The Keys to the White House
September 22, 2020 1:45 PM   Subscribe

Professor Alan Lichtman has correctly predicted every* American election since 1984, says that polls are a snapshot in time, but are not predictive of Election Day results. Voters are rational, practical decision makers who are not really swayed by rallies, campaign tactics, platforms, or promises. Rather, American Presidential elections are a referendum on the incumbent party and their performance governing over the previous 4 years. His methodology is inspired from earthquake prediction; either there will be stability (incumbent party retains the White House) or there is an “earthquake” (the challenging party takes the White House.) There are 13 true/false “keys”, and if six or more of them are “false”, the model predicts an electoral “earthquake”.

The thirteen true/false keys are:
1. Party Mandate: After the midterm elections, the incumbent party holds more seats in the U.S. House of Representatives than after the previous midterm elections. (FALSE)
2. Contest: There is no serious contest for the incumbent party nomination. (TRUE)
3. Incumbency: The incumbent party candidate is the sitting president. (TRUE)
4. Third party: There is no significant third party or independent campaign. (TRUE)
5. Short-term economy: The economy is not in recession during the election campaign. ( FALSE)
6. Long-term economy: Real per capita economic growth during the term equals or exceeds mean growth during the previous two terms. (FALSE)
7. Policy change: The incumbent administration affects major changes in national policy.(TRUE)
8. Social unrest: There is no sustained social unrest during the term. (FALSE)
9. Scandal: The incumbent administration is untainted by major scandal. (FALSE, deserves EXTRA POINTS)
10. Foreign/military failure: The incumbent administration suffers no major failure in foreign or military affairs. (TRUE)
11. Foreign/military success: The incumbent administration achieves a major success in foreign or military affairs. (FALSE)
12. Incumbent (party) charisma: The incumbent party candidate is charismatic or a national hero. (FALSE)
13. Challenger (party) charisma: The challenging party candidate is not charismatic or a national hero. (TRUE)

7 FALSE keys predicts the 2020 election, as a referendum on the incumbent party, will result in a win for Biden, regardless of all the polling and campaigning between now and election day.

Two things that Professor Lichtman are worried about which the model does not account for: voter suppression and interference from Russia.


Prof. Lichtman’s model predicted Gore’s popular vote win in 2000. At the time, the winner of the Electoral College not being the winner of the popular vote had not happened for over 100 years. Since 2000, Prof Lichtman has predicted the overall election winner.
posted by Pirate-Bartender-Zombie-Monkey (61 comments total) 21 users marked this as a favorite
 
The incumbent party candidate is charismatic or a national hero. (FALSE)

Unfortunately there are in fact a lot of people for whom this seems to be true.
posted by obfuscation at 1:52 PM on September 22 [96 favorites]


I feel like 12 is very much up for debate depending on who you talk to. Trump's a great man and a compelling speaker, and a bold crusader against the child-molesting Deep State, if you talk to his supporters.
posted by egypturnash at 1:53 PM on September 22


12 is "TRUE" to many. Too close to call. (edit on preview: ok, "ditto")

7 may be a positive or negative for the campaigns as well. I wonder why that's predictive as a preference for the incumbent. Maybe bias for action is enough to give the benefit of the doubt...
posted by I'mSoBlue at 1:55 PM on September 22


I, too, have successfully predicted every* election result.

*except for the ones I haven’t
posted by robcorr at 1:58 PM on September 22 [17 favorites]


This whole thing seems like a cute relic from back in the days when politics were fairly "normal".
posted by Liquidwolf at 2:03 PM on September 22 [44 favorites]


This guy does seem to have distilled key factors in electoral wins, but he's been wrong twice and the subjectivity of his model (charisma?) makes it impossible to disprove
posted by latkes at 2:03 PM on September 22 [24 favorites]


It seems like 9 is also, sadly, up for debate in certain quarters. :(
posted by eviemath at 2:07 PM on September 22 [2 favorites]


A couple of these, more than just #12 depend on who you support already.

"9 - Scandal: The incumbent administration is untainted by major scandal" - Lots of trump supporters consider him untainted by scandal. There have been many scandals, all of which trump supporters consider fake.

"11. Foreign/military success: The incumbent administration achieves a major success in foreign or military affairs" - According to trump supporters, he de-nuclearized North Korea and single-handedly saved the U.S. economy from China. Surely that counts as a success. Also, to them, the concentration camps on the southern border count as a success in foreign affairs.

In short, I don't think this is as predictive as claimed.
posted by mrgoat at 2:08 PM on September 22 [21 favorites]


I would like to share wisdom from Virginia Lee Burton’s book “Maybelle the Cable Car:” “No more speeches...no more talking...just one vote from each and everyone and no one could tell what the answer would be until the polls were closed and all the votes counted.”

I realize this is a simplistic take (it is a children’s book), but I find myself thinking about it often as a resist the urge to get sucked back into pundits, professors, and politicians offering me predictions.
posted by CMcG at 2:13 PM on September 22 [3 favorites]


So, 2000 doesn't count because Gore won the popular vote, but 2016 does count even though Clinton won the popular vote?
posted by ckape at 2:29 PM on September 22 [39 favorites]


For the people this model is supposed to model, none of these facts matter, because no facts matter anymore, only loyalty. Trump is charismatic, the economy is doing great, there have been no scandals, and social unrest is the fault of The Enemy.

This isn't a regular democratic election, it is -- at best -- a referendum on Fascism in America. Attempting to model anything other than that, especially based on past democratic elections, is foolhardy.
posted by Imperfect at 2:31 PM on September 22 [29 favorites]


Given that about 40% of America is in a different reality tunnel, where Trump has achieved peace in the Middle East, the scandals are “fake news” spread by the liberal elites, and the biggest threat to America is anarchist rule in coastal cities, this model's predictive value looks questionable at best.
posted by acb at 2:35 PM on September 22 [6 favorites]


Voters are rational, practical decision makers who are not really swayed by rallies, campaign tactics, platforms, or promises.

[Spit take] Has he met these people??
posted by gimonca at 2:36 PM on September 22 [35 favorites]


These all look like really good inputs to a successful predictive model but I think they probably need to be weighted rather than indiscriminately summed. If there was an empirical method to convert economic data/primary results/opinion polling into floating point values for the majority of these inputs (which is a roundabout way of echoing/expanding the complaints re: #12 above), I'd bet good money you could make a highly successful machine learning algorithm for predicting national election results with accuracy matching or exceeding 538-style statistical methods.

On preview: part of the reason you'd want to get floats for primary results is precisely because of the "two different realities" objections being raised.
posted by Ryvar at 2:37 PM on September 22 [2 favorites]


I'm old enough to remember when we'd sit around the campfire, stick a flashlight under our chin, and tell the story of the Curse of Tippecanoe.
posted by gimonca at 2:41 PM on September 22 [5 favorites]


Ugh.

I can predict where lightning will strike. Really! It's a fairly simple rule: If lightning has struck in the area in the last 15 minutes, I predict there will be another lightning strike. You may be less impressed with my performance when you realize this is another way of saying "is there currently a thunderstorm?" But it's a good analogy to think about about when evaluating models. Not, how impressive does the headline sound, but how does is it compare to a simple heuristic?

There've been 9 elections since '84. He's gotten seven correct, with two (2000 and 2016) outliers. If you were placing the boring, safe bet each time, which of the seven would you have gotten wrong? Any of them? (Maybe Dukakis and Kerry, depending on when you made the prediction. No one else, really.)

And note that since these aren't objective criteria ("charisma", "scandal", "success", etc.) you are necessarily using your gut to tweak the prediction, based on how people respond to the candidates and their platforms in irrational ways.

Pundits and news coverage always try to make things seem like a horserace and reward contrarianism, so compared to them this is a good run. But they really suck at the prediction stuff! They don't want to be boring or safe!

I actually *do* think as a decision making exercise, listing simple & obvious things out and assessing them is OK strategy. It'll help you catch yourself if you've fallen in love with a pet theory/idea/candidate. And you'll do better than random if you do it. But it's not super accurate science; it's the equivalent of my lightning prediction model or investing in an index fund.

tl;dr; "My simple checklist accurately predicted the release and impact of the Comey letter and its impact in swing states" is kind of a stretch.
posted by mark k at 2:41 PM on September 22 [13 favorites]


I happen to think that the peace agreements between Israel and UAE and Bahrain aren't huge deals (more of a formalizing of an existing, tacit, relationship), but it would be very easy to argue otherwise. They are unquestionably a success, but whether they are a major success is rather open to debate, which leaves 11 up in the air.

12 is also debatable. Trump's charisma clearly doesn't work on everyone, but he's definitely got something.

One thing that makes me skeptical of this whole methodology, is that he says he's predicting the winner of the election, not the popular vote winner. If Florida's panhandle were part of Alabama rather than Florida, Hillary Clinton would be President. We all know that the state borders play a critical role in all of this and his model can't take that into account. If we assume anything, we must assume that it's picking the winner of the popular vote because anything else is nonsense.
posted by It's Never Lurgi at 2:47 PM on September 22 [5 favorites]


Shade from 2020 Nate Silver. Apparently, Lichtman no longer predicts the popular vote? I wonder why that could be.

More detailed criticism from 2011 Nate Silver.
posted by Huffy Puffy at 2:58 PM on September 22 [6 favorites]


Professor Alan Lichtman is full of crap, and can safely be ignored. He either got 2000 right and 2016 wrong, or 2016 right and 2000 wrong. But he wants you to accept that he got both right.
posted by Frayed Knot at 3:00 PM on September 22 [22 favorites]


It'd also be more honest, given the granularity of the model, to frame it as getting 2000 right and 2016 wrong. I mean there's nothing specific to electoral vote weighting in there! It's clearly a nationwide model, so popular vote is the best it can do.

But he wants credit for predicting Trump.
posted by mark k at 3:10 PM on September 22 [1 favorite]


Whatever his model says, go vote.

And rest a little easier at night since RCP (which is pretty decidedly right of center) has Georgia, Texas, North Carolina, and Florida as inside the margin of error battlegrounds. FUCKING. TEXAS.

The result isn’t guaranteed even though the numbers look good. The numbers looking good certainly feels good. That they’ve mobilized voters with amateur hour EO shit like the “anarchist jurisdictions” designation also helps. Angry people vote.

Make sure you vote.

Please make sure you vote and that everyone you know votes.
posted by Slackermagee at 3:32 PM on September 22 [15 favorites]


Voters are rational, practical decision makers who are not really swayed by rallies, campaign tactics, platforms, or promises. Rather, American Presidential elections are a referendum on the incumbent party and their performance governing over the previous 4 years.

These two statements are contradictory. Failing to gauge the relative merits of the opposing candidate, and whether he is (for example) a notorious infomercial con-artist, is neither rational nor practical.
posted by East Manitoba Regional Junior Kabaddi Champion '94 at 3:40 PM on September 22 [2 favorites]


10. Foreign/military failure: The incumbent administration suffers no major failure in foreign or military affairs. (TRUE)

This is another "depends on your perspective (or propaganda intake)" - being in Putin's pocket and failing to actually accomplish anything with North Korea and having the leaders of the world mock him pre-pandemic and then the rest of the world look on in shock and horror and pity as we hit over 200,000 deaths from COVID all seem pretty big failures to me. Also Mexico still hasn't paid for that damn wall.

13. Challenger (party) charisma: The challenging party candidate is not charismatic or a national hero. (TRUE)

I know Biden is hardly Metafilter's favorite, but saying he's got no charisma is just . . . what rock are you hiding under? Were you asleep during the Obama years where he was everyone's Cool Tell-It-Like-it-Is Uncle Joe and thus often Obama's attack dog? Hell, if anything, AFAICT most objections to him are, "He's got plenty of charisma, like all politicians, he just doesn't use it to create and push good legislation."

Paraphrasing somebody somewhere on Twitter about one of his recent trips; "He turned the folksy charm up to 11 and broke the knob off."
posted by soundguy99 at 3:44 PM on September 22 [3 favorites]


Lumping that factor in with "national hero" means you're looking for a level of charisma beyond what I'd think even Biden's long-time fans ascribe to him. I see it as more of a 2008 Obama/1960 Kennedy thing.
posted by Holy Zarquon's Singing Fish at 3:45 PM on September 22 [1 favorite]


There have been 9 presidential elections since 1984. A pundit "predicting" each election by flipping a coin to decide between the Democrat or the Republican candidate would have a 1 in 512 chance of a perfect prediction rate. If there are thousands of pundits predicting presidential elections (as there are) and their methods are independent and no better or worse than random guessing, we should expect there to be many pundits with perfect records, but in fact they have no special insight and their future predictions are no better than anyone else's.

There's a scam based on a similar principle. Send thousands of people free investing advice, telling half of them to buy and half to sell a certain stock. Whichever one turns out to have been the wrong advice, remove those people from your pool, then repeat the process on those who remain. After several iterations you have a small group of people for whom you've given perfectly accurate investing advice a remarkable number of times, who, not knowing that you've been wrong an equal number of times with other people, might think you're an investing genius. Then offer to continue giving stock tips for a fee.

Political punditry is often a version of this scam, except instead of a single person scamming a few people into thinking they have special insight, it's a whole system of pundits and news-consumers scamming itself into thinking that some individuals have special insight. (Of course in the modern media, being consistently wrong is no barrier to people believing you have special insight, either.) This isn't to say that expertise and insight don't exist, but the simple fact of a good track record doesn't provide any evidence for it.

Also we can't predict earthquakes.
posted by biogeo at 3:45 PM on September 22 [44 favorites]


Also, I visit enough news & politics websites that I'm sure various algorithms have noticed so on the occasions when there are unavoidable clickbait ads there's a good chance at least one of them is "THIS PERSON HAS CORRECTLY PREDICTED ALL OF AMERICA'S ELECTIONS!!! CLICK TO SEE WHO WILL WIN THIS TIME!!"

I assume it's this guy.
posted by soundguy99 at 3:54 PM on September 22


Lumping that factor in with "national hero" means you're looking for a level of charisma beyond what I'd think even Biden's long-time fans ascribe to him. I see it as more of a 2008 Obama/1960 Kennedy thing.

Fun fact, Obama counted as charismatic in 2008 but not in 2012!
posted by Huffy Puffy at 3:56 PM on September 22 [6 favorites]


I would interpret this algorithm not as predicting the winner but the odds. So Biden has a 7/13 to Trump's 6/13 chance of winning. Or 54% to 46% chance of winning, which actually seems fairly reasonable.
posted by storybored at 4:30 PM on September 22 [2 favorites]


So, like, even if his model was reliable, I think it contemplates a free and fair election. This is not going to be a free and fair election. They're going to press the courts to decide this the night of the election, when there's going to be a huge mail in turnout. They're doing their damndest to disenfranchise everyone they can. They're trying to wreck the post office so ballots don't arrive on time. "Protestors" are literally showing up at polling sites trying to keep people from voting.

This is not going to be a normal election and shouldn't be treated as one. If the UN were observing this vote there will likely be cause for it to be considered invalid.

We all know that in a fair election Biden wins. That's why they're trying really really really hard to not make this a fair election.
posted by mikesch at 4:53 PM on September 22 [20 favorites]


Also, the methodology of polling analysts like Five Thirty Eight is far better vetted, as it's been applied to hundreds of state and local elections, not just one presidential election every four years. There's enough data there to have some confidence that we're not just dealing with selection bias: it's unlikely that any analyst would have as good a track record as Five Thirty Eight if those predictions were no better than random chance. So his claim that polls are not predictive of election day results is demonstrably false.
posted by biogeo at 5:18 PM on September 22 [4 favorites]


The Keys to the White House don't mean anything when somebody's changed the locks.
posted by oneswellfoop at 5:28 PM on September 22 [4 favorites]


There's a scam based on a similar principle. Send thousands of people free investing advice, telling half of them to buy and half to sell a certain stock. Whichever one turns out to have been the wrong advice, remove those people from your pool, then repeat the process on those who remain. After several iterations you have a small group of people for whom you've given perfectly accurate investing advice a remarkable number of times, who, not knowing that you've been wrong an equal number of times with other people, might think you're an investing genius. Then offer to continue giving stock tips for a fee.

Ah, the Nate Silver Principle. All you have to do is publicly get it very right exactly once, and then your word is gold. Doesn't matter how many times you subsequently fuck up, as long as your group of investors (or the WaPo and the NYT and every other major media outlet) keeps on drinking the KoolAid.
posted by Mayor West at 5:45 PM on September 22 [3 favorites]


10. Foreign/military failure: The incumbent administration suffers no major failure in foreign or military affairs. (TRUE)

Man, that Russian bounty story sank like a stone. Any negative coronavirus coverage that involves the military (plague ships, VA hospital lawsuits, fund misappropriations, etc.) doesn't get sustained national attention, either. And Fort Hood: “The numbers are high here. They are the highest in most cases for sexual assault, harassment, murders, for our entire formation in the U.S. Army,” Army Secretary Ryan McCarthy said when he visited Fort Hood in early August. Since January 2016, there have been 159 noncombat deaths of Fort Hood soldiers, including seven homicides and 71 suicides. Congress launched an investigation two weeks ago, after the 28th death this year.

Professor Alan Lichtman is full of crap, and can safely be ignored.

But he's the only one who knows when the band uniforms will arrive.
posted by Iris Gambol at 5:48 PM on September 22 [5 favorites]


Also, I visit enough news & politics websites that I'm sure various algorithms have noticed so on the occasions when there are unavoidable clickbait ads there's a good chance at least one of them is "THIS PERSON HAS CORRECTLY PREDICTED ALL OF AMERICA'S ELECTIONS!!! CLICK TO SEE WHO WILL WIN THIS TIME!!"

I assume it's this guy.


No, it's these guys.
posted by dannyboybell at 6:54 PM on September 22 [2 favorites]


/me stamps hoof on floor five times
posted by flabdablet at 7:03 PM on September 22 [1 favorite]


/me stamps hoof on floor five times

I'm sure you will be a wonderful Senator
posted by StarkRoads at 7:21 PM on September 22 [16 favorites]


Statistics doesn't model Republican cheaters very well — until well after the fact, when it is too late.

We need to get out the vote. Voter turnout will defeat cheating and make voter intimidation by Republicans harder. At the very least, it will raise correct suspicions about their tampering with ballots and tallying results.
posted by They sucked his brains out! at 7:35 PM on September 22 [4 favorites]


There are no swing voters any more. It's all about turnout. Go vote now.
posted by heatherlogan at 7:44 PM on September 22 [2 favorites]


/me stamps hoof on floor five times

I'm sure you will be a wonderful Senator


If a horse's ass can be a senator, why not the whole horse?
posted by His thoughts were red thoughts at 7:53 PM on September 22 [1 favorite]


Ah, the Nate Silver Principle.
I'm torn between pointing out that on one level this is totally unfair, and on another it's totally fair, but look, I just had it both ways, too!
posted by aspersioncast at 8:35 PM on September 22 [4 favorites]


why not the whole horse?

Who's that at the door now?
posted by flabdablet at 9:37 PM on September 22 [1 favorite]


biogeo, that kind of thing happens, but Philip Tetlock's work on prediction suggests that people mostly read pundits for entertainment and don't keep track of whether predictions are accurate.

For what it's worth, I thought Obama became less charismatic in his second term, maybe earlier, or at least his speeches became perfunctory and repetitive. I didn't have a problem with that, he was busy.
posted by Nancy Lebovitz at 2:06 AM on September 23 [1 favorite]


We're getting to Drake Equation level of uncertainties here, but I really want to see some kind of estimates for a formula like:

Probability of Trump winning election fairly + (Probability Trump disputes election * Gets to Supreme Court * Supreme Court rules for Trump) + (Probability states submit phony electors * Congress certifies phony electors) + Probability polls rigged or votes suppressed enough for Trump to win.
posted by TheophileEscargot at 4:56 AM on September 23


I don't think he's taking into account whether Mercury is in retrograde on election day or not.
posted by octothorpe at 5:48 AM on September 23


I am the absolute worst at making predictions about the future, but I did get this right.
posted by The Card Cheat at 6:33 AM on September 23 [1 favorite]


I think I might sit out the 2020 Election Prediction MeTa.
posted by box at 9:29 AM on September 23 [4 favorites]


Bailing out of the Paris Agreement, JCPOA, and WHO render 10 false IMHO.
posted by HillbillyInBC at 11:47 AM on September 23 [1 favorite]


2020 version

Candidate agrees that facts exist T/F

Party agrees that facts exist T/F

Fox News exists T/F
posted by zerobyproxy at 11:55 AM on September 23 [2 favorites]


I think I might sit out the 2020 Election Prediction MeTa.

Oh man, I just went back and re-read some of that thread. There was such confidence that Trump was going to lose, and the predictions on various other things were so interesting to read. God, little did we know just how bad it would get. It made me cry, actually.
posted by sundrop at 12:02 PM on September 23 [2 favorites]


work horse/show horse.

rodeo clown
During the bull riding event, the barrelman supports the bullfighters, including taunting the bulls by calling them names and waving props at them, usually from within the safety of the barrel. Barrelmen may also tell jokes and use topical humor.

posted by clavdivs at 12:49 PM on September 23


A pundit "predicting" each election by flipping a coin to decide between the Democrat or the Republican candidate would have a 1 in 512 chance of a perfect prediction rate. If there are thousands of pundits predicting presidential elections (as there are) and their methods are independent and no better or worse than random guessing, we should expect there to be many pundits with perfect records

I'd argue that while statistically true, it's the wrong way to look at it.

The stock market analogy is the day trader who makes money during a bubble--basically everyone makes money during a bubble. The problem is no that the trader is unusually lucky (as you say) or smart (as they imagine); those can be hard to distinguish in any event. The real issue is that their performance is unremarkable.

In stocks comparing profits to index performance is mainstream, but people don't apply this to other predictions consistently.

that kind of thing happens, but Philip Tetlock's work on prediction suggests that people mostly read pundits for entertainment and don't keep track of whether predictions are accurate.

And the corollary is that pundits are trying to entertain, establish a brand, and be memorable. They say different things if they are asked to make an accurate prediction.
Apparently, "even the most opinionated hedgehogs become more circumspect" when they feel their accuracy will soon be compared to that of ideological rivals.
(That's from your link, thank you. I'd read about his earlier work but don't think I'd seen a summary of some of the later claims.)
posted by mark k at 1:03 PM on September 23


Ah, the Nate Silver Principle. All you have to do is publicly get it very right exactly once, and then your word is gold.

But of course Silver was wrong in 2016 just like everyone else. He just somehow spun it that because he was slightly less wrong (71% Clinton odds vs 85% (NYT), 89% (PredictWise), 98% (HuffPo), 99% (Wang)) he somehow counts as the winner, as opposed to everyone being very deeply wrong.
posted by chortly at 7:22 PM on September 23 [1 favorite]


Not really. If I tell you there's a 1 in 6 chance you'll die if you play Russian Roulette, and then you play Russian Roulette and die, I wasn't wrong just because that was the less likely outcome. I honestly still don't understand why so many people looked at those odds and decided that Five Thirty Eight was saying that Clinton would win.
posted by biogeo at 9:53 PM on September 23 [11 favorites]


I agree with biogeo on the 538 predictions. Given the data available what would you think the correct odds of a Trump win were?

Almost everyone actually predicting a Trump win at various times did so for reasons that did not come to pass, including Lichtman. (I know no one is defending him but Lichtman overestimated the Johnson vote share by a factor of two, which was a critical decision point in his system; had he accurately pegged it his system would have called it for Clinton. He also hedged then, too ["this election could go either way"] just as he's doing this year.)
posted by mark k at 10:38 PM on September 23


I agree with biogeo on the 538 predictions. Given the data available what would you think the correct odds of a Trump win were?

What do you think they are now? 45/55 Trump seems way more accurate than 22/78% Biden.

If we spent our day assuming 1/6 odds happen, then the world would literally be a standstill. That's why we create odds and use them in the first place.
posted by The_Vegetables at 7:24 AM on September 24


If I tell you there's a 1 in 6 chance you'll die if you play Russian Roulette, and then you play Russian Roulette and die, I wasn't wrong just because that was the less likely outcome.


And if you believe that, what's the cutoff? 1-10? 1-50? 1-1000? Listing the odds is not the same as making a making a prediction.

With the same logic, anti-vaxxers are 'not wrong' that vaccines are dangerous because they do very rarely kill people.
posted by The_Vegetables at 7:28 AM on September 24 [1 favorite]


Listing the odds is not the same as making a making a prediction.

Yes, exactly. Five Thirty Eight never made a prediction that Clinton would win the election. That's not what responsible analysts do. Or even irresponsible analysts who care about getting called on their bullshit. That's the difference between this nonsense from Lichtman which claims to predict election outcomes, versus polling analytics which claim to describe the probability distribution of election outcomes.

Anti-vaxxers are wrong not because they claim that vaccines can kill people or have other adverse outcomes. They're wrong because they vastly inflate the risk of those outcomes. Five Thirty Eight and the Princeton Election Consortium both estimated that there were better than even odds that Clinton would win in 2016, but PEC was wrong because it used a flawed methodology that failed to account for correlated errors in different polls and thus vastly inflated the probability of a Clinton win. 538's methodology, on other hand, appears in retrospect still to have been pretty sound, and their estimate of the odds given the polling data was probably pretty close to correct.

Saying that there's a 1 in 6 chance of dying if you play Russian Roulette shouldn't convince you that it's fine to play Russian Roulette, it should convince you of exactly the opposite. Risk analysis involves both the probability of an outcome and the severity of that outcome. The polling analytics from 538 shouldn't have been reassuring to anyone (megathread link from August 2016), they should have been a call to action that our democracy was in terrible danger.
posted by biogeo at 10:45 AM on September 24 [7 favorites]


And again, at a 22% chance of a Trump win, Five Thirty Eight is currently predicting that even without unprecedented vote suppression, fraud, and other tactics that might skew the results to be less well predicted by the polls, the odds of Trump winning the election are currently greater than the odds of dying when playing Russian Roulette with a six-shooter. If you wouldn't be happy playing Russian Roulette, don't be happy now. That's why we create odds and use them.
posted by biogeo at 10:57 AM on September 24 [9 favorites]


For checking against previous elections (going way back), the closest these aggregated polling results compare to for Biden, is Reagan in 84. As my long ago British plasterer boss used to say . . . 'just sayin'.
posted by Harry Caul at 1:43 PM on September 24


If we spent our day assuming 1/6 odds happen, then the world would literally be a standstill. That's why we create odds and use them in the first place.

I go into work every day (well, virtually speaking) and work on projects that have a much lower than 1 in 6 chance of ever being successful. These are pretty well funded because my corporate masters think that they're the best odds they can get--and some of them do pay off. In fact, if we didn't know that 1 in 6 odds happen, that's what would put my job at a standstill.

Then I go home (well, log off) and a lot of evenings play a table top game of one sort or another. If I'm in a position where the opponent can win if if they roll a five or six I certainly don't think "I have this in the bag."

The game analogy was what I used before the election to explain how worried I was about it. I could easily call to mind many examples of coming out on the short end of rolls like that.

What do you think they are now? 45/55 Trump seems way more accurate than 22/78% Biden.

Why do you say that?
posted by mark k at 10:08 PM on September 24


Mayor West: "Ah, the Nate Silver Principle. All you have to do is publicly get it very right exactly once, and then your word is gold. Doesn't matter how many times you subsequently fuck up, as long as your group of investors (or the WaPo and the NYT and every other major media outlet) keeps on drinking the KoolAid."

Their track record is actually quite good, and I'm not just talking about the early high-profile elections where they called every state right (or the fact that they were closer to being right in 2016 than every other model). If you review every prediction they've made in every race in every election so far, their forecasts line up quite closely with what actually happened. So of all the candidates they said had a ~65% chance of winning, those candidates won about 65% of the time. Add in sports forecasts and they're even closer to the mark.
posted by Rhaomi at 7:08 PM on September 26 [2 favorites]


« Older Discover a World of Sounds!   |   The Humbling of Dane Cook Newer »


This thread has been archived and is closed to new comments