October 9, 2020

The polls are wrong again: expecting the unprecedented, and ruling out the amply attested, in a status quo election

By this point, the results of presidential opinion polls reveal more about their own methodology, and the broader climate of opinion that is heavily shaped by the media, than they do about the outcome of the election. I'm not going to micro-analyze any single poll, because the failures are systemic and require the pollsters themselves -- not me -- to inspect each step of their process to locate where all the failures are.

How do we know that the polls are broken? Because their results are off in outer space, with little to no precedent throughout the entire political era we're currently in (the neoliberal / Reagan era, beginning in 1980). If a poll's design, methodology, and execution produce such risible results, then they are in need of one hell of a "de-bugging," to borrow a term from software programming.

Imagine if their result said that "Hillary Clinton, the Republican, will win X percent of the vote in 2020, while Donald Trump, the Democrat, will win 100-X percent." We would dismiss everything that this poll yielded because it is so fundamentally broken that it got the candidates' parties reversed, and even then it thinks that Clinton, not Biden, is the 2020 Dem nominee.

These would not be isolated flukes, like typos, that could be easily corrected and put our worries to rest. They would only be the tip of the iceberg -- the most starkly visible symptoms of a broader underlying syndrome. For example, if the guy coding the candidates' parties was stoned out of his mind, he likely introduced other errors into the process, just ones that will take a finer-grained investigation to root out and correct.

The results of the polls act as a test of the underlying design, methodology, and execution. If it is ridiculous, we don't accept it just because it came from some design and methodology -- it means we should be suspicious of the design and methods that yielded it. To blindly accept whatever any old model generates is to practice cargo-cult science. We would then subject each step of the process to ruthless scrutiny, to be sure that the process generating an extreme result is sound. In all likelihood, it means the process is not sound.

In fairness, maybe it would pass the ruthless inspection, and we are about to experience an extreme result. However, none of the polls yielding these ridiculous results are in fact being subjected to such de-bugging, to fully reassure what ought to be a highly skeptical audience. Instead the results are being accepted at face value -- not that they take any single one to be correct, but that in some aggregate, they are correct. But averaging a bunch of risible results does not make their processes any more sound or reliable. The average of a pile of garbage is still garbage.

What are some of the most flagrant examples of these results? I might add others in the comments as I come across them, but here are the big whoppers.

1. The challenger (Biden) is going to unseat a one-term incumbent (Trump) from the dominant party (GOP) of the era (Reaganism).

This has never happened in our era, and the last time it did was the realignment election that started it all -- Reagan unseating Carter in 1980. The 2020 election is not one of realignment -- indeed, Biden is adamantly campaigning on returning to the status quo ante Trump, who is disjunctive, rather than trying to steal Trump's anti-Reaganite themes from 2016 and promising to deliver on them in a way that Trump himself has been mostly unable to. (Reagan out-Cartered Carter on dismantling the New Deal.)

2. The winner is going to win the popular vote by 9-10 points.

Also has not happened since the realigner Reagan. In '84, he won by nearly 18 points -- but that was a re-election, not his first win, and he was from the dominant party of his era, not the opposition. Plus he was a realigner. None of those conditions is true for Biden in 2020. Reagan also won by nearly 10 points in '80, when he was an opposition-party challenger to a one-term incumbent from the dominant party of the era. So again, this reduces to how analogous 2020 is to 1980 (not at all).

3. The Democrat in the Reagan era will win voters aged 65 and older -- and if the flashy polls are to be believed, by something like 60% to 40%.

Qualitatively, this does have some precedent, albeit 20 years ago or longer: Clinton won them in both '92 and '96, and Gore won them in 2000. But quantitatively, the result is without precedent. Gore only narrowly won them (51% to 47%), and both of Clinton's elections had a major third-party candidate (Perot) who drew more from Republicans than Democrats. Putting Perot voters back to their usual party, Clinton would still have won seniors, but again only narrowly (low 50-something percent range). That is also the typical range for how this age group usually votes in the Reagan era, i.e. Republican (Trump won them 52% to 45%).

Once again, you'd have to go back to realigner Reagan to find seniors voting 60% for either party. Even if Trump did narrowly lose them, that would not prevent a win -- Obama lost them both times and won both times, and Bush lost them in 2000 and still won. The 45-64 age group is more important.

4. Third party vote share will be in the 5-10 point range.

Much of the reason why polls show Biden leading by so much is because there are so many saying they'll vote third party, are unsure, etc. There is some precedent for that in our era: 6% voted third parties in 2016, Perot exceeded that in '92 and '96, and Anderson got over 6% in '80. But we know that's not happening this time because third parties are effectively invisible and inactive, and each major party is far more unified than four years ago.

There's no Jill Stein leading a march of angry Bernie bros and babes outside the DNC, nor a Gary Johnson being constantly platformed by MSNBC to target suburban moderates, nor a McCuckin platformed by CNN to steal Utah (he ended up robbing Trump of Minnesota instead). Most people could not name the Green or Libertarian nominee this time around.

Perhaps 0.5% will go to the Greens, 1% Libertarian, and no more than another 0.5% combined for the others. Most of the supposed non-major party support from polling will wind up in Trump's column on election day: the Libertarians are far more numerous than the Greens, wary right-wingers no longer consider Trump a wild card or beyond the pale, and Republicans are the only ones with an incentive to say "not sure" when polled, since the climate of hysteria and fear makes them, not the Democrats, the evil witches whose homes must be burned down to save humanity.

The parties are more unified than in 2016, but the GOP is more united than the Dems, since there was no contest for the primary on the GOP side (incumbent president), whereas there was a decent contest from the Bernie bros on the Dem side (albeit far less enthusiastic and effective than in 2016). Bitter or disillusioned voters, who make up the target for third parties, will be more common on the Dem rather than GOP side.

5. Battlegrounds: Arizona flipping blue, no Rust Belters remaining red

The current RCP average has Biden up over 3 points in Arizona, a state that has only voted Democrat once in the Reagan era, in '96 -- when there was a big third-party wild card (Perot), and when the Democrat was the first-term incumbent president, neither of which applies today. Rationalizers try to paper over this ridiculous prediction by pointing to Trump's narrower margin there in 2016, compared to Romney's or McCain's performance in the two elections before that. But that was due to "respectable" Republicans chafing at Trump and voting third party -- Clinton got no more of the vote than Obama did either time. Since those wary Republicans are coming home this time, Trump's margin will be closer to McCain's or Romney's than it was last time. Arizona is not a battleground.

On the flipside of the loyal Republican Sun Belt, there was the disaffected Rust Belt that defected from blue to red last time. Current polling averages show none of them remaining red this time. This systemic model failure is the flipside of expecting the unprecedented -- ruling out the amply attested. The Rust Belt voted for Reagan twice, and while Minnesota, Wisconsin, and Iowa voted blue in '88, Michigan, Ohio, and Pennsylvania voted for George Bush Sr. Iowa voted for Bush Jr. during re-election in '04, and Trump is now up for re-election. Not to mention the even more recent history -- namely, the very last election, when Trump swept the region.

The models do not have to predict that Trump is going to sweep the region to the same degree as last time, but ruling out even a single hold (including the purple / swing state of Ohio) is an absurdity. And worse than that, a biased absurdity. These results are not all over the place, some ridiculously pro-Biden and some ridiculously pro-Trump -- they are uniformly in the pro-Democrat direction. They are not predicting Trump to win New Jersey, or to expand his popular vote share by 10 points.

These failures interact with the problem of the third party estimates. On the eve of the last election, polls in the Rust Belt suggested that 5-10% would vote third party. In reality, only half of the estimate actually did, and the other half voted Trump, while hardly any voted Clinton. This kept the estimates of Clinton's share pretty accurate, but then "delivered" Trump a "bump" of several points over his polling. In most states, that was enough to overtake Clinton, although in Minnesota the degree was not enough to win outright (though it was qualitatively the same pattern).

The terms "deliver" and "bump" are in quotes because those people did not actually make up their minds last-minute, or "break for" Trump. They were going to vote for him, but did not want to tell that to the pollsters. Clinton supporters felt no pressure to keep their intentions secret, so they saw no apparent last-minute bump of undecideds in their favor.

This pattern is repeating itself this election: Biden's estimates on the eve of the election will be pretty accurate, but half of the undecideds will be Trump voters and give him an apparent last-minute bump. We're still weeks away from the eve of the election, so the polls will narrow in Trump's favor over that period, just as they did last time. The narrowing in polls, plus the apparent bump, will win Trump at least some of these states again. Regardless of which ones are held, the net effect will be a Trump re-election, although perhaps by a smaller Electoral College margin than last time.

Without going into a whole 'nother post about the causes of these failures, suffice it to say that the media themselves are to blame for the systemic failures of their polling models. They have bombarded the culture with unrelenting hysterical propaganda not only against Trump but against everyone who voted for him, or even sympathized with his campaign. Naturally that will make a lot of his supporters unwilling to respond to the pollster in the first place, or to guard their answer behind a non-committal "third party / don't know" response. They have not only not hammered the same propaganda campaign against Democrat voters, they have sanctified them, so there is no counter-balancing pressure for them to hide their intentions.

Republican elites do not control the informational sectors of society, but the material sectors, and they have not used them to intimidate, demonize, or otherwise pressure the voters of the other side. The military is not withholding national defense, say by waving foreign armies in to freely occupy urban Democrat strongholds. The agriculture cartel is not starving the cities of food shipments. Big oil and big coal are not turning off the power in blue zip codes. And the manufacturing cartel is not withholding cars, clothing, or other products from liberal customers.

And even if they were, the military et al. don't control the polling firms -- why would a hypothetical Democrat who got harmed by GOP elite sectors feel intimidated to reveal his voting preference to the media, which is on the other side of the political spectrum? Maybe they wouldn't want to say they're Democrats if they faced a military checkpoint outside of a voting station, but to a pollster from the mainstream liberal media? They'd feel perfectly fine venting to fellow Democrats.

Because the media, entertainment, and info-tech sectors have only ramped up their psychotic attacks against half the population, do not expect presidential polling to get better anytime soon. Rather, the results will only reveal in what ways and to what degree the polls are systematically broken, due to the climate of hysteria created by those sectors themselves.


  1. Bush Sr. in '92 was a 3rd-term incumbent, in case that's not clear. Parties, not individuals, control the government. Although Bush Sr. had only served one term by the '92 election, his party had served 3 in a row. The Democrats broke that 3-term streak of the Republican party -- not so much Clinton unseating Bush after 1 personal term.

    Trump is a 1st-term incumbent, since he and his party have only controlled the WH for one term. They are not in the weak position Bush Sr. was in for '92.

  2. I made a bet that Trump wouldn't win last time, and Theo can confirm I paid. I'm willing to do the same again this time. If agnostic wants a non-linear bet based on the electoral college, I can do that, but again my zero-point is 269 so I pay nothing if Trump loses.

    On a minor note, Firefox is no longer permitting comments here because of some embedded window issue.

  3. I believe Trump will lose also, due to wide numbers of working-class people simply sitting out the election(in order to teach the Republicans a lesson). You point out that the Dems have still alienated their traditional working-class voters - true, but it doesn't matter if those people don't vote Republican, either. Yuppies will show huge turnout - the black agitation does not offend them(Kamala Harris comes across like a nerdy yuppie, not even a whiff of 'black power'). And blacks will also show huge turnout.

  4. Betting on politics is spergy cosplay stuff. The whole BIG DATA larp took it up the ass last time, and I have even less interest in taking part this time (despite the fact that I called it in detail, early).

    All those individuals in the crowd-sourcing team, all those dollars wagered, and they still had zero clue that a Republican was going to win an election in the Reagan era that they dominate, after two consecutive terms for the opposition party.

    Clinton winning would've been literally unprecedented in all of American history, but they thought it was a lock, while Trump winning would've been par for the course, yet they ruled it completely out of possibility.

    None of these pseudo-quants see structure in anything, even when the structures are simple. Elections are grouped into eras, each era has a dominant and an opposition coalition, there are incumbent parties vs. challenger parties, duration of incumbency, etc.

    Fill in these abstract variables with particular values: Reagan era, dominant GOP and opposition Democrats, incumbent oppo pres for 2 terms, GOP is challenger = GOP win.

    No machine learning, no millions of variables over-fitting the data, no distributed crowd-sourcing of any kind. That's all just clever-silly show-off junk for nerds.

  5. The reason for betting is to put skin in the game for those who influence outcomes. But I'm just a blogger on the internet, and I don't shape reality thereby. (Other than making Aimee Terese giggle and blush with seranades below her cyber-window.)

    When nerds get heavily invested in online betting about politics or economics, they're just self-aggrandizing.

    "We in the cognitive vanguard must wield our powerful influence responsibly, and ethics demands that we put some skin in the game for our prognostications."

    Like they're some hedge fund honcho with a billion dollars of assets under management, whose investment decisions must be held accountable. FOH.

  6. The only way I'd structure a bet in politics is not over the outcome of an election, which means nothing per se, except to the media-addled political theater fags.

    To normal people, what's important is what the winner of the election is going to do upon inauguration -- more spending on healthcare, more wars, etc. The electoral outcome does NOT summarize those outcomes into a single convenient variable, because winning the election only reflects the promises made on the campaign trail -- not what they actually deliver in office.

    I laid that out last time, and said I'd only bet on the number of Americans in Afghanistan, the trade deficit, the number of deportations, etc. If Trump won, I figured they'd go in the direction of his campaign promises.

    Lucky for me, I did not bet on those real-world outcomes because although I nailed the electoral outcome long in advance, Trump's admin delivered the OPPOSITE of his campaign. Americans in Afghanistan surged, the trade deficit exploded, deportations fell off a cliff / border crossings surged, the executive branch is witch-hunting the KKK boogeyman rather than the roving gangs of BLM / antifa, and so on and so forth.

    Losing a shitload of money on those real-world outcomes would not have made me correct my views. It would've just made me angry since there would be no way to win back those losses. I'd just have to eat shit.

    But pursuing this as a cerebral hobby, motivated me to figure out what I'd gotten wrong, and who or what to support in the future in order to see a shrinking trade deficit, demilitarization, etc. That's what I wrote on in 2017 and '18 -- elite sectors of society control the two parties, which sectors are with which party, the dynamics of the regime cycle (I thought Trump would be a realigner, but he was a disjunctive), etc.

    Serious people aren't motivated by loss of money, they're motivated by being wrong and not getting the real-world outcomes they thought they were going to.

  7. Do you think there will be a Civil War (which Peter Turchin) predicts?

  8. I don't think Turchin is predicting a Civil War with two clearly delineated teams, as in the American Civil War or the two civil wars in Rome before the Imperial age. I'm not predicting that either.

    It's more like civil breakdown, chaos, collective unrest or violence -- but without two cohesive teams waging war against each other. That requires a degree of national cohesion that we don't have. It's more like the Crisis of the Third Century, from Roman history, not the earlier two civil wars.

    The difference, referring again to Turchin, is in asabiya -- solidarity, or the potential for collective action -- which accompanies (and drives) territorial expansion. Rome was expanding during the first two civil wars, but had pretty much reached saturation by the 3rd C., so that period of unrest devolved into internecine chaos.

    The US was expanding during the 19th C., so our crisis period in the 1860s tapped into the high levels of asabiya and made it take the form of two cohesive teams. We've stopped expanding since our peak in WWII, and asabiya has been falling off a cliff for decades. Our collective violence crisis will not take the form of Team A vs. Team B.

    The English Civil War took place against a background of imperial expansion, which gave them high enough asabiya to stay cohesive even amidst a disintegrative phase. Not so cohesive that breakdown was avoided entirely, but restricted to two national teams. Then by the time the British Empire was past its peak, and another collective violence crisis hit, circa WWI, they didn't have Team A vs. Team B anymore. It was just generalized chaos, because their asabiya levels were falling.

  9. "The reason for betting is to put skin in the game for those who influence outcomes."
    People bet on sports all the time, without having influence over the outcomes. The lack of "skin the game" is something Taleb complains about all the time for people who talk/write about an issue. It's especially a problem when people who influence outcomes don't have "skin in the game", but randos on the internet can be subject to the same sorts of errors if they don't put their money where their mouth is. And if you really were that confident in your system for predicting elections, you could make some easy money off a fool like me while also establishing your own credibility as someone who actually believes what he's saying.

    "The only way I'd structure a bet in politics is not over the outcome of an election"
    You did structure a bet over electoral outcomes in 2016. I'm trying to take that bet, just not using Romney's outcome as the zero-point.

    "Serious people aren't motivated by loss of money, they're motivated by being wrong"
    There are almost no serious people, because people are wrong so often without that making much of a difference. We see in experiments that people say very different things when they are vs aren't betting on outcomes. When you were laughing off the notion that anyone other than Jeff Sessions would be VP, it didn't require you to stop and reconsider all your beliefs about Trump when the talk about it being Pence turned out to be true.

  10. Taleb is talking about influential prognosticators, experts, etc., who have the ears of those with their hands on the levers of decision-making. The class of consultants, advisors, talking heads, and so on.

    That's not me. Do you really think Taleb would suggest that normies on Facebook who are posting their thoughts ought to be putting something at stake? I'm not part of the media, even informally or at the margins, and clearly nobody powerful is following my advice.

    And one rando on the internet does not own enough wealth to enforce discipline. Strange as it may seem, I'm not a millionaire, nor are my parents. If I wagered what I actually could, would anyone take that seriously as a check on my thoughts? "Oh, well, he wagered a couple hundred or thousand dollars, he's really going to watch what he says now..." Moronic, and just cosplay.

    I would have to be like skilled poker players -- raising shitloads of other people's money in order to put my skills to use in a high-stakes bet. Strange as it may seem, I'm not going to go around to eccentric millionaires or billionaires to raise the funds necessary to make my "stake" in the outcome big enough to convince the audience.

    I establish my credibility by being right about major things, and wrong about minor things. Don't quote Taleb and then pretend that the only thing we're interested in is probabilities -- it's the probability TIMES the effect size of that outcome. That's the expectation.

    That's why I don't care, and no one else cares, that I miscalled Trump's VP pick. VP's don't matter for the election, or any outcome upon inauguration. Breaking a tie in the Senate -- any R would break for the other R's.

    I called the map of the GOP primary in nearly perfect detail, early on, and ditto for the general election, as well as illuminating the themes (rooted vs. rootless, industrialized vs not, etc.). That was the major-effect-size stuff, and I got it right when everyone else got it wrong.

    Miscalling the VP nom is small potatoes, of no significance. "Meh, he's not perfect, but right where it matters."

    What I did get wrong, and was of a large effect size, were the actual policies that would result from a Trump win. I thought the main campaign themes would lead the way, but they all went in the opposite direction.

    Unlike other Trump fans from 2016, though, I actually admitted what happened, whereas most of them just lied about what was happening. And more than that, I figured out why the '16 themes were heading in the wrong direction, based on which elite sectors control the GOP vs. Democrats, and what phase of the regime cycle we were in.

    No more spergy comments about the LARP value of betting. Your whole model was spectacularly falsified by the most recent election. Online betting does not improve intellectual discipline, and does not give us more accurate estimates of important events.

  11. Most accurate poll: Trump up among poor, non-college, Hispanic, urban demos than 2016 exit polls. Meaning, up among each one of those groups.


    Apparent decline in the other groups -- cushy, college, white, suburban -- is likely the "pressured Trump voter" effect of intending to vote for him, but not wanting to get dragged into another "conversation about Trump" by the media. Some of it real, of course.

    Proles aren't locked in the media / striver bubble (wealthy educated metropolitans), so they're less pressured to lie or hide or not respond. So those increases are real.

    Will write up more in detail. Hilarious.

    Can't wait to hear wealthy educated WASPs and Ashkenazi Jews lecturing the working-class Latinos this time, not just the white proles, about how they're not allowed to "vote against their economic interests".

    Chapo Trap House will have to come up with a new bit dismissing "taco truck proprietors" as petite bourgeoise reactionaries, unlike the unionized workers of the media elite who live in 1% zip codes and are going to lead a proletarian revolution.

  12. inb4 Biden calls them los dey-plor-ah-bleys at a La Raza fundraiser.

  13. It reinforces that working-class Hispanics are not tied to the Dems. Are they voting for Trump because he's so soft on immigration? I still believe he's going to lose, though.

  14. You mention policies of the 2016 Trump win going in the wrong direction. As an intellectual exercise, would you mind spelling out for me what you mean? This is a recurring critique I hear from a lot of ppl I greatly respect, but it doesn’t go much deeper than surface level and I’m curious about the details.

  15. You can browse the archives from May 2017 through 2018 or '19 for more detail. Briefly, though, there were 3 big themes that set Trump against the Reaganite status quo: re-industrialization, shrinking military global footprint, curbing immigration / increasing deportations.

    On all three, his admin headed in the opposite direction by the summer of his first year. Trade deficit soared far higher than under Obama, NAFTA was merely re-branded as the MC-whatever-A. He bombed, invaded, and occupied Syria -- a new war (Obama admin only supported local jihadist proxies through the CIA). And immigration went through the roof, including illegal border crossings, while deportations fell.

    Understanding why requires looking at which sectors of society control the GOP vs. the Dems.

    The GOP is controlled by the manufacturing cartel, which explains why they vetoed Trump's theme of re-industrialization (which would eat into their profit margins, since off-shoring cuts costs for them).

    It's also controlled by the military, which explains why they expanded the military footprint by starting the war with Syria, ended the Iran deal and assassinated their highest general, added a new nation to NATO (Montenegro), and sent tens of thousands of Americans back into Afghanistan.

    And it's controlled by the agriculture cartel, along with various labor-intensive small businesses, who depend on cheap unit labor costs to maximize profit. That's why they opened the floodgates of immigration -- to get more cheap foreign labor.

  16. Thanks so much for the response.

    So in a sense, the toxic re-rooting of what keeps this nation stable is too risky to weed out (I.e. excess cheap immigrant labor vs. union/local, domestic capital vs. hegemonic suppression, etc). Feels like each regime is trying to slow the spiral of civic degeneration down instead of actually trying to reverse the trend.


You MUST enter a nickname with the "Name/URL" option if you're not signed in. We can't follow who is saying what if everyone is "Anonymous."