Different Perceptions of Reality in the US Presidential Election

[Jacob]

I thought this was pretty interesting: http://www.theatlantic.com/politics/archive/2012/11/how-can-that-be-more-on-the-they-cant-both-be-right-saga/264024/

What do you all make of it? What are the consequences in either case?

[DGuller]

If it looks like either American left or right have to be perceiving reality in a highly alternative way, my money's on the left having it closer to the bullseye.  Sometimes past performance does predict future behavior.

[Eddie Teach]

If it looks like either American left or right have to be perceiving reality in a highly alternative way, my money's on the left having it closer to the bullseye.  Sometimes past performance does predict future behavior.

Reality is that the votes(well, most of them) haven't been cast yet, and all the polls are showing a nearly even split yet again. Of course, even if their guy was behind 10 points, pundits and party operatives would be publicly optimistic about their chances.

[derspiess]

Yeah, if you're a public figure (or better yet involved with either campaign) you can't be negative about your own side's chances in these things-- otherwise you'll depress your own side's vote and lessen your chances.  On the other hand, you have to keep your voters and donors 'scared' just enough so that they'll make sure to vote/donate.

[Scipio]

The reality is that despite all of James Fallows's protestations, he's still in the bag for Barry, and he's a partisan hack.

Do most people not understand how probability forecasts work?  Yes, including many people doing those forecasts.

[Strix]

The reality is that elections have become big business with big money. Those people raking in the cash have to make it appear close so that more money is spent. And, like what was previously stated, you have to have swings back and forth to scare donors and encourage more spending.

[OttoVonBismarck]

There has been more argument about polls and how to interpret them and probability models in this election than in any before it, that much is for sure.

A big part of it is an intrinsic misunderstanding of what it means when Nate Silver says Obama's likelihood of carrying 270+ electoral votes on e-day is say, 65% or 70%. Some people just can't in their mind grasp that this means essentially that in Silver's model, Romney still wins 3-3.5 times out of 10. A lot of people on the left are exaggerating what such a probability means, for one, and them some people on the right are rejecting the whole premise of probability modeling and just keep repeating "it's a close election." I don't really like Nate Silver on a personal level, but even his own model considers 60% probability or less to essentially be a "tossup" (and that kind of almost literally makes sense), so the representation of his percent probability as being a sign of an election that "won't be close" is more the lefties who have run with those numbers to a place where those numbers don't lead.

[DGuller]

I recall a while ago OvB mentioning how models come and go, being able to perfectly predict the past, but then failing to predict the next election, and going away.  Turns out Nate Silver is very keenly aware of the problem of overfitting:  http://fivethirtyeight.blogs.nytimes.com/2011/03/24/models-can-be-superficial-in-politics-too/ .  This is a very good layman explanation of the pretty subtle but fatal error a statistician can commit.

My confidence in Nate Silver has gone up even higher.  Sure, his model may still get it wrong (though I think it's least likely among all candidates to do so), but at least it won't get it wrong because of him being a statistics noob.

[OttoVonBismarck]

As is typical DGuller failed to understand my earlier point. Models like the Fair Model have not failed to predict elections because their creators were "statistics noobs." Professor Fair is a statistics professor at Yale. They have failed to always be right because no model is always going to be right.

[DGuller]

The first job of the model is to not always be right in predicting the future, that's impossible.  Its job is to be least wrong out of all the alternatives on average.

[OttoVonBismarck]

It's also worth mentioning Silver is different from Fair. Fair looks at economic data and tries to correlate it to election results, with good success generally but over the years (he's been at it a lot longer than Silver) his model has been wrong.

Silver basically creates probabilities based on polling. But any student of history is aware that periodically there is "massive polling fail", since any serious model based on aggregating and analyzing polls is going to be spitting out numbers based on how those polls have worked in the past anytime there is a large deviation between the polls and the actual result these models will break down. In 2008 Obama won comfortably in almost every State he won, and the polls more or less had reflected that.

Fair has been doing his gig since 1980, Silver had one really good election and it was not a hotly contested election. If Silver gets 49 out of 50 States in 2012 I'll be a lot more impressed, but the reality is an honest Silver would probably admit that such a result would be luck, that every "coin flip" in all 50 States went for his most likely result. Any statistician would admit that is in fact, unlikely.

[OttoVonBismarck]

FWIW I'm a lot more impressed with Fair's models because he actually makes real predictions of popular vote totals. Silver just signs his name to "probabilities" of a candidate winning a given State and a candidate winning the overall election. Fair's model is thus a lot more aggressive in what it is trying to predict, it's a true prediction model whereas Silver is basically the equivalent of someone who sets the odds at a horse race.

[DGuller]

Silver basically creates probabilities based on polling. But any student of history is aware that periodically there is "massive polling fail", since any serious model based on aggregating and analyzing polls is going to be spitting out numbers based on how those polls have worked in the past anytime there is a large deviation between the polls and the actual result these models will break down. In 2008 Obama won comfortably in almost every State he won, and the polls more or less had reflected that.

Fair point, but the science of polling has advanced very far as well.  You have way more polls now, the results of which if used correctly would give a better estimate.  You also have polls that are much better designed than in the past.  IMO, the big reason why polls in the past failed to predict the results sometimes is because polls in the past were very poorly designed.  Take Gallup:  they don't know their ass from their face, and yet they're the ones who are the venerable pollsters, and whose numbers are often looked at throughout time.

Fair has been doing his gig since 1980, Silver had one really good election and it was not a hotly contested election. If Silver gets 49 out of 50 States in 2012 I'll be a lot more impressed, but the reality is an honest Silver would probably admit that such a result would be luck, that every "coin flip" in all 50 States went for his most likely result. Any statistician would admit that is in fact, unlikely.

Silver did state as much numerous times.  If you call every single state as a 60% likely to vote for Obama, and Obama wins all 50 states, you technically did call all 50 state results correctly.  However, you nevertheless did a very horrible job handicapping them, because Obama was supposed to win only 30 of those state if you did the math correctly. *



*  This is a simplification, because state results are almost assuredly not independent from each other, but that would muddy up the point.

[CountDeMoney]

I don't really like Nate Silver on a personal level

Is it the New York Times thing?  The Jew thing?  Or is it the PECOTA thing?  It certainly can't be his academic pedigree from Chicago.

[DGuller]

FWIW I'm a lot more impressed with Fair's models because he actually makes real predictions of popular vote totals. Silver just signs his name to "probabilities" of a candidate winning a given State and a candidate winning the overall election. Fair's model is thus a lot more aggressive in what it is trying to predict, it's a true prediction model whereas Silver is basically the equivalent of someone who sets the odds at a horse race.

Aggressive is not necessarily good.  A competent statistician is as confident as the data lets him be, and not more than that.  It may sound like hedging, but reality is uncertain, and overconfidence doesn't really change that.