Locking in Election Predictions

Avatar

Jason Kuznicki

Jason Kuznicki is a research fellow at the Cato Institute and contributor of Cato Unbound. He's on twitter as JasonKuznicki. His interests include political theory and history.

Related Post Roulette

223 Responses

  1. Avatar Ethan Gach says:

    I know NPR has swing state score card, though I’m not sure if anyone has a complete down-ticket layout. http://apps.npr.org/swing-state-scorecard/

    My first prediction for tomorrow: One man will win the Presidency, the rest of the nation will continue to lose.Report

  2. How about this way? (Link is to my prediction.)Report

  3. Avatar Plinko says:

    I love the idea, but I’m sure I’d just crib whatever Nate Silver has today and call it better than whatever I could muster.Report

  4. Avatar Tod Kelly says:

    That sounds fun. Do you want us to put them up here, or are you going to throw up another post for it?Report

  5. Avatar Shazbot5 says:

    Prediction?

    Pain.

    —-

    Sorry for the necessary Mr. T quote.

    My actual prediction?

    Obama 290, Romney 248

    Obama loses Virginia and Florida in close races ( I mean they go all night, close to a recount.) But he holds OH, NH, CO, NV, and IA. And PA and MI and WI will be closer than hoped, i.e. 2-3 points for the Dems,

    I’ll put the popular vote at, say, Obama 50.5 and Romney 49.2 with Johnson at a fraction of a percent.

    I’ll guess turnout will be at 58 percent.

    Now I see you weren’t asking for predictions here. But whatever.Report

  6. Avatar KatherineMW says:

    Don’t know of one, but if Obama wins both Virginia and Colorado by more than one percentage point, I will officially acknowledge that Nate Silver is smarter than me.Report

  7. Avatar Burt Likko says:

    270towin lets you share your map with a url. I’m mobile and can’t generate one of my own at the moment. But mine looks just like Will’s at our sub blog with the exception that I call Colorado for Romney instead of Obama.Report

  8. Avatar Tod Kelly says:

    My map is here. Essentially, though, here is my own “pulled out of my ass & wouldn’t bet more than a beer on it” prediction:

    The obvious states will go the obvious way, models by the Examiner, Daily Caller, Unskewed Polls and Mr. Farmer not withstanding.

    The sluggish economy will hurt Obama much more in swing states than those polls now indicate. As a result, Romney will easily win Florida and North Carolina, and will also pull out victories in Virginia, Colorado and his running mate’s home of Wisconsin.

    Obama will win Nevada, Iowa and Ohio, which will give him enough for a victory.

    Scare stories run by the media that this election will be up in the air for months/decided by the Supreme Court/etc. will be exactly that: scare stories. We’ll know who will be the next president by Wed morning.Report

  9. Avatar LarryM says:

    Hmm. Really I have to go with what one might call the consensus of the conventional wisdom. At this point, if one gives all toss ups to the guy ahead in the polls, the RCP map and the 538 map are identical. Romney gets Florida and North Carolina, and nothing else among the contested states, for a comfortable 303-235 Obama win.Report

  10. Avatar KatherineMW says:

    Thanks, Burt! Here’s my map (Obama 281, Romney 257). Once again, it comes down to Ohio.

    http://www.270towin.com/2012_election_predictions.php?mapid=bkcnReport

  11. Avatar DRS says:

    Obama – 294. Romney – 244.

    Ohio, Virginia, Michigan, Massachusetts – Obama. Colorado, Utah – Romney. (Romney loses 2 out of 3 of his “home” states.)

    Why? Because I have faith in John Scalzi.Report

  12. Avatar Jaybird says:

    As Colorado’s Unofficial Spokesperson, allow me to say something useless:

    Disclosure: I live in the bluest part (downtown) of the reddest part (Colorado Springs) of Colorado and I’m sure that that tints my view. With that said,

    Colorado has several different kinds of Conservativism. The kind you’re probably familiar with is the Focus on the Family version made popular in Colorado Springs (back in the 90’s, we bragged about having 300 Christian Organizations headquartered here!) but, once you get a couple of miles West of Denver, there’s also the “Leave Me The Hell Alone” conservativism that the Mountain West is more generally known for and the pro-business conservativism found in places like Fort Collins.

    The three reliably blue cities are Denver, Boulder, and Pueblo and they tend to politely ignore each other because they have surprisingly little in common past not liking Republicans. Boulder is overwhelmingly white, hippie, and “Progressive”. Denver is Cosmopolitan. Pueblo is mostly blue collar conservative Democrats and even more Conservative Latino/Hispanic Democrats.

    I can see Colorado going Romney.

    But, as I said, I live in Colorado Springs and that could easily tint my view of the state.Report

  13. Avatar Tod Kelly says:

    Serious question for number crunching election-wonks:

    I keep seeing NH showing up as a swing state, despite it going D in every election but one since the 80s, a D gov, and a solid lead in the polls for O. Why is NH considered a swing when, say, Delaware is not?Report

    • Avatar Burt Likko in reply to Tod Kelly says:

      They’re just kind of ornery up there.Report

    • Avatar Kim in reply to Tod Kelly says:

      Same reason PA is. If a state is near enough to the edge that national momentum COULD sway it, it becomes a “swing state.”

      I feel like PA has been overcalled as a swing state, for a few reasons:
      1) It’s electoral numbers make it a big prize.
      2) It doesn’t have early voting, so McCain concentrated heavily here (and Romney’s throwing his last nickels here too).Report

    • Avatar Jaybird in reply to Tod Kelly says:

      Free State Project ended up moving a substantial number of folks to New Hampshire. While I’m sure that non FSP-folks have voting patterns commensurate with Delaware’s, I’m also pretty sure that the FSP-folks vote. Like, 90%+ of them. While a huge chunk is likely to go for Johnson, I’m pretty confident that another huge chunk will go for Romney.Report

  14. Avatar trizzlor says:

    Assuming a systematic oversampling of D’s, Romney 281/Obama 257 with Romney taking all of the Bush 2004 states except for Nevada:

    http://elections.huffingtonpost.com/2012/custom-presidential-election-map#nkammnjannabjenje

    BTW, my hunch is that the polls are systematically biased somehow and Nate Silver’s confidence intervals are much wider than they appear, so I would be shorting him if I could right now.Report

  15. Avatar Marchmaine says:

    Obama wins 275 to 263.

    Romney carries these toss-ups: FL, VA, NC and CO. That’s it.

    Democratic turn-out will be ordinary (but down from 2008) but Republican turn-out will also be ordinary (only slightly up from 2008). There are a lot of sparks from this cycle, but not a lot of heat; Romney ultimately won’t get the nose-holder vote he needs to win. Obama will limp into the White House with no mandate, and no clear reason why he’s there rather than the other guy.

    (If Romney wins, just invert everything I wrote above).Report

  16. I’ve got Obama winning all the swing states except North Carolina, although I expect NC to be closer than advertised. That seems to give me 332-206.

    For the Senate, I’ve got Maine, Massachusetts, Indiana, and Nevada flipping blue (or blue-enough in Maine), with Nebraska and North Dakota flipping red. That gives me a final count of 55-45.

    And I’ve got 225-210 in the House (Republicans control).Report

  17. Avatar Jesse Ewiak says:

    http://www.270towin.com/2012_election_predictions.php?mapid=blMl

    Obama 303 – Romney 235 ; PV in Obama 50.9 – Romney 48.1Report

    • This is excellent, 5 equally valid scenarios:

      Scenario 1: Nate Silver is right

      Scenario 2: RCP is right

      Scenario 3: Rasmussen is right

      Scenario 4: Gallup tracking poll is right

      Scenario 5: Gallup electorate poll is right

      #5, the overrated Dem/hidden GOP scenario, makes the most sense to me, although not the extravagant EC vote. If the phenomenon doesn’t exist, Obama should cruise. The secret of the Nate Silver formula is that if you keep getting Obama 50 Romney 49, despite being within the 3% Margin of Error, 75% of the time, the 50% beats the 49% [if you follow that].

      This is Joe Trippi’s prediction as well, that it should be comfortable one way or the other— if there’s a hidden GOP vote, it’ll tip a lot of states, if not, Obama wins ’em.

      Which is OK by me. Really. I think another Bush/Gore 2000 would be worse for the country than a clean Obama win. There are still people crying about 2000 and I just can’t take any more.Report

      • Avatar Morat20 in reply to Tom Van Dyke says:

        Question: Why aren’t there seperate scenarios for every other pollster’s tracking polls and electorate polls?

        Why are you singling out Gallup? Why are you only addressing Rassmussen as a single unit, and not two scenarios?

        What logical basis is there for singling out two pollsters and one turnout model over all the others?Report

        • Avatar Ryan Noonan in reply to Morat20 says:

          I believe we went through a phase where the appropriate response to things like this was OPRE.Report

        • Avatar Tom Van Dyke in reply to Morat20 says:

          Geez, man, believe ’em or don’t. There are 2 Obama scenarios and 3 Romneys. Pick one you like. They’re all valid, none would surprise me, except the Romney with 377.

          You’re not going to get Obama elected in a comments box, so let’s chill, bro.Report

          • Avatar Shazbot3 in reply to Tom Van Dyke says:

            Another scenario is that absolutely no one votes for either Romney or Obama or any presidential candidate. Every state is a 0-0 tie and thus the electoral colleges is either a tie or somehow yields no result. Then the presidential election goes to the House where, again, everyone forgets to vote. We end up with no president and the Republic is dissolved.

            That is a 6th possible scenario.Report

          • Avatar Morat20 in reply to Tom Van Dyke says:

            You missed answering the question. There are, what, six tracking polls? All with likely voter screens.

            Why are you singling out ONE of those polls as an “entire scenario” and it’s likely voter model as another “likely scenario” and Rasmussen as a third?

            Do you think Gallup’s tracking poll is so superior that it is equally as likely as the aggregate of all other polls? Or that it’s likely voter model is so superior that it is as likely as the aggregate of all other polls?

            I’m not getting the logic here.

            If you say “Scenario 1: Gallup is right” how is Scenario’s 2-X not “Rassmussen is right”, “Pew is right”, “Pollster Whatever is right”.

            The setup there elevates Gallup above all others. Why?

            I don’t give a flying fish about Obama winning or Romney winning right now. I’m trying to follow your logic. You’re making it surprisingly difficult, almost as if you don’t want to defend it or explain it.Report

            • Avatar Tom Van Dyke in reply to Morat20 says:

              No, I really don’t, Mr. Morat, and only a moronity can’t see why. 😉Report

              • Avatar Morat20 in reply to Tom Van Dyke says:

                Again, you’re not answering a simple question: Why are three of the five scenarios taken from two pollsters. Why those pollsters? Why not Pew, or any of the other tracking pollsters?

                Heck, why is gallup there twice?

                You posted that to be read. You obviously want people to understand your point — to communicate. So please, again: Why is Gallup there twice? Why Rasmussen? Why not any other tracking polls?Report

              • Avatar James Hanley in reply to Morat20 says:

                You posted that to be read. You obviously want people to understand your point — to communicate.

                Objection. Assumes facts not in evidence.Report

              • Avatar George Turner in reply to Morat20 says:

                The point in breaking things down into scenarios isn’t to let all the polling companies vote, with the majority winning, it’s to break things down into scenarios reflecting different, well, scenarios of how the electorate could break.

                Who would waste their time actually writing up identical scenarios?

                “And scenario #2, which is based on tracking poll data that matches the tracking poll data used in scenario #1 because it uses similar methodology, looks exactly like scenario #1. Similarly, scenario #3, which uses aggregated poll data that matches the poll data used in scenario #1, looks exactly like scenario #1, but let me completely repeat the thirty minutes of explanation I gave in the previous hour, when I detailed scenario #1.”

                “Um, I just got beeped. A patient needs a brain transplant stat.”

                “Yeah, I just got beeped too. The wife’s cycle just started.”

                “Beep here. OECD’s quarterly aluminum smelter data just came out and I’m urgently needed in the office to move it to the recycle bin.”Report

              • Avatar Morat20 in reply to George Turner says:

                And yet, you can’t answer: Why those three? Why Gallup, Gallup Likely Voter, Rasmussen as three of the five?

                Why those three, as opposed to any other three?

                I ask because I don’t think Tom pulled those out of a hat, and he’s trying to make some sort of point about liklihood, but I can’t even get a clear answer as to why he chose THOSE scenarios over any others.Report

              • Avatar Tom Van Dyke in reply to Morat20 says:

                Thx Messrs Turner & Morat. You got it, between you.

                Mr. Morat, I did pull those 5 scenarios out of a hat so to speak, a blog post I found interesting and informative. Mr. Turner gets it that you can have 10 scenarios if that seems fair, it’s just that they’re largely redundant.

                The prosecution can call 9 witnesses and the defense just one.

                I also cited Dem stategist Joe Trippi and gave a nod to Nate Silver’s method, that consistent 50-49 results move to 75% probability despite a 3-pt margin of error. I believe I’ve been completely logical about the whole affair. It’s just a comment box.Report

              • Avatar Shazbot3 in reply to Morat20 says:

                “It’s just a comment box.”

                If you make better comments (more intellectually honest, with better arguments) into the comment box, then the comment boxes that you read in reply won’t call you out on poor arguments and intellectual dishonesty.

                Or maybe you don’t care about truth and intellectual honesty and good argumentation. It’s just a blog, who cares about the posts. It’s just a comment box, who cares if I say false, silly things and refuse to own up to my errors.

                Maybe that should be your contribution to the blog. (as you are on the masthead, your ideas are associated with the blog.) You can make the blog’s credo “It’s just a blog, so you don’t need to justify what you say or admit mistakes.”Report

              • Avatar Tom Van Dyke in reply to Morat20 says:

                The intelligent people understand me just fine, sir. And those with a record of honesty and civility will get an explanation. As for the rest, you’re right: I don’t care.Report

              • Avatar Shazbot3 in reply to Morat20 says:

                “The intelligent people understand me just fine”

                Your ideas are so complicated, like special relativity or cosmic string theory, that poor old ordinary folks can’t understand. These ideas need pages and pages to be explained. They look like simple misunderstandings of the most basic tenants of statistics and inductive reasoning, but in reality it is a jazzy, Coltrane-like display of mathematical genius.

                Please.

                You got caught saying something really ill-informed and obviously wrong and you refuse to admit it. Period.

                I will take no explanation of your position as admission that you were wrong but that you refuse to admit it.Report

              • Avatar Tom Van Dyke in reply to Morat20 says:

                Somewhere you got the impression people owe you the time of day. They don’t.Report

              • Avatar Shazbot3 in reply to Morat20 says:

                You everyone an explanation of the bizarrely false thing you said or an admission that you are wrong.

                If you do not live up to your obligation, you are accepting your intellectual dishonesty, and you do not belong on the masthead.Report

              • Avatar James Hanley in reply to George Turner says:

                You’re missing the issue, George. TVD claimed each scenario was equally valid. As a matter of basic statistical validity they are not equally valid.Report

              • Avatar Morat20 in reply to James Hanley says:

                Well, that’s assuming by “equally valid” Tom meant “have equal odds of happening”, which I didn’t think Tom meant, because that’s moronic.Report

              • Avatar James Hanley in reply to James Hanley says:

                Moronic or not, since he goes on to argue which makes the most sense to him, and implies things about likelihood, I think he is arguing that the single polls have at least equal odds of being right as the poll aggregator.Report

              • Avatar Morat20 in reply to James Hanley says:

                Yes, which is why I keep asking him to clarify. I’m certain that’s not what he meant, even though that’s how it’s coming across.

                But he won’t clarify it, for reasons clear only to him.Report

              • Avatar James Hanley in reply to James Hanley says:

                But he won’t clarify it, for reasons clear only to him.

                He never clarifies. That’s his style. That it’s a form of intellectual dishonesty to refuse to make oneself clear then blame others for not understanding , or, as often happens, accuse them of purposely misinterpreting, one seems not to matter to him.Report

              • Avatar Kim in reply to James Hanley says:

                rofl. Someone’s walking around nekkider ‘n me.
                I may make some crazy arguments…
                I may bring some stupid facts to the table…
                But I try like hell to get the math right!Report

              • Avatar Shazbot3 in reply to Tom Van Dyke says:

                On the masthead folks.Report

              • Avatar KatherineMW in reply to Tom Van Dyke says:

                Yes. Because you picked the tracking polls that are best for Romney and left the others out. But that’s no reason to insult Morat for presuming nonexistent good faith.Report

              • Avatar Morat20 in reply to KatherineMW says:

                I found it funny. I was chastised for being rude by asking him to clarify, rather than assuming he’d made a mathematically illiterate statement.Report

            • Avatar George Turner in reply to Morat20 says:

              There are other elements, such as the note in the Tom’s link that polling over-estimated the Democrats’ votes by 5 to 7% in the last two elections where a Democratic incumbent was faced by a Reublican challenger.

              Apparently Republicans are more motivated to get rid of Democrat incumbents than assumed, and Democrats are less likely to defend them – compared to what the polls say, and if you think Republicans didn’t like Clinton in 1996, it’s nothing compared to Obama in 2012 where they feel they’re almost facing and end-times scenario if he wins. Cats and dogs in soup lines together, and neither one even likes soup.Report

              • Avatar Kim in reply to George Turner says:

                So Dole won? (seriously, is this just stupid averaging math, where Carter vastly swamps Clinton?)

                I’d exempt Carter from polling, simply because we hadn’t gotten to modern political landscape yet (aka one without wild moodswings).Report

              • Avatar Morat20 in reply to George Turner says:

                Hehe. Did you follow the link on that cite? It was to 538, to a post titled “Poll Averages Have No History of Consistent Partisan Bias”.

                The guy apparently didn’t actually read the post he linked to, as he used it to make a point that was the complete OPPOSITE of the point 538 made.

                That’s hilarity. I mean, you gotta have either big brass ones or just be totally dense to cherry-pick out a fact like that, but leave in the link to the post that was “Why cherry-picking facts like that is really stupid and historically inaccurate”.Report

              • Avatar George Turner in reply to George Turner says:

                No, Dole didn’t win, but he outperformed the polling expectations by a wide margin. Perhaps it’s right to ignore that election, blaming the error on Perot, or perhaps it’s not. But when you start thinking in such terms, the idea of predicting the odds of an Obama victory to three significant figures seems pretty ludicrous, no?Report

              • Avatar Morat20 in reply to George Turner says:

                Are you still on that significant figure thing? You do realize polls are done on normalized data, which have standard deviations accurate to three figures, minumum?

                And since he uses a Monte Carlo simulation, that does tens of thousands or millions of runs, three significant figures is actually far less than he can scientifically use?

                Quick question: If I simulate a hand of poker a million times, and I say the odds of getting two pair are 5.75%, are you going to laugh and say “There’s only 52 cards! How can you get three significant figures!”

                If the answer is yes, stop talking about math. If the answer is no, why are you harping on something you’re obviously wrong about?Report

              • Avatar George Turner in reply to George Turner says:

                If you don’t know which 52 cards are still in the deck, and in this case we don’t, then yes, your statistics are bogus. This isn’t even the second time Obama has faced Romney, much less the millionth. Our demographics, preferences, and opinions shift between each election. The way we respond to pollsters changes. The way the media portrays the race changes. The election rules change. The spending levels change. Almost as much changes as stays the same, if not more.

                Silver’s techniques come from baseball, where nothing is allowed to change. If the statistics even start to shift, the game adjusts the pitcher’s mound and other factors to try to cancel the change, so that the statistics from one era can be reliably compared to the statistics from another era. In baseball, or in cards, there’s nothing remotely equivalent to the questions, “Should we include the two games with Ross Perot in the statistics?” “Should we include the Carter/Reagan election because some argue that it took place before the modern breakdown?”

                As an engineer, it’s very important to know when a number is BS, because people often die if you convince yourself that you can extrapolate vastly more certainty than the raw data justifies, allowing you to push your design to imaginary margins when the calculations are beset by unacknowledged unknowns, assumptions, and guesses. The field allows for very precise calculation of planetary trajectories, and very imprecise knowledge of when an O-ring is going to blow out.

                Just because your technique can produce a number to three significant figures doesn’t mean that all three figures are, in fact, signficant. When there is doubt about the validity of the assumptions, you have to test, test, and retest, to establish their validity. In the case of elections, we don’t get to run the same candidates over and over under the same circumstances. They’re always different, always changing. The candidates themselves change their behavior according to polling data in a feedback cycle.

                Silver is like McNamara, thinking a cmputer program can quantitize human conflict and spit out a victory probability to three significant figures. Vietnam didn’t go so well.Report

              • Avatar Morat20 in reply to George Turner says:

                Right. So you understand neither significant digits OR statistics.

                Or rather, you understand them the way a Young Earth Creationist understands thermodynamics.Report

              • Avatar Patrick Cahalan in reply to George Turner says:

                Silver is like McNamara, thinking a cmputer program can quantitize human conflict and spit out a victory probability to three significant figures.

                I think you misunderstand statisticians in general. Silver is nothing like McNamara, because he’s not actually trying to enact policy based upon his analysis of the numbers. He’s just calling it as he sees it.

                As near as I can tell, anyway.Report

              • Avatar James Hanley in reply to Patrick Cahalan says:

                Silver’s also not predicting an outcome to three decimal points certainty, which I think is what George believes he is doing.Report

              • Avatar Morat20 in reply to Patrick Cahalan says:

                Right. he’s just reporting the outcome of a Monte Carlo simulation to three significant digits, which is just fine and perfectly acceptable, because Monte Carlo simuluations are simulated tens of thousands or more times.

                George is thinking like an engineer, who uses his least precise measurement. Not as a mathematician or statistician, whose use of significant digits is different.

                The playing card example was fairly choice — to calculate the odds you’re using an analysis of very large numbers (52!) which means you can and should report your results to several significant digits.

                Claiming you can’t even reach three? That’s effectively claiming you can’t calculate the odds on a Royal Flush, because you cannot express 650,000 to 1 odds with as anything other than “zero” that way.Report

              • Avatar James Hanley in reply to George Turner says:

                And since the numbers can’t be perfect, my gut feeling–which is even less testable–is obviously superior!

                Therein lies the problem, George. You’re good at critiquing the statistical models (well, not really, but let’s say you are), but you aren’t willing to provide any scrutiny to the alternative modes of predicting.

                I really need you to extend your insightful critiques to other predictive methods. See, this morning I checked my tea leaves, and they said Romney was going to win. Then I checked my Tarot cards, and they said Obama was going to win. Then I had a bowel movement and examined my turds carefully, and their message was that it was going to be a tie. My friend is asking me to bet on the outcome of the race. Please critique each model so I know which way I should wager.Report

              • Avatar Patrick Cahalan in reply to James Hanley says:

                Always go with your colon. It can’t be bullshitted, it sees too much of that stuff to fall for it.Report

              • I’ll go ahead and be the annoying contrarian and say this:

                This “whose model is right” was a lot more interesting four weeks ago with a month left before the election. With 12+ hours to go, it seems a little tedious.

                We’re just about to find out if Nate’s right and Obama will win with some electoral votes to spare, or if the guys Tom, George and others find convincing are right and Romney’s going to win in a small landslide.

                Really, the quality of arguments made aren’t going to change that.Report

              • Right, Tod. Latest polls via Drudge:

                EXCLUSIVE: Romney internal polls put him UP in OH, TIED in PA and WI…

                national:

                GALLUP: R 50% O 49%
                ABCWASHPOST: O 50% R 47%
                RASMUSSEN: R 49% O 48%

                And FTR, I have other reasons for leaning to the “missing GOP” theory more than “liking” Rasmussen or Gallup but there’s not much point going through the grinder any more since the real thing goes down in 24 hours.

                And OTOH, I was at an African American bookshop the other night [a prof friend of mine is on a book tour] and saw a quiet, cold intensity for Barack where 2008 was more hot and heavy. The lower black vote theory may not hold either, despite the fact that the Obama presidency has been very bad on the economic front for them.Report

              • Avatar Shazbot3 in reply to James Hanley says:

                Tod,

                The outcome of Obama winning is logically compatible with Tom being right and the outcome of Romney winning is logically compatible with Silver being right.

                The single outcome of the election will prove nothing (almost nothing) about the reliability of Silver’s methods.

                So, I disagree. No?Report

              • Avatar James Hanley in reply to James Hanley says:

                We’re just about to find out if Nate’s right and Obama will win with some electoral votes to spare, or if the guys Tom, George and others find convincing are right and Romney’s going to win in a small landslide.

                But that’s not correct, Tod. We’re about to find out whose prediction is right, not whose model is right. Whether Silver’s model is right or not actually has little to do with whether Obama wins or not. In fact as I noted above, he’s not technically even making a prediction–he’s just providing a statistical analysis of odds. Silver gives Romney almost a 20% chance of winning the election–do you know what happens when you have 1 in 5 odds? You win 20% of the time.

                That some folks here can’t get that, but think they understand these things better than Silver is, as Morat notes, very much like being a young earth creationist.Report

              • Avatar Morat20 in reply to James Hanley says:

                Wow, Tom, if you believe that Romney’s internals have him “Tied in PA and up in Ohio”, well…

                I’ve got a bridge to sell you.

                Although the anecdotes from what your friend saw…that’s stirring stuff. You should take all this poll denialism and skepticism and gut instincts to Vegas, get rich, and buy a private island or something.Report

              • You don’t read me carefully or charitably enough to have a conversation, Morat. For instance, ’twas I who was in the bookstore.

                I guess I’ll leave George to put up with this nonsense. If I’m to do the unnecessary for the ungrateful, I need to get paid.Report

              • James:

                Good point, and poorly phrased on my part. And you are of course correct.Report

              • Avatar MFarmer in reply to James Hanley says:

                “That some folks here can’t get that, but think they understand these things better than Silver is, as Morat notes, very much like being a young earth creationist.”

                Spanish male penguin: Will you go out with me?

                Beautiful female penguin: Fat chance!

                Spanish male penguin: I have a chance, and it’s FAT!Report

              • Avatar James Hanley in reply to James Hanley says:

                Romney internal polls

                Anyone who believes a public report of an internal poll–regardless of what party or candidate it comes from–is either hopelessly gullible or is clinging to their partisan binders with the strength of desperation.

                If the Obama camp released an internal poll showing them up in Ohio, that’s when I’d begin to believe Romney was well ahead.

                But seriously, the Romney internal poll allegedly shows him up by one point. One point. A U. Cincinnati poll shows Obama up by one point. You know what we call that? A statistical dead heat. You know what folks have been saying about Ohio for quite a while now? It’s a tossup.

                You know what we call a statistical dead heat? Yeah, a tossup. So Romney’s internal poll is showing us just what all the other polls have been showing us. Wow, big scoop there, Mr. Inside.Report

              • Avatar George Turner in reply to George Turner says:

                No, I understand both. I think it’s you who is confused. Engineers constantly have to deal with measurement errors from raw sensor data. Sometimes those measurement errors are well understood, and sometimes they’re not. You can’t wash it away statistically if you don’t have either a characterization of the error or know the source. It could be that the individual sensor errors are random and all cancel out, or it could be that they all swing together because the error is actually in a common reference quantity, or that it comes from an unanticipated factor that isn’t even being measured, such as EMI.

                What you can’t do is act like an undergraduate in a liberal arts statistics class and assume it’s just random noise. You actually have to prove that.

                The worst trait in any engineer is hubris, because that’s a very dangerous trait for someone who’s designing things like the airliners you ride in. Pretending you’re certain, or believing with all your heart that you’re certain when you actually don’t even have good data, is a recipe for disaster, so we put a lot of time and effort into knowing when we really know something, and knowing when we’re spinning yarns with a spreadsheet filled with raw data based on bad measurements, unjustified assumptions, and other unknowns.

                Put more simply, we have to know where our numbers are accurate and where we’re making assumptions, and the design margins have to allow for the questionable assumptions being wrong. If we just resorted to statistics without understanding the uncertainties in the raw data (and those are usually not random) you wouldn’t get on an airliner after you noticed how we spout dizzying reliability statistics for completely untested systems. We wouldn’t even bother with test flying aircraft before putting them into service, confident that everything would work without even running a systems check on the prototype.

                Or as was famously said, “If your experiment relies on statistics, you should’ve designed a better experiment.”

                In short, if you could measure and predict human preferences as accurately as you claim, the phrase “marketing disaster” wouldn’t produce a single Google hit because nobody would’ve ever heard of one.Report

              • Avatar Morat20 in reply to George Turner says:

                George, here’s a question:

                Is 52 * 51 = 2652, or 2700? One has four significant digits, the other 2.

                Is (52 * 51)/100 = 26.52 or 27.00?

                You’re claiming Nate’s wrong because he says “26.52” instead of “27.00”.

                you wouldn’t get on an airliner after you noticed how we spout dizzying reliability statistics for completely untested systems.

                My day job is working with the software aerospace engineers use to determine when aircraft parts and frames have to be checked or replaced.

                The number of flight hours required before a check or replacement are a function of pure statistics, revolving around median time to fail and standard deviations of it.

                You should stay off of planes.Report

              • Avatar George Turner in reply to George Turner says:

                Hrm… Do I know an integer from a real, a number from a number that represents a measurement that had an error range? I think I do.

                Your reliability data was gathered at great trouble and expense, with lots and lots of materials tests to back it up, and then built on a huge amount of maintenance data.

                What do you do for a protyoype that’s flown six times and had major design reworks after each? You inspect everything you can easily get to, every flight, and tear down anything that the engineers were worried about. If only they had Nate Silver to tell them that he’d modeled the aircraft based on in-flight photographs of the six test hops, and in 78.6% of the simulations, no stress cracking was observed in the first 2,000 hours of operation.

                Perhaps this is why airlines don’t hire baseball statisticians to run their maintenance programs.Report

              • Avatar Chris in reply to George Turner says:

                Quibble: medians don’t have “standard” deviations.Report

              • Avatar bookdragon in reply to George Turner says:

                Please, as one engineer to another: stop making engineers sound as though we don’t know how to design using statistics.

                In my field the materials do not have the sort of massive databases you posit and in fact frequently show both batch-to-batch and manufacturer-based variations. Does that mean we don’t design with them? No. It means that we build in factors to account for uncertainty and use statistical methods to determine A and B-basis allowables.

                There is no reason that Silver or Wang or Linzer or any of the other statisticians (whose predictions largely agree, btw) can’t do the same when dealing with batch-to-batch, i.e., election-to-election variability. And even where human variability is concerned, there are still trends and other known factors that can be applied to pool the data usefully.

                Now, engineers would build safety factors into using such data for design. That is certainly true. However, now of these election prediction stats guys are designing vehicles. They are only calculating odds, which requires a much lower standard of certainty. (If they were doing this for life-or-death, no one would predict anything over 100 EVs for either candidate).Report

              • Avatar George Turner in reply to George Turner says:

                Bookdragon, I’m not saying engineers don’t know how to use statistics, I’m saying we know how to use them very, very well, and most importantly know that the output is meaningless without traceability data all the way back to the real world, so that the assumptions, caveats, and warnings that start with the raw data are traceable all the way to final output, so that you also know when the output becomes invalid because of a change in conditions (such as temperature) or other externalities.

                The raw data we’re dealing with in this case comes from polling, and how much statistical analysis would you do if your part’s MTBF numbers came from not from the engineers, but from calling a contractor’s sales reps, HR people, accounting people, and receptionists and asking them, “Will your part – A) last not very long. B) Last a long time C) Last a really, really long time.”

                Given that fundamental method of collecting data, I really don’t think any amount of math can push the answers to an accuracy level beyond “Somewhat likely, pretty likely, and very very likely”, no matter how much you apply cutting edge statistics, signal processing algorithms, neural nets, fuzzy logic, or wavelet analysis.

                There’s just no precision in the input, and the lack of precision isn’t precisely known, either.Report

              • Avatar Chris in reply to George Turner says:

                The gist of this is, “I don’t like it, so it’s not good.”Report

              • Avatar Shazbot3 in reply to George Turner says:

                “Silver’s techniques come from baseball, where nothing is allowed to change. If the statistics even start to shift, the game adjusts the pitcher’s mound and other factors to try to cancel the change, so that the statistics from one era can be reliably compared to the statistics from another era. In baseball… there’s nothing remotely equivalent to the questions, “Should we include the two games with Ross Perot in the statistics?” “Should we include the Carter/Reagan election because some argue that it took place before the modern breakdown?”

                Huh?

                I’m pretty sure that “Silver’s techniques” don’t come from baseball, nor do they come from Silver (as if he is the father of this basic kind of inductive reasoning.) Sure, baseball is where Silver first applied his predictive skills, but other statisticians, like Sam Wang, have been doing using “these techniques” (nothing fancy, BTW) for a while, and have been trained and train people who use statistics in a variety of fields to predict human behavior.

                Also, I’m no expert, but I’m pretty sure there are era effects in baseball statistics. (I know there are huge era effects in hockey stats.) I’m also pretty sure that there are questions about whether data from a player’s injured seasons should be used when using past performance to predict future performance. There are also questions about how to weight all sorts of contextual statistics (I am more familiar with hockey than the baseball stats.)

                I mean, obviously, the data about past performance in baseball will look somewhat different than in hockey, which will in turn look different from past election and polling data in the U.S., which will look different than past election and polling data in other countries, which will in turn look different from data about whether patterns, which will in turn look different about past uses of subway lines during the holidays. But the “techniques” used to predict future can be used on a variety of different data sets.

                I don’t get anything you are saying.Report

              • Avatar Morat20 in reply to Shazbot3 says:

                You know, one of my friends was always just flabbergasted when I talked about basic stats problems — the kind you cover in Probability 101.

                Things like, you know, take a bag full of 50 red marbles, 30 blue ones, and 20 yellow ones. Remove 25 at random, and set them aside without looking at the color.

                Remove five from the bag. What are the odds at least 3 are red? At least one is blue? That you get one of each color?

                He was shocked you could actually solve that.

                Of course, these days I wouldn’t bother doing the proper analysis. I’d whip up a program and simulate it two million times and let you know the percentages, which would be pretty much exactly the same as the “true” (analyzed) value.

                The funny thing is — he’d never dream of saying that it COULDN’T be calculated, just because he didn’t know how — or that the simulation couldn’t have three significant digits because I was only drawing out one signficant digit’s worth of marbles.Report

              • Avatar George Turner in reply to Shazbot3 says:

                Take a bag with 45 to 55 red ones, 25 to 35 blue ones, and 10 to 15 yellow ones, loaded non-randomly by someone trying to prove a point. What happens to the accuracy of your prediction?

                The marble loading is being done by people trying to effect the outcome of what comes out of the bag. Most are partisans, some are independents, and some are flakes. All your statistics about the probabilities of what comes out of the bag is based on your not wholly uninformed guess about who loaded the bag, and what they were thinking when they did.

                The errors in your knowledge of which people showed up to put the marbles in, and which marbles they picked, since those factors are not random, necessarily limits the accuracy of your prediction about which color marble is most likely to come out.Report

              • Avatar Stillwater in reply to Shazbot3 says:

                The marble loading is being done by people trying to effect the outcome of what comes out of the bag.

                So, your sticking with the story that pre-polls are inaccurate because a statistically significant number of people who will actually vote Republican were lying to pollsters about it?

                I mean, that’s the only way for this comment to cohere with everything else you’ve said about this topic.Report

              • Avatar Kolohe in reply to Shazbot3 says:

                That the GOP has lost its marbles recently is a valid analysis.Report

              • Avatar George Turner in reply to Shazbot3 says:

                That’s just one of many sources of uncertainty. For example, in one of the latest polls (Gallup, I think) Romney and Obama were tied, but in a poll that had +11 Democrat sampling, whereas other polls have found R’s and D’s evenly matched. That’s a pretty big error bar in the data.

                Keep in mind, if the polls were always accurate we wouldn’t have upsets.

                Or, put more simply, you and I have a bet to see who can get the most friends to come over for a party. We call all our usual friends up and note how many say they’ll be there. We can go look at our previous four or five parties held over the past twenty years and make some estimates as to who will win. But since we can both do that, and we really want to win the bet, we start doing all sorts of crazy things, like driving around to our friend’s houses and trying to drag them with us. We work the phones to try to get our friends to bring some of their friends. We come up with clever tweets to try and cause a cascade of party goers to show up. We do what’s called “the ground game.”

                There’s only one real certainty involved in this bet, that the geek sitting on his iPad building a probability spreadsheet is not going to get laid.Report

              • Avatar George Turner in reply to Shazbot3 says:

                Put more simply, if Silver’s technique didn’t have some fundamental problems because of the uncertainty of the data inputs and the swings in voter turnout, even assuming that Republicans couldn’t count high enough to get to the center of a Tootsie Roll Tootsie Pop, and given that the Democrat Party has always included hundreds of statistics professors from Harvard, Yale, Stanford, and Princeton, why haven’t they been using this technique for the past fifty years?

                Probably because Democrat mathematics professors aren’t usually dumb enough to believe it would work.Report

              • Avatar Kolohe in reply to Shazbot3 says:

                “why haven’t they been using this technique for the past fifty years?”

                Because Silver’s Monte Carlo simulations are computational intensive, and if they had run on computers back in the day, those same computers would still be running the LBJ Goldwater race?Report

              • Avatar Patrick Cahalan in reply to Shazbot3 says:

                Yeah, I smell positivism on you, George, and a comfort with deterministic algorithms, which is totally normal given that you have an engineering background.

                What Nate is doing isn’t that. Now, you can discount Monte Carlo simulation methodology as being inferior to deterministic algorithms, but that misses the point. You can’t have a deterministic algorithm in this problem space.Report

              • Avatar Stillwater in reply to Shazbot3 says:

                Yeah, I smell positivism partisanism on you, George

                Fix’t.Report

              • Avatar Shazbot3 in reply to Shazbot3 says:

                George is jumping the shark in 5, 4, 3, 2, 1,…..

                “and given that the Democrat Party has always included hundreds of statistics professors from Harvard, Yale, Stanford, and Princeton, why haven’t they been using this technique for the past fifty years?

                Probably because. Democrat mathematics professors aren’t usually dumb enough to believe it would work.”

                Or those partisan Democrat(ic) mathematicians didn’t care to bother making a prediction because they wouldn’t get anything out of it. (Making the prediction doesn’t help the campaign win very much, BTW. And the campaigns themselves likely believed, as you seem to, erroneously that the outcome was entirely unpredictable with any degree of reliability.) And maybe mathematicians had other things to do. Or they did make the prediction, at home for fun, but didn’t bother publishing it. (The internet has resulted in lots of ideas getting aired publicly that used to be known in narrow circles.) Or maybe just no math savvy person thought of bothering. (People sometimes don’t do things that are useful and easy for decades and then realize how stupid they had been for not doing it earlier. Etc. Etc.

                I will now treat all of your future postings as ridiculous. I encourage others to mock you as well.Report

              • Avatar Shazbot3 in reply to Shazbot3 says:

                Indeed, before the internet existed and before it became widely used, I suppose polling data was published, but it wasn’t exactly a click away. You’d have to work to collect all the data, especially crosstabs and state polls and sample sizes and house effects and LV models, and demographics, and econometric data. Moreover, complex calculations took more time and more resources the further you go back in time. Nate can run this stuff on a laptop pretty quickly, but it wasn’t always so easy.

                You could have had a Nate Silver decades ago, I guess, but he wouldn’t have sprung up organically on his own the way the actual Nate Silver did. Some university, newspaper, or campaign would have to hire a mathematician full-time and give him assistants to collect data and help make computations. Maybe some campaign -D or R, mathematicians would take the cash to do the work- should’ve spent the cash to make a Nate Silver, but the fact that they didn’t see it that way doesn’t mean jack, except that campaigns have always believed in local political gurus with no evidence to back up their beliefs over data-driven conclusions.Report

              • Avatar James Hanley in reply to Shazbot3 says:

                why haven’t they been using this technique for the past fifty years?

                Why didn’t George Washington just drop a nuke on the British?Report

              • Avatar George Turner in reply to Shazbot3 says:

                Because Silver’s Monte Carlo simulations are computational intensive, and if they had run on computers back in the day, those same computers would still be running the LBJ Goldwater race?

                Wrong answer. We used them in the 1940’s to design bombs. Any of Nate Silver’s techniques could be run on an 4.77 MHz IBM PC/XT just fine. You could run them on an ENIAC, but you couldn’t get the card’s punched fast enough to get the program done.

                A few minutes ago Obama and Romney both appeared for a half-time interview on a sportscast I was watching. Needless to say, it was like watching the head of the Olympics debate Britney Spears about sports. It was so bad, and will be so fresh in the minds of sports fans, that it will probably generate a 60 point shift in the election, putting Romeny at +110 and Obama and -10, vacating Obama’s 2008 victory over McCain and giving us a do-over of the last four years, depending on the ref’s decision. If that happens, we get four years of McCain and then four years of Romney.

                The statistical models don’t include breathtakingly stupid decisions taken on election-night eve, and all should really include the caveat “Assuming Obama doesn’t totally embarrass himself on national TV at halftime the night before the election, in front of every sports fan in the nation.”Report

              • Avatar Pat Cahalan in reply to Shazbot3 says:

                We used them in the 1940?s to design bombs. Any of Nate Silver’s techniques could be run on an 4.77 MHz IBM PC/XT just fine. You could run them on an ENIAC, but you couldn’t get the card’s punched fast enough to get the program done.

                Er, no.

                Well, maybe.

                As far as I know, though, nobody ever wrote a credible pRNG for the XT, which was an 8 or 16 bit processor. The XT was 16, IIRC. Either way, you can’t back port Linux or FreeBSD to run on that, so you’re writing your OS from scratch or you’re trying to write a credible pRNG in BASIC. Good luck with that one.

                I guess you could buy a copy of Rand’s million random digits book (I understand they still sell it) and use that, but that’s a manual input method for 100,000 runs of the simulation. I think you’d get RSI.

                You neglected to mention that it took an awful lot of people to run those simulations on the 40s to make the bomb.

                At any rate, I’ve just read 10 criticisms of Nate’s work at FiveThirtyEight on various conservative blogs, and it’s clear from those 10 that those 10 authors didn’t even read his methodology page, because they claim outright that he doesn’t do something that he clearly includes in his model (http://fivethirtyeight.blogs.nytimes.com/methodology/).

                I’m not saying the guy is the end-all beat-all or anything, but most of the criticism of his work seems to clearly not understand it.

                FWIW, Tom’s link up above to the Bob Krumm site has the most credible counter; Silver’s model is more or less correct, but…

                It’s just that he focused on the wrong parts: instead of looking at state races, he should have watched the underlying issues and demographics that were pulling Obama down.

                That’s at least a debatable argument. The thing is, underlying issues and demographics do matter, but they matter for election trends over decades (IMO).

                In any event, we’ve got tomorrow.Report

              • Avatar Stillwater in reply to Shazbot3 says:

                nobody ever wrote a credible pRNG for the XT, which was an 8 or 16 bit processor. The XT was 16, IIRC. Either way, you can’t back port Linux or FreeBSD to run on that, so you’re writing your OS from scratch or you’re trying to write a credible pRNG in BASIC.

                Patrick, I’m glad you wrote this comment so I didn’t have to.Report

              • Avatar George Turner in reply to Shazbot3 says:

                Pat, Monte Carlo simulations don’t need a good random number generator, at least not by any standard we’d use to describe one. The random number generators for those can have all kinds of herringbone patterns, repeats, bit flips, and other things that would get any encryption algorithm ripped wide open and left with its guts hanging out.

                For a Monte Carlo, you can use the simplest random number algorithms (as in VIC-20 BASIC simple) and get results that you won’t improve on, because the only requirement for a Monte Carlo’s random number is that it covers the space from 0.0 to 1.0 evenly, and doesn’t have patterns that might show up across sequential calls (like high-low-high-low) where your code might be using the numbers for alternating X and Y inputs.

                Most applications that use a Monte Carlo method could also just iterate each input and get the same effect, but as complexity builds it’s often faster to randomly sample the space instead of exploring it completely, like when you need a neutron path and don’t really care about all the possible paths, you just need a couple thousand random looking neutrons to see how many get lost and where the fissions happen.

                GOOD random number generators are generally only important for encryption, where any underlying pattern in any bit, across millions of messages, can create a weakness that can be exploited.

                And when your raw data has error bars like 4%, any random number generator that doesn’t get stuck is appropriate, even if it’s a 20-sided dice.Report

              • Avatar George Turner in reply to Shazbot3 says:

                Oh, and I wrote my own Forth interpretter in assembly language on an 8080 (which got me my first university programming job). Random numbers aren’t a problem, especially when you’ve got a copy of Knuth handy.

                Randomly shifting state voting percentagess around a centroid of state polling data and counting how often a candidate wins wouldn’t tax a TRS-80. Heck, you should’ve seen what the horse industry was doing back then with the jockey club data, much less the video capture and gate analysis.

                That’s why I’m saying that if the method was inherently robust, any university person, even an undergrad, could’ve been using it at least since the early 1980’s. Since everyone knew about such methods, the most likely reason they didn’t use them is that either they’re pretty obviously not robust, or even more likely, if you try them on any electoral data sets from that era they probably fail miserably, leading to their immediate rejection. This raises an interesting question. If Nate Silver only goes back X number of elections, and if nobody in the early PC era used these methods (back when the programmers were trying out their new found mathematical powers on everything from horse racing to stock picks), perhaps that eras elections do create staggering failures with the method, which is why Nate Silver doesn’t try to use them.

                If so, that might bear a similarity to the modeling failure that led to the collapse of the housing market, where the innovative (and stupid) method they were using looked great on all the input data they had, but their short sample of input data had never included a downturn in the housing market because the last downturn was too far in the past.Report

              • Avatar Patrick Cahalan in reply to Shazbot3 says:

                For a Monte Carlo, you can use the simplest random number algorithms (as in VIC-20 BASIC simple) and get results that you won’t improve on, because the only requirement for a Monte Carlo’s random number is that it covers the space from 0.0 to 1.0 evenly, and doesn’t have patterns that might show up across sequential calls (like high-low-high-low) where your code might be using the numbers for alternating X and Y inputs.

                I will admit that I’m not an MC expert by any means and you may have a point, here. But given the frequency with which Silver puts out his simulation results, and today’s modern computing power, and Moore, if I go backwards I get a computationally unfeasible amount of time to reproduce his model on a XT. Sure, there’s a lot of assumptions baked in there, but the difference is so much greater than what you’re saying here that I suspect that we’re both largely talking out of our ass.

                That’s why I’m saying that if the method was inherently robust, any university person, even an undergrad, could’ve been using it at least since the early 1980?s.

                This doesn’t follow, George. Sabermetrics weren’t popular in baseball until recently, either.

                Just because something works doesn’t mean people use it. Institutional inertia must be overcome, and “We’ve always done it this way” is a fucking huge boulder to move, not to mention additional factors.

                Like, right now, in major data centers you could save an enormous amount of money by eliminating a/c power supplies at the individual node, and you could reduce your heat load by… what, 15% or so?

                But there are very, very few data centers that run d/c power supplies on their nodes, because the supply chain for nodes is rigged as a distribution method for a/c power supply machines. The fact that d/c machines would be better, and cheaper, is not relevant, because you’d have to find a unique vendor or build them yourself, and the first one comes with a huge supply chain risk and the second just simply isn’t feasible for most places.

                I mean, come on, if people only did stuff that worked, there would be no political punditry industry. Those guys have horrible track records.Report

              • Avatar Kim in reply to Shazbot3 says:

                There’s at least one datacenter in NYC that’s running on tequila right now…Report

              • Avatar Patrick Cahalan in reply to Shazbot3 says:

                Kim, I have to ask…

                … are you medicated?Report

              • Avatar Morat20 in reply to Shazbot3 says:

                I’m still working on the statement that you don’t need a good pRNG to do Monte Carlo simulations.

                You, in fact, do — even back in the 90s people doing real work were having to out of their way to get decent random numbers (well, psuedo random) — it’s not like rand() was doing it, you know.

                Even the better algorithms for it weren’t all that great, most people used noise sources to at least salt their data, if not generate the numbers. (A friend of mine, in the 90s, cobbled one together using a supermarket scanner and a lava lamp)

                I think a lot of people just don’t realize how much both computers and information availability has changed in the last two decades.Report

      • Avatar James Hanley in reply to Tom Van Dyke says:

        Holy cow. He thinks the odds of a single poll being right are the same as the odds of a poll aggregator being right.

        Basic mathematical literacy came knocking, but nobody was home.Report

        • Avatar Chris in reply to James Hanley says:

          Mathematical models that calculate probabilities based on current data and past performance are exactly the same as a single poll. Everyone who’s ever taken a statistics course knows this!Report

          • Avatar Morat20 in reply to Chris says:

            One time, I had a bag full of marbles — 20 blue, 10 red. Since there are two colors, the odds of me pulling out a blue marble is 50/50.Report

            • Avatar James Hanley in reply to Morat20 says:

              Seriously, nothing about this election is irritating me as badly as these asshats who’ve apparently never taken a single stats class in their whole lives thinking they’ve got some insight into it that the math geeks who dream in matrix algebra don’t.Report

              • Avatar MikeSchilling in reply to James Hanley says:

                The Stupid Party isn’t just an insult.Report

              • Avatar Morat20 in reply to MikeSchilling says:

                To be fair, Democrats are just as mathematically illiterate and didn’t cover themselves in glory in 2004.

                They just don’t really have a reason to doubt the polls.

                OTOH, some blogger or pundit joked that when Democrats see a losing poll, they get depressed and start blaming each other. Republicans blame the pollster. 🙂Report

              • Avatar Kim in reply to Morat20 says:

                These are still close polls. Democrats tend to get hoppin’ mad and start marching their friends off to the polls when this happens.

                It’s when they’re really far behind that they vote libertarian.Report

              • Avatar Stillwater in reply to James Hanley says:

                Seriously, nothing about this election is irritating me as badly as these asshats who’ve apparently never taken a single stats class in their whole lives thinking they’ve got some insight into it that the math geeks who dream in matrix algebra don’t.

                Heh. Just saw this! Awesome.Report

        • Avatar trizzlor in reply to James Hanley says:

          Question for the stats people here: Silver is aggregating over many polls/samples taken over partially overlapping/correlated data. Each sample also has a different amount of variance/sampling error which, given the data is correlated, are also expected to be correlated. Each sample also has a bunch of biases (house effects) that silver attempts to correct for with a single weight. From what I can see, Silver gets an estimate of the overall expectation by running a series of permutations with random starts.

          So … what’s the variance on his aggregate expectation and how robustly can it be calculated? In the best case (all polls completely independent) I imagine it’s just the inverse-variance weighted sum of the polls, what about the worst case (all polls identical copies of each other)?Report

          • Avatar Chris in reply to trizzlor says:

            His methods are a bit more complex than this. He uses multiple weights for each poll (recency, sample-size, prior accuracy, house effects, and some others), plus a poll produced by looking at conditions on the ground (he uses a regression that gives vote share and then counts as another poll with its own weights), plus a further regression to weight the standard error based on the actual error of poll’s in the past (I think it weights the standard error of the mean of the polls, but he does make it sound like it actually produces that standard error, so I’m not sure what he means by that), and a further refining of the standard error by a sum of squares calculation that combines the errors at different grains (local and national), and then a simulation on the resulting distributions for each of the states, with some extra shit thrown in there to allow for the noise that the time between the election date and the simulations produces. His calculation of the error associated with the predictions from the simulation, based on previous results, is +/- 1 point. Because it’s a simulation based on the distributions around the weighted poll averages, I’m not sure there’s a theoretical way of producing a margin of error for the predictions. It’s just a matter of looking at how accurate they are over multiple tests. Tomorrow will be another test for his presidential general election model.

            By the way, I believe he has Romney’s chances at around 14% tomorrow.Report

            • Avatar Burt Likko in reply to Chris says:

              Those odds have been… revised. Go see for yourself.Report

              • Avatar Stillwater in reply to Burt Likko says:

                8%.

                Whoa. That’s bleak.Report

              • Avatar Jaybird in reply to Stillwater says:

                No upside to that. If Obama wins, 85% is just as good as 92%. If Romney wins, it becomes “AND AT THE LAST MINUTE, HE EVEN SAID THAT ROMNEY HAD ONLY AN 8% CHANCE!!!”Report

              • Avatar Stillwater in reply to Jaybird says:

                No upside … on the assumption that Nate’s a pre-neo-post-partisan hack pretending to be neo-post-partisan.

                Which he might be.Report

              • Avatar Michael Drew in reply to Jaybird says:

                No upside just increases his credibility (maybe that’s what you’re saying). Notice, I say: *his* credibility. The model won’t look so good if it was lowering the winner’s odds as closing time approached ever closer. But Silver, who surely understands that, looks better for having stuck with his formula. He’s a guy with an algorithm. That’s it.

                Also, you heard it here first: next time, they’ll be saying that his admission that he changed his model from its 2012 settingto account to some degree for what it got wrong that year just PROVES he’s a hack and a fraud and incompetent and gay.Report

            • Avatar trizzlor in reply to Chris says:

              This is really really helpful Chris, and breaks down the model a lot.

              In terms of data-mining, I think Silver is doing the right thing, but still, the whole issue of correlated error really bothers me and, I feel, keeps the “unskewed” concern trolling somewhat valid. Given that his previously results are, at best, a decade of elections (Rassmussen started in 2000 I believe) the se of 1% is hard to swallow (especially since, in many elections, a single state off by a percent swings the whole thing). A 90% Obama win with a true sd of 25%, for example, would basically mean nothing; and it would mean less than nothing if the whole distribution isn’t even normal but shifted towards Romney. I get that it’s a NYTimes blog, but I’m still very dubious when the results are not shown to be normal and have no clearly derived standard errors.Report

              • Avatar bookdragon in reply to trizzlor says:

                Go look at Sam Wang’s predictions then. Another stats guy who runs and give results based on several prediction methods.

                http://election.princeton.edu/electoral-college-map/

                Given that his models have generally been slightly more favorable to Obama, I don’t understand why everyone has been jumping on Silver and ignoring Wang. Maybe because someone from Princeton who specializes in stats is harder to dismiss?

                In any case, these are all discussing probabilities, not hard predictions. I’m not sure why I can’t reply to posts beyond a certain point at this website, but if I could have replied to Mr. Turner’s last one to me, I would have pointed out that comparing their stats to what engineers use for design is apples and oranges.

                After all, even the current high-probability prediction on 538 is still almost a 1-in-10 chance that Obama doesn’t win. No engineer would ever sign off on a design with a 1-in-10 chance of failure. Most gamblers, however, would be pretty happy with a 9-in-10 chance of winning.Report

              • Avatar Patrick Cahalan in reply to bookdragon says:

                The nesting closes after a bit because threaded reading will wind up with

                c
                o
                m
                m
                e
                n
                t
                s

                l
                i
                k
                e

                t
                h
                i
                s.

                It’s a drawback of threading the comments. It doesn’t come up often, because usually people don’t go back-and-forth that much.Report

              • Avatar Kim in reply to bookdragon says:

                Book,
                My numbers (over the weekend) were Obama at 98.5% chance of being reelected. Maybe that’s not good enough for you engineers, but you know us physicists? We crazy, man.

                And engineers could and did sign off on designs with 1 in 10 chances of failing catastrophically during the BP spill.Report

  18. Avatar North says:

    I’m predicting Obama wins 294. CO goes to Romney with libertarian pot legalizers representing the margin. Obama and Dems sit up and take notice (you idiots).

    Romney gets FL, NC and CO, Obama gets the rest. There’s kvetching about vote counting but the win is decisive enough that there isn’t too much controversy.
    http://elections.huffingtonpost.com/2012/custom-presidential-election-map#nkamdnjannaajenae

    The popular vote margin is super close so Obama is said to have no mandate. Hopefully he ignores this, he doesn’t need a mandate. He doesn’t have to do anything. If the GOP try to double done on their “No on everything” position from the last four years then we go off the “fiscal cliff” and the GOP’s neocon/defense contractor and Tax warrior wings tear into each other like a pack of feral hyenas when the tax rates reset and the defense sequestrations land (we’d likely get a recession too though, this kind of austerity would be pretty hard on the economy but the bond markets would love it).

    In North’s that’d be nice imagination Obama and his Senate buddies craft a tax cut bill for middle to lower income groups that eases their pain and dares the GOP to block it, the GOP folds and taxes are moderated somewhat, this prevents full on recession.

    In North’s ideal lala-land Obama and the GOP cut a deal. Tax rates go up a little with some exemptions phased out, spending is cut in the long term with short term boosts in safety net spending, hurricane relief and other stuff. Some sort of compromise is made. The market surges and then the GOP implodes as the Tea Partiers turn on their treacherous leadership.Report

    • Avatar Shazbot3 in reply to North says:

      “Obama is said to have no mandate.”

      That would be said to be true even if he won by a similar (or slightly greater) margin to 2008.

      I think to “be said to have a mandate” Obama would need to win by 10 or more.

      I mean, he ran on healthcare reform as his main policy objective, won massively, and tried to pass a more centrist version of what he ran on, and then he was told he didn’t have a mandate.

      Mandates are meaningless in the new Republic.Report

      • Avatar Ryan Noonan in reply to Shazbot3 says:

        According to Politico, you can only have a mandate if you win the white vote. Which is why every Republican president has one, even if he loses the popular vote overall.Report

        • Avatar MikeSchilling in reply to Ryan Noonan says:

          The white male vote, because everyone else is a special interest.Report

          • Avatar zic in reply to MikeSchilling says:

            It’s all Norm’s fault, isn’t it?Report

          • Avatar Shazbot3 in reply to MikeSchilling says:

            I’m part of the general interest, dagnabit!

            You and your special interests, Mike. “Oh, I’m so special. I need money for kidney research because I have diseased kidneys. I’m special.” “Oh, no I’m special because I’m a teacher and I want to be middle-class in pay plus benefits. I’m special because I’m trying to stop polluters from poring mercury in lakes and rivers, too much CO2 in the air, and AIDS infected lead paint on sharp needles into children’s eyes. So special.”

            Not me. Not a darned thing that is special about me. No sirree.Report

        • Avatar Morat20 in reply to Ryan Noonan says:

          Ah, actually not the white vote. You have to win the middle-aged white male vote.

          18-29 year old white votes don’t count. Female white votes don’t count.

          Only middle aged white male’s count. If you don’t get them, you don’t have a mandate.

          Actually, let’s be honest. Only Republicans get mandates.Report

    • Avatar KatherineMW in reply to North says:

      If either side loses Colorado with Johnson voters providing the margin of error, hopefully they will pay attention. I wouldn’t put money on it, though.Report

    • Avatar Scott Fields in reply to North says:

      My map looks just like yours does, North.

      I was tempted to put Indiana down for Obama, since I didn’t want to be completely unoriginal and that was the one state Nate Silver got wrong in 2008. Perhaps an anti-Mourdock wave would turn the tide. But, in the end, I figured that was just too much to ask of my home state.Report

  19. Avatar Chris says:

    Here’s my prediction: Tyson will knock out Buster Douglas in the 4th.Report

  20. Avatar zic says:

    My only prediction is that we will once again prove our exceptional ability to fuck up voting.

    American exceptionalism. Gotta love it.Report

  21. Avatar KenB says:

    No prediction, but I do have a question for anyone who’s been following the polling closely. I just saw a CNN poll that showed the national numbers as very tight, but I noticed that the question that they asked was “Suppose that the presidential election were being held today and you had to choose between Barack Obama and Joe Biden as the Democratic Party’s candidates, and Mitt Romney and Paul Ryan as the Republican Party’s candidates. Who would you be more likely to vote for?” Obviously that’s not the same choice that the voter will be faced with in the voting booth — seems like that would introduce inaccuracies for people who have some chance of voting 3rd party. Are any or many other polls set up this way as well?Report

  22. Avatar Randy Harris says:

    Electoral Vote: Obama 303, Romney 235
    Popular Vote: Obama 49.98%, Romney 48.86%, Others 1.16%Report

  23. Avatar Patrick Cahalan says:

    Prediction for tomorrow, Obama wins with 290+ electoral votes and 52+ % of the popular vote. Mourdock loses in Indiana. King wins in Maine. Warren wins in Massachusetts. McCaskill loses to Aiken, guaranteeing late night talk show jokes about Aiken for another 6 years. Tester holds on in Montana due to low GOP voter turnout (nobody likes Mitt in Montana, as near as I can tell). Berkley wins i
    n Nevada, but it’s so close it goes to recount and we get to hear about that for at least a month. No call on Virginia: George has name recognition, but being attached to the GOP in 2000-2006 is nearly as bad of a mark as being the DNC chairman under Obama. I’m not dumb enough to try to predict anything in Wisconsin this year; that’s the USC-UCLA game, right there.Report

  24. Avatar bookdragon says:

    I’ll go with Obama 303, Romney 235 – Silver and RCP agreeing makes it seems most likely.

    Of course, some of this is wishful thinking – my folks still live in Ohio and it’s election is going to be a complete mess with all sort of legal issues (dear Mr. Husted, Start packing. If you turn OH2012 into FL2000, expect even republicans in the state to want to string you up). Therefore, I hope Obama has enough margin with EVs for him to win even if Ohio’s are questionable.

    Other prediction: Romney will win the popular vote because NY and NJ voter turnout will be much lower than usual due to Sandy.Report

  25. Avatar Kazzy says:

    Normally, I’d bet on the white guy winning.

    That didn’t work last time. So I’ll assume it won’t work this time.

    I think Obama holds Ohio, which will prove the decisive state, and takes the EC and PV.Report

  26. Avatar KatherineMW says:

    Because this is so endlessly relevant to our discussions of polls, I’m leaving it here:

    http://www.xkcd.com/1130/Report

  27. Prediction for tomorrow:
    I decide to stay in the States for another 4 years (or so).Report

  28. Avatar MFarmer says:

    As always I’m voting for Professor Irwin Corey, and if he doesn’t win it will be due to a last minute announcement from Pynchon declaring their decades long love affair or last minute concerns with his height. In either case, it will be last minute associated.Report

Leave a Reply

Your email address will not be published. Required fields are marked *