Part 2 of 3 — you can read Part 1 here

As we’ve been working on this series, new theories have continued to emerge to explain why the polls were so wrong in the swing states, and swing states only.  Even the Democratic establishment seems determined to find something other than voter suppression to explain it.  Curious.  It’s as the two parties have a shared interest in not having too many of certain types of people voting.  Apparently aware that throwing state-level pollsters under the bus again isn’t going to work, and that they need to come up some kind of American voter less mythological than, for example, Trump supporters who mysteriously stop answering their phones when they cross interstate lines into “electorally important” states (while their equally bombarded Democratic counterparts eagerly seek out every pollster they can find), they’ve hit on the concept of affixing a scarlet (R) to so-called ‘low-information’ voters, a class of Americans that clearly does exist, arguably in larger proportion than any other Western democracy.  The claim is that this type of voter is (a) particularly unlikely to respond to pollsters (b) strongly pro-Trump (a cognitive layup when preaching to the converted) and (c) particularly concentrated in the swing states.  But while these theorists certainly know their ABCs, they seem to have forgotten their Ps and Qs.

Let’s take as a given that the consultants and pundits are right about the relationship–or lack thereof–between pollsters and these voters, even though, like so many other recent assertions about voter psychographics, we’ve never seen it proven except by tautology, and even though these poor unfortunates (the voters, not the pollsters) sound like the kind of people who can never get out of jury duty, not to mention the last to get pollster-proofing technologies like state of the art caller ID and call blocking.

Wikipedia, of all sources, includes, in its definition of “low information voter,” George Lakoff’s tart observation that the term is primarily just “a pejorative mainly used by American liberals to refer to people who vote conservative against their own interests and assume they do it because they lack sufficient information.”  Indeed, for pollsters, the field marks required to properly identify and poll them are a bit murky and, as a result, educational attainment is what’s typically used as a proxy; in fact, in the just-so story de jour, they are referred to as “lower education, low trust” voters (aka LELTs).

As someone who has done high profile new media & technology online survey-based research for going on three decades,¹ I would be among the last to say educational attainment isn’t an important consideration–it’s either the Rolls Royce of demographic variables or the silent killer, depending on the client and what they hope to prove.  But are “lower education” voters really “heavily concentrated” in the swing states?  The Census Bureau doesn’t seem to think so, at least as recently as 2019, as can be seen from the chart below:

You can see, probably without even looking at the numbers, that there’s almost literally no difference between swing and non-swing voters when it comes to high school completion, and the differences at the college (1.6%) and advanced degree (1.5%) levels are nowhere close to substantial enough to account for the polling “errors” observed in the swings.  In fact, if you remove the outlier that’s distorting these results, the District of Columbia, which we should (sorry DC), since it’s purely an artifact of government and gentrification (sorry DC), there’s less than a point of difference at the college level (0.9%) and only a tenth of a point more at the “high information” marker.

But let’s let them cherry-pick a little.  Ah, they might say, but when we talk about the “electorally important,” we’re not talking about all swing states, just those in the upper Midwest, where the polls were most inaccurate. If, in response, we limit the swings to MN, WI, MI, IA, OH, and PA, the percentage of residents with a college degree… actually increases, to 31.6%, reducing the alleged cognitive deficit (sans DC) to 0.8%.  But let’s let them further set aside the Bings and throw out Minnesota.  The remaining five states still have a college degree attainment of 30.5%, a difference of only 1.9% w/o DC.  Let them make us settle for Montmorency–throw out Pennsylvania, too, so we’re only looking at the four Midwestern states where the polling misses averaged 7.2%.  That drops degree attainment to 30 percent-even, which makes for a gap of 2.4%, which still only accounts for a third of the “miss.”

But as the ad says, that’s not all.  Wikipedia describes low-info voters as typically “moderate,” less likely to have “developed clear cut ideological preferences,” less likely to vote, and more likely to vote for candidates they find personally appealing (Donald Trump??²).  Furthermore, while there may be differences between the definition of a “low information voter” and what the Democratic consultant/pundit class means by “lower education, low trust,” if educational attainment is the primary marker being used, the argument faces a more fundamental challenge:

That’s right, and as everyone knows, Black and LatinX voters heavily favor Democratic candidates, which leads to one of two possible conclusions, the first probably accurate, the second a softball over the heart of the plate. Either:

(A) It’s really harder to get low education voters to respond to polls, they’re disproportionately Democratic, and therefore pollsters in the swing states (where LELTs are allegedly “heavily concentrated”) actually underestimated Biden’s real level of support, not Trump’s, which only further supports the case that only voter suppression can explain our skewed polling results.   OR

(B) Thanks to their long and positive experience with the government and other authorities at all levels, Black, Latinx, and Native American voters, especially those without college educations, are actually what we call “high trust” in nature, eager to pick up their phones and answer questions posed by people they don’t know.  Much more so than whites, who have experienced so much more discrimination and hardship they no longer trust anyone anymore. [insert sarcastic/sarcasm emoji]

Which do you believe? [insert rhetorical question emoji]

Here the gurus manque may be tempted to interject that Blacks and Latinx voters aren’t remotely as Democratic as we think, as “evidenced” by their “sharp drop” in support for Biden vs. their support for Hillary Clinton in 2016–in fact, they might say, they’re exactly the low education, low trust voters we meant all along!  Maybe this is one of the reasons the Democratic punditocracy keeps harping on the loss of support among minorities, typically by referencing specific counties like Miami-Dade.

So let’s take a look at the ethnicity of the 2020 electorate as pieced together by the Washington Post from national and state exit polls, plus a telephone survey of 25,000+ voters to account for early and mail-in voting, and how it changed from 2016 (albeit from a different, but generally reliable, source):

A couple of things seem worth noting here.  The first is that the most-educated ethnicity, Asians, apparently supported Biden less than they did Clinton, even with Kamala Harris on the ticket, which isn’t actually that surprising, given the higher income levels of this ethnic group, unmediated by both parties’ less-than-stellar history of equal-opportunity racism where their race is concerned.  Whites, the group that includes the prototypical barkaloungers and veriphobes liberals envision when the magic words “low information Trump supporter” are incanted, supported Biden more than they supported Hillary in 2016. And while (a) ethnic analyses of elections depend on exit polls and other surveys, which means “results may will vary,” and (b) irony threatens to runneth over when polls are used to tell us what’s wrong with polls, as if the eye sees itself, after all, the ballyhooed losses suffered by Democrats among Black and Latinx voters appear to have been so small it seems highly unlikely that any other respectable analysis can make them sizable enough to explain what happened to the polls.

In fact, based on the proportion of the electorate each group represented, even if you stereotypically assume all the Black and Latinx voters Biden “lost” were of the less educated variety, this converts–as shown on the chart below–to just a 0.39% ding on his total. If lower education voters were merely represented among the defectors in the same proportion they are in Black and Latinx populations in general, the impact would have been even lower, just 0.3%.  Even if you believe Biden lost as much as 8% of Clinton’s Latinx support, this translates to only 0.84% of the overall vote, or 1.04% with Black “losses” included.

Of course, this exposition’s not really about general results, but what happened in specific “electorally important” swing states, and it must be confessed that there is a fairly strong correlation (0.62) between the proportion of white voters without a college education in the swing states where the biggest polling misses took place and the magnitude of the miss involved, which is more than the GRE or SAT can say.

That said, the Brookings Institution’s analysis of swing state exit polls provides little to no comfort for any claim this correlation metamorphs into causation; quite the opposite.  In swing after state, white support for Trump fell precipitously, including, and often especially, among those without a college education, e.g..

• In Michigan, Trump’s support among non-college educated white men fell by 14 points from the previous election; his support among white non-college educated women fell by 2.
• In Wisconsin, non-college educated white male Trump support dropped by 13; non-college educated women of the Caucasian persuasion were 9 points less enthusiastic about four more years.
• In Arizona, his margin among white men without a college education declined by 18 points, though his support among women of the same attainment increased by 3
• In Georgia, white men who never graduated from college dropped him like a hot potato head³ by 8 points, women without BAs or BSs freedom-fried him by 7.

It appears many of the “poorly educated” whom Trump professed to love for their lack of education in 2016 got their BAs in BS during the four years since.  But let’s stop parrying these just-so s.o.s.’s–and destroy them instead.

(ed. note: Since we published this piece, Pew’s much–anticipated validated voter analysis lit up the echo chamber, showing that while Biden lost less ground among Black Americans than previously believed, he fell short of Clinton with Latinx voters by a whopping 21 points. As some Wile E’s in the consulting class claimed vindication on this basis, we decided to look more closely at the potential impact of this shift, specifically on the swing states, as well as Pew’s data on “low information” (i.e. non-college) white voters–since the concept of “low information” voters in general formed the basis of the original claim, not just a pejorative definition of Latinx as coincident with this epithet.  To our surprise/not surprise, the new data actually makes the “low information” explanation for polling error even less tenable; in fact, it destroys it entirely. You can take our word for it, or check out our analysis here)

Exhibit B, C, D, E

While Exhibit A in our prosecution of the real conspiracy to throw the 2020 election (i.e. the fact that polling distortions were highly specific to the swing states, which voter suppression can easily explain, and none of the pollsters’ other theories can explain at all) is, as you’d expect, highly 2020–specific, Exhibit B isn’t an artifact of last fall’s election at all; it’s data from the election before–and the six national elections before that.  It may be hard to remember now, but 2018 was a very good year for America’s pollsters, especially after the “debacle” that was 2016.  The overall polling bias towards Democrats was 0.4-0.5%, making that year’s polls the most accurate since 2006.  Said date, emerging out of ancient history (in both political and Internet time) is significant, because of several key events that occurred in the interim.

In 2007, the Supreme Court, in Crawford vs. Marion County Election Board, upheld Indiana’s new strict voter ID law, opening locks along the political Love Canal for similar measures across the country.  Both the Supreme Court Justice who wrote that decision, Republican appointee John Paul Stevens, and even more so–and more significantly–the giant of conservative jurisprudence, Richard Posner, who wrote the appellate court opinion that swayed the Supremes, came to regard the decision as a serious mistake (Posner, in particular, has been quoted as considering it the worst of his career) as they came to realize that the alleged “legitimate state interest” justifying these laws was just a cover for their real purpose, aggressive, widespread voter suppression of the demographic groups most likely to have difficult acquiring ‘proper’ ID: young voters, minorities, and the poor.

But it took time for the depths of Crawford to be plumbed, and the states historically most eager to suppress were still held back by the Voting Rights Act (aka the VRA), which required them to get permission from the Justice Department before making any changes to their election laws that could disproportionately affect Black voters (and therefore, in most cases, the other Democratic groups referenced as well).  Until 2013, that is, when the (then) most conservative and partisan Supreme Court in our history decided, in Shelby County v. Holder, that we were now living in the “post-racial era” (as we all know, Jackie Robinson eliminated the ghettos, Jim Crow, sundown laws, redlining and other racist practices decades ago, and of course the election–and even re-election–of our first biracially Black president signed, sealed, and delivered the new new deal), rendering such protections no longer necessary.  Within 24 hours of this decision, Alabama and Mississippi began enforcing photo ID laws that had previously been blocked by the DOJ, and Texas announced it would be implementing a strict voter ID law as well. Well, not that strict; you can use a gun license as your ID in TX–even though neither residence nor citizenship is required to acquire one–but not a student ID, which de facto requires residence, nor an EBT (“food stamp”) card, which de jure requires citizenship.  Because, you know, people who depend on their EBT cards for food are always willing to “loan” them to complete strangers to vote on their behalf at election time.

Research by the Brennan Center showed that states previously held in check by the VRA began purging voters from the registration rolls at a significantly higher rate than others.  States canceled online voter registration, early voting, “Souls to the Polls” Sunday voting, same-day registration, pre-registration for those about to turn 18, and by the time of the 2016 election, nearly 900 polling places had been closed, many in predominantly Black counties.  Not surprisingly, the heaviest voting restrictions were implemented in the states that had the highest Black voter turnout in 2008–apparently they didn’t like the sound of “post-racial.”  As a result of Crawford, Shelby, and the machinations of the nation’s shadow government/confederacy (a.k.a. ALEC), half the states passed restrictive voting laws between 2010 and 2016.  As the 2016 election campaign kicked into high gear, democracy was a limousine serenely gliding towards its Dealey Plaza shooting gallery, blithely unaware that it had already been hollowed out by parasites.  And the crowd waiting for it there included tens of thousands of Russians or Russian bots masquerading as Americans, championing a much more useful idiot (charitably speaking) than they may have had available to them in 1963.

Yes, the Mueller Report was a dense, dry tome, pitched by a reluctant front man, but that excuses no one from reading the first sentence, which was completely unambiguous, requiring no interpretation from compromised Justice Department employees or pundits of any persuasion: “The Russian government (not merely ‘Russians,’ not a 400 pound kid sitting in his bedroom in China, the Russian government) interfered (not tried to interfere, interfered) in the 2016 election in sweeping and systematic fashion.”  As seventeen US intelligence agencies had previously unanimously concluded, after being alerted by the intelligence agencies of at least six US allies, Mueller concluded–still on page 1–that this interference wholly “favored presidential candidate Donald J. Trump and disparaged presidential candidate Hillary Clinton.” Unlike the “Whitewater” investigation of Democrat Bill Clinton, a multi-year fishing expedition by a right-wing partisan hack, virtually everyone who came to be vilified for this report’s genesis and execution–Robert Mueller, James Comey, Andrew McCabe, Peter Strzok, Rod Rosenstein, Christopher Wray–was a lifelong Republican.

We could write a column about Russia and 2016 rivaling The Report in length (and heck, go ahead and throw in the five volumes produced by the bipartisan Senate Intelligence Committee as well, especially four and five), but we’re specifically focused on polling and suppression. Trumpeting the unknowable “fact” that “no votes were changed,” as if this were the rock-bottom standard for the leader of the free world to uphold (not to mention the electoral equivalent of “nothing up my sleeve” laced with “plausible deniability”), is a bit like crowing “NO COLLUSION” in response to a report that says, explicitly–we’re on page two now–that it doesn’t address the question of collusion at all. But one vanishingly small positive effect on our economy the Trump administration may have actually had is that it regularly reinforced the value and necessity of making lists, especially in a nation with a globally unrivaled capacity to move from finding to forgetting with breakneck velocity. With that in mind, let’s review just a few of the ways the Russians almost certainly–no, more than certainly–significantly negatively impacted Democratic turnout:

• The DNC Surprise. They leaked stolen materials just before the Democratic convention that blunted the typical momentum this critical event is designed to generate, divided the party exactly when it was gathering to unite, and drove substantial numbers of Sanders supporters to vote for Trump, a third party, or no one at all in the fall election. And after the last four years’ worth of revelations, even Trump supporters who sanctimoniously defended this crime at the time as “truth the electorate deserved to hear” have to admit (or turn in their intergalactic “sentient life” cards) that if the Russians had hacked and leaked email from the Trump campaign, losing the election would have been the very least of its worries.  As for anyone who seriously believes the only reason this didn’t happen was that the RNC and campaign were impervious to such an attack, well…[and well…]
• The October Drip, Drip, Drip: Their leaks of more emails and other materials, starting hours after the Access Hollywood bombshell caused even leading Republicans to call for Trump to step down, and continuing throughout the final month of the campaign, distracted voters who would have voted for Hillary in outrage over his behavior, and completely smothered her strong debate performances–in fact, perversely, the better she performed in the debates, the more this convinced viewers of everything negative they were reading from the Russians about her, in effect turning her debating skills against her.
• Target Theft Deployment. They used and shared stolen modeling data to identify individuals, precincts, and psychographics the Dems worried were most likely to defect from their coalition and used social media to target-bombard them with anti-Democratic messaging in a variety of ways.
• Hi, I’m An American.  In recognition that we’re an age of growing (and what’s worse [or better, if you’re a hostile foreign autocrat] justified) establishment skepticism, to the point we’ve become more convinced by each other than anyone else, Putin’s Army created thousands of fake Americans posing as members of key demographic and psychographic groups, even created entire fake organizations for the same purpose, to astonishing effect. For example, a mere six fake American Russian Facebook accounts, leveraging the platform’s propensity to push extreme posts from “friends” to the top of user feeds, were responsible for generating fake news propaganda shared more than 340 million times, and a fake African-American group they created, Blacktivist, generated more hits to its Facebook page than Black Lives Matter.
• Cordycepifying Comey. We now know the Reds planted fake intelligence related to the Hillary email probe–specifically, forged emails that appeared to come from Attorney General Loretta Lynch, promising to go easy on Clinton–that led FBI Director James Comey to conclude she was compromised, therefore compelling him to seize the role of investigation spokesman, which, in turn, compelled him to make a series of public announcements about the probe, each damaging to the Democratic candidate, especially a leaked re-opening of the investigation late in the campaign that turned out to be completely unnecessary and caused support for her to bleed away without sufficient opportunity to recover.
• And more, much more. The link at left is to a book by Kathleen Hall Jamieson, the scrupulously non-partisan–even according to Republicans–co-founder of Factcheck.org; there are, of course, many, many other sources on the subject as well.

Virtually everything above is illegal, all of it unethical. Essentially Trump outsourced all the dirty work he needed for “victory” to a country beyond the reach of U.S. prosecution. With 270+ contacts between the Trump campaign and Russian government figures and cut-outs that we know of,⁴ all of them lied about, plus an unusually lengthy and detailed indictment from the exasperatingly cautious Mueller showing the Russians shared the Democratic strategy documents and voter data they stole with the GOP, plus a bipartisan Senate committee’s finding that the Trump campaign shared polling and strategy with a known Russian agent, who, we’ve just learned, passed it on to the intelligence agencies running Putin’s interference campaign, frankly, no matter what the definition of a “criminal conspiracy” and “beyond a reasonable doubt in a court of law” are, it just prostitutes and destitutes belief to claim (let alone believe) that the campaign didn’t help the Russians in their efforts, and vice versa.  And frankly, if they didn’t, who cares? Voter suppression is theft, and a stolen election is illegitimate, no matter who steals it; RICO statutes were made for situations like this.

It’s been pointed out by the Washington Post that Russia’s efforts weren’t particularly targeted–only tiny amounts of money were spent on WI, MI, and PA in the campaign’s closing weeks, for example.  But this makes the laughingly naïve assumptions that (a) we know the identity of all the Russian accounts buying Facebook ads (because they never use intermediaries or shells, nor do they ever represent themselves as someone other than who they are) and (b) Facebook has done everything it could possibly do to find all of these accounts (as opposed to, say, the minimum required, as defined by its lawyers, to avoid downstream consequences in the unlikely event of a thorough federal audit) and can be counted on to be as scrupulously honest and transparent as the Russians.  We’re talking about a country whose signature is stealth (Solar Winds, anyone?), that never leaves fingerprints unless it wants to be caught (and they did want that–up to a point).  The assertion that they supposedly never spent much money in those three critical states (whose importance they could have learned from picking up any newspaper on virtually any day in 2016), and stopped spending right when they might most be expected to go all-in, should be absorbed with deep suspicion, not exhaled with relief.

Moreover, anyone who knows anything about how social media works (members of our team literally wrote the book on online community-building for all incoming producers at social pioneer AOL; were the first to coin and define the concept of an “online influencer” as well as associated psychographics; brought together more than 1,000 education organizations to promote, discuss, and advance the concept of connected education at more than 6M online locations, etc. etc. etc.), literally anyone with this expertise knows that it’s word of mouth, not ads, that drive traffic online and have the greatest levels of engagement and influence (something the Russians showed they clearly understand in the most recent campaign, when they stopped buying ads altogether). So where’s the Post’s evidence and analysis showing that the tens of thousands of Russians, their Potemkin pages, and their bots, posing as American voters and organizations, weren’t working WI, MI, PA (and Big 10 country) in general–and working it hard, to disinform and discourage Black voters, for example, who were a key target of the Kremlin’s campaign?  Answer: there isn’t any such–evidence or analysis–of course, which is more than a little bizarre when we know that just six of these fake Americans generated “content” shared more than 340 million times.  Though to be charitable, this may just be the result of naiveté or ignorance, given how little hands-on experience our media and political elites have with online community-building of any kind; most can’t even handle comments on their articles.

At this point, you might justifiably wonder why we’re going into this level of detail jousting with the Post again and “re-litigating” Russia 2016 in a series on voter suppression (beyond the fact that Putin et al were heavily involved in said suppression, in a game-changing way).  A: Because it’s part of a pattern very much reflected in discussions of suppression among experts and influencers as well, centered on an attitude of not really not wanting to look too closely, for fear, if we’re being charitable again, of what might be uncovered, and the impact it might have on the credibility of democracy. An attitude, further, of not really wanting to take these dual, overlapping, and therefore mutually amplifying threats to democracy–foreign interference and suppression–as seriously as they deserve and need to be.  Of adopting a pose of world-weary skepticism and hoping no one sees through to the cowardice behind it. Or worse, believing it’s “all in the game,” and that to suggest otherwise is like complaining about the refs–sour grapes. Or worse still, believing election interference by a hostile foreign power and ballot burning are OK as long as it’s on behalf of “our side’ and “keeps that b**** (Hillary in ’16, Kamala in ’20) out of the White House.”

It’s part of a pattern of saying things that are literally thoughtless, putting forward opinions that otherwise sane, rational, even insightful pundits, condultants, and other pooh-bahs floating them like lead balloons would never do if they thought for even a few seconds about how truly ridiculous their analysis is (or if they thought/feared enough of us would read more than the headlines of their pieces before jumping to comments or tweeting away). Just/much like the Anything But Suppression explanations of the 2020 polling results we’ve been putting into our thought processors since the second paragraph of Part 1 of this series (by the end of Part 3, we hope to have made at least a three course meal out of them for you).

Moreover, sometimes, as we re-re-reⁿ-learned last summer, even in a world that over-rewards hand-waving hot takes and quick hits, sometimes you have to take a deep breath, put a microscope on a very specific example, and let it be a microcosm for the whole, to help people really get it–viscerally–an analogy we’re willing to make only because its victims are many of the same people.

In that spirit, and to conclude the Россия portion of this discussion–we promise–we present to you this Russian-generated Facebook ad, apparently the most successful in their 2016 campaign, meaning it received the most “customer engagement”:

The Post includes it in its ‘nothing to see here’ piece, followed by this back-handedly triumphant observation: “Not exactly a strong exhortation to vote for Trump.” Really.  Well, that rather depends on who it was sent to, doesn’t it? The population most of Russia’s ads went to may have been have been relatively untargeted geographically, but it surely wasn’t untargeted in other ways; to literally send these ads to “all of America” would not only have been prohibitively expensive, it would have been foolish. Facebook itself, wanting to retain them as customers, would have told them this–in the highly unlikely event they didn’t already know targeting is by far the most impactful reason to use FB and Google as ad platforms at all.

And whether Russia’s primary goal was to “sow chaos and division” or to elect Donald Trump (plainly a distinction without a difference), it seems pretty clear where a country that’s had serious problems with its own Muslim population would have sent an ad like this. Especially when deception and impersonation were not only core tactics of the Russian campaign, but a capability their GOP allies have been showing off since at least the early ’70s (e.g. do the original 3 AM phone calls in the days before the 1972 Democratic New Hampshire primary bring back any memories?), and how powerful, in particular, it is to represent yourself as someone you know your audience despises, as opposed to presenting yourself straight up, which triggers the production of antibodies against politics, causing a protective glaze to cover the eyes.

It’s been pointed out that the Russians stole so much from the DNC and DCCC that they likely didn’t need any help from the Trump campaign, beyond the occasional “what do u think of the info on the turnout model for the democrats entire presidential campaign,” as if this were somehow exculpatory.  Sure, and if the mob boss doesn’t tell the hitman how to do his job, it means he never ordered the hit, either.  Maybe it even means the hit never happened.  I’ll have to try this the next time I need some plumbing work done; I’ll bring in the guy, let him do the work, and when it’s time to pay up, I’ll just tell him that since I didn’t tell him how to fix the problem, he didn’t really fix it, so I don’t owe him anything (in fairness, a tactic Trump has probably used more than once).  I wonder what will happen if he’s connected to the Russian mob.

Even more nonsensical are the arguments that the Trump campaign couldn’t have colluded because it was too incompetent.  As if we’re most likely to hire plumbers to solve problems we can easily handle ourselves, and least likely to do so when we really, really can’t.  The reality is that the only competence required to bring in a plumber is the ability to pick up a phone; 270+ Trump-Russia contacts convincingly demonstrates that, if nothing else, the campaign had this competence at least. And its broader incompetence only provides even more motive for outsourcing, especially given that the RNC’s hands for plumbing were largely tied.

In the end, the fact–if it is one–that the Trump campaign didn’t tell the Russians, in detail, what to do absolves Trump of nothing (except the obligation to pay FICA taxes and provide other employee benefits) if there’s a quid pro quo involved, and there’s everything but a numbered Swiss account to prove that Putin was rewarded handsomely for his help.  Moreover, even the claim that no actual instructions changed hands should not be accepted until we know how his zelyonye kiberchelovechki knew to attack targets like VR Systems, a small, obscure nine-employee Florida (of course) firm whose poll book software just happened to be used in 23 counties in North Carolina and 64 out of 67 counties in Florida, both crucial swing states.  Which may require finding new grounds to prosecute all the Trump minions who he bribed into silence with get–out–of–jail–free pardons, exactly as we all knew he would.

Are we saying Russia gave Trump the election? I happen to believe this, yes, which is why I’ve never associated his name with the P word and never will, even if he’s one day “elected” again (on this, both the Bible and the law agree). On the other hand, as someone with deep faith in my fellow Americans, I also believe it takes a village of adversities for anyone to lose an election to Donald Trump.  In any case, I’m not asking you to believe it; this is an exegesis of polling bias, not which of the plethora of misfortunes the Clinton campaign faced or self-inflicted really cost her the election.

And when it comes to the question of whether “polling bias” is really “just” suppression made quantitatively visible, it should be much less than controversial to assert that the suppressing efforts of the Russians, combined with those we’ve described on the part of Republican state governments, not to talk of the Trump campaign’s execution of the GOP’s standard suppression handbook, had what can only be considered fairly profound effects.  A Pew analysis shortly after the election found that voter-eligible non-voters actually favored Clinton over Trump by a whopping seven points, more than three times her margin of victory (at least by percentage) in the so-called “popular” vote (here at CP, we simply call it “the vote,” like every other country in the world claiming to be a democracy does, and it will be so referred to from this point forward).  And this doesn’t include the countless number of Clinton voters who were told–or thought they were told–by the Clinton campaign and/or its surrogates that they could vote by text or tweet, believed it (as we surely know now from the ongoing aftermath of 2020, it’s not hard to get partisans to believe far less likely things emanating from voices of authority), and so may not know, to this day, that they were actually “non-voters” in that election as well.

Moreover, as can be seen from the chart below, the non-voters who did know they hadn’t cast a ballot were (wildly) disproportionately from exactly the groups Republican voter suppression tactics have targeted for decades: the young, the Black, the brown, and the poor:

It’s only natural to wonder why this didn’t result in widespread outrage over voter suppression and demands that it be eliminated, especially given Pew’s status as the non-partisan gold standard for this kind of research.  The most benign explanation is that the study came out right after the election, before we knew much, if anything, about the extent of Russia’s interference, and by the time we did, it had been long forgotten; nobody ever really put two and two together.

Outrage was directed instead at the pollsters for “getting it wrong,” outrage Trump used to pave the path to an alternative reality where he really was the most popular president in history and “the people” backed all of his policies, no matter how extreme.  The first guardrail in a democracy was thus breached: subsequently, every time public opinion polling showed that his programs were, in fact, wildly unpopular, Trump turned back the clock to say “the polls were wrong in 2016, they’re wrong again, you can’t believe the polls.” It is difficult to overstate the damage this did to the polity, or the new stretches of Route 666 (aka the road to perdition) it paved in turn, which is why it was so enervating (to put it politely) to hear the same baying for pollster blood again on the night of November 3rd, 2020.

A Tale Of Two Elections: 2018

So what happened on the suppression front between 2016 and 2018?  The creation of a pre-pandemic awareness of the size of the cricket population, mostly.  In other words, not much.  Perhaps the GOP:

• Got cocky once it controlled all three branches of government (the GOP? Under Trump? Cocky?)
• Started to believe its own press clippings
• Started to believe it really spoke in the voice of the people and had won all the elections it did legitimately

Or perhaps it just believed, in consequence of the fat stock portfolios of its leadership, that “pigs get slaughtered,” that one more turn or twist of the suppression vice would cause the electorate to explode in its face.

Perhaps the party leadership–Trump and Putin, that is–just didn’t care about 2018, selfishly in Trump’s case, strategically in Putin’s, not enough, at any rate, to “release the kraken,” because the only candidate they really both cared about wasn’t on the ballot.

Perhaps V&D (VD, for shorts) thought a Democratic House looked like a win-win for the dark side by: (a) potentially causing the rest of us to let our guard down before the election that really mattered, thinking we’d figured out how to stop the overseas hacks and attacks, while (b) providing an excellent target (and excuse for not getting anything done) for the man who’s at his best (and only) running Against, not For, even when he’s the incumbent.

Whatever the reason(s), as the seventh election of the postmodern era of suppression (initially kicked off in Florida, 2000) approached, the party that believes it has a direct line to God rested.

Which isn’t to say there was no suppression in 2018.  The Brennan Center is no doubt correct that it was worse than ever, because none of the previous suppression measures were rolled back and more were added.  But while between 2010 and 2016, half the states enacted new voting restrictions, in many cases sweeping and multiple, between 2016 and 2018, only nine did, and in most cases these were one-offs, and familiar ones at that. Furthermore, the Trump administration, and more importantly, the not-quite-yet-corrupted intelligence agencies, were telling the truth (be still, your heart) when they said the Russians had not targeted the mid-terms for suppression activities.  Above all, there were no new game-changing events like Crawford in 2007, Shelby in 2013, or Russian engagement in 2016.

All this is significant, because in the postmodern era, suppression is an arms race, dependent for its success on moving goal posts faster than suppressed populations and their supporters can adapt, turning the definition of “likely voter” pollsters use to make predictions into a platinum-plated brass snitch.  When the GOP took its  Fuß aus das gas, it allowed the rest of us to catch up, permitting record numbers of young and minority voters, especially the young, to stream into polling places and actually vote in numbers that reflected their preferences.  In fact, Pew found that more than half the difference in Democratic support between the 2016 and 2018 elections came from voters who did not vote in 2016, which is historic, considering that the often Sisyphean challenge for politicians in mid-terms has always been to get even your most reliable voters to vote at all.

Furthermore, like low tide, low turnout elections often reveal much that might otherwise be concealed, in this case a coup de verace for any theory other than suppression purporting to explain polling error. Take a look at what happened in the nine states that did impose new voting restrictions between 2016 and 2018, how their poll numbers compared to actual results in a year when so-called “Democratic polling bias” virtually disappeared nationwide:

Particularly telling is what happened in Indiana–where multiple suppressive laws were passed between 2016 and 2018–and Missouri, where only one such law went into effect, but a huge one, turning the state from one where no ID was needed to one where strict photo ID was required to cast a ballot.  Despite more polls being conducted in both states than any of the other seven, the “polling error,” aka “Democratic bias” was at least three times greater in both than in any other state, not to mention remarkably consistent.

Moreover, some of the states where less “bias” was observed are classic rule-proving exceptions we’ll be discussing in more detail in part four of this series. In Georgia, 2018 was Year IV of Stacey Abrams’ heroic, and ultimately successful, fight to help Democratic voters overcome egregious suppression in her state, efforts that already, in 2018, came within a few dumpsters worth of burned ballots of electing a black woman governor of a southern state, of any state, in U.S. history. In Wisconsin, Ben Wikler moved on–or rather, back–to his home state and town (he would be elected chair of the Democratic Party in a landslide early in 2019) to help Tony Evers take down Scott Walker and break the GOP stranglehold on state politics. On the flip side, it should be noted that the result in Arkansas, because it’s based on a single poll, is comparatively meaningless.

Is it coincidence that five of these nine states had vulnerable Democratic incumbent senators up for re-election in 2018?  If you think so, you may know jack, but you don’t know ALEC.  Overall, we’d call this little set of 60+ state polls our smoking gun, except that when it comes to suppression, the voting booth is so smoke-filled, it’s hard to tell which of the weapons has been fired when; in fact, it’s hard to see the weapons at all, which, as with Trump’s daily spews of flak, is likely the point.

A Deeper Dive

The data source for the analysis you just read is part of a much larger mother lode of 6,000+ polls fielded in the three weeks before national elections between 1998-2018 helpfully provided by CNN’s Harry Enten in a piece he wrote about 2018 polling.  We decided that rather than just taking the media’s word for it, we wanted to see for ourselves what’s happened to polling accuracy over time.  Our results may differ from others’ you may have seen because we’ve included every poll in the database, and simply averaged them, rather than applying various weights based on perceived (or actual) bias of the pollsters involved.  We’ve done this for three reasons:

• The ultimate purpose of polls:  In our view, the ultimate justification for public polls to exist, despite their potential to warp the political process, is as a check on political corruption–they provide data to the people, beyond the reach of national, state, and local authorities, about our true preferences.  When polls diverge substantially differ from results, we believe it should be seen as a red flag to the electorate and the media that something untoward may have happened, something that, as a minimum, requires thorough investigation and explanation–and we know from international experience and our own recent history that such forceful investigations will not happen unless we, the people, are fully informed and demanding them.  In this sense, making distinctions between “good” and “bad” polls is meaningless, even counterproductive–the public is just as likely to see (and therefore be influenced) by “bad” polls as they are by the “good” ones; in fact, for reasons we’ll discuss below, in a presidential year, in particular, they’re more likely to see the “bad” ones.
• The raison d’etre of our series/analysis:  We aren’t trying, in arrears, to make the polls look better or more “accurate” than they were; we’re trying to assess the impact of a known evil in our political system on a vital part of the political process, an evil that’s an existential threat to the work of all pollsters, both good and bad.  Moreover, if we’re right about the true cause of polling inaccuracy, those pollsters who’ve found a way to “fix it” to get “more accurate” results are only making the real problem worse, as we’ll discuss further downstream.
• Total transparency: We have a crisis of expertise in our country and, sad to say, the experts bear a large portion of the blame for it.  As one of those “experts” myself, at least where online surveys are concerned, I’m well aware that there are advanced statistical tools I could use to make pretty much any dataset say whatever I want it to say (and the temptation is strong!); the problem is, at this point, the public is well aware of this reality too. So, as much as possible, we want to limit our manipulations (which, in modern expert culture, are too often based on unspoken and unrevealed assumptions) to the most simple and basic calculations, like averaging a set of results or, at most, typing =CORREL into an Excel spreadsheet (try it!).  In particular, we know this approach will be critical to persuade the persuadable on the right side of the aisle in the church of democracy; it’s one of the reasons why, to liberals’ never-ending bafflement, conservatives trust Real Clear Politics (RCP) more than they do FiveThirtyEight and feel non-partisanly justified in doing so.
  • For similar reasons, in any cases where we do have to do manipulations of any kind, we’re going to do our best to spell them out completely so you can decide for yourself whether you think what we’ve done is legit.  In other words, we manipulate, you decide.  Seems like that should be somebody’s motto already.

In that spirit, here’s exactly what was done to the historical data set you’re about to see, initially by CNN, then us (feel free to skip this if you already know, trust us, or both–admittedly, under other circumstances it would only be a footnote).  For each of the 6,000+ polls:

• CNN first subtracted the pollsters’ projected Republican result from the Democratic.
  • This results in a positive number if the pollster thinks the Democrat is going to win, a negative number if they believe the Republican will prevail.
  • We call this “the polling gap.”
• The network then looked at the actual results of the election in the same way–subtracting the percentage of the vote the Republican received from the Democrat’s percentage.
  • Again, if the Democrat wins, this results in a positive number; if the GOP triumphs, the number is negative, which isn’t meant, even subconsciously, to reflect negatively on our Republican friends; it’s only this way because we’re trying to measure polling bias in favor of the Democrats, so gaps in the Rs favor should be positive numbers.
  • In any case, by performing this exercise on all the polls, we get an “electoral gap” between the candidates covered in the poll.
• Next, Enten & Co. subtracted the electoral gap from the polling gap for each poll, and the following scenarios come into play:
  • If the pollster is dead-on, not only in terms of percentage but which party prevailed, the result of the calculation is 0, and the poll unbiased
  • If a Democrat who was predicted to win wins, but by less than expected, the difference between the pollster’s prediction and the election’s result is a positive number, which indicates a Democratic bias
  • If a Democrat who was predicted to win wins by more than expected, the result will be a negative number, reflecting a Republican bias to the poll, since the ultimate electoral gap in favor of the Democrat is larger than the original polling gap was.
  • If a Republican who was predicted to win wins by less than expected, the result will again be negative, meaning there was Republican bias in the poll, since the electoral gap is smaller than the polling gap and subtracting a negative number is the same as adding a positive one
    • E.g. if the Republican was favored to win by 10 but won by only two, the Republican bias of the poll would be -10-(-2)=-10+2=-8, and the Republican bias is 8 points. 
  • If a Republican who was projected to win wins by more than expected, the result will be positive, reflecting Democratic bias.
    • E.g. if the Republican was favored by 2 and ends up winning by 10,-2-(-10)=-2+10=8; the Democratic bias is 8. 
  • In the rare cases where the Republican is projected to win and the Democrat does instead, or vice versa, the pollster is a short-seller who just got squeezed
    • E.g. if the Republican was expected to win by 10 and the Democrat wins by 2 instead, -10–2=-12, and the poll was Republican-biased by an embarrassing 12 points.
• Having graciously let CNN do all the heavy lifting, we next took all the polls from the database for a given election year and added them together, which really means taking a bunch of positive numbers (polls with Democratic bias) and subtracting a bunch of negative numbers (polls with Republican bias).  If the net number, the sum, is a positive number, the polls for that year had a Democratic bias; if the sum is negative, the pollsters overestimated the Republicans instead.
• Finally we divided this sum by the number of polls taken that year to get an average bias per poll, which allows us to compare different years to each other on an apples-to-apples basis even though the number of polls may fluctuate from year to year.

So what do we get if we perform these steps on every one of those 6,000+ polls? This chart:

As you can see, for most of this time period, from 1998 to 2012, there actually wasn’t a “Democratic polling bias;” if anything, the polls were overly sanguine about Republican fortunes.  The polling industry, always trying to take responsibility and fix things, would probably say it was during this time period that it developed sophisticated methods to better capture hard to reach populations that were traditionally, in the main, Democratic (and still are!).

But it was also a period when the Democratic “ground game,” still relatively unfettered by suppression tactics, reliably turned out what were then extraordinary numbers of voters, especially after Barack Obama and his team seized control of the job from the party, culminating with an Election Day exclamation point in 2012 when Mitt Romney, by far the most successful businessman the GOP has ever nominated, a proven organizational wizard, savior of Olympics and failing companies alike, suffered a complete meltdown of his GOTV software and campaign.

If even Romney couldn’t beat the Democrats on the ground, thought the GOP, well, the autopsy was nice, but something clearly had to be done––or more accurately, ALEC–accelerated––to start shaking the good earth beneath certain unfriendly zips (no doubt those old redlining maps came in handy).  In the words of Tom Lehrer’s old nuclear proliferation/arms race ditty:

The lord’s our shepherd says the psalm
But just in case, we’d better get a bomb

I vualya, by 2014, as shown, Democratic voters had begun dramatically slip sliding away on their way to the franchise, a propensity reaching Richter scale levels (via helpfully manufactured fault lines) in 2016, before regaining their balance in 2018 for the reasons we’ve discussed.

Since our focus is on the Biden-Trump tale of the tape, we thought it might be useful to separate presidential from congressional polls to see if any further insights fall out:

Amusingly, not only did so-called “Democratic polling bias” commence in earnest nearly a full year before the phonom conventional wisdom has claimed as its cause had even declared himself in the running, but if the CW’s logic is to be believed, in the bellwether 2016 contest, apparently it was Mitch McConnell and Paul Ryan who deliveranced those Panhandle backwoods voters from the political dead. Because based on our no-frills analysis of 800+ 2016 pre-election polls, pollsters underestimated the GOP more in the congressional elections than in the presidential race (and if you believe McConnell’s charisma, rather than suppression, was the cause, we have anywhere in Florida to sell you–get it before it’s hot). 

It’s also worth noting how little daylight there’s been in general, in the past two decades, between congressional and presidential polling “errors” during the presidential years when they coincide, notable because 90%+ of the congressional polls in CNN’s database were conducted within a single state or congressional district (91.5% to be precise). Notable also because the mantra after the 2016 election–protecting pollsters and establishment alike from confronting the real reason for that year’s alarming polling disconnect–was that “the national polls got it right, it was just the state polls in a few critical states that spoiled the whole bunch, baby” 

Whether to help this explanation go down with a minimum of ego, or simply as a matter of charity, this alleged poltergeist in the machine was, in turn, typically chalked up to the inherently smaller sample sizes in state polls (as opposed to, say, local yokel incompetence). And it’s true, smaller sample sizes result in wider confidence intervals/error ranges, as well as some potentially wonderlandian distortions, b/c pollsters working with smaller samples are less likely to be able to fill the required demographic slots, resorting to “weighting” instead.  In one particularly infamous 2016 case involving a popular online panel-based national pollster, for example, a single nineteen-year-old Black male voter in Illinois, who happened to be a fervent Trump supporter, was regularly pushing up Trump’s national numbers with all African-Americans nationally by a full point in the poll every time he participated.

But when even polls limited to a single Congressional district have been holding their own (in comparative accuracy) with national presidential surveys for the last five–spoiler alert, six–presidential elections, it’s really past time to stop scapegoating state and local pollsters to avoid facing the truth.

Diving Deeper

Whenever you get results from a data analysis that support your hypothesis as cleanly as what we’ve just presented–even as it ties alternative explanations up in Gordian knots–it’s only natural to listen for the sound of shoes as they fall, possibly even whistling past your ear.  It’s indeed rare to see a chart so fulsomely resemble a narrative line. So we had to ask ourselves: what if this is all a mirage, an artifact of the work of some other external force? What if, in particular, it’s just a dance of veils concealing the most obvious candidate, changes in the quality of the polls themselves?

If you’re not particularly fascinated by the polling industry, we’ll just tell you the answer (spoiler alert again): it wasn’t and it isn’t. And we’ll help you skip ahead if you’d like.  But otherwise, to answer this question, we:

• Sorted the entire data set from 1998 to 2018 by year and pollster 
• Looked up the rating of every pollster in the data set on Five Thirty-Eight
• Converted Five Thirty-Eight’s ratings, which take the form of letter grades, into numerical values, using the crazy quilt of grading systems we’ve collectively been exposed to in public and private high schools, colleges, and universities.  You can see what we ended up with on the right.
• Assigned a numerical value to each poll based on the pollster’s current rating⁵
• Added up the poll values for each election year and divided the sum by the number of polls fielded in that year to arrive at a “poll/pollster GPA” for that year.

We could just share those GPAs with you at this point, but in the interest of our commitment to total transparency, and because when you’re talking about a People of the Numbers, even dry charts reveal a rollicking, bouncing yarn ball of an industry whose recent history seems positively cinematic (with Gerald McRaney playing the head of Remington Research, maybe playing them all). If Nate Silver and Michael Lewis pass on writing the page-turner and screenplay, maybe I will–so here’s the my coverage and initial draft dramatic treatment:

That’s a lot to take in without the benefit of the hours we had assembling it, so let us share some quick impressions we gleaned from our could-be canary, aka the polling industry, before we cut to the horse race:

• Good God, at least 356 organizations fielded presidential and/or congressional polls in the last two decades in the last three weeks before a national election alone.
• The dynamism of the sector makes monopoly capitalist apologia look glacial by comparison–less than one in six of the polls fielded in 2018 (16.6%) were fielded by pollsters who also fielded them in the last three weeks of the 1998 election, suggesting a highly competitive free market with low barriers to entry in the space (which is good, provided it doesn’t mean a lot of folks continually trying to re-invent the wheel–keep reading).
  • To save you time, the ’98ers still standing included ABC News/Washington Post, CBS News/The New York Times, Fleming & Associates, Fox News, Gallup, Harris Insights & Analytics, Marist College, Mason-Dixon Polling & Strategy, NBC News/The Wall Street Journal, Public Opinion Strategies, Quinnipiac University, SSRS, and Survey USA.
  • There are only eight organizations (most of them collaborations) that have fielded polls in the stretch run of every national election since 1998 inclusive (which ain’t ideal for longitudinal analysis)–ABC News/Washington Post, CBS News/The New York Times, Fox News (and various polling firm partners), Mason-Dixon, NBC News/The Wall Street Journal, Public Opinion Strategies, Quinnipiac, and Survey USA.
  • There’s fortunately a bit more stability, at least when it comes to repeat defenders (of democracy), in presidential years: Nearly one in four (23.8%) of the polls fielded in 2016 were conducted by (nearly twice as many) pollsters who were involved in the 2000 election as well.
• There are a lot more polls (more than double–+130%+) fielded by more pollsters (+20%+) during presidential years than during midterms.  This isn’t surprising, given the much higher levels of interest in presidential elections, but the ensuing scramble for mindshare, with everyone and their psychographic coming out to poll, has its costs.
• As a general rule, the more polls a pollster fields (presumably with a primary focus on maximizing eyeballs), the lower their quality;⁶ as a result, other things being equal, the average voter sees more bad polls than good ones in any given election year.
  • There are notable exceptions, but they may only help prove the rule–for example, A-rated pollster Survey USA almost single-handedly rescued polling GPA in the 2004 & 2008 elections by churning out 117 and 101 in the last three weeks of those election years, respectively, but has declined significantly in the number of polls it fields since, down to 13 in 2016, and 8 in 2018, suggesting that keeping up with the Kardashians of polling can lead to serious burnout or reconsideration.
  • The combination of more (and less-experienced) pollsters churning out lots more polls suggests there ought to a Bourdainian rule about looking at polls in a presidential year, especially since, unlike the fish entrée on the Sunday brunch menu, the multitudinous grains of salt required aren’t already baked in.
  • Election polling appears to be a growth industry, especially in off-year/midterm elections: there were 70%+ more pollsters fielding more than double the number of polls (+139%) in 2018 than in 1998, while 20% more pollsters fielded 30%+ more polls in 2016 than 2000.

• At the same time, there are so many one-and-dones in the space, including the aptly named Singularis, you can almost hear them saying “blimey, this polling thing is more difficult than we thought”
  • At regular intervals, it seems, a behind-the-machinator, like Steve Bannon, Richard Wirthlin, or Doug Schoen, pops to the public surface briefly, then disappears again.
• A number of the best pollsters are relatively unknown colleges, but many better-known schools have–much to conservatives’ delight, no doubt–proven to be shockingly bad. We love you, but [insert Brown joke here].
  • Perhaps this also accounts for the number of schools in the past twenty years who have conducted one with reasonable accuracy, then strolled away stretching their legs like a cat walking away from its bowl.
• Pew gets only a B/C grade? Really? Miaow?
• Like clockwork, it seems, every presidential election some new moda appears, with some new theory of the case, presumably, and spews out a massive number of poorly constructed instruments that drag that year’s GPA down (e.g. Survey Monkey and Lucid in 2016; TCJ Research, Pharos Research Group & Google Surveys in 2012; Research 2000 in 2008, eight years after its expiration date; Strategic Vision in 2004; and, to a lesser extent, Harris, Rasmussen, and Zogby in 2000)
• There are some words that, when they appear in a pollster’s name, apparently inherently flag their work as not top-shelf.  No firm with the word “marketing” in its name gets better than a B/C grade, for example (though Market Shares Corp gets a provisional A/B), and a B- is the best any group with “opinion” in its moniker can do.  Is any potential customer reassured by knowing that its pollster is “lucid?”  And don’t you just know a company who feels the need to slip a buzzword like “big data” into its brand probably doesn’t know what it’s doing?
• There are potential storylines everywhere you look.

And what all this (and more) boils down to is the distinctive pattern shown in the graph below, tracking pollster GPA against election year:

As you can see, it appears pollster performance, or rather the quality of the mix of pollsters polling the last three weeks before national elections (on which their performance was judged), was in slow, steady decline for the first decade, plateaued during the Obama years, then started gyrating wildly like weather in response to a warming atmosphere; in other words, as if an outside/external force were involved…such as….? 🙂  Note also the consistent differences in GPA/quality between presidential and midterm elections–there’s literally no case in which pollster GPA in a presidential year is better or equal to that of the midterm before or after; in every case it’s worse.

If we now map pollster GPA against “Democratic polling bias,” here’s what we get:

The key data points here are 2012 and 2014, the inflection period when “polling bias” decidedly flipped from Republican to Democratic.  If this were simply a function of the quality of the pollsters involved, not a real phenomenon caused by forces outside the polling process, you’d expect the relatively large increase in Democratic bias in 2014 to be accompanied by a corresponding decline in pollster quality; instead pollster GPA in ’14 was the highest it’s been since 1998, when few beyond polling’s titans were involved.

This alone is a fairly fundamental proof that changes in polling quality can’t explain the pattern and history of “Democratic bias” (or lack thereof), but you can also see a number of instances where GPA and bias seem to be, at best, loosely connected, which is reflected in the correlation between the two (-0.23), i.e. the higher the GPA, the lower the bias, and vice versa.  This is a level of correlation that does indicate that polling bias and pollster quality are connected, and later you’ll see us prick up our ears at lower numbers, but given that:

(a) 538’s pollster grades are determined by cumulative past performance–therefore so is pollster GPA, and

(b) evaluation of past performance, in turn, is based on the accuracy of the pollster (i.e. each pollster’s level of bias),

pollster GPA is something of a (semi-)dependent variable determined by bias levels and, as such, the correlation between the two ought to be a lot higher than -0.23 if changes in Democratic bias are really being determined by changes in polling quality.  Especially given that we already know some level of cause and effect runs in the other direction, i.e. that bias (up to, but not including, the election year one is looking at) impacts GPA, which, in turn, is ultimately determined by the mix of pollsters and the number of polls conducted by each (not just by each pollster’s past performance), which, in turn, is why the bias v. quality comparison as defined is viable, not just a fool’s errand.

Out of curiosity, we also decided to compare GPA against “absolute bias,” meaning the magnitude of bias in either direction, Democratic or Republican, above or below zero. Here’s what that looks like:
Interestingly, in this case, the connection is even weaker, as you can easily see, both from the graph and the much weaker correlation in the lower right hand corner, which argues further against any possibility that pollster quality, rather than an outside force like suppression, is determining, or even shaping, polling bias.

A Tale Of Two Elections: 2020

Which brings us to the election we just survived. By 2020, Republicans had:

• Controlled a majority of governorships since the 2010 election and enjoyed a 2020 edge of 27-23
• Controlled a majority of state senates and legislatures since 2010 as well, with a 59-39 chamber advantage heading into election year,
• Also held a majority of secretary of state offices, typically responsible for running elections, for the past decade, and effectively led the Dems 27-21 in this category as campaign season got underway (in three states, the Lieutenant Governor carries out the SoS’s duties; in two of those states, the Lt. Gov. was a Republican).

Philosophically, the goal of the postmodern GOP has been to win elections before they start, so let’s look at the lay of the voting landscape as 2019 rounded the turn into January 2020, to see how well they did over their decade of opportunity to rig the decennial contest for all the marbles (which we were also at risk of losing) by making it more difficult to vote, particularly in the swing states that would decide the election.

Isn’t it a little harsh, a little partisan to call it rigging? asked the pot. Not really, not if you’re truly an “originalist:”

rigging: (v) 1. to fix in advance for a desired result (Webster’s, dictionary of the Founders) 2. manage so as to produce a result or situation that is advantageous to a particular person (Oxford, dictionary of the world). cf. perhaps Scandinavian, see Norwegian rigga, to bind or wrap up.  (emphases mine)

“To bind or wrap up;” that sounds about right, though a little cozy or kinky for suppression. “Fixing in advance for a desired result,” yes, most of what ALEC wrought for states the GOP controlled would certainly pass the Anatole France bitter laugh test, though much of the handiwork has been so brazen it more likely would leave said Nobel laureate, known for his nobility of style, grace, and compassion, tight-lipped or slack-jawed, flushed and speechless.

Keeping 50 states, and potentially dozens of parameters, in our heads for the analysis we wanted to conduct was an exercise that had COVID brain written all over it, so we decided to look for a data set we could turn into a simple color-coded map for us all.  We found one in The Guardian, in which they assessed ease of voting–state by state–by looking at:

• The nature of its voter ID laws (no ID vs. no photo ID vs. photo ID vs. strict photo ID requirements)
• Its voter registration mechanisms (automatic; same day/Election Day; before Election Day; proof of citizenship/exact match required)
• How it treats ex-felons and prisoners (they can vote when they get out, or even while they’re still in; there are parole or probation restrictions on when they can vote; some or all are permanently disenfranchised)
• Availability of early/absentee voting (available with no restrictions; available only with approval; no early voting at all, but absentee with approval)

We converted each of these metrics into either a 1-4 or 1-3 scale, depending how the publication rated them, with 1 as the easiest state of affairs, 3 or 4 as the hardest where that metric is concerned, then added and averaged each state’s numbers across the measures.

• So, for example, Utah requires a photo ID (3) and requires registration before Election Day (3), but allows all ex-felons to vote without restrictions (1), and offers early/absentee voting with no restrictions either (1).  3+3+1+1=8, 8/4=2

Finally we color-coded each state’s average in a shade of green, with light green representing easy, and dark green representing hard. States averaging 1-1.25 were the lightest, then those averaging 1.5-1.75, 2-2.25, 2.5-2.75, and 3-3.25, for five shades in all.  If you do all this, this is the map you get:

Pretty, pretty dark (except for Puerto Rico, which is just hanging out)–especially in the Midwest and Southeast. And if we next put the focus specifically on the swing states, coloring the rest with the optimal shade of green, things look very dark indeed:

Quantitatively, not a single swinger makes it easy to vote, 11 of the 13 are in the middle category or worse (which doesn’t mean any of them are “middle of the road”–overall, the states aren’t ranked, so there’s no reason other than immoral why they couldn’t be all light green), and three are in the bottom two categories, which just shouldn’t be the case in an allegedly democratic election where they could be decisive. All of which suggests 2020’s polls actually started skewing while there was still driven snow on the ground.

Of course, the four factors The Guardian singled out aren’t the only ones that determine how easy or hard it is to vote in a state, and when analyzing something as loaded as voting rights, it’s always good to get a second opinion, especially when the first one might be shrouded and obscured by fog on the pond between the special relationship. Fortunately, Northern Illinois University has been conducting election year studies of the “total cost to vote” in every state for some time now, by definition producing a more comprehensive and sophisticated, possibly significantly more accurate, view of the frontlines of the franchise.

Unlike The Guardian, NIU ranks the states from 1-50, which in theory allows for finer gradations and distinctions, but in interest of comparability, and with healthy respect for the corpus of cognitive research showing that juggling more than five pieces of input/info simultaneously is way past our bedtime, we’ve again divided their rankings into fifths, so states ranked 1-10 are the lightest green, followed by those ranked 11-20, 21-30, 31-40, and 41-50.  Because the states are being compared to each other, not some ideal standard, the map isn’t as dark and gloomy as The Guardian’s (the Brits are such Eeyores), but we think it’s even more revealing:

It’s more than a little damning, for example, to see swing-state New Hampshire as a little island of near-black in a sea of light green all around it, and side-eye swing-states Pennsylvania and Ohio keeping company with the Deep South, all while the West, which led the way on women’s suffrage more than a century ago, continues to lead largely on its lonesome–trailblazing in support of what should have been settled once both/all genders had the right to vote in 1921–leaving the rest of the country, including most of the swings, in its woke wake.

And when we isolate said swing states, the picture gets even clearer–and darker still.

Once again, no swing is in this scale’s version of the “easy to vote” category; eight of the 13 are in the bottom half of the rankings, with two others, Wisconsin and Iowa, where the pollsters were most “wrong,” on the cusp of joining them at #24 (IA) and 25 (WI).  Three of the five hardest states to vote in are swing states as well, including the very bottom of the barrel, #49 and 50. Not surprising, while the average non-swing state’s ranking is 23.4 (23rd of 50), the average swing stumbles in at 31.4 (31st out of 50).

At the very bottom is the second biggest electoral prize of all (and growing), Texas, one of six states plus DC in which Black and Latinx voters are in the majority, yet seemingly constitutionally incapable of electing a Democrat to a single statewide office, a state of affairs made all the more remarkable by the fact that our other majority-minority states are California, Hawaii, Maryland, Nevada, and New Mexico, in all of which virtually every statewide office is held by a Dem (Maryland’s governor being the exception that proves the rule).  There could not be a better casus belli for both the severity and impact of suppression than this contrast.

Yet astonishingly, Democrats and the media continue to insist this is just an artifact of a different color, that Texas Blacks and Latinx voters are somehow a different species (“not monolithic” is the preferred euphemistic finesse these days) from their ethnic brothers and sisters in the rest of the country, even those in neighboring Nuevo Mexico, that the problem is “well, those [insert pejorative here] POCs just don’t want to vote,” uttered with a helpless shrug (and the same level of conviction with which it used to be argued Blacks don’t really want to be free).

Meanwhile, as this argument is made in earnest, the industriousness with which the state, aka the Republican Party, has striven to remain the world’s largest plantation, even going so far as to pioneer unconstitutional⁷ re-redistricting in the middle of a decade after a single election in the their favor (like a thief claiming nine tenths of the law), no longer even merits a raised eyebrow; it’s just Texas being Texas, we seem to reckon.  And in fairness, in many ways it looks like TX is really just in the flow of traffic formerly known as white flight.  After all, the next state likely to go majority-minority, Georgia (52.7% white), ranks next to last in ease of voting, and has been making some Texas-sized news itself on the subject in recent weeks, and if not GA, it could be the dark continent of Florida (#40, 53.9% white), aka the Christopher Columbus, George Washington and godfather of modern suppression, and if not FL, then maybe Arizona (#30, 54.7% white), ground zero for the ongoing efforts to overturn the results of the 2020 election and some of the most cynical new voter suppression laws in the country.

Still, Texas is where it always wants and believes itself to be in everything, #1 (a 50 being just a 1 turned upside down), and in light of this, it’s with extra authority (and irony) that the state of play heading into 2020 may have best been summed up by the first Texan to become president, Lyndon Johnson, once reviled but now regarded as one of our greatest, when he said:

“You do not take a person who, for years, has been hobbled by chains and liberate him, bring him up to the starting line of a race and then say, ‘You are free to compete with all the others,’ and still justly believe that you have been completely fair. Thus it is not enough just to open the gates of opportunity. All our citizens must have the ability to walk through those gates. And this is the next and the more profound stage of the battle for civil rights.”

He was speaking in support of affirmative action, and what more poignant testimony could there be to how far we’ve fallen from our ideals that we can read it, instead, as a call for something so much more basic, a true right to vote.  Thus was the table set, as primary season got underway.

We’ve emphasized the importance of “game changers,” given the arms race dynamic that makes modern suppression possible, an endeavor that can as well be likened to trapping and containing anti-matter with electromagnetic fields in constant motion. In early 2018, there was a development Republicans themselves called “a huge, huge, huge, huge deal“: the end of the DNC vs. RNC consent decree, originally signed in 1982, precipitated by the GOP’s Ballot Security Task Force.  This infamous organization, in the name of preventing “voter fraud,” had posted signs all around minority voting precincts, warning residents that they were “PATROLLING” the area, and reminding residents that “IT IS A CRIME TO FALSIFY BALLOTS OR TO VIOLATE ELECTION LAWS.”  To ensure would-be voters got the message, members of the group, made up of off-duty police and sheriff’s’ department officers, and organized by none other than Roger Stone, made themselves highly visible, wearing BSTF armbands (like you know who) whilst brandishing “prominently displayed” revolvers and two-way radios.

In an eerie, but now all too familiar, preview of 2016 and 2020, not only did pre-election polls show the Democratic candidate, Jim Florio, with a healthy 8 point lead, but exit poll results caused the networks to call the race for him shortly after the balloting closed, only for Republican Tom Kean to emerge as the winner by 1,700 votes when all were counted.  Yes indeedy, ask any dictator or strongman who’s won with 90% of the vote or higher: this intimidation stuff really works.  When Kean was declared victorious, the Democrats sued (not, it should be noted, to overturn the results–some things never change), claiming voter intimidation prohibited by the Voting Rights Act, and the Republicans were pressed into an agreement that they would not engage in “activities that suppress the vote, particularly when it comes to minority voters” (which is why they were only too happy to have the Russians do this for them in 2016), and to submit all “ballot security” programs, including “voter caging” endeavors, to the federal courts for review before implementation.

By total coincidence, the consent decree that’d been in place ever since was scheduled to expire shortly after the 2016 election. Democrats attempted to extend it, the extension was denied, and 2020 became the first presidential election since 1980 when Republicans had free reign to engage in intimidation tactics.  Even worse, because Democrats, in seeking to extend the decree, had specifically cited Trump’s 2016 cries that the election was “rigged” and his incitements to his followers to watch polling places for voter fraud, a plea that was rejected, the GOP took the decree’s demise as an ‘already litigated’ legal open license, even though the judgment (written by an “Obama judge”) was not that such tactics were OK, only that the Dems had failed to prove that the RNC, the signatories to the original agreement, had colluded with the Trump campaign in Trump’s, or his campaign’s, execution of them.

The chortling establishment/MSM wisdom about voter intimidation in 2020 is that Trump’s army of poll watchers turned out to be “more like a small platoon.”  No biggee, much ado, etc.  Like other such smug, derisive post-election sighs of relief, this is dangerously naive. The goal of a voter intimidation campaign is not to literally prevent the opposition from voting, to physically blockade their access; it’s to deter them, of their own “free will,” from ever showing up. When the GOP distributes ominous flyers, as they’ve done, illegally, decree or no decree, for years, warning “undesirables” law enforcement will be at the polling places looking to nab citizens with outstanding warrants, back taxes, or unpaid parking tickets, the goal isn’t to get those at the receiving end of these missives to come out to the polls to find out if it’s true.  All the publicity sought by Don Jr for his Army, and reliably provided by the corporate media, is no different than those flyers, except that it’s worse, as it comes from a more trusted source than an anonymous flyer, through a medium (television) where it’s more likely to be seen and processed.

Sure, if Trump and his campaign weren’t so incompetent, they would have preferred that the army turned out in force–nobody believes more than Donald Trump that nothing succeeds like excess, but the truth in advertising of the matter is that by Election Day, the most potent mix of ingredients for an effective voter intimidation campaign in decades had long since been brewed, shaken, and stirred, to the point where it didn’t really matter whether Trump’s Army showed up or not. Specifically:

• There was the end of the consent decree itself, the anticipated impact of which the GOP made no secret of, even if the infamous “huge, huge, huge, huge deal” quote was (allegedly) secretly recorded–and why would/should they?  They knew they could count on the “liberal” media to run with that (it was new(s), after all), triggered by four years of trauma and abuse to repeatedly spray a toxic blend of their own alarm and top-shelf hype, especially the key core message that the game had changed, that shackles had been removed, fists ungloved, that something this way was coming that no one had ever seen before; project onto that what you will, dear voter, ideally from your subconscious or horror flick conditioning.
• There were five years’ worth of incitements to violence, regularly and skillfully compiled (by the media) into montages that began to be widely shared as early as the Republican primaries during the 2016 campaign, and continuously thereafter, montages that helpfully often included not only his incitements but his rally–goers acting on them in real time, and a disturbing number of examples, also well–cataloged and distributed, of individuals and groups acting on those incitements in more extreme, more violent ways.
• Additional air cover was provided by another skillful and endlessly repeated Trump quote, well–reflecting his background in organized crime, that he had “the support of the police, the support of the military, the support of the Bikers for Trump” that he had “the tough people,” who “don’t don’t play it tough — until they go to a certain point,” at which point things were going to get “very bad, very bad.”
• Coprageously, even the more widely shared montages of police brutality collected by BLM activists at George Floyd demonstrations doubtlessly fed fears among voters, as did:
  • The full-throated endorsements he received from police, sheriff, and other state and local law enforcement organizations
  • The brutality of his actions against demonstrators in DC, in particular for the sake of a ridiculous photo opp, which bring to mind what was said of the top gangster at the heart of The Sting, i.e. that he’d “kill a grifter over a sum of money that wouldn’t feed him for two days,” an approach to power that anyone who has lived in a neighborhood with significant gang activity is not only familiar with, but has likely at least brushed against and/or seen the consequences thereof
  • The involvement of the military and use of aggressive military tactics (e.g. those low-flying helicopters, also endlessly replayed)–with leaks of worse, such as the proposed use of a microwave-based weapon that the US Army had declined to deploy in Iraq and Afghanistan out of concern it would be viewed as torture
  • The deployment of tactics in Portland, OR that are time-honored only among the shock troops of dictators, such as snatching up of protesters into unmarked vans by unidentified, out-of-uniform law enforcement officers, accompanied by emphatic support of right-wing paramilitaries, such as the well-armed Proud Boys caravan that descended into the city, conspicuously cheered on by the so-called president of us all, a line extension of his earlier call for militias to “LIBERATE MICHIGAN!” (not to mention Virginia and Minnesota, all blue states), which, in MI’s case, led first to a preview of the January 6th armed insurrection at the Michigan state capitol, then a plot to kidnap its Democratic governor.

Then there was Trump’s poll watching initiative, touted for months in advance as 50,000+ strong, and not even the only force talked up in the coalition of the willing against democracy, which became increasingly martial as Election Day approached, with Don Jr. tweeting out video calls for every “able-bodied” man and and woman to “enlist” to “defend” the vote as part of the freshly rebranded “Army For Trump.”  Like any would-be dictator seeking to impose himself, there were, of course, related shows of force, beginning with the “platoon” of Trump supporters that descended out of nowhere to disrupt a polling place in deep blue and DC-adjacent Fairfax, Virginia at the start of early voting, which ensured maximum coverage, by virtue of location, and maximum impact, by virtue of timing and improbability, like terrorists hitting up a grocery store in Des Moines. And ending with a caravan of Trump supporters running a Biden campaign bus off the road in Texas just before Election Day, again celebrated by the Occupant.

Beyond all the above, there were intimidation tactics deployed that were unavailable by technology or ethics to the pre-decree GOP.  For example, the mysterious robocall, believed to be foreign in origin (who could that be?), that went out to at least 10,000,000 voters in the final pre-election days encouraging them to “stay home and stay safe.”  I say “at least” because while news reports described this as Election Week phenomenon, Redditors began reporting they were getting these calls in July. And the interference campaign in which Iranians called swing state voters pretending to be Proud Boys (how the mighty have fallen–back in the day, they could’ve just identified themselves as Iranians) telling the unfortunates who picked up what would happen if they didn’t vote for Trump, an effort we’ll have more to say about in Part 3 of the series.

A friend very familiar with both higher education and Democratic campaigning in Texas told us weeks before the election that rural students of color there were being contacted in droves on social media by entities who represented, with proof, they had enough of the collegians’ personal information–and that of their families–to back up threats of what would happen to them if they showed up to vote.  And lo and behold, didn’t Trump end up doing much better than expected with Latinx voters in places like the Lower Rio Grande Valley, to the point where this was viewed as key to his success in the state?  Must be yet another one of those “coincidences” the heroes of right-wing law enforcement shows are always telling us they “don’t believe in.”

From our perspective, this is just a small sample example of suppression–and especially intimidation–efforts that were never covered by the media at all. It’s clear from Rep. Clyburn’s pre-election Fox interview, for example, that there were many such, which is why we’re working to build a Wall Of Shame (currently under development) that, through stories and visuals, will make them more accessible to the media and general public going forward.  If you or someone you know experienced suppression, or experiences it going forward, please encourage them to check it out.

On the flip side, there was at least one intimidation effort that would have to go down as the most public and “above the radar” in American political history, Trump’s instruction to the Proud Boys during the second presidential debate, with more than 60 million watching, to “stand back and stand by,” not “down,” just “back” and “by,” and the aftermath of same.  Never let it be said the man didn’t cover all bases and leave it all on the field, even if he “worked harder” in the three weeks after the election than he’d ever “worked in [his] life.”

All of the above took its toll, especially on the media, which found itself diabolically impickled.  On the one hand, to ignore or even merely underreport wave after wave of these outrages would have to be seen by anyone with a conscience as irresponsible, and not just from a journalistic perspective. On the other hand, to report it with the appropriate level of alarm clearly risks helping the terrorists win.

Under normal circumstances, such a dilemma never arises because of a longstanding symbiotic relationship between press and government–the fourth estate raises the alarm, and the government reassures the public. But what happens if alarm, if fear, is exactly what the government wants the public to feel? When, for example, said government’s hopelessly compromised Department of Homeland Security helpfully issues a widely quoted pre-election “threat assessment” that polling places are shaping up as “likely flashpoints for potential violence?” What happens? We just found out, I think.

You may be an American after my own heart; when a political foe threatens you, you say “bring it on.”  I can’t tell you, for example, how much I’m looking forward to encountering the first Trump supporter who acts on Tucker Carlson’s urgings and tells me to take off my mask.  I truly love my Trump-supporting friends; I actually think they were right to be angry about a lot of what enraged them: the swamp, the size of government, the taxes they were paying, command and control regulation, federalism, how screwed up the immigration system is, how we keep subsidizing countries that ought to be able to pay their own way, and more. “Right message, wrong messenger” I tell them, when I can.  But I feel like I’ve been waiting for four years for a moment like that, for a Trump supporter to just try getting in my face like that.

And sometimes, normal times, when the electorate isn’t as engulfed in flames as California, and turnout is at its usual completely embarrassing levels for the worldwide leader in democracy, one of those periodic fabled backlashes against oppression that warm the heart and make you remember why you’re proud to be an American–Florida 2012, Wisconsin & Georgia 2020–can indeed carry the day.  But in this election, friend, the yous and mes all voted, leaving the fate of the nation in the hands of others.  And we all have what we probably all consider a shocking number of friends, neighbors, colleagues, who don’t have the same life-and-death attitude about politics we do, who care more about things like the personal safety of their loved ones and how they’re going to make ends meet.  Shocking.

Picture one of those less politically passionate (or just confrontation-averse) friends, neighbors, colleagues reading sentences like those we’ve collected below, drawn from just a few of the hundreds and hundreds of articles written about the election in the campaign’s final weeks, focused specifically on what to expect on Election Day:

• “U.S. polling sites are likely to be a battleground”
• “The United States faces its most perilous election since the Civil War”
• “We can expect widespread and potentially violent conflict focused at polling sites on November 3”
• “The specter of people who are violent in nature and have violent agendas, and often come armed with long guns is becoming a very real possibility.”
• “It is likely that significant numbers of people will bring guns to polling places under the guise of preventing election fraud.”
• “This is a case where the last thing we want is for law enforcement to show up”
• “Cities nationwide bracing for potential violence on Election Day that’s peaceful–so far”
• “Concerns about potential polling place violence come at a time of immense stress for election officials”
• “It’s the self-appointed poll watchers — those who might show up at the polls unannounced — that have some people more worried.”
• “We’re going to have sheriffs, and we’re going to have law enforcement. And we’re going to have hopefully US attorneys, and we’re going to have everybody and attorney generals, Trump said.”
• “[The] executive director of the Coalition to Stop Gun Violence said he’s also worried about what might happen if some of those who come to the polls are armed. He noted the increased presence of armed militia at protests and political rallies around the country…
• “Guns Down America issued a report noting that few states have laws expressly prohibiting guns at polling sites.”
• Etc. etc. etc.

Can you not picture people you know who’d read things like this (x100), further amplified emotionally on television and social media, slathered on top of all the mass shootings and other political violence of the Trump years, thinking maybe I’ll just sit this one out? Lulled further to inertia, perhaps, by all those polls showing Biden way ahead?  Oh, and by the way, was the friend, neighbor, or colleague you pictured when reading all those quotes about “likely” “widespread” “armed” political violence Black or Latino/Latina,  by any chance? No?

Furthermore, voter intimidation was, of course, only one tool in a suppression army knife being maniacally wielded with all blades out at once, with effects both cumulative and greater than the sum of parts.  In Part 1, we cited Trevor Potter, a former chair of the Federal Election Commission who served as general counsel to Republican John McCain’s two presidential campaigns, and as we we haven’t found anyone with a greater level of expertise and bipartisan credentials who’s laid out the situation on the ground better, we quote him again, more fully:

“What we have seen, which is completely unprecedented … is a concerted national Republican effort across the country in every one of the states that has had a legal battle to make it harder for citizens to vote… There just has been this unrelenting Republican attack on making it easier to vote.  It puzzles me … I’ve never worked for a Republican candidate who thought it was a good idea to make it hard for people to vote.”

Put more plainly: “Never before in modern presidential politics has a candidate been so reliant on wide-scale efforts to depress the vote as Trump,” and as aggressive as the efforts to suppress were in general, they were that much more so in the swing states where the polls were so “wrong.” Compare the Wikipedia article on voter suppression in 2018 with this accounting of just some of what was going down in swing state Pennsylvania alone in the last two weeks of the campaign alone, including:

• Litigation to prevent ballots arriving after Election Day from being counted, in a state where, like other swings, mail delivery had been badly delayed via political sabotage of the post office (more on this below)
• Challenges to the installation of drop boxes for mail-in voters who, for some reason, didn’t trust TrumpMail.
• Blocking the counting of mail-in ballots until Election Day, as part of what was a widely, and accurately, predicted effort by Trump to take the lead in the state via the results of in-person voting, then declare the deluge of mail-in ballots still being counted late into the night to have been manufactured by the Democrats in the dead of night and illegitimate.
• Blocking even the pre-processing of these ballots–not counting them, just opening them and flattening them so they’d be ready to count, which the GOP was only willing to agree to if Dems had agreed to get rid of all drop boxes, impose new signature-matching requirements, and allow poll watchers to cross county lines so they could flashmob polling places.
• Litigation to block counties from giving mail-in voters any help in correcting mistakes they may have made that would invalidate their ballots
• Litigation to block counties from even informing voters that they’d made mistakes and/or giving them the opportunity to cure them. The GOP would later sue to prevent cured ballots from being counted, using, fittingly, the Bush v Gore decision that handed them the 2000 election, a decision so rank that even those who wrote the opinion said it should never be used as precedent for any other case
• Blocking election officials from accepting ballots that arrived without their inner envelopes, a.k.a. “security sleeves,” which Philadelphia officials warned could result in the disqualification of more than 100,000 votes (20,000 more than Biden’s ultimate margin of victory)
• Widespread videotaping of voters by GOP operatives in a clear, and illegal, effort to intimidate them
• Posting of surveillance images of voters for the same purpose (a ballot is hardly secret if its casting is broadcast to every extremist online).
• Dispatching “poll watchers” to early voting sites and even mail-in drop boxes, where they pressed election officials to block voters from depositing more than one ballot, even if the second one was for a family member too ill (some with COVID) to drop it off themselves
• The scouring of social media for “I voted” images that could be positioned as showing Pennsylvanians depositing more than one ballot at a time
• Repeated false public statements from Trump, as he barnstormed the state, about election security in the commonwealth, accompanied by ominous warnings, like “A lot of strange things happening in Philadelphia…We’re watching you, Philadelphia. We’re watching at the highest level.”
• Robocalls blanketing urban areas warning that voting by mail could subject voters of color to arrest, debt collection, and forced vaccination. This wave of incidents actually happened a couple of weeks earlier than the others above, but was doubtless still reverberating and being spread in the final weeks. Those responsible (may they be sentenced to at least ten years each) were eventually forced to leave a retraction voice mail, on October 29, for an election taking place on November 3rd–ask Hillary Clinton whether that was timely enough.

Interestingly, and more than a little suspiciously, despite all we’ve documented and detailed, with more to come, there have been a number of post-election surveys like this one about how easy it was to vote. Curiously, unlike the post-election Pew study in 2016, the respondent pool, at least for every such survey I’ve seen, has been limited to fellow citizens who actually voted.  While this is more than understandable if the purpose of the poll is essentially usability testing the efficacy of the mechanics deployed to enable voting in a microbiological war zone, it’s questionable, bordering on unprofessional, to leap to conclusions about how easy voting was when you only talk to the self-selecting bunch who did it, not the 33.9% of eligible voters who didn’t, let alone the 2.5% of the population who were denied the right to do so entirely as the result of felony convictions, even after release, unlike nearly all the rest of the developed world where even those still in prison can still cast ballots.

Is “easy” even the right question, when so much more than the basic mechanics are involved? If you didn’t vote because you were afraid of Trump’s Army, for example, are you going to say it was “hard” (especially after you found out the army was AWOL in the end) or are you going to, more honestly, pick door number three and check the box all responsible pollsters know they must provide to such a question: “don’t know?” Even if you did vote, isn’t “easiness,” like happiness, as least as much a measure of expectation as it is of reality? If you’re a Black man in Georgia who Republicans have repeatedly purged or otherwise stymied, but this election, thanks to Stacey Abrams, while hey, you had to stand in line for seven hours (insuring there’s a good chance you won’t show up next time), isn’t it still going to seem “easy” compared to everything you had to deal with in prior elections, when you repeatedly tried and failed? What do you think said Black man would say about how “easy” it was after you tell him that in every other Western democracy, registration is automatic and Election Day is a holiday?  Funny, there don’t appear to have been any questions like that in these survey instruments.

But, as the ad says, again and again, that’s not all.

COVID-’20

To all of the above, add the mother of all game-changers, the pandemic, which Trump played like a Roman fiddle. The conventional wisdom is that but for COVID, Trump would have ridden a strong economy to easy re-election. There’s a lot of reason to believe the conventional wisdom is wrong.

First of all, the economy really wasn’t that great. Tax cuts for big business and the wealthy had turned out to be even more ineffectual than their opponents believed they would be, producing nothing but a sugar high that dissipated after just a year, having never achieved the growth promised, leaving nothing but trillion dollar deficits as far as the CBO could see. The repatriation of overseas money had stalled at 25% of what was promised as well. Growth in 2019 was just 2.3% and was projected to dip to 2.0% in 2020; 1.7% in 2021. Business investment was down, manufacturing was in recession, bonds had recently inverted, and corporate debt was exploding like it was two thousand zero zero seven (oops, out of time).

Furthermore, it’s just not the case, if it ever was, that voters automatically credit–and therefore vote for–a president who presides over a good economy; ask Barack Obama, who was more entitled to that kind of approbation than any president since FDR (yes, racism had more than a little to do with the lack thereof, but…)  In reality, the politico-economic relationship is more nuanced than that, the evidence more correlation than causation. For some reason, voters these days seem to think that they (and the companies they work for) have more than a little something to do with the economy when it’s good. When it’s good, they just want government not to screw it up. Manage to achieve this, and they “approve” of “your” “handling” of the economy, as they would a Pottery Barn employee who succeeds in restocking the store without dropping anything. Then, if you’re an incumbent, all your other advantages come into play.

Republicans peddling the stab-in-the-mitochondria theory of COVID in Trump’s defense should understand this better than anyone–after all, they’re the ones who taught us “government isn’t the solution; government is the problem.” When big numbers rang up for Trump on the economy, even in September and October, they were typically framed as giving him “at least some credit,” what we survey researchers call “top 2 box” metrics, combining those who say he deserves “a lot of credit” with a much larger number who (often grudgingly) concede he deserves “some credit.” They were also, like polls comparing Trump’s approval to the media’s or Hillary Clinton’s, skewed by current asymmetries in the electorate.⁸

Eyes Adjusted To The Dark

So, what was the impact of all the GOP intrigues we’ve described, plus many we haven’t?  At this point, we hope we’ve convinced you that the first thing to look for in this regard are significant disconnects between polling and results, that this is prima facie evidence of voter suppression, the real voter fraud.  And sure enough, when we add 2020 to the history books or, in this case, to the bias–tracking chart we previously constructed and shared, the needle jumps like a polygraph:

If we further break this out to distinguish between presidential and congressional polling, you can see that unlike 2016, the “bias” of the polls was greater in the presidential race, which might argue for the existence of the vaunted “Trump effect” if we didn’t have a mountain of evidence just around the virtual corner that we’ll be dumping from the world’s largest virtual wheelbarrow in part 3 of the series to bury that mythology once and for all.¹⁰

Beyond this, beyond the swing vs. non–swing differences we keep coming back to, beyond all the other qualitative and quantitative proofs (often discovered in the course of trying to disprove our own theories) that would normally satisfy many a Nancy Grace viewer, once your eyes have adjusted to the lack of light, you can see signs of suppression nearly everywhere you look, none of them definitive alone, but collectively producing a picture of the elephant even seven or fewer blind men might have easily seen had they all worked together, 21st century-style.  Here are just a few examples:

Macropolinomics. There was a record $14.4 billion spent on the 2020 election, more than double the record–breaking sum spent in 2016.  More than $5.7 billion of this was spent on the presidential race directly, and a lot of that money, in turn, was spent on the swing states––the Biden and Trump campaigns and their allies spent more than $1 billion just on TV ads in the swings, an amount that doesn’t include any spending on anything other than TV (the ground game, digital, etc.).  It also doesn’t include money spent on other races in the swings; for example, nine of the ten most expensive Senate races in history took place in 2020, and six of those nine were in swing states (Georgia [2], North Carolina, Iowa, Arizona, Michigan) to the tune of another nearly $1.9 billion, an amount that, itself, doesn’t include Senate races in Minnesota, New Hampshire or Texas, governor’s races in New Hampshire or Minnesota, or any House races in any of the swings.

You would think, with that kind of money coursing through these states’ arteries and veins, that turnout in the swing states would dwarf that in the rest of the country, but even though these states turned in their mail-in ballots at a higher rate–clearly the spirit was willing–to paraphrase the immortal Warner Wolf, if you had the swings +4, you lost, as you can see below:

Billions and billions of dollars spent, and turnout barely budged.  It gets curiouser.  We decided to take a look at how the turnout in each state compared to the national average, how much money was spent per eligible voter in each state (and how this compared to the average) and how much money was spent per vote in each (again, comparing it to the mean).  In chart form, that looks like this (red numbers in brackets are negative):

One can’t help but notice how ridiculous some of these numbers are.  When you look at Wisconsin, Michigan, Florida, North Carolina, Pennsylvania, Arizona, and Nevada, in particular, it’s hard not to wonder if it wouldn’t have been more efficient/effective for the parties to just tell everyone eligible they’ll give them a Jackson to vote and hand out cash at the door, especially when you consider these tabulations are based solely on TV ads and appearances by the presidential campaigns, not the entire kaboodle.

Unless, of course, no amount of money could address the real reason turnout in the swings was nowhere close to what scientific or econometrics would project, which is suggested by the accompanying correlations, or lack thereof, that we calculated.   For example, in a capitalist, market-based economy, you’d normally expect that the more money you spend per eligible voter (vs. the average), the better your turnout should be, but in the swingin’ 2020s this correlation (0.15) is weak at best.  And you’d think the more money you spend per actual vote, meaning your dollars had actual quantifiable ROI, the correlation would be even stronger, but in fact, it’s practically non-existent (0.09).

If we look at spending in the context of what motivated this piece to begin, polling “error,” the correlations get stronger but even less intuitive.  As you can see in the chart below, there’s actually moderately strong positive correlation between spend per voter (as well as per vote), and the size of the Democratic polling “bias” in the state, meaning the more the candidates spent per voter and vote, the more inaccurate were the results:

But things really get strange when we focus in on the six states both sides determined would decide the election, into which, by October, they were pouring 85% of their media dollars, and perform the same analysis  Consider, in particular, as you look below at the final chart in this sequence, that:

• Biden badly outspent Trump in five of these six states (the sixth was North Carolina)
• Trump strongly outspent Biden in six of the other seven swings (and the seventh was Texas, which Trump assumed he had in the bag, probably for reasons similar to George Bush’s in Florida in 2000, and so basically spent nothing, despite polls showing a close race in a big state he had to win. Hmm…)

As you can see:

• The polls were more inaccurate, more “Democrat-biased” in the “Big 6 states” where Biden outspent Trump than those where Trump outspent Biden, i.e. exactly the opposite of what one would normally expect.  Specifically, if Candidate A outspends Candidate B and the “polls are wrong,” we’d expect this to be because the polls weren’t able to properly take into account Candidate A’s spending edge, and the impact this has on his ability to get his voters to the polls; we’d expect, therefore, the polls to underestimate Candidate A and therefore be “Candidate B-biased,” not the other way around.
• In these states where Biden largely outspent Trump, there was a strong (0.59) correlation between the amount the campaigns spent per voter and the size of the polling “error”/Democratic “bias.” Put another way, the more money Biden spent per voter, the more likely the polls were to overestimate his strength and underestimate Trump’s. Taken to its literal absurd extreme, the data strongly suggests Biden would have done better against Trump in these states if he’d cut back on the resources he was deploying in his campaign.

Absurd, like I said.  But factually correct.  Absurd unless there’s another explanation for the data.

Outspending your opposition doesn’t guarantee victory–ask Mike Bloomberg and Tom Steyer–but our much maligned pollsters knew the jig was up before Bloomberg and Steyer did, and nobody was saying “the polls were wrong” during the Democratic primaries.  It’s a lot more unusual to have a big lead in the polls, substantially outspend your opponent (and to be clear, Biden didn’t just outspend Trump on television–overall, he outraised and outspent Trump, $1.06B to $785M, and it’s likely his allies, led by Bloomberg, substantially widened the gap), only to fall dramatically fall short of your numbers when the votes are counted.

To put this in another, less polite way, put yourself in Trump’s shoes.  If all the issues are trending against you, if a clear majority of Americans oppose you on every one of them (if you can find a Trump policy or position most Americans supported other than his modest criminal justice reform, please post in comments below), if all the polls show you trailing badly, have shown you trailing so badly–for months, years even–that the day after Robert Mueller tosses you out of the frying pan, you immediately jump into the fire and, on top of all that, because you’ve completely blown your campaign war chest, you’re being dramatically outspent into the bargain, what do you do? What do you do?  We love her, but ask Tonya Harding.

Voter suppression, especially with taxpayer dollars and other government resources behind it, is a lot cheaper than running TV ads, and a lot easier than raising enough money to not just equal, but exceed, your opponent’s haul and give you a fighting chance to make up lost ground.  That’s the simplest, easiest, most Ockhamian way to explain why all the spending failed to boost turnout in the swings more than it did, and why, perversely, the more Biden spent, the worse he did relative to the polls.

Under some circumstances you could explain the latter away as a correlation of a different currency, much like the correlation between drinking diet soda and obesity, i.e. that the reason Biden did worse in the states where he spent more/voter was because he knew he was doing worse in those states, he spent more to make up the difference, and it didn’t work.  This would still go against everything we’ve been taught about the impact of money on politics and political economy more generally; still, at least it’s plausible.  But not if the polls were telling him he was winning those states going away, including, apparently, the internal ones.

The Kids Weren’t Alright. Here’s another way to grip the elephant.  After young voters launched what the GOP clearly considered an un-American surprise first strike on them in the 2018 midterms, getting ahead of the suppressors in the arms race for once, Republican state governments dedicated an unhealthy, if not unseemly, portion of their calendar to pass a raft of legislation specifically designed to make it more difficult for our future to get in the way of the past.  E.g., e.g., e.g.s.

This was compounded by the “unintentional,” yet highly predictable and very demograph-specific, suppressing impacts of the pandemic, e.g.:

• No capacity for campus organizing (because campuses were closed and/or largely emptied)
• No capacity for groups like Rock The Vote and Headcount that rely on music and other venues for their voter registration drives (because all concerts and events were canceled).
  • In general, COVID brought voter registration drives to a virtual halt in many states for weeks, and the fact that non-voters tend to “not like either party” doesn’t mean, in an existential election year, where one side’s population was on the rise, and the other’s in decline (and, as it turned out, was not, in any event, as passionately pro-Trump as expected) this equally affected both parties, and that’s before only one of the two parties decided to run a responsible ground game.
• More generally, no hands-on support of any kind available throughout the process, which is critical because so many young voters are also first-time voters (which incremental Trump voters, no matter how infrequently they voted, were less likely to be, nor have to re-register because they’d been wrongly purged by GOP partisans)
• Ongoing, paralyzing confusion and uncertainty about where–home or school?–and how to register, in a population that has always been less likely to vote if they’re registered in a locale differing from the one they reside in when it’s time to cast their ballots.  In a normal year, this confusion is regularly exploited by GOP operatives spreading a student-specific version of their “police will be waiting” suppression lie, telling collegians that if they vote in the “wrong place,” their federal aid will be taken from them. With the vast majority of students forced off campus and isolated, one can only imagine how much more effective and widespread disinformation like this was this year.
• Etc. etc. etc.

Since 2000, Harvard’s Institute on Politics has been conducting youth polls on voting intent and other issues among potential voters ages 18-29.  In their fall 2020 poll, 63% of “likely voters” under 30 said they were “definitely” going to vote (or already had), another 9% said they were probably going to do so, and another 11% said they were “50/50” to do so, for a total of 80% telling Harvard’s pollsters there was at least a 50/50 chance they would be exercising the franchise (63% of the “likely” boiled down to 60% of “all”), with nearly 70% (69) saying the odds were substantially better than that.

But when the ballots were all counted, only 53% in this critical age group had ended up casting theirs.

Now if you’re a Boomer-aged voter, you reaction may well be something along the lines of “of course they didn’t show–these kids never follow through on anything.”  But if so, OK, that would be just another demonstration of how clueless our generation (I’m one of you) is about Millennials and Gen Z.  Because as you can see from the chart below, in the last presidential election, more under 30s showed up than said they would, and while they fell short of their goals in 2012, it was by a little, not a yawn:

In fact, you have to go back to 2008 to find the last time young voters lagged expectations as much in a presidential election, and 2008 is yet another exception that proves the rule, because ’08 was the election when GOP suppressionists, buoyed by Crawford, made their first big push, their first preemptive strike in the arms race, with a particular emphasis on shutting down the youth vote because it was rightly seen as such a critical and enthusiastic part of the new Obama coalition.

If that’s too small a dataset for you, here’s a comparison between intent (as expressed in the Harvard poll) and results for every election from 2010-2020, which is even more striking, for two reasons:

• It shows that the gap between intent and action in 2020 was more than seven times greater in 2020 than the average from 2010-18
• It shows there was a steady rise in intent and results from presidential to presidential, midterm to midterm, throughout most of the decade until 2020, when suddenly the road of good intentions cratered.

We Approve This Message. One more example (for now, that is; like all CP pieces, we’ll be continuing to add to this one, hopefully with your help).  Historically, Election Eve presidential approval ratings have been strong predictors of re-election results, and a valuable “second opinion” metric because they’re independent of horse race polls and national in scope (so nearly always have robust sample sizes, with less demographic weighting that can potentially introduce distortions, and not subject to vicissitudes of state-level polling). When we compare Trump’s election day approval and disapproval ratings with the election’s results, the differences with previous re-election campaigns are striking:

As you can see, in a typical election, a slightly lower proportion of voters support the incumbent than approve of his performance, and many more vote against incumbents than said they disapproved of him as they headed to the polls–apparently, being the eternal optimists we are as a nation, even if we rather like the guy in office, we tend to think the grass could always be greener, and in the privacy of a voting booth, a substantial portion of us are willing to bet on it.  Except last year.  In 2020, if the official results are to be believed, against a backdrop of right track/wrong track numbers of 23-70 in August (by which point the vast majority of voters have typically made up their minds, with their opinion not yet colored by anticipation of change to come) and 32-61 on Election Day (by contrast, it’s 43-50 today) and the least popular incumbent (by average approval rating) in modern history, this is the year we broke with American cultural tradition and decided there was no better alternative than to stay the course with the Occupant at a rate nearly twenty times higher than average.

That could really have happened, I suppose, just as there’s no definitive proof Dominion Voting Systems didn’t ship all of our swing state ballots out to Venezuela–or was it Germany–or was it Italy. Or we could draw the much simpler, more rational conclusion: that a substantial portion of Americans who could have been expected, based on Trump’s approval rating, to be voting against him, never made it to the polls for some surely “unfathomable” reason (which, btw, also bumped his percentage of the vote above his approval rating).  On the third hand, we could say Americans have been so dug in on their polarized positions for so long, the result was an approval rate for Trump that was unusually accurate in predicting his share of the vote.  Except that would mean admitting “fake news media” polls are legitimate–all the approval rating pollsters conduct horse race polls as well, after all–and would make it even harder to explain the disconnect between election polling and the official results as anything but the product of suppression, leaving Trumpists well and truly hornered.

Is the kind of reasoning we’re using here different, in kind or degree, from Republican “anomalies” we all ridicule?  E.g. Biden “couldn’t have won” because “no Democrat has ever won without winning Ohio, Florida, and Iowa?”  Well yes, it is different, because first of all, in a sizable not-so-surprising-anymore proportion of these officiously-intoned cases, the “anomalies” being “found” have clearly been made up out of whole cloth and/or never researched. It’s simply not true, for example, that Dems have never won without winning Ohio (1944, 1960) or without winning both Ohio and Florida (1960), or without winning Ohio, Florida, or Iowa (1960).  And before any Trump supporters object that this is the rule-proving exception because “everybody knows” the Democrats stole 1960, too, the claims that Illinois and Texas were stolen that year have been thoroughly debunked. Furthermore, Texas was only the 10th closest contested state that year, and among those closer was California, which Kennedy appeared to have won until late absentee ballots (you know, the kind that this year were inherently rife with fraud) put Nixon over the top, not that the landslide majority of Americans–i.e. the 60%+ of us, including more than 1 in 5 Republicans, who believe the Electoral College should be abolished–care, given that nobody has (credibly) questioned whether JFK received more votes overall.  On the other hand, it is true that Republicans have never won while losing Arizona, a streak that continues, unblemished, to the present day. No wonder they fought that result especially fiercely.

More generally, our “anomalies” are almost always based on averages and correlations, data sets, not Ns/sample sizes of 1; we’re not calling out the political equivalent of astrological alignments that “have never happened before” as evidence they can’t be happening for the first time–something, btw, that not even astrologers, for whom everything runs in cycles and circles, would do–nor are we resorting to devices like Benford’s “Law” that, as proofs of voter fraud, belong more in the realm of numerology than mathematics.  And we’re providing rational, historically established reasons for the anomalies we’re highlighting, whereas there’s no rational reason why a Democrat shouldn’t be able to win while losing Ohio, Florida, and Iowa, since between them these states have only 53 of the 270 electoral votes required to win, and the only state that voted for the same party every presidential election since 1964 isn’t even a state (sorry DC) and has never voted for the GOP.

Where They’re At

You may have noticed that we’ve taken a lot of what might seem like gratuitous shots at 44.46 without apparent fear of bias, and we promise you we’ll take a lot more between now and the end of this series.  It’s not because we’re dyed-in-wool Democrats–I’m not a Democrat at all.¹¹  We merely feel justified in doing so because because what we’re presenting are just the cold hard numbers, a true no-spin zone.  Besides, Trump has no ideology, only an idiology of I*deo*logy (i.e. I know I’m God). Of course, evidence based on actual reported data isn’t likely to impress Trump supporters.  They know that isn’t going to help their case, so they’ve been busy tunneling under it.  A lot of the evidence is quite impressive-looking, until you drill down.  Take this chart, for example:

What’s nefarious about what’s shown here?  Based on Trumpist talking points in general, it’s apparently supposed to convince us that Biden won Pennsylvania by registering a large number of dead or demented 90+year-olds. It’s not clear whether the registrations being shown are the total number of 90+year-olds registered in PA or the new registrations of 90+year-olds this year, but a little math and logic makes pretty clear it has to be the latter.¹²

When I first saw this chart, it was in the context of a imperifying video where it was flashed up to represent one of the “many other anomalies” not included in the screen screed, and apparently shot by a Ken Burns wannabe with a short attention span, because he didn’t pull back far enough to reveal the chart’s legend.  Fortunately, we were able to screengrab it before Google removed the vid from YouTube, then use Google Image to get to the Twitter feed of a well-known Trumpist who posted it in full, and confirmed, via his outrage, what Trump supporters believe about it.

So what does this chart show, when you’re given more than 2.5 seconds in a video to inspect it? It appears to indicate the number of new 90+year-old registrations jumped from about 400/year in 2018 and 2019 to 1,900 in 2020. 2018 was a mid-term and 2019 was an off-year, so you’d expect there to be a significant jump from those years to 2020. A fairer comparison would be between 2016, the last presidential year, and 2020. In 2016, it looks like about 600 90+year-olds registered for the first time, vs. 1,920 in 2020.

Is this problematic? Absolutely not. First of all, there are between 160,000-320,000 90+year-olds in Pennsylvania, so we’re talking about an “excess number” of registrations of about 1,300 in that population, or less than 1%, which also makes one wonder why it’s even being raised in a state where more than 7 million votes were cast, Biden’s margin of victory was 80,000+, and those 1,300 “excess registrations” were a total across the entire state, not just “Democratic” counties. Second, unlike many other populations, the pandemic made it easier for relatively immobile 90+year-olds to vote, because restrictions on mail-in voting were loosened, which could have incented them to register, and motivated them more to vote, because the virus was a greater existential threat to them than anyone else.

Third, given the aging of our population and the total size of the 90+year-old cohort, even if registrations were completely normal, you’d expect significantly more 90+year-olds to register to vote in 2020 than in 2016; in fact, you can see this playing out in the Trumpists’ own data–in each election cycle, the number of new 90+year-old registrations is significantly higher, and the growth appears to be exponential. Based on this simple analysis, it looks like there were actually fewer new 90+year-old registrations in PA in 2020 than would be predicted by the previous years of data on display–there should probably have been at least 2,400 new registrations by demographics alone.  Finally, if you’re a Trumpkin, and even after reading all the above you still think those 90+year-old registrations were ‘anomalous,’ here’s the kicker: look at the chart, closely–there were substantially more new 90+year-old registrations in what the creator euphemistically calls “Other” counties (why not call them Republican counties?) than there were in Democratic counties (A: That’s why), so if anyone should be accused on manufacturing votes from dead or dementing 90+year- olds, it’s the GOP, not the Dems.

If I reasoned and argued like a typical Trump supporter, I’d take this analysis, based on one frame of this one 16 minute video, and tell you it discredits the entire GOP “voter fraud” case from all sources. Too many Trump supporters, when they’re faced with fifty factual arguments against their position, act as if the entire world is just one big conspiracy theory, and think that if they can successfully attack just one of those fifty, they’ve pulled a logical string that unravels all the rest. But although I love a good conspiracy yarn myself, my whole world doesn’t revolve around that point of view and its implications for debate.

At the same time, I have to acknowledge that a number of the deep state dives Trump supporters are taking are also pretty disturbing.  Take this piece by Mac Cutchins, for example, which, unlike many, appears to be based on legitimate sourced data, presented with full transparency, and with full acknowledgment that not all of what he’s presenting–namely massive, apparently inexplicable vote shifts from Trump to Biden–fully supports the manipulations he is strongly suggesting occurred in some states. In fact, he goes out of his way to show contested states that don’t.  If you’re having a hard time understanding how even fairly rational (i.e. probably not charismatic Christian) Trump supporters can cling to the idea that the election was stolen from him, this post is worth taking a few minutes to click on and read.  Really.  For a fuller picture of the universe of such claims that have been made, check out Michael Lindell’s Absolute Proof (scroll down the page to see the full video).  It’s long (nearly two hours), the production values are cheesy, and he’s not much of a narrator, but it may be the only place you can see a lot of it, given all the social media scrubbing going on, unless you’re willing to take your virtual life into your hands and venture onto Parler.

To be clear, I can read Cutchins’ piece, which focuses on what look like massive vote switches from Trump’s column into Biden’s, and ask a number of questions that would appear to puncture it.  For example:

• If this was done to swing the election, why was it deployed in states Trump had no chance of winning (with the attendant greater risk of exposure)?  Shouldn’t we see it in swing states only?
• Why was this technique deployed only in some of the key states Trump lost and not in others?
• Why was it done so openly, as reflected in data available to any member of the general public?

But as a bit of a c-theorist myself, I can also answer those questions:

• It could have been done in non-swing states to deflect suspicion and make it look more routine
• It could have been done in some swing states and not others for the same reason, and also because it’s easier/more legal to do in some states than others (since under our Constitution, every state has its own election rules)
• It could have been done openly for the same reason that Trump openly “admits” to his crimes–as an effective form of intimidation (yes, we did this, and the fact we’re doing it so brazenly should tell you there isn’t a damn thing you can do about it) and/or inoculation (there can’t be anything wrong with it if we’re telling you we did it, without a hint of apology for doing so).

Frankly, I should be able to read a piece like Cutchins’, do a search, and more or less immediately find a response to each of the specific claims made, every instance, state by state, from official sources. If you’ve read this far, or read our other pieces, I think you know we’re pretty good at searching, and can assure you that such an explanation doesn’t exactly pop to the head of the search results, as it should.  Instead, in general, the response, such as it is, to such accusations appears to take one or more of four forms:

• Ignore them, because to respond only “dignifies” the original claim, and gives it “a legitimacy it doesn’t deserve.”
• Deny them without explanation or evidence, because explaining only always gets you in more trouble.
• Apply continued pressure on social media companies to delete them and thereby “deprive them of oxygen.”
• Provide one or two examples of similar cases where everyone agrees the original claim was mistaken, ideally cases of straw man magnitude, like Michigan’s Antrim County, and imperiously awesume this addresses all such cases.

All of which can be justified by the infamous Backfire Effect, the view, based on, it must be admitted, considerable evidence, that providing someone with information contradicting their beliefs only causes them to ‘double-down’ on those beliefs, a hypothesis which has been conveniently cited by so many over the last four years to justify life in the bubble, free of unpleasant conversations we’d just rather not have.

Such are the tactics and rationalia that have led us to where we are today, and worse.  Deleting election fraud claims is less than futile: they get around anyway, even on the platforms doing the deleting (e.g. posts deleted on Facebook simply circulate via Facebook Messenger–I get them that way from Trump-supporting friends all the time). It’s a form of censorship that reeks of repression, cover-up, and guilt, and it leaves all who might want to help the authorities engage with these claims blind to what “the other side” is hearing or what might (fairly rationally) be making them so (irrationally) angry.  Meanwhile, debunking just a claim here or there, assuming this suffices to debunk the rest is, ironically, remarkably similar to what conspiracy theorists do themselves; if one is concerned with “providing oxygen,” fighting fire with fire is probably not the best response, unless you’re looking to create a political environment in which no one can breath.

The backfire effect defense is perhaps the most insidious and wrong-headed of all.  Yes, there’s plenty of evidence validating its reality, but the same theorists who cite it also bemoaningly support the efficacy of “the Big Lie.”  How can both of these phenomena be real?  There’s only one way we can think of: the backfire effect represents an initial response to contrary information, but if the recipient is repeatedly presented with that evidence, a la big lie theory, it eventually gets through.  After all, if only those already primed to accept “big lies” were affected by them, there would be no need for (endless) repetition (first in the media, then in the history books).

At the end of Election Day, we’re not touting the presentation of contrary facts that neither party has firsthand knowledge of in a vacuum; we’re talking about giving ‘the other side’ home field advantage and beating them at their own game, taking the ‘facts’ they they throw at us and turning them aside with credible alternative explanations, an approach that’s already producing positive results in reducing the related problem of vaccine hesitancy.  For those of you angry with ‘them,’ we’re talking about “letting them lead with their chin.”  If the powers-that-be won’t do this, we may have to do it for them, each and every time, everywhere, drawing on a collectively created, ever-growing database of responses, a la AOL’s original “Teacher Pager” homework help service, and on our much greater credibility as fellow citizens in the trenches than any group of experts has anymore (click here if you’re interested in joining us).

More broadly, we believe there’s a need for a social media meta-platform–call it Comity–that brings red and blue voters together on a sustained, sustainable, and scalable basis revolving around shared interests outside politics.  We’re pretty sure that by leveraging strong relationships with groups like Braver Angels, connections with leading thinkers in the social & emotional learning/emotional intelligence arena and, above all, decades of successful online community-building experience, we know the keys to making that happen, bigly (and would love to have your input and involvement here as well).

The Latin root of the word education, educare, means “to lead out.” Back in the place we’re all ultimately from, mother Africa, the words for ‘teacher’ and ‘leader’ are the same in many local languages.  Those who want to be leaders, or even just persuaders, need to understand, like all teachers do, that explaining, tirelessly if necessary, is an integral part of the job description.  Stonewalling and censorship are not.  Thanks to decades of condescension and obscurantism, change won’t happen overnight, but that’s no reason not to start, to begin treating fellow Americans as at least as intelligent as we’re learning our fellow denizens on earth to be.  Of course, there are a substantial number of post-modern deniers who believe the Mandela effect is proof of the existence of alternative universes, with parallel election data, too, and all we can do in such cases is suggest this might mean they’re at the leading edge of the “illegals” who keep voting in “our” elections. Nobody’s perfect.

And that goes for our analysis, too.  Anyone who thinks we’ve misused or misinterpreted the data is welcome to do so in comments below or by writing to us directly.  There’s nowhere for political discourse to go in our country but up :::knocking vigorously on wood:::, and having everyone, all the ‘experts’ lay out their assumptions behind their interpretation of the same data we’re looking at, for once, would be a big step in the right direction, so all can learn from all and decide for ourselves what to believe, with the help of the best tools and thinking available to us.

In fact, we’d go broad, deep, and long on this: in our opinion, all major media outlets should be requiring any person or organization purporting to be expressing a view that’s data-driven should be required, within the limitations of applicable privacy laws, to make available to the public the raw data on which their opinion is based, assumption- and weighting-free, accompanied by a complete accounting of and rationale for the assumptions they made and/or manipulations they conducted to reach the conclusions they did, with universal access provided (via government, foundation, tech funding) to a standard package of tools and educational materials that make it realistically possible for members of the public to do analyses of that data themselves, for the benefit of all.

Does this sound naïve and hopelessly optimistic? It isn’t. It can’t be. As the century continues and the rate of change only increases, we are rightly returning to the original definition of “amateur,” derived from the Latin amare, to love, and simply referred to someone with a passion for the subject in question, not “immature,” “inferior” “unskilled” or “inept.” More broadly, we’re re-learning what the rest of the animal kingdom has never forgotten, that generalists, not specialists, ultimately lead disciplines, companies, and nations. You can see this reflected at the bottom line in the rise of companies like Innocentive, through which industries and academic fields are leveraging the fresh eyes of outsiders to find solutions to problems experts in the field are too close and hidebound to see.  Even at the highest levels of intellectual thought and rigor, what was once seen merely as satire, e.g. the Ig Nobels, originally created by the Annals of Improbable Research, and at one time popularly considered the Razzies of science, are gaining influence and prestige, thanks in large part to an increasing number of scientists working far, and farther, afield, becoming “deliberate amateurs” in their practices, letting their curiosity and interests carry them where they will.

We say this with emphasis in hopes that you’ll be inspired, given confidence, feel compelled, pushed (or aggravated) to do the same, which, in addition to serving the just cause of radical transparency, is why we’ve been walking you through every step of our own research. Because there’s clearly a lot of Sophia conventionalis that needs to be weeded or cleared so we can see–and see each other–more clearly in politics, and the “experts” don’t seem to be nearly as aware as you–or we–are of how little they’re wearing.

For our part, we’re always looking for writers and creators–with a name like Creative Politics, we’ve got to be good–and open to a wide variety of arrangements (with as many as 30+ years of new media success behind some members of our team, we’ve seen and made them all). You don’t have to be all fired up for writing articles/posts to contribute as a writer, and there’s a veritable cornucopia of other roles, both big and small, you can play in our projects, including coming up with your own project(s), role(s), etc.  Oh, and btw, if you agree with our stance on universal access to data and the tools to analyze it, come join our Proof project, and help make it so.

Will all this, any of this, by itself, restore trust in science and expertise?  No, of course not–but again, the lack of a silver bullet should never be an excuse for inaction, because what we cannot accept any more, from the ‘political pros’ of either persuasion, are ‘trust us’ appeals to their authority, experience, and expertise; not from the Republicans who swallowed and supported 30,000+ lies alone from the country’s leader in the last four years, and not from the Democrats who, over the course of the last four decades, have repeatedly demonstrated an inability to shoot fish in a barrel even when the barrel is made of fish, yet continue to proclaim they understand “the voter” better than the rest of us.

One More Thing…

Let us end where we began, with the question of whether there’s any sensical explanation other than voter suppression for the apparent difference between swing and non-swing polling accuracy, but go one level deeper in our analysis. Specifically, as huge as the differences in polling vs. results between the two state types appear to be, is it at all possible in any way that this is really just an artifact of the generic pro-Biden orientation of the “electorally unimportant” parts of the country?

Having stumbled upon CNN’s data jackpot (lo, so many scrolls ago), we wondered if there was a similar data set available for the 2020 election, and there is, from Five Thirty-Eight, of course (truth be told, that’s probably where CNN got theirs, too, but let’s share the love).  Here it is, to play with as you see fit.  As with previous years, we decided to focus on polls taken in the last three weeks before the election, and in this case, on state-based presidential polls only.  As before, we did no weighting of any kind–massaging them into accuracy isn’t the goal–but where the swing states are concerned, to maximize credibility, we used the widely reported polling averages of each swing state, and took as given, therefore, the gaps between the polls and results that have been reported, and the “Democratic polling bias” this appears to reflect.  Starting with these assumptions and process, and ranking all 50 states plus DC on this basis, we get a chart that looks like this:

Which makes it look like maybe you wasted at least some portion of the x minutes of your life you took to get to the sentence you’re reading now.  Except.  Except that comparing Five Thirty-Eight’s swing state polling averages to raw sets of non-swing state polls is more like an apple-to-radish comparison than it is between any two of the 7,500 varieties in the genus Malus, for two related reasons.  The first is that the quality of the polls done in non-swing states is, understandably, though regrettably, abysmal, as you can see in the chart below, which compares the proportion of polls in swings vs. non-swings at each level of quality identified by FiveThirtyEight at the time of polling (full disclosure: they’ve since revised their pollster ratings/grades), using their dataset for the associated analysis as well:

And you can confirm this qualitatively just by glancing at a few of the polls in the non-swing aisle that frankly don’t pass even the ROTFL or LMAO tests:

• Trump eking out a 6-7 point win in Utah (as 12 polls predicted, and no, they don’t like Mitt that much out there)
• Trump hanging on to beat Biden by 2-4 in South Carolina (according to 7 polls, with another 12 going out on a limb to pick 44.46 by 5-6, and no, they don’t hate Lindsey that much down there, either)
• Trump holding off Biden by 8 in South Dakota (7 polls)
• Trump pulling off a win in Nebraska by 4-6 (19 polls, and we’re pretty sure we eliminated all the District 2-only polls from the mix)

The second reason is that in constructing its swing state polling averages, Five Thirty-Eight is trying to make accurate polling predictions, using three techniques in particular:

• Adjusting for “house effects”— With twenty+ years of data, Nate & Co. have a pretty good idea of the degree to which each pollster they include in their database is typically Democratic or Republican-biased (you can see this in their pollster ratings) and they make adjustments for this with respect to each pollster when constructing their polling averages
• Weighting based on pollster/poll quality — Famously, they don’t just average the polls like we (and RCP) do; they give more weight to pollsters and methodologies with strong track records, and less to the kind of pollsters who are thrilled to have the non-swings more or less to themselves
• Adjusting/weighting for likely voter samples vs. others  — Which they do on the theory that in our unique form of “democracy,” polls limited to likely voters only are more likely to be accurate than those sampling registered voters or, God forbid, everyone who’s eligible to vote

To give you a sense of the kind of impact these kinds of “corrections” can have, we’ve taken the raw data from the swing states and applied each technique (or a reasonable facsimile) to create the polling equivalent of “before and after” pictures.

First we took the full swing state poll dataset and, for each state, added up all the polls and averaged them, just like we did for the non-swings above (and all polls in previous years), so as to see how these polls in raw form, i.e. before Nate’s adjustments for “house effects” are applied, compare to the Election Eve polling averages (the Refined column) that were reported for each swing.  If you do that, you get something like this:

As you can see, these kinds of adjustments can have huge impacts in improving the accuracy of the polls, meaning, in our case, reducing their “Democratic bias,” and, more often than not, they do.  But just as interesting here are the cases where they don’t.  Note, for example, the relatively strong (for social science at least) positive correlation (0.33) between the size of states’ original polling “error” and the impact of the adjustments made.¹³  In plain English, what this means (because error values are both positive and negative) is that the bigger the size of the original error, the less impact the adjustments have in reducing it. In fact, in four of the five states with the biggest polling “errors,” the adjustments intended to improve the polls’ accuracy, themselves based on more than two decades of data and well-proven expertise, apparently actually made them less accurate, and for six of the seven states with the largest “errors,” the adjustments either made the situation worse or had no impact at all.

When it comes to 538’s second technique, weighting results by poll quality, we’re a bit in the dark of a black box in trying to show its impact to you, because we have no idea how FiveThirtyEight’s algorithm handles this, but we can create what we think is a pretty good proxy/approximation by taking the following steps:

• • • Determining the average Democratic bias at each pollster grade level (A, B, C, D) for each swing state
    • Altering the number of polls at each grade level in each swing state to match the proportion of polls at the same grade level in the average non-swing state (rounding up or down to the nearest whole number in each case)
    • Multiplying the number of polls at each grade level in each swing state by the average bias of the polls at that grade level in that state, then dividing by the total number of polls in that state.

Which gives you a good picture of what the Democratic bias in each swing state would be if each swing state had the same proportion of polls at each 538 grade level that the average non-swing state did, i.e. the same “subprime” poll quality profile.  A picture that looks like this, with the adjusted bias for each state in the far right column:

When you then compare these numbers to the “Democratic bias/polling error” reported for each state, you can get a real feel for just how much off-target swing state polling averages would’ve been if the same quality (or lack thereof) mix of pollsters (the Non-Swing Q column) had been doing the polling there as polled the rest of the country:

In fact, as with “house effects,” the average impact would have been even greater if there weren’t some states, like Ohio, Florida, and Nevada where, perversely, it appears election forecasting would have been more accurate if less competent pollsters were doing the work, a claim further supported by the fairly strong negative correlation (-0.53) in general between the official magnitude of the polling error in these states and the extent to which lower quality pollsters make the problem worse, i.e. they don’t.  In fact, it appears that in the case of the worst misses, it would have been better to throw darts at a roulette wheel.  You can see this less abstractly if we roll back to the raw data used to produce it–here’s the chart of that data, this time with all the cases where, on average, better pollsters produced worse results highlighted on it (with blue representing more Democrat-biased, red less so):

Curious enough to kick the tires on this one a little more?  We did, by calculating the GPA of the polls run in each state and comparing it against the magnitude of that state’s “polling error.”  In a world subject to ordinary laws, as opposed to reality distortion fields, you would expect that the higher the quality of polling, the smaller the polling error would be.  Here’s what we found instead:

Not only is there no sign of the expected correlation, there’s a small positive correlation instead, meaning that the higher the GPA (quality) of the pollsters, the bigger the polling error was in these states.  To be fair, it’s a correlation that’s probably too small, even for social science, to hang a supremacist on, but here’s another pass we made at the data, this time involving a more fundamental, virtually unchallenged principle that extends to all science, not just polling, with none of the subjectivity inherent in even the best of pollster rating systems (except those that are literally based on past performance only): regression to the mean. Everyone knows that the more measurements of something you make, other things being equal, the closer to the truth, the true measurement you’re supposed to get.  Unless you’re a group of pollsters surveying swing states during the 2020 election, as you can see here:

Once again, not only does this universal scientific principle get no love, no validation, but there’s actually a small positive correlation in the data, meaning that, if anything, the more times pollsters polled a swing state, the more inaccurate the poll results were.

When you combine these results with all the rest, it seems clear we’re at a crossroads where polling and even basic science, both natural and social, are concerned. From our perspective, everything we’ve presented throughout this piece–the real anomalies, as opposed to the ones Trumpsters keep churning up in their moms’ basements–is irrefutable evidence of an external force at work, a small but obese thumb on the scale.  In fact, we’ve created a little Occam’s checklist with 30+ questions, drawn from the research that’s informed this series to date, which any other “polling error” theory ought to be able to answer, convincingly, because voter suppression can account for them all, with ease. And we’re confident, based on what we’ve found by just turning over a rock or two (in a field that looks like Colonial-era New England, circa 1776), it’s a list that’s only going to grow until the establishment cries ‘uncle Sam.’

When polling results violate basic scientific principles, proven polling practices, and widely accepted scientific and social science research in as many ways as we’ve described in this series, to believe this is the result of some problem with polling that needs to be fixed is like watching a rocket ship successfully blast off into space and thinking this means the law of gravity needs to be revised or repealed. And positing various mythical psychographics hiding in the demography is like believing the personal grooming habits of the astronauts have a determinative role in whether the rocket reaches escape velocity and/or what direction it goes. P.S. Our overall choice of metaphor is not entirely coincidental.

But we’ve digressed again, and it’s time to bring it home.  Not surprisingly, the third way 538 cleans up after the pollster community, namely weighting polls of likely voters more heavily than other samples, also has a substantial impact, as you can see when we compare the accuracy of the raw dataset of swing state polls with the same dataset when all polls except those based on “likely voters” have been removed:

As you can see, restricting the polls considered to those using a “likely voter” poll of respondents reduces polling error in every case, and in many cases substantially.  Because “registered voter” and “all voter” polls in the data set were mainly conducted by the lowest quality pollsters, we wanted to make sure that the reduction in polling error we were seeing wasn’t just a result of removing a lot of low quality polls from the calculus, which would make the apparent impact of “sample type” (likely vs. registered vs. all) merely a proxy for pollster “house effects” and other adjustments for poll quality. So, in addition, we looked at the D grade polls in isolation.  If “likely voter” adjustments were just a proxy, not an independent variable, you’d expect there to be little or no difference between the Democratic bias of all D grade polls and “likely voter” D grade polls only, but as you can see, pulling “registered” and “all” voter polls out of the D poll mix reduces bias even more for that lowly set than it does for swing state polls in general.

So all three “massages” Nate’s algorithm applies to polling results have significant, independent, positive impacts on polling accuracy in the swing states.  What if we apply just one of them to the heretofore raw non-swing polling data?  So that we’re at least comparing apples to oranges, if not Honeycrisp to Red Delicious?  What if, for example, we apply pollster “house effect” adjustments to every poll in every non-swing state?  We find this, Mitch:

And with that, and just that, the non-swing state polls become what we thought they were all along, once again, strikingly more accurate than those in the swing states, a nearly insurmountable reality for any phenomenon other than voter suppression to explain.  In fairness, we don’t actually know what Nate Silver means when he says he “adjusts for house effects;” maybe he doesn’t apply them fully, as we did.  On the other hand, we haven’t even begun to explore how much non-swing poll accuracy would be improved if non-swing state polls had the same mix of quality pollsters fielding them as the swing states do, or if we down-weighted all non-swing polls except those surveying “likely voters.” Furthermore, it’s fairly obvious, from some of the howlers we led off with, that non-swing pollsters, understandably, weren’t putting the same level of effort into QA in these states as they were in the swings; there was only one “Biden by 17 in Wisconsin” in the swings, not fifty-seven such whoppers as there were in ND, SD, SC, and NE alone.

The results above lead us to try to put one more nail in the coffin of any polling explanation that doesn’t start with a “v” and end with “ion,” comparing polling error against NIU’s rankings of how difficult it is to vote in each state.  In map form, with the easiest states in light green, the hardest in near-black, here’s what she wrote:

You can see visually the extent to which “Democratic polling bias” syncs with voter suppression, and this appearance is supported mathematically in the form of a positive correlation of 0.24, which, again, is fairly high for social science, meaning that the harder it is to vote, the greater the polling “error” overestimating Democratic support.¹⁴  And this is after applying only one of 538’s three main digital reagents to the non-swing data; the correlation would almost certainly increase substantially if we applied the other two as well.

It’s also suppressed (pun intended, I guess) along our southern border by a string of exceptions that prove the rule–South Carolina, Georgia, Mississippi, Louisiana, Texas, and New Mexico–and more importantly prove that we, the people, can overcome suppression, even if the politicians who are supposed to be on the side of democracy and their consultants refuse to acknowledge the problem even exists.  Georgia, of course, has been ground zero for the operations of the greatest political organizer of our time, Stacey Abrams.  What’s less known is that since it was founded in 2018, her voter registration and turnout organization, Fair Fight, has spread to twenty other states, chief and first among them, wait for it…Louisiana and Mississippi.

In South Carolina, Jaime Harrison’s massive registration and turnout organization may have failed to unseat Lindsey Graham, but it impressed those self-same jaded politicians and pros enough to make him the chairman of the national party. Texas in 2020 was the beneficiary of Beto O’Rourke’s massive organization and turnout operation in 2018 that nearly took down Ted Cruz.  New Mexico is a majority-minority state whose government has been compelled to be responsive to the people–in the early days of vaccine distribution, when “vaccine hesitancy” on the part of minority populations (a.k.a a lack of access parallel to their lack of access to the polls), NM led the nation in vaccination rate.

In any case, one thing is for sure–FiveThirtyEight has been saving the polling industry’s clipboards, random digit dialers, and whatever constitutes its hindquarters for a while now, and there probably should be a statue of Nate Silver in the lobby of 529 14th Street.  But what’s good for pollsters in the short-run may not be good for anyone in the so-called long-term rapidly telescoping towards us. It should be clear from everything we’ve looked at so far that voter suppression didn’t become an existential threat to our country on March 25th, 2021, and Georgia’s SB 202 isn’t “the new Jim Crow.”  It’s just another incremental brick in the wall, and to claim more than this is a dangerous exaggeration, particularly if we’re somehow successful in getting it rolled back, which could be the archetypical ixodid problem, the political equivalent of removing a tick without getting the head or mouthparts.

Because as we’ve shown and will continue to show, the real new Jim Crow is everything Georgia and other states have done in the last twenty years to make it harder for “the wrong people” to vote, as well as everything they’ve not done in the face of increasingly aggressive partisan suppressive tactics.  It’s not even clear whether ‘ol Jim ever really went away; after all, the clocks in his house all run backward, and we didn’t start looking for him in the dark until the 2000 Florida recount.  It’s going to take more than a little selective online shopping to deport him to where he belongs, to go get radical in the best, original sense of the word and not just halt the latest round of tactics but tear suppression out by the roots like the invasive noxious poisonous weed it is.

So perhaps it would have been better for Nate & Co. to let the pollsters really face-plant, years ago, so badly they’d have had their backs to the wall of a business banana republic courtyard, so badly the industry–and all of us–would have been compelled to recognize and address the real problem, even at the risk of angering one of the two major political parties and the powerful forces aligned behind them, not to mention bringing into open question whether we really believe in the ideals and principles the country was founded on, before voter oppression, repression, denial, deception, obstruction, obfuscation, restriction, intimidation, coercion, aversion, subjugation, deprivation, nullification, cancellation, expropriation, defilement, discrimination, harassment, rooking, screwing, duping, deluding, bamboozling, swindling, defrauding, hampering, hindering, profiling, bullying, robbery, abduction, assault, and battery–ballot burning, the real voter fraud–had thoroughly and insidiously wormed its way into the woodwork of democracy that many can’t even see it, let alone figure out how to tear up the floorboards we’re standing on to extract it.

The good news, pollsters of the world, and those who love them?  Donald Trump, the wannabe leader, the warm-up act who should never have been beyond, has laid the groundwork and paved the path for you.  For years, decades, he has done nothing but blame others for his failings, normalizing gutlessness and irresponsibility the way the Clintons normalized throwing subordinates under the bus, and in the process, won the loyalty of the very people most likely to accuse you of just “whining” and “making excuses.”  There has literally never been anyone in history bearing the mantle of leadership who has cried so self-piteously so frequently; in fact, if you’ve ever known anyone in life, no matter how unfortunate, who has whinged as much as the Snowflake-In-Chief, by all means post their histrionics in comments below–we could all use a good laugh.  And in robust defense of your profession for a change, pollsters, stop abjectly apologizing for the inconvenience (which resembles nothing so much as a perverse humble brag asserting your ability to control that which you clearly cannot), and start pointing your fingers in the direction they belong.

In part 3, we demolish the main alternative explanation for pollstergeddon, the so-called ‘Trump effect,’ and do so with what would be an unhealthy mix of extreme prejudice and glee, were it not for the fact that this alternative reality also continues, in the eyes of millions of his supporters, to justify the destruction of democracy. Along the way, we crush Democrats’ self-interested, self-sabotaging mirroring of Trumpist reasoning.

 

Creative Politics is the world’s first community-based political incubator, synthesizing the best of liberal and conservative ideals with technology and history to generate policies, strategies, applications, and actions for the post-modern era that are well outside the beltway, and well beyond just talk.  All Creative Politics blog posts are collaborative, living documents, the way Madison and Hamilton would create them if they were writing The Federalist today.  We welcome, nay urge, your feedback in the comment/discussion section below, and will be using it (with credit) to make what you just read more and more real–thanks much for your time and insights; they will go unpunished!

¹ Much, though not all, of this work was done with Grunwald Associates, focused on new media & technology, especially ed tech, and you can see a lot of it here. GA’s work has been called “the gold standard” in the field by ed tech leaders like John Katzman and Linda Roberts. Full disclosure: My role was to develop the instruments and analyze the results; I was not the “stats guy,” but was in the kitchen where it happened. Back

² The poll cited is from late 2019.  Go to Zoom, open a room, and raise your hand if you think people liked Trump personally more by Election Day, 2020, than they did in 2019  Back

³ Memo to cancel culture conservative friends: you do know that Hasbro has been subverting your values with this toy since 1953, when Mrs. Potato Head was added to the mix, right? That while you were apparently asleep at the wheel for far longer than Mr. Van Winkle (or Van Winkle, as he’s known today), kids have been using Hasbro’s idle hands and the spuds’ interchangeable parts to make trans potato heads for decades now, and you’ve just given a ringing endorsement to that state of affairs? Back

⁴ As Mueller made clear in volume 2 of his report, he was blocked at every turn from truly getting to the bottom of the Trump-Russia relationship; it’s a miracle he was able to document as many contacts as he was.  The two individuals, Paul Manafort and Roger Stone, who had the most contact with the Russians, and likely knew of many more communications with them, their own and others, both completely stonewalled him.  Mueller believed Trump had abused his powers by dangling pardons in front of both to buy their silence, and to no one’s surprise was proven right on December 23rd, 2020 Back

⁵ The data set we had from CNN didn’t include pollster ratings, which is, in one way, unfortunate, because those change over time. But they change in both directions, and at least some of them change because the raters (Five Thirty-Eight) realize, either through a fuller body of work or re-evaluation of best practices, that the pollsters and their polls were either better or worse than they thought at the time, rather than as a result of a subsequent improvement or deterioration in their process and product. In reviewing 2020 data, which likely came from the same source, we’ve come to realize that pollster rating was likely included in the original data set before CNN adapted (and added to) it, and at some point in the near future, we plan to get this rawest data and see if the grades originally associated with each pollster differ enough to potentially change the conclusions of our overall analysis, and if so, how, but in the interest of time–yours as well as ours–we decided to press on with what we have, with the belief that whatever differences there are likely balance out in both direction and rationale.  Back

⁶ We ran correlations on each election year that generally support this, even though the data isn’t ideal for that kind of analysis because so many pollsters share the same letter grade and therefore are semi-unranked. We also don’t know the point at which “too many” starts to degrade quality; no doubt there are sophisticated statistical tools we could use to determine this, but we want to stick to using the basic techniques anyone can understand and replicate.

The average correlation between poll quality and the number of polls fielded in the last three weeks of any given election season is fairly weak, but in the “right” direction, -0.05, meaning the more polls fielded, the lower the letter grade of the pollster.  This is at least partially the result of the large number of groups that fielded only one or two polls, who are all over the map in their ratings because, frankly, fielding one or two or even three (one/week) polls is unlikely to have any impact on the quality of your work.  If we remove all the pollsters who fielded only one poll, the correlation strengthens to -0.08, if we eliminate all those who fielded only one or two, it strengthens to -0.14, limiting it to pollsters who put 4 or more polls out into the public domain raises the correlation to -0.19.

Overall the pattern–the more polls you field, the lower your grade–holds for eight of the eleven elections in our sample. Interestingly, the three election years where it’s reversed are all mid-terms, where polling quality is consistently higher, including the last two (2014, 2018). But of course, our primary interest is in presidential years, not mid-terms, and if you strip the mids out of analysis, the correlations become even stronger: -0.11 with all pollsters in presidential years included, -0.13 with the one-offs removed, -0.20 with the analysis limited to pollsters who fielded three or more, and -0.27 if only those who fielded four or more are included. Could they be even higher if we kept eliminating pollsters by volume?  Perhaps, but they could also become more random, and we think our point’s been made.  Back

⁷ Except according to the same court, but more right-wing, that decided Bush v Gore. Trump’s talk of “Obama judges?” Just one more example of “I know you are, but what am I” playground projectionism. Back

⁸ In the case of the media, for example, the 40-45% of likely voters (as opposed to all Americans) who support Trump monolithically support him and monolithically despise the media. Meanwhile many of the rest of us, while we may monolithically disapprove of Trump, have our own problems with the media, so many of us wouldn’t give the press a thumbs up either. If 45% approve of Trump and disapprove of the media with the same fervor, and just 20% of the rest of us share their disdain for the Fourth Estate, Trump’s approval ratings will exceed the media’s. The same dynamic holds where Trump vs. Hillary are concerned, maybe even more so, given the proportion of Democrats and independents who reflexively “disapprove” of Hillary for what they believe was a fecklessly run campaign that saddled us with Trump for the past four years (though that disapproval is almost certainly softening now that they’ve seen Biden deal with the outrageous level of un-American cheating and fraud the modern GOP and its allies in dictatorships around the world are willing to engage in). In any case, if the media or Hillary ever come within 10-20 points of Trump in approval, it should be considered either a miracle or a sign that respondents are getting more savvy about what they need to say to keep these kinds of poll results from being badly misused, as they have been for years now.  Back

⁹ If, in making this comparison, we use the high end of the historical average (20%) and, in 2020, consider only the states that did not automatically send ballots out to everyone (which results in an average non-return rate/state of 25.7% ), then non-returns were 28.5% above average.  If we use the low end of the historical average (17%) and compare it with the average proportion of non-returns of all states (28.2%), then unreturned ballots were 65.9% above average. Average 28.5% and 65.9%, and you get an increase of 47.2%  Back

¹⁰ We took as a given that the presidential poll bias in the 2020 election was 4 points; 538 had Biden up by 8.4 points on Election Eve, and he won by 4.4. At the time we did these calculations, we could only find one piece of published research laying out the average “Democratic bias” across all House (2.5) and Senate (2.25) races (Panagopoulos, 2020).  We then looked at the average proportion of all polls in the last three weeks of each of the previous five presidential election years (2000, 2004, 2008, 2012, 2016) that were presidential, Senate, or House-focused, then averaged these to derive weights to estimate the overall Democratic bias (across all national races) and the Democratic bias in Congressional races (House + Senate).  Since then, as you’ll see later in the piece, we’ve found a full trove of 2020 data that we could have used instead of this extrapolative approach.  But because we don’t know how much massaging CNN did with the data we used from previous elections (if, indeed, 538 was their source for it), and we wanted our 2020 estimate to be as apples-to-apples with the data we used from the previous 11 elections (which is also why we ignored Panagopoulos’ analyses of the 2016 and 2018 elections in his paper), we elected to stick with the extrapolation rather than end up comparing raw 2020 data with somewhat refined data from previous years.  That’s the appropriate justification anyway.  The truth is also that we could not see fully analyzing yet another 2,000 polls and reanalyzing the 6,000 polls we worked with previously to get a fit between the two datasets, all to make adjustments to two data points on a graph before putting the rest of this work out there.  Consider that a “proof of sanity” decision–we’re bone-tired, and willing to admit it. We’ll try to work with the 2020 data in this way down the road and update this piece accordingly, i.e. by comparing 538’s raw data from the previous 11 elections against the CNN dataset we used to find the differences, and make appropriate adjustments to the raw 2020 data, before (re)calculating the 2020 biases in the same way as we did the previous 11 years’ worth. Thanks for your patience and understanding, which we assume we must be grateful for in the case of anyone dedicated enough to this topic to be reading our footnotes. Back

¹¹ Over the years, I’ve given at least as much in contributions to Republicans as Democrats and supported multiple GOPers for president. I support smaller government, lower taxes for all but the wealthy, less regulation, more federalism, the 2nd Amendment (especially after the last four years), and Citizens United (for the same reason as I support strict Net neutrality). My position on immigration is what the GOP’s was prior to the rise of Trumpism, and while I’m an ardent environmentalist, so were many Republican leaders until the mid-1990’s. But I am what Trump supporters call a ‘hater,’ meaning I despise all authoritarians and dictators; Trump was already the former, and was aspiring to be the latter.   Back

¹² First, it seems clear, both measures have to be either all registrations or new registrations because the number of registered 90+year-olds per 100,000 is the same for both Democratic and Other counties, about 15 per 100,000, which given that PA has 12.8M residents comes to 1,920 total 90+year-olds, and the total number of 90+year-olds registered comes to about the same number. Presumably, the only reason we see both metrics on the chart is to show each of them vouches for the veracity of the other. Fine.

2.5% of PA’s population is 85+, which means there are 320,000 residents of the state 85 or older. Let’s assume that half of those people are between 85-89. That leaves 160,000 Pennsylvanians 90+. It wouldn’t pass the laugh test to believe, based on everything we know about the propensity of older generations to vote, that only 1,900 out of 160,000 of them–1.2%–are registered to vote. On the other hand, for the same reason, it’s very reasonable to expect that only 1,900 of them would be registering for the first time, because the vast majority of 90-year-olds are already registered. So it’s safe to assume what we’re looking at is a chart showing new registrations by 90+year-olds, both in total and per 100,000 Pennsylvanians. Back

¹³ If we were to look at the change in error by comparing the raw to the refined, rather than the refined to the raw (i.e. raw/refined rather than refined/raw), the correlation is actually even stronger in the negative direction (-0.51), but as we’re trying to avoid all forms of statistical sleight of hand and this one makes my head spin a little, we’ve opted to show you the weaker of the two correlations   Back

¹⁴ Neither NIU nor The Guardian’s rankings include Washington, DC, and in this case excluding the District just doesn’t sit right with us.  So we looked at how DC measured up against The Guardian’s criteria–top marks across the board (in fact, DC not only lets ex-felons vote, like ME and VT it lets prisoners vote too), then took a look at the five states who earned the same marks across the pond (OR, WA, IL, VT, and ME), then looked at where they fell in NIUs rankings (1st, 2nd, 4th, 9th, and 14th respectively), and averaged those rankings to come up with one for DC, 6th, the same as Hawaii, which we think is actually conservative, as we believe that had DC been included in the NIU rankings, it would have come closer to OR/WA/IL than VT/ME Back

——————————————————————–

Our Analysis of Pew’s Validated Voter Data On “Low Information Voters”

First we found out what proportion of the population in each swing state is Latinx, as well the proportion of Latinx individuals in the US population as a whole. We assumed the proportion of Latinxers under 18 (as well as the proportion of children in all other ethnicities) was the same from state to state (relative to the total proportion of Latinxers). Noting that while 18% of the population is Latinx, only 13% of the 2020 electorate was, we reduced the Latinx population under consideration in each state by the same proportion (thereby accounting not only for historically lower turnout in this ethnic group but also the higher proportion of the group who are non–citizens, and therefore ineligible to vote). And because the consultants’ claim is about “low information” (i.e. non–college) voters, we removed a proportion of the remaining Latinx voter pool equal to the proportion of Latinx citizens who have college degrees or higher.  Finally we took 21% of this adjusted Latinx proportion of the electorate (representing the 21 point decline of Latinx support for Biden vs. Trump compared to Hillary’s margin of victory, a shift we took, for simplicity, as the same in every state) and declared it to be the maximum possible impact of this shift on votes cast, other things being equal (which they never are). Finally, we took this (wholly unexpected?) shift of voters from Biden to Trump and compared it to the polling error in each state to determine what proportion of the “polling error” it could potentially account for.  When you do all this, you get the chart below:

As you can see, in only one state of the six with the greatest “polling errors” does the Latin shift explain as much as half the “error;” in four of the six it explains only 10% or less.  Since these are the states with the most egregious “errors,” this alone should tell us it’s (again) not a causal explanation of what happened, but let’s go on: in nine of the full set of thirteen, it doesn’t explain as much as half, in eight it doesn’t explain as much as a third, in seven (more than half) it doesn’t explain as much as 20%.  Of the other four states, all with smaller polling “errors” to begin with, three have large Latinx populations, so you really have to believe the just-so “I didn’t go to college so I don’t answer the phone or talk to pollsters” party line to give that explanation the credit, and the other one is Georgia, where the margin of victory and polling error were both so small, the results have a thousand fathers and a thousand deadbeat dads.

But again, as the ad says, that’s not all.  We can leave Black voters aside because the differences between Clinton and Biden were so small, but if we’re going to accept the “low information” tale of the tape, it needs to apply to all ethnic groups, and especially to the traditional poster child for the argument, the angry white working class voter who doesn’t trust nobody and slams the phone down every time he or she sees “Out Of Area,” “Private Caller,” or “Blocked,” if he/she even has a phone that shows this, replacing the violent response with a few choice words for the stranger on the other end of the line if it doesn’t.  And unfortunately for the consultant class, as we’ve foreshadowed, Pew finds this group swung 11 points in Biden’s favor vs. Hillary Clinton.

So we conducted more or less the same exercise again, with a few extra hoops thrown in.  Already knowing the proportion of whites nationally who lack college degrees, we found out how many whites there were in each swing state, how many whites have college degrees in each, what proportion of the residents of each state are children (assuming it’s roughly the same across ethnicities, and would, in any case, be closest to the proportion within the white population, except in states that are ‘majority minority’), then combined this information to get the proportion of adult whites in each state who lack college degrees.  Next we found out what proportion of the adult population in each state is white and used this, in combination with the proportion of whites without degrees to determine the proportion of the electorate in each that’s white non–college. Then we compared the proportion of the adult population nationally that’s white without a college degree, and compared it to the proportion of actual voters who were white non–college to arrive at a turnout adjustment we applied to each state.  Finally we took 11% of the adjusted number that results to arrive at the maximum impact the 11 point swing among white non college voters could have had, assuming none of it was already baked into the polls.

No, actually, then we did one thing more.  To get the full or net impact of all the shifts in the voting patterns of low information voters–in Trump’s favor in the case of non-college Latinx, in Biden’s in the case of non-college whites–we subtracted the max impact of the non-college white shift in each state from the max impact of the shift in non-college Latinx.  If this resulted in a positive number, then shifts in the low information population could account for some portion, however small, of the polling “errors” seen, but if negative, not only does this theory have no explanatory value, all the consulting Wile E’s that believed in it just found out there’s nothing but air beneath their feet and nary the slightest breeze beneath their wings.

We rest our case. Again.   Back