florida would it have significantly tarnished Silver’s accuracy. It was very close either way and that is what makes his analysis valuable.

55

You know which chart I mean. As I said, you’re LOLing at the format of the blog, not the analysis Silver did. Why?

I’m talking about you. You have looked at the chart we are talking about. You already posted about it.

I see. Certainly plenty of other people made the mistake of taking 313 for an exact prediction; as noted, we discussed the difference between that average number and the actual EV number many times on this forum. But since it’s been explained I don’t know why we still have to talk about what it is.

One sample on August 7: “But North Carolina just isn’t that important to the electoral math. Mr. Obama currently holds leads in the forecast in states totaling 332 electoral votes.”
I’d have to subscribe to get more of the blog posts, but I see he also discussed it in late October.

I don’t care at all. I want to discuss their analysis and what they said, not what their sites look like. Looking at Wang’s site again, I see that right at the top, he has a Snapshot number with 312 electoral votes for Obama. Surely a careless visitor could mistake that for his prediction. Here, a day before the election, Wang inveighs against “just watching the EV estimator” and talks about Florida and North Carolina moving back and forth and affecting his estimate.

Wrong yet again!

I understand what he’s saying: that by that point all the poll data is already in and the result should be obvious. But it’s worth noting that Wang and Silver were saying the same things on Election Eve that they had been saying for weeks. They didn’t just move things in to place once all the polls were in.

Who are these people? Here’s how I explained it over and over:

56

Sam Wang nailed the Senate, which I give him massive props for, calling 54-56 Dems (with the assumption that King would caucus with them), with 55 as the best point estimate. I think his edge over Silver with respect to the Senate swamps any difference between them with respect to the Presidency.

I didn’t do too badly in predicting the Senate myself - I said back on September 24 that the GOP would wind up with only 46 seats.

57

Should be obvious. But George Will was predicting a more than decisive Romney victory and Dick Morris was calling a Romney landslide. And out of all the members of Team Data, Nate Silver had the positioning to take on the hacks.

Qualitatively, the forecasts and predictions (yes, there is a difference) of the serious data boys were pretty damn similar. I’d give props to all of them, but grant Silver a special award for schooling Team Bozo (showcased on Fox News, Morning Joe and CNBC) and showcasing their viewers as rubes.

58

Whoops, you just happened to miss one. Just an innocent oversight, I’m sure:

Except that of course when Silver had the opportunity to put one number up for everyone to see, 313 rather than 332 is the number he chose, and a visitor to his blog in the frenzied days leading up to the election would not see the number 332 anywhere. I guess we don’t know if he was “asked” but he’s a rather odd duck if the number he voluntarily featured up top under the heading “Nov. 6 Forecast” was not the number he’d give if he were “asked” to give one.

Now I’ll make like O’Reilly and give you the last word as this has become boring and repetitive. I feel fine with just standing behind what I already wrote upthread (except that there was one thing you were right and I was wrong about: you were second, not first, to talk about Wang, as you picked up that thread from **Derleth **and ran with it.)

59

I’m not quite sure why the New York Times crappy blog design should be attributed to Silver. It is obvious to any mathematically knowledgeable reader that 313.0 is a mean and has no bearing on any actual election results, but is merely a one number summary of his last Monte Carlo simulation run.

I’m not finding where I saw him say that 332 was the most likely election result, but I came close. On November 3rd, he tweeted “Conditional upon Obama winning, his most likely outcomes are 332, 303, 347, 348, 333, 290, 281, 294, 318 electoral votes.” Considering that Silver never once predicted that Romney would win, there’s the straight-forward forecast that you wanted. As I mentioned earlier in the thread, compiling his state-by-state forecasts also leads to a total of 332.

60

Here’s another angle. Nate provided the full probability distribution: anybody who has taken introductory statistics (and most have not) would be familiar with that. If you wanted him to give you a single estimate, then you are actually asking him to provide less information.

Ok, but how do we evaluate the output of Nate Silver or Sam Wang? That’s not an easy question actually: what you need is some sort of scoring technique. Generally speaking though, you can evaluate a forecast along multiple dimensions: one treats assigned probabilities as bets, another looks to internal consistency, another matches assumptions to our current scientific knowledge.

To show I’m not just blowing hot air, here’s a syllabus on the subject, via Duck Duck Go: http://www.nssl.noaa.gov/users/brooks/public_html/feda/