MOBILE USERS: m.isthmus.com
Connect with Isthmus on Twitter · Facebook · Flickr · Newsletters · Instagram 
Saturday, September 20, 2014 |  Madison, WI: 73.0° F  Mostly Cloudy
Collapse Photo Bar

Nate Silver's Flawed Model

Races for the Senate, U.S. House, etc. and other issues of national importance.

Re: Nate Silver's Flawed Model

Postby Huckleby » Tue Oct 30, 2012 10:01 am

I don't doubt Nate Silver's integrity and ability. He could prove to be correct.

This election is so close and unknowable that Silver's predictions should be taken as just one hopeful scenario. In particular, his projections on the electoral college and chances of victory have no reliability. A small surge in the national mood (popular vote) will be reflected in large movement of electoral votes in the swing states.

The Republican theory (great hope?) is that undecided voters tend to break more for the challenger. I don't know if this is true, but it seems credible. What I know for sure is that anybody who is still undecided is thinking in ways outside of my imagination, they might as well be martians. Nate Silver has no insight into their psychology.

This election may not be as close as people are saying. Remember the Walker recall election? Something like 5% of the population decided in last couple days and there was a landside defying all the polls. Either candidate may yet win comfortably in electoral college.

2012 is far more unpredictable than 2008 or 2010. As I said before, I treat the Nate Silver predictions as a shot of whisky, a stiffener in an anxiety-provoking, high stakes, unknowable election.
Huckleby
Forum God/Goddess
 
Posts: 6574
Joined: Thu Oct 20, 2005 5:12 pm
Location: parents' basement

Re: Nate Silver's Flawed Model

Postby kurt_w » Tue Oct 30, 2012 10:51 am

Huckleby wrote:This election is so close and unknowable that Silver's predictions should be taken as just one hopeful scenario. In particular, his projections on the electoral college and chances of victory have no reliability.


I think you're overstating that.

Let's say you woke up tomorrow and the positions in Nate's analysis were reversed -- Romney was given a 73% chance of winning, with a projected 295 electoral votes.

Would that overnight change affect your guesses about the outcome of the race at all? If so, you can't really say that his predictions have "no reliability".

Yes, something could happen in the next six days that would cause the balance to shift. But right now Nate Silver's model -- and all the electoral college predictions -- show Obama winning. Regardless of what might or might not happen in the future, I'd rather be starting from that position than from the opposite configuration. Right?

By the way -- even the conservative websites' projections are starting to converge on Nate Silver's numbers.

FiveThirtyEight: Obama 295, Romney 243
RealClearPolitics: Obama 290, Romney 248
ElectionProjection: Obama 290, Romney 248
Karl Rove: Obama 292, Romney 246*

In other words, since Friday Nate has Obama gaining 1 electoral vote, while RCP has increased his vote by 9, ElectionProjection has increased it by 13, and Karl Rove has increased it by 8.

Let's go back to Bludgeon's comments at the top of this post:

Bludgeon wrote:fivethirtyeight.com is a wet dream on wheels. Especially this year. An orchard full of cherries to pick. Selective reasoning at its finest. [...] I would say Nate's out of his mind, but the truth is I don't believe he thinks it's an accurate representation either. My suspicion is that Silver knows he's churning out the universe's ultimate push-poll.


Since the night Bludgeon posted that -- less than a week ago -- the EC predictions from conservative websites have risen to almost the same level as FiveThirtyEight.

Does that change Bludgeon's thinking at all? Is Nate Silver looking a little better now? I'm assuming that even Bludgeon doesn't believe that Karl Rove, RealClearPolitics, and ElectionProjection are trying to throw the election to Obama.....

* Assigning close states to whichever candidate is leading and splitting the votes of states that are precisely tied.
kurt_w
Forum God/Goddess
 
Posts: 5083
Joined: Tue Jul 22, 2003 3:11 pm

Re: Nate Silver's Flawed Model

Postby Huckleby » Tue Oct 30, 2012 11:35 am

kurt_w wrote: Would that overnight change affect your guesses about the outcome of the race at all? If so, you can't really say that his predictions have "no reliability".

Yes, something could happen in the next six days that would cause the balance to shift. But right now Nate Silver's model -- and all the electoral college predictions -- show Obama winning. Regardless of what might or might not happen in the future, I'd rather be starting from that position than from the opposite configuration. Right?


Sure, I take some small comfort from Silver's predictions.

The problem is that we may see a surge that easily overwhelms Obama's consistent but small popular vote leads in the swing states. The electoral college is highly volatile this year. I think the mood of the margins of the electorate, best seen as *trends* in the national popular polling, best predicts any surge that may come in the last few days.

I'm watching any tiny trends in RCP national average of polls very closely. I don't care if the absoute numbers are right, what's important is that I trust the polling is done consistently from day to day. Romney's 1% lead has ticked-down a couple tenths of a percentage point the last three days. To me, that tiny bit of encouragement is about equivalent to a Nate Silver-induced boner.
Huckleby
Forum God/Goddess
 
Posts: 6574
Joined: Thu Oct 20, 2005 5:12 pm
Location: parents' basement

Re: Nate Silver's Flawed Model

Postby kurt_w » Tue Oct 30, 2012 11:53 am

OK, if you like RealClearPolitics, and like to look at trends, then here's their trend in the electoral college prediction:

Oct 19 through Oct 21: Obama leading by 16
Oct 22 through Oct 25: Obama leading by 24
Oct 26 through today: Obama leading by 42

So, the trend has certainly been in Obama's favor lately.

I am also rather amused at the way RealClearPolitics is converging on Nate Silver's model ... given Bludgeon's angry rant against Nate Silver at the top of this thread.
kurt_w
Forum God/Goddess
 
Posts: 5083
Joined: Tue Jul 22, 2003 3:11 pm

Re: Nate Silver's Flawed Model

Postby Talon Newsman » Tue Oct 30, 2012 11:56 am

We have no choice but to attack Nate Silver.

He tells us things we don't want to hear.
Talon Newsman
Forum Addict
 
Posts: 478
Joined: Thu Mar 02, 2006 6:28 pm

Re: Nate Silver's Flawed Model

Postby Bludgeon » Tue Oct 30, 2012 12:05 pm

Talon Newsman wrote:We have no choice but to attack Nate Silver.

He tells us things we don't want to hear.


Nate Silver has always been at war with Eastasia... uh, wait...
Bludgeon
Forum God/Goddess
 
Posts: 1291
Joined: Tue Apr 06, 2010 2:27 am

Re: Nate Silver's Flawed Model

Postby Bludgeon » Tue Oct 30, 2012 12:21 pm

kurt_w wrote:OK, if you like RealClearPolitics, and like to look at trends, then here's their trend in the electoral college prediction:

Oct 19 through Oct 21: Obama leading by 16
Oct 22 through Oct 25: Obama leading by 24
Oct 26 through today: Obama leading by 42


Another interesting trend: your view of the election entirely through the prism of the electoral college. Yes, electoral votes are the determining factor. But somebody's got to do more than be ahead in state polls: they've got to win the majority of votes in the state. That's something that a poll of state polls is definitively not going to show you when the incumbent is under 50%. Especially not a weighted poll of polls.

Looking at ALL the polls, state and national, one thing that's becoming clear to everybody is that nobody knows what's going to happen, although the fundamentals favor Romney.

Nationally in tracking polls from Gallup to Rasmussen to IBD: These three are a great example of the uncertainty about this election. Three accomplished polling companies, each doing their best to create an accurate representative sample, three consistently varying opinions. One of them is going to be right. There are a lot of questions about who will have the turnout advantage - will there be more Democrats or more Republicans? Because Romney is winning independents by all accounts.

Gallup has Romney up something like 52% to 46% in early voters. Nearly identical to the margin Romney leads by overall.

http://www.gallup.com/poll/158420/regis ... llots.aspx
Bludgeon
Forum God/Goddess
 
Posts: 1291
Joined: Tue Apr 06, 2010 2:27 am

Re: Nate Silver's Flawed Model

Postby kurt_w » Tue Oct 30, 2012 2:27 pm

Bludgeon, a simple question:

If this is what you had to say about Nate Silver when he was projecting a 294-vote electoral college victory for Obama:

Bludgeon wrote:fivethirtyeight.com is a wet dream on wheels. Especially this year. An orchard full of cherries to pick. Selective reasoning at its finest. [...] I would say Nate's out of his mind, but the truth is I don't believe he thinks it's an accurate representation either. My suspicion is that Silver knows he's churning out the universe's ultimate push-poll.


then does all of that same stuff apply to RealClearPolitics, when they project a 290 vote electoral college victory for Obama?

Or is the difference between 294 votes and 290 votes the precise dividing line between "dishonest cherry-picking liar" and "honest conservative poll-watcher"?
kurt_w
Forum God/Goddess
 
Posts: 5083
Joined: Tue Jul 22, 2003 3:11 pm

Re: Nate Silver's Flawed Model

Postby Huckleby » Tue Oct 30, 2012 2:49 pm

kurt_w wrote:OK, if you like RealClearPolitics, and like to look at trends, then here's their trend in the electoral college prediction


To my way of thinking, electoral college predictions in a tight race are uninteresting. Those electoral numbers could swing wildly based on a small surge in popular vote.

Excuse the broken record, but the outcome of the presidential race is largely unknowable and unpredictable, it will be decided in last three days, and there is no statistical evidence today offering any clues about the late deciders. I don't think we even know with high certainty that the final result will be tight. I would bet on Obama based on where he sits today in electoral college polling, but it is a shaky advantage.
Huckleby
Forum God/Goddess
 
Posts: 6574
Joined: Thu Oct 20, 2005 5:12 pm
Location: parents' basement

Re: Nate Silver's Flawed Model

Postby kurt_w » Tue Oct 30, 2012 3:24 pm

Huckleby, there are two separate issues in play in this thread:

1. Is it reasonable to project the winner of the election based on statistical analysis of state polls?

2. Is FiveThirtyEight basically honest, or is Nate Silver putting his thumb on the scale to skew the results?

I'm happy to discuss the first question, but it's kind of an open-ended question and not easy to answer. In contrast, there seems to be an easy way to resolve the second one. Is Nate's model an outlier? Does it give radically different predictions than other models?

I've shown that FiveThirtyEight's predictions are actually right in the middle of the pack. They're not an outlier at all. In particular, the predictions from various conservative-leaning analysts are now showing results that are more or less indistinguishable from those of FiveThirtyEight.

So, if Bludgeon is right and Nate Silver is a dishonest lying cheater, what does that say about all the other sites whose predictions match those of FiveThirtyEight pretty closely? Perhaps it's a Vast Bipartisan Conspiracy.

Personally, I think Nate is basically honest. I think Bludgeon didn't like the *result* that Nate's model was predicting, so he leapt to the conclusion that Nate must be lying. He then posted a bunch of intemperate accusations. Now he's having trouble walking those accusations back, because in this kind of online back-and-forth it's always awkward to admit that you went overboard or were so vehemently wrong about something.
kurt_w
Forum God/Goddess
 
Posts: 5083
Joined: Tue Jul 22, 2003 3:11 pm

Re: Nate Silver's Flawed Model

Postby kurt_w » Tue Oct 30, 2012 3:54 pm

Something else to consider:

Obama needs 270 electoral college votes to win. He has 201 pretty much locked up, so he needs 69 more.

Looking at the RealClearPolitics poll averages, Romney has never, not for one day, had the lead in any of the following states:

Iowa, Michigan, Ohio, Pennsylvania, Nevada, Wisconsin

Those states have 76 EC votes, more than enough to put Obama over the top.

In other words, Romney could win every single swing state where he's *ever* been ahead in the RCP polling average, and he'd still lose.

Romney needs to radically turn the race around in at least one state where he's never yet had the lead. Now, that could certainly happen -- and it explains why he's running this blast of obvious lies in Ohio right now. But his time is running out.
kurt_w
Forum God/Goddess
 
Posts: 5083
Joined: Tue Jul 22, 2003 3:11 pm

Re: Nate Silver's Flawed Model

Postby Bludgeon » Tue Oct 30, 2012 6:40 pm

kurt_w wrote:Bludgeon, a simple question... does all of that same stuff apply to RealClearPolitics, when they project a 290 vote electoral college victory for Obama?

Or is the difference between 294 votes and 290 votes the precise dividing line between "dishonest cherry-picking liar" and "honest conservative poll-watcher"?


No. Unlike fivethirtyeight, Tom Bevan from RealClearPolitics has not spent the last six months issuing a steady stream of articles claiming one candidate or the other has a sure lock on the election.

This is where Silver fans like yourself go astray: you list off a compilation of aggrigators, note Silver's position on the list; or cite Silver's (s-h-o-r-t) track record of prediction rates...

...then you make an unfounded leap in logic to assert that because of his position in the list of aggregators or because of his track record, "therefore, his sunny stream of cheerleader articles must also be accurate."

In this case, one does not equal the other. The current "NO TOSSUP" electoral count is fine; his official odds rating is fine, though not to be taken with the same level of seroiusness as Intrade. But you cross the line from being a respectable armchair quarterback like the rest of us, to being a real life cheerleader, when you start sipping the soup that he's cooking in articles.

Note: none of this is to say I agree with his weighting system. If you review the OP, those reasons are clear.
Bludgeon
Forum God/Goddess
 
Posts: 1291
Joined: Tue Apr 06, 2010 2:27 am

Re: Nate Silver's Flawed Model

Postby Huckleby » Tue Oct 30, 2012 6:48 pm

kurt_w wrote:Looking at the RealClearPolitics poll averages, Romney has never, not for one day, had the lead in any of the following states:

Iowa, Michigan, Ohio, Pennsylvania, Nevada, Wisconsin


It really doesn't matter so much that Romney never lead in those states, what matters is that they are within striking distance for both candidates. The late deciders were never counted before.

Romney has a decent shot in all 6 of those states to ride superior Republican enthusiasm, plus fence-sitting apoliticals swayed by the "Obama had his chance" argument in an 8% unemployment playing field.

I take Nate Silver's projections to mean that Obama has a little better chance of winning election than Romney, ONLY because I don't have a strong feel for which direction the electorate will break next Sunday, Monday and Tuesday. The late deciders could end up looking much like the rest of the voting population.
Huckleby
Forum God/Goddess
 
Posts: 6574
Joined: Thu Oct 20, 2005 5:12 pm
Location: parents' basement

Re: Nate Silver's Flawed Model

Postby kurt_w » Tue Oct 30, 2012 7:25 pm

Bludgeon wrote:Unlike fivethirtyeight, Tom Bevan from RealClearPolitics has not spent the last six months issuing a steady stream of articles claiming one candidate or the other has a sure lock on the election.


This right there makes it clear that you've never actually read one of Nate Silver's columns. My guess is your opinion of him was formed by reading all the hit-pieces in the National Review, etc.

Silver says basically the same thing I've said here: this will be a close election, and either candidate is quite capable of pulling off a win, but as of right now, Obama has an edge over Romney.

For example, here's his most recent column about the state of the race overall:

Oct. 28: In Swing States, a Predictable Election?

Read it! It's chock full of interesting information. It's also remarkably dispassionate, neutral in tone, and cautious in its conclusions.

I think your rant at the top of this thread was way off base. My guess is that you really thought FiveThirtyEight's predictions were some kind of far-left outlier. I can see why you would have thought that, considering the kind of stuff that's been written about him on the National Review website and elsewhere.

But ... maybe the problem isn't with FiveThirtyEight. Maybe the problem's with a right-wing blogosphere that is psychologically incapable of accepting news that it doesn't like.

I'm starting to get the impression that your mind is pretty much closed on this topic. You're not willing to listen to someone from the other side of the aisle. That's too bad, because in my experience one can generally learn a lot from listening to people who might be coming from a different point of view.
kurt_w
Forum God/Goddess
 
Posts: 5083
Joined: Tue Jul 22, 2003 3:11 pm

Re: Nate Silver's Flawed Model

Postby Bludgeon » Tue Oct 30, 2012 8:19 pm

Joe Klein wrote:- I don't know. Over the past week, everyone’s been asking me who’s going to win. Beats me. I really don’t know. The polls seem stalled, hilariously inconclusive. The race is frozen, more or less, for the next few days–except for the advertising.

...And it's the truth.

Really the main reason there's so much discrepancy in the polls is that nobody can agree that there's a consensus on what the turnout is going to look like. Each polling outfit makes their best guess, I think it's clear the voters themselves are actually cooking up a little surprise this year.

Take a left leaning poll (state or nationally), they're operating on the idea that 2012 will be a lot different than 2010 so it "must" be like 2008, and some of these polls actually weight in a Democratic advantage to their sample that's greater than what the president actually won four years ago. To make it happen they have to bring the white vote down to 74%, give Obama 80% of the minority vote, then tweak the white vote again to include more young voters and less seniors.

Items of note:

    >Even the lefty polls show Obama winning less than 40% of the white vote.

    >Even with the statistical weight and warp that comes from assuming a D+9 turnout advantage, those polls only show Obama with a slight lead, or often tied or behind.

    >If Democrats turn out not to have a 2008 level turnout advantage, then all the polls including one are way off the mark.
Romney is on track to win 60%-61% of the white vote, something no Republican has done in a long while. He can win the whole election with 59% (+20%/minority vote). If Obama wins a few less than 80% of the minority vote, Romney could win with 58% of the white vote. If the white share of the electorate equals 77% of the electorate this year and Romney wins 60% of it, another 4% of the vote puts him over 50% of all ballots cast. 20% of the minority vote would do it.

What if there's more give in the minority vote than is generally observed? It's certainly been more pliant in the past.

These are the reasons that very few experts are inclined to say they know or can predict what is going to happen; nobody wants to (or should) go so far as to say the 2012 electorate will look like 2010, or 2008. But in the polls where you don't assume that young voters or black voters are going to show up in the numbers they did four years ago, Romney is ahead.

Today's national NPR poll has a sample where whites make up only 74% of the electorate and Romney is ahead by 1 point. So - what if they really make up 75% of the electorate?

I'm not making a prediction for you either. NPR's 48% to 47% still leaves a few points up for grabs; so do the other polls. My feeling is that the voters have a little surprise cooked up for everybody.
Bludgeon
Forum God/Goddess
 
Posts: 1291
Joined: Tue Apr 06, 2010 2:27 am

PreviousNext

Return to National Politics & Government

Who is online

Users browsing this forum: No registered users and 3 guests

moviesmusiceats
Select a Movie
Select a Theater


commentsViewedForum
  ISTHMUS FLICKR

Promotions Contact us Privacy Policy Jobs Newsletters RSS
Collapse Photo Bar