America’s on notice that, if Trump’s in trouble, the post-election fur will fly.
Listen To This Story
|
With the polling deadlock glued in right up to E-Day, it remains to be seen whether the consensus GOAT sore loser will be moved to outdo himself in making the holidays a living hell for the nation. He might well win the Electoral College, and then, aside from some desultory grousing about having his popular vote landslide stolen from him, we’ll be spared any short-term upheavals.
But should things not be dormie for Donald Trump, America has been put on notice that the post-election fur will fly.
Stop the Steal 2.0 has long been under construction, with legions of lawyers and operatives ready to pounce on every apparent anomaly, however benign the explanation; every inevitable human error, however quickly addressed and corrected; and everything that might look “fishy,” even if only through a special fish-eye lens.
We saw it in 2020 — from Antrim, MI; to Atlanta, GA; to Maricopa County, AZ, and a hundred other venues — a slew of charges of fraud that turned out to have no basis. Stop the Steal 1.0 rested heavily on a stubborn, bordering on wilfully moronic, refusal to comprehend that, because mail-in ballots skewed heavily Democratic (because Trump instructed his MAGA voters not to vote by mail!), the slower counting of those ballots in many venues meant that early Trump leads would be gradually eroded and ultimately overcome. “Look, they’re stealing it!!”
Over 60 lawsuits were filed by Team Trump — all (with one very minor exception) dismissed, including quite a few by Trump-appointed judges. Accompanying those charges and suits were a few ‘damning’ statistical analyses, so easily punctured that it would have been laughable, had not so much damage been done.
Lost in the gyrations of the St. Vitus dance set off by Election 2020 and Stop the Steal 1.0 was the actual data, the polling and voting patterns that, when objectively analyzed, painted just about the opposite picture from the grand theft Trump was screaming about.
The damage to the trust levels on which our democracy depends was massive. It would be hard to miss the “something in the air” that, over the past four years, has set America on edge and Americans too often at each other’s throats.
And the damage continues, because somehow, in spite of the baselessness of 2020’s Stop the Steal 1.0, in spite of the fact that none of its many challenges and allegations panned out, Trump got MAGA-World to swallow it whole.
It is the primary article of faith among the faithful — cynically reinforced by the officer class, for whom election denialism became a litmus test — that Dear Leader’s “sacred landslide” was stolen. And it is that bone-deep conviction, false as it is, that is set to provide the jet fuel for Stop the Steal 2.0: “Damned if we’ll stand by and let them do it to us again! And, hell, if necessary, why not just steal it back!? They owe us!”
Lost in the gyrations of the St. Vitus dance set off by Election 2020 and Stop the Steal 1.0 was the actual data, the polling and voting patterns that, when objectively analyzed, painted just about the opposite picture from the grand theft Trump was screaming about.
What the numbers showed beyond argument was that, if there was in fact any major error or malignant interference with the 2020 election, its beneficiaries were Donald Trump and the GOP, up and down the ballot.
I published a version of that analysis here in August 2021 and I recommend the updated version for anyone who wants a data-based view of what actually went down in our last presidential election.
I want to recapitulate it here, in summary form, because I believe it is vital to both an understanding of how warped and mythical Stop the Steal 1.0 was and, at the same time, a recognition of the things that look like they did go bump in the night of our last presidential election, and may do so once again, for the benefit of the very candidate (and his party) who is now running as the furious, vengeful victim of fraud.
Let’s look at the numbers.
Presidential
Taking it, as it were, from the top, Joe Biden’s final popular vote margin was just slightly over 7 million votes. It is no secret that the pollsters got it very “wrong”: The final polling aggregate (in which polls are combined, with more recent and more historically reliable polls weighted heavier) predicted a Biden margin of 12.6 million votes, roughly 5.5 million greater than it turned out to be. The national exit poll, with a sample size of 15,234 and a margin of error (MOE) of less than 1 percent, corroborated the pre-election polling, putting Biden’s margin at just over 13 million votes.
Of course, the whole polling industry, including the exit pollsters, came away with great gobs of egg on their face for what were assumed to be huge misses. How could they possibly miss all those Trump voters!?
Drowned in the laughter and catcalls was a very telling metric that suggested the pollsters hadn’t missed the Trump voters at all.
The key question, asked in all the polls, was Trump approval. In the aggregate pre-election polling, Trump approval stood at 44.6 percent; in the national exit poll it was 46.0 percent. The Trump approval baseline at the end of October 2020, just before the election, was quite stable, hovering between 42 and 44 percent. So there were, if anything, slightly elevated levels of Trump approvers in both sets of polls that gave Biden close to double the margin he wound up with.
In effect, the Trump approval metric validated the polls, effectively ruling out anti-Trump bias of their samples.
Of course the popular vote is not what wins presidential elections. Stop the Steal 1.0 targeted the swing states in which Biden’s margins were slim (several of them the same states that Hillary Clinton had declined to challenge in 2016, when Trump’s margins were slim). Both pre-election and exit polling in every one of these states showed Biden winning and, in all cases, with margins greater than in the official results.
In Wisconsin, for example, the “red” shift from pre-election polls to official vote shares was 7.6 percent; from exit poll to vote share an even heftier 9.8 percent. In Michigan, those numbers were 6.4 and 5.2; Arizona 5.3 and 2.3; Nevada 3.8 and 3.7; Pennsylvania 1.8 and 3.5; New Hampshire 2.8 and 3.3; Georgia 2.6 and 0.7.
It bears repeating: In every case, there was no numerical support whatsoever for Trump’s Stop the Steal claims, rather just the reverse — a statistical pattern of expressed Biden voting intentions not fully translating to Biden votes.
Hold that thought while we work our way down the ballot.
The Senate
The Senate, I think we’re all aware, is constituted to over-represent small, rural states, which tend to be red. At the last census, 17 percent of the US population was represented by half the senators, 83 percent by the other half. It is something of a miracle, from the Democrats’ standpoint, that the Senate remains roughly evenly divided.
But so it is and, biennially, with a third of its membership up for election, control of the Senate has inevitably come down to a handful of competitive contests. This year the “map” is favorable to the GOP, with the Democrats defending more seats in swing states like Michigan and red states like Ohio and Montana.
In 2020, the map somewhat favored the Democrats and pre-election expectations were for an outright Democratic majority. The Senate wound up 50-50 (with Vice President Kamala Harris’s tie-breaking vote giving the Democrats a bare working majority), but only because Trump’s Stop the Steal strong-arming and caterwauling in Georgia cost the GOP the state’s two very winnable Senate elections that went to January runoffs.
Back in November, three states — Iowa, North Carolina, and Maine — shifted from Democratic leads in final pre-election polling to Republican victories, the most egregious being Maine, as detailed in the full analysis linked above; no states shifted the other way. Three Senate seats may not sound like much but, given the delicate balance of the Senate map, it’s the whole game.
The House
In the House the results were — there’s no other word for it — shocking. Because of precision gerrymandering, only about one in five House contests are remotely competitive, of which perhaps a third are truly competitive — a total of 30 to 40 seats in any given election. In 2020, the nonpartisan and highly reputed Cook Report categorized the House contests as “Expected to Win Easily,” “Expected to Win Narrowly,” and “Toss-up.” There were 27 toss-ups.
The GOP won every toss-up, all 27.
If you flipped a penny 27 times and it came up “tails” every time, you’d certainly wonder about that penny. But the rout didn’t stop there: Of the 36 races Cook expected the Democrats to win narrowly, the GOP picked off 7; of the 26 races the GOP was expected to win narrowly, they lost none. The overall impact was to take a projected Democratic net gain of 15 or so House seats and turn it into a Democratic loss of 14 seats, leaving the Democrats with a narrow 222-213 majority ripe for overturning in the 2022 midterms, in which the party in the White House traditionally bleeds House seats — and did so, giving the GOP control of the House despite that party’s poor showing.
Even at the state legislative level, where the Democrats were primed to cut into the GOP advantage in control of chambers, the Republicans instead defied expectations and increased their advantage.
A rough sketch of Election 2020: Biden won, with a much lower than expected margin; Trump screamed “Fraud!” from the rooftops with no evidence to back it up; and down-ballot Democrats were unexpectedly clobbered, Biden’s “coattails” being profoundly negative.
But, since all of these strange departures from expectations were functions, in large part, of polling, all we have really been able to establish to this point in our examination, is that there were either problems of some sort with the election or a massive polling failure, or perhaps a bit of both.
Which is what makes this next (and last, I promise!) analysis so important.
A Key Poll-Independent Indicator
The default position regarding US elections has long been that when the polls do not match the vote counts, however egregious the disparity, it’s the polls that will be deemed to be “off.” Trump’s all-out and continuing assault on democracy and the legitimacy of our elections has only strengthened that resolve and the blanket refusal to consider the possibility that the fault may not lie exclusively in the polling.
With that in mind, I combed the E2020 data dump for an indicator of some sort that was not poll-based. In examining noncompetitive contests for the US House, I found one.
The scatterplot below includes only those 2020 US House contests that were rated either Solid Republican (red dots) or Solid Democratic (blue dots) by FiveThirtyEight. Altogether there are 324 contests; the 74 competitive and 37 uncontested US House races have been excluded. For each of these 324 contests, FiveThirtyEight generated a predicted result: a winner and both a vote share and a win percentage (likelihood of winning) for each candidate.
However, because these contests were all seen as noncompetitive, virtually none of them were polled.
The predictions, therefore, were based almost entirely on other factors known in the biz as “fundamentals,” such as voter registration, demographics, prior results, prior candidate performance, campaign expenditures, etc. — none of which involved sampling and questioning of respondents. This is, as will be seen, of great significance.
In the scatterplot, the x-axis shows the margin of victory, Democratic being positive (+x); the y-axis shows the disparity between predicted and actual results, with “red shift” (results better than predicted for GOP candidate) being downward (-y). The crossed circles represent mean x,y values for each group, Solid-R and Solid-D.
The first thing to note is that no dots cross the center line (y-axis): Blue are all +x and red are all -x, which means that all 324 predictions correctly identified the winners.
The next is that the distribution is clearly (and oddly) bi-modal with respect to y — that is, the blue and red clusters are visibly distinct in their y-values or degree of red shift. The Solid-R mean is just barely below the x-axis, a y-value of -0.9 percent, a very minimal red shift. The Solid-D mean, on the other hand, is well below the x-axis, a y-value of -5.7 percent, a major red shift.
We naturally asked what might account for such a distinct pattern. Why were the predictions so much worse in places with high concentrations of Democratic voters (and votes) than in Republican strongholds?
When I first showed this pattern to one-time pollster (CBS) and long-time polling expert David Moore, his first thought was a particular type of sampling bias in which GOP/Trump voters would be more comfortable responding to pollsters in Republican strongholds where they were in the majority than in Democratic strongholds where they were a dwarfed minority — call it the “shy Trump voter away from MAGA-land” hypothesis.
There was a working theory out there that Trump/GOP voters in general are more likely to be “shy” and refuse to respond to polls (which they associate with the despised, liberal, “fake-news” media) than are their Democratic counterparts — a theory often trotted out to explain the major red-shift disparities of 2020 (and prior elections) as a massive polling failure.
It certainly is no great stretch to extend that theory to take into account the politically friendly or hostile environments in which such voters find themselves. It’s quite plausible that Trump voters would be more shy when approached on “enemy” turf.
But recall that, with very rare exceptions, the predictions that generated this scatterplot were not based on polls. There could hardly be any “shy” Trump/GOP respondents if there were virtually no respondents!
The scatterplot above consists of a large number of data points and the statistical significance is very high. We are not surprised by the relatively wide dispersion of the y-values in both clusters — we expect a fair amount of variance or “noise” with non-poll-based predictions. But we would expect roughly the same mean y-value (or red/blue shift) for both clusters — there should be no correlation between partisanship and prediction-to-result shift.
The line connecting the crossed circles in the plot above should be horizontal, not a slant.
So something clearly was happening in Blue America that was not happening in Red America. Were these particular contests targeted for manipulation? Of course that makes no sense: These were blowout races and, as we noted, every winner was correctly predicted. No rational rigger would target any of these contests.
How, then, to read this riddle?
Our hypothesis is that these contests resided on ballots in which other contests may well have been competitive and of national significance, and therefore attractive targets for interference of one form or another.
Many of the blue dots above represent gerrymandered urban congressional districts in presidential or senatorial battleground states; some even contain within them or overlap with competitive state legislative districts. These Solid-D US House races were not targeted but the scatterplot pattern suggests that at least several million whole ballots on which such races resided were either not successfully cast, went uncounted, or were mistabulated (i.e., flipped).
The “safe” US House contests that “went along for the ride” were, in other words, collateral and politically insignificant damage, but a strong signal of process disenfranchisement of predominantly Democratic voters living in Democratic strongholds.
Many of these would be the very voters of color, residing in primarily urban and easily identified zip codes, who have been the targets of both open and, we suspect, also less visible GOP suppression tactics.
They might be seen waiting on hours-long lines — many leaving, by necessity or in discouragement, before voting. They might be dropped from the voting rolls in sweeping, targeted county- or state-wide purges. Their mail-in ballots might be delivered late or not at all. Their signatures might be rejected as “not matching.”
The fact is there are many ways to tamp down on the casting and counting of whole ballots — especially in venues where the votes to be suppressed are heavily clustered and readily identified by the intersection of demography and geography. The same “Big Data” used for ruthless, precision gerrymandering easily flags precincts and zip codes for suppression. As Willy Sutton put it when asked why he robbed banks, “Because that’s where they keep the money.”
The scatterplot above suggests that “banks” holding lots of Democratic ballots were “hit” in 2020.
This signal of large-scale interference with the casting and/or counting processes jibes with the investigative reporting of Greg Palast into voter suppression, specifically the systematic suppression of the votes of traditionally Democratic constituencies. Throughout the computerized voting era, Palast has uncovered illegal purges of voter rolls as well as disqualification and spoliation of ballots in numbers large enough to produce the scatterplot’s extra 5 percent red shift in the Democratic areas. Needless to say, voter suppression is currently alive and well — ramped up, in fact — wherever MAGA rules. The courts have been, and will continue to be, very busy.
In addition, the surge in mail-in ballots spurred by the pandemic and the tweaking of voting protocols to accommodate it made the US Postal Service a potential choke point for vote casting, whether through delayed delivery of requested ballots to voters, delayed delivery of completed ballots to counties, or the loss or destruction of ballots in USPS custody. It remains unclear whether Trump’s postmaster general, Louis DeJoy, still on the job, put his thumb on the scale in any of these ways.
What is clear is the signal flashed by the scatterplot: a significant red shift of vote counts relative to non-poll-based predictive baselines, collectively impacting venues with a high concentration of Democratic voters — millions of votes, nearly enough to account for the 6 million vote red shift in the presidential race, and likely impacting competitive Senate contests and state legislative contests as well. If the same level of interference were extrapolated to the competitive US House contests, it would also provide an explanation for the Republican 27-for-27 table-run of “toss-up” races and flipping of seven Democratic leaners.
Can 2024 Be Rigged Without Being Challenged and Challenged Without Being Rigged?
There’s more in the full analysis, including county-level drilldowns and a suspect pattern we’ve dubbed “the Cuba plot” for its distinctive shape. Like virtually all statistical forensics, it falls short of proof of “rigging,” at least if we define that term as covert manipulation of election outcomes.
But it does emphatically put the lie to Trump’s Big Lie, to his still thriving claim that the last election was stolen from him. And it does paint a disturbing picture of what has become a chronic shift to the right from predictions, expectations, and just about every measure of the electorate besides the count of votes.
Stop the Steal 2.0 is a major upgrade over Stop the Steal 1.0, far better prepared to challenge any unfavorable results in any elections of significance — at the precinct, county, and state levels; in the courts; in Congress; and on the streets.
This time around, for all the bashing, polls have been more prominent than ever. Not just the junkies, but much of the public has been clinging to every zig and zag as America awaits its fate.
In one camp are the pundits who point to 2016 and 2020 as evidence that Trump habitually outperforms the polls and is on course to win on Tuesday.
In another camp are those who suspect the polling “herd,” stung by its massive miss in 2020, has over-corrected, nudging their polls too far in Trump’s direction. They see the polling deadlock as auguring a Harris win.
There are even some, including myself, who wonder whether the same factors that might have caused Trump’s outperformance in 2016 and 2020 are still lurking, such that the pollsters’ “correction” will get it just right.
Which of course would mean a lot of close results. Which would play straight into the plan of Stop the Steal 2.0.
I believe that election security has been enhanced over the past four years — somewhat better audits, more votes cast on paper, certainly more scrutiny.
But it’s also evident that Stop the Steal 2.0 is a major upgrade over Stop the Steal 1.0, far better prepared to challenge any unfavorable results in any elections of significance — at the precinct, county, and state levels; in the courts; in Congress; and on the streets.
Driving those challenges will be the MAGA Gospel Truth that the last one was stolen, Trump’s “sacred landslide.” I hope what I’ve presented here, and the more comprehensive analysis linked, makes clear just how Big that particular Lie is, while also alerting you to the tendency of our elections to, for whatever reason, chronically defy expectations in just the opposite direction.
I will be collecting and analyzing the 2024 data as objectively and as rapidly as I can. The truth is that right now, as the late William Goldman once put it, nobody knows anything. I include myself in that. I’m waiting, as I’ll bet you are, on tenterhooks.
There’s not much more to be said, but there’s still much work to be done. My colleagues and I will be putting on our election forensics hats and preparing to snag and crunch the numbers, looking for the truth and the signal amidst all the noise. Buckle up and stay tuned.