Why The Election Was Closer Than It Looks
(Slim Pickens riding the thermonuclear bomb onto Russia, the ultimate symbol of America. Dr. Strangelove 1964)
December 4th, 2020
The final count of the 2020 presidential election will be a national popular vote margin of approximately 4.5%, which will be a win of 7 million more votes out of about 158 million total. The Electoral College will, in the absence of faithless electors, be 306-232 in favor of Biden. There is an improbable symmetry in this. That was the margin of the 2016 election. What Trump once called --- dishonestly, and now ironically --- a "landslide" victory. While it is no landslide, it is a comfortable and clear, decisive win by President-Elect Biden.
It is especially relevant that Biden's victory is robust with a few different paths to 270, considering the election could only be overturned by flipping three battleground states. Whether this is accomplished by recounts, court decisions, or illegal actions by state legislatures or governors or election officials, is immaterial.
In the event that Michigan and Pennsylvania were unable to send electors, Biden would still be at the winning threshold of 270 electoral votes. If it were instead Michigan and Wisconsin, Biden would be at 280. If it were Pennsylvania and Georgia, Biden would be at 270. If it were Pennsylvania and Arizona, Biden would be at 274. If it were Arizona and Georgia, Biden would be at 279. Two states would be insufficient for overturning the election. It would be within reach for faithless electors or a single blocked slate, however, which is perhaps the goal (if there even is one) of the Trump legal team's strategy.
While this is the context for the low incentives Republican state and local elected officials have to "go rogue" and try to steal the election for President Trump --- and they have shown far more integrity and backbone on this matter than their sitting federal counterparts, becoming the subject of violent threats and deranged conspiracy theories --- it is difficult to know whether they would have behaved as well if the election was more singularly hinging on them. In spite of the considerations above, where Biden has a comfortable buffer, the election was dangerously close to a full blown constitutional crisis.
I. The Long National Nightmare
The country would have been plunged into months of turmoil and power struggles if the three closest battleground states had gone to Trump or otherwise been rendered indeterminant. While the alternative paths opened by Biden winning Nebraska's 2nd district were a source of robustness, with its single electoral vote keeping him at the edge of chaos in some outcomes involving two state losses, there was also a dark mirror of fragility in his loss of Maine's 2nd district. Catastrophe would have flowed from its single electoral vote if the three closest states had been lost or blocked.
The winning margins for Biden were +0.30% in Arizona, +0.26% in Georgia, and +0.63% in Wisconsin. These were 10,457 votes in Arizona, 12,670 votes in Georgia, and 20,682 votes in Wisconsin --- 43,809 votes in total. When you consider that a swing voter counts double for vote margins, subtracting one from Trump and adding one for Biden, the election was decided by as few as 22,000 voters across three states.
(Stealing Michigan but Biden holding Georgia is an identical scenario to this nightmare map, keeping Biden at 269 without Arizona and Wisconsin. While Biden holding Arizona, but having Georgia and Michigan overturned, is a hung 264-264 without Wisconsin.)
With Trump winning in Arizona and Georgia, Biden would be at 269 electoral votes. The whole election would have then depended on Wisconsin. With Wisconsin, Biden wins: 279-259. In the scenario where Trump wins Wisconsin, there is instead an Electoral College tie: 269-269. The third scenario is that, by hook or by crook, Wisconsin is sabotaged from sending its electors. This would result in a 269-259 split. In this situation a single faithless elector would be sufficient to swing the election to Biden, or else swing it to Trump, but only in the 269-269 scenario.
The Supreme Court ruled unanimously in Chiafalo v. Washington (2020) that states with binding laws can force electors to vote as mandated. Whether the states without binding laws then intervened on faithless electors could have resulted in another presidential election decided by the Supreme Court.
This would have been highly controversial as the Senate Republicans are regarded by Democrats as having abusively stacked the Supreme Court with a 6-3 conservative majority, that they "stole" Merrick Garland's seat and removed the Senate confirmation filibuster and even seated a judge accused of sexual assault by multiple women, and President Trump explicitly framed the rushed appointment of Amy Coney Barrett in terms of the Supreme Court giving him wins in election lawsuits. The Supreme Court would have had no choice but to either allow or stop the faithless electors. In the situations where Biden stays at 269 in the Electoral College votes, the House of Representatives would decide the winner in January with a contingent election. This would be done by state delegations, where Republicans hold the majority.
Thus, Trump would have been kept as President through a hyper-partisan vote in Congress, after a highly controversial Supreme Court ruling, and a Florida 2000 style crisis in Wisconsin. This would quite likely have happened through legal sabotage even if Biden won Wisconsin by the 537 vote margin Bush had in Florida. In this situation you would have had Biden winning about +4.5% and +7 million votes nationally --- and perhaps even winning the vote in Wisconsin --- but Trump keeping the White House through partisan courts and legislatures, blocked state certifications, and under the cloak of wild and baseless conspiracy theories about election fraud.
Imagine an ideal world with absolutely no "irregularities" or fraudulent votes where Trump loses the election. What would happen? Trump would immediately --- indeed, pre-emptively --- declare there was massive, systematic voting fraud. That the election was rigged by Democrats. His lawyers would try to get judges to throw out millions of legal votes on the thinnest pretexts. There would be many affidavits from people reading malicious behavior into processes they do not understand. The Internet would be filled with fabricated evidence, misleading videos, and shameless misinformation. Voting machines skewing vote totals, dead people voting, double voters, ballot stuffing, unsolicited mail ballots returned illegally.
Allegations of individual voter fraud, stacking redundantly on assertions of count rigging, would only significantly affect the margin if Democrats were much more likely to do it than Republicans. A dubious proposition at best. If it were a partisan voter fraud split of 55% Democrats to 45% Republicans in a single state, for example, closing a 10,000 vote margin would require throwing out an impressive 100,000 fraudulent votes.
Nevermind that all of these would be demonstrably false, that would only prove the Republican election officials are "in on it." There would be extraordinarily elaborate conspiracy theories without any credible evidence supporting them. Serious lawyers would back away from the Trump legal campaign, and only crank tinfoil hat wearers would remain. There would be a pressure campaign to get state legislatures, governors, and election officials to illegally overturn the vote to "stop the steal." It would look much the same as what we have seen over the past month. The difference is that in a truly up-in-the-air election, it would be much more likely to succeed.
This would have effectively ended democracy in presidential elections, as the paths taken to overturn the results would become precedent for any future erroneous claim of fraud. While Trump's lawyers are still trying to do this in the election, and the pardoned felons Roger Stone and Michael Flynn have both called for martial law, success with the absence of evidence means evidence would no longer matter. Biden would have as much grounds for falsely asserting he won Texas, Florida, North Carolina, and Ohio, alleging massive voting fraud against him orchestrated by the Russian G.R.U., a cabal of international bankers, and the lizard people of the Flat Earth Society.
This is not an original sin of a single party. There has been an ongoing current of institutional delegitimization of elections on both sides of the political spectrum for the past twenty years. Many on the left believe the Supreme Court stole the 2000 election, even though Bush still won Florida in later recounts. The controversy over Ohio in 2004 on the left sounds remarkably similar to the right in 2020.
Infamously, Barack Obama was baselessly presumed to be illegitimate by tens of millions of Republicans (including notoriously by Trump), with racist claims that he was a Muslim born in Kenya. Which would have been irrelevant, even if it were the truth. His mother was born in Kansas. Children born of a United States citizen are eligible to run for President regardless of where they were born themselves. John McCain was born in the Panama Canal Zone, George Romney was born in Mexico, Ted Cruz was born in Canada. Barry Goldwater was born in Arizona before it was granted statehood. In spite of popular myth, Alexander Hamilton was eligible to run for President even though he was born in the Caribbean, as he was a citizen of New York when the Constitution was ratified.
Hillary Clinton and Jimmy Carter have said Trump only won in 2016 due to Russian interference in the election. Left activists campaigned for faithless electors to overturn the outcome in the Electoral College, appealing to Hamilton's early conception of it as a "vote of conscience" safeguard against the mistakes of democracy. While this would have been a legal subversion of democracy, it was also deeply anti-democratic. Though not pursued by Hillary Clinton herself, who conceded the election, nor by Obama who cooperated with the peaceful transfer of power.
There are grains of truth in these things. But they are overstated and corrosive to the legitimacy of our elections. The Russians did interfere in the 2016 election, but it was not the reason Trump won the Electoral College. They were interfering in the 2020 election in various ways as well, along with Iran, who threatened Florida voters (against voting for Biden) by posing as the neo-fascist Proud Boys. There was a G.O.P. absentee ballot mill in the 2018 9th congressional district race in North Carolina, but it was easily caught and its operative was charged with felonies. Stacey Abrams merely "acknowledged" the Georgia 2018 gubernatorial race on the grounds of voter suppression. Matt Bevin falsely claimed voter fraud in the Kentucky 2019 gubernatorial election. Republicans have been searching for, and failing to find, voter fraud for decades.
But all of this is an order of magnitude less severe than the total and utter disregard for elections and the integrity of democracy demonstrated by the Trump administration. The State Department would be classifying it as an attempted coup, what is known as an "autogolpe" or self-coup, by an authoritarian strongman if this were any other country. Trump is brazenly trying to steal the election, and framing it as though he were the one trying to stop it. Much as any "populist" dictator who is the protector of true democracy.
However puerile it might be, his rhetoric to the effect of "I'm not [doing whatever], you are" works, at least on those who want to believe it. It is the formula that has given him a blank check for every crime and act of corruption. His supporters will then say: "Well, why not steal the election, since they are trying to steal it anyway?"
What Trump did with Ukraine, and now in the transition period, is far worse than what Nixon did in Watergate. The second term of the Trump administration, following a successful self-coup and his failed impeachment for election interference, would have seen widely expanded abuses of power. The end of civil service protections. The prosecution of political opponents and journalists. Secret police out of federal law enforcement and perhaps even right-wing paramilitaries as modern day brownshirts. Freikorps reprising yet another element of 1918.
It is not surprising that in the view of his supporters it is instead the Democrats, the radical left who want to destroy America, who are trying to end freedom and impose authoritarianism. Nor is the silence of the Republican establishment in Washington, even if it is more instrumental lack of support than complicity. They will not help him overturn the will of the voters, but they will also not condemn it, nor will they to try to stop it if it actually works. It was the center-right of the Weimar Republic, after all, who allowed the Enabling Act. Hitler was a buffoon. But he was a useful idiot. He would stop the communists. The world has not ended yet with him in power. What is the worst that could happen?
II. The Polls: Not Great, Not Terrible
The question of the "closeness" of the election is related to the issue of the accuracy of the polls. In truth the polls performed about as well as they usually do in presidential elections. There is a vicious cycle where people expect more accuracy than the polls claim, cherry pick the outlier polls in individual states, and then say "the polls were wrong" whether tight races tilted slightly the other way or the outliers were way off. In reality the polling aggregators average the state polls, which only presume to be "right" within several points either way, and then the mean absolute error will be around three or four points.
Illustrating this is a table constructed from the polling averages on the RealClearPolitics and FiveThirtyEight websites from the night before Election Day. Hypothetical cases were calculated in advance, showing what the results would be if the polling error from 2016 repeated itself exactly, and what each state result would be if there was a 5% error in Trump's favor. This number was chosen because the average error is about 3%, and most states will be within roughly 5% of their polling averages. One month later the actual vote margins were included.
(With the 2016 polling error Biden wins a handful of battleground states by less than 1%, but not necessarily the same states.)
This illustrates how the polls were no more "wrong" than 2016, that this outcome was not surprising, and that the polling errors in each state were fairly normal. Florida was more wrong than it has been historically. But the states that were wrong by more than 5% were the same ones the polls missed in 2016. Florida had an eyebrow raising polling error in favor of Republicans in 2018 as well. There is something in that for pollsters to fix. Consistent errors in the same direction suggest a causal root that may or may not persist in future elections. It is possible this may represent a growing skew of "anti-institutional" voters who pollsters cannot reach, or it might be something that goes away without someone like Trump. But historically normal error in a very abnormal election is no great sin. As Dyatlov said within the meltdown of Chernobyl: "3.6? Not great, not terrible."
The polling aggregators, for what it is worth, correctly "called" 48 out of 50 states. Most states were not close. Close states were, in fact, close states. While this was not a surprising result, the aggregators or pollsters cannot simply throw in a bias factor, because the polls do not consistently underestimate either party. They underestimated Obama in 2012. But he was the incumbent and leading in the polls, so it was not "shocking" like 2016. The winner won by more than he was expected to win. Meanwhile the 2018 midterm elections were, overall, the most accurate year for polls ever.
Forecasters such as FiveThirtyEight essentially do the opposite of assuming the polls are right. They make a probability distribution of outcomes based on the variation you would expect from the historical "wrongness" of the polls. The forecasts are agnostic about the direction of polling error, including the outcomes you would expect from polling errors favoring either party. They simulate the election tens of thousands of times with randomness, then count up the number of times the two candidates won, and the Electoral College spread. This represents the range of plausible outcomes with the available information, which means having a visible fraction of a dangerous outcome is a serious hazard.
Trump was never going to win a landslide, for example, that was far outside the range of outcomes. But it was possible Biden might have, even though Trump might narrowly win. Forecasters only give certainty in the sense of setting the scale of uncertainty. Beyond that it is up to the reader to understand that uncertainty and live with it instead of replacing it with higher certainty.
In the presidential election of 2020, Biden held a 7% or 8% national polling lead constantly for half a year, which meant plausible outcomes ranged from +4%-5% to +10%-11% popular vote margins for Biden. These would have been normal sized polling errors. There was no volatility or excesses of undecided voters to suggest the likelihood of last minute swings toward Trump, it was more of a question of how high the turnouts of infrequent or new voters might be by the end. The battleground states were a few points narrower, which meant a normal polling error for Trump would probably (but not necessarily) result in a narrow Biden win, and a normal polling error for Biden would be a "landslide" (by contemporary standards of +10% constituting a landslide) where he wins states like Texas.
In the much more unstable 2016 presidential election there were other forecasters, such as modelers for Daily Kos and the Princeton Election Consortium, who had Hillary Clinton with absurdly high win probabilities such as 98% or 99%. The New York Times Upshot model had Trump around 15% chance of winning. Those are Russian roulette odds. FiveThirtyEight had Trump at 35% odds in the week before the election, which is more like Russian roulette with two bullets. Those were better odds than Romney was given in 2012.
It is not terribly different from the 50% odds of a coin flip. Keep in mind that these are win probabilities rather than polls. The 15 point margin in a 50%-35% poll is huge. But it is only a minor difference if it refers to odds. 50% is 1:1 odds, 33% is 2:1 odds, 25% is 3:1 odds. Those are all very good odds in horse racing.
Counter-intuitively, 50% is twice as likely as 25%, while 95% is twice as likely as 90%, and 99% is twice as likely as 98%. Representing odds as percentages is part of why forecasters are misunderstood. There is also a problem with people ignoring the literalness of forecasts. When the National Weather Service says "10% chance of rain", it means 1 out of every 10 times they say that, it will rain. If it happens more often, or less often, the forecast is wrong. There is no rounding down to 0% or up to 100%. While you might say this is unfalsifiable as a presidential election only happens once, remember that the model is making projections for hundreds of races and fifty states, and they make forecasts for every national election.
If the presidential election were decided by a Dungeons & Dragons die roll, then FiveThirtyEight was saying if Hillary rolled less than an 8 on a 20 sided die (or a 6 with the Election Day polls), Trump would become President. If Biden rolled less than a 3 on a 20 sided die, Trump would stay President. Biden rolled a 5 or 6 out of 20. Equivalently, if it were a 100 sided die, Biden needed at least 11. He rolled something like a 25 or 30. Failing on such rolls is not unusual. In the case of Biden, you would expect Trump to win 10% of the time, and ten percent is not zero percent.
Hillary was only slightly favored to win the election after the Comey letter if you were following the numbers. It was not surprising, even if was "unthinkable." In contrast, FiveThirtyEight's forecast model had Biden at about 90% odds of winning (on the edge of "clearly favored"), and these odds were still almost unchanged late on election night at the height of the "red mirage" when Trump was falsely declaring victory. The Economist's model was more bullish. But the forecasters in general were consistently giving Trump around 1-in-10 to 1-in-20 odds of winning. This is ultimately because he would probably still lose the Electoral College even if there was a normal sized polling error in his favor. That is what happened.
It is a disservice to empirical reality to act as if "the polls are broken" and the "forecasters were wrong" when the result was within the normal range of expected outcomes. Nor is it valid to say the higher range was wrong in hindsight, now that we know there was a polling error in Trump's favor. When future knowledge is allowed, Biden had a 100% chance of winning, and Democrats had a 100% chance of keeping control of the House.
You might sniff that it matters more if a state is "called" wrong, even if it was only a tiny polling error, or that we only care about having certainty in very tight races. That we "know" there would not be an upset in the wider races regardless. (Nevermind that forecasters do not "call" races, no more than bookies do when setting odds.) But that is ignoring the fact that partisans often have unrealistic, or even highly unrealistic, expectations for the vote margins. Trump supporters in Pennsylvania were quoted before the election as saying the polls were wrong (or "lies") and that Trump would win 70%-30% or 80%-20%. It is no wonder when such people believe the election was stolen.
When confronted with huge landslide expectations, polling average errors of 3% or 4% are irrelevant. Issues polls, polls of public opinion, have similar errors. Their imprecision is perceived as irrelevant, they reveal the bulk "thereabouts." With horse race polls every variance is a sin. It is an unreasonable expectation. Sometimes an outlier poll, such as Biden +17% in Wisconsin, is embarrassingly wrong. Other times, such as Trump +7% in Iowa, it is right. Pollsters need to report their outlier results, it is up to the averagers to weight and smooth it out.
Polls are independent reality checks on expectations for vote results. Imagine how much more shocking Trump winning the election in 2016 would have been without polls. Who in the establishment press would have imagined he was even remotely close to Hillary? What we are seeing now is consistent with what you would expect if there was no widespread systemic voter fraud. Democrats in contrast are disappointed that the gains they would have made in the House and Senate, if the polling error had been in their favor, did not happen and they actually lost seats in the House. They failed to gain in state legislatures, except for New York. They lost every House race the Cook Political Report categorized as "toss ups."
However, the middle of the FiveThirtyEight forecast had them around 50 Senate seats, which they still have a chance of pulling off. They were at a high water mark after the 2018 midterms, trying to hold House seats in reddish districts. In actuality they did about as well as Biden, winning roughly as many House seats as districts where Biden won the vote (around 223 to 225), which means down-ballot ticket splitting had little to do with it. This is a situation of gerrymandering and Democrats being geographically concentrated in ways that hurt them in the House and especially the Senate. They were never likely to win substantial control in the Senate, and every single voting member House seat is up for grabs every two years. No one judges games by who beat the point spread. Wins are wins. The Democrats won the House and the White House. They might even win the Senate. Trump supporters want to boycott the vote in Georgia.
III. The Future of Democracy
The internal sniping between progressives and the center-left is missing the point because the Democratic coalition indispensably needs turnout from both wings. The Republican party is unified to a fault, surviving in a kind of ideological nihilism. But the "silent majority" is a double lie, it is neither silent nor the majority. The Republican party has now lost the popular vote in 7 out of the last 8 presidential elections. The only exception was 2004, where Bush was riding on the War on Terror, and held 45% of the Hispanic vote. The Trump style of elections through white grievance falls far short of this, he only improved relative to himself.
The Electoral College does not consistently favor either party over the long run. It happened to screw over the Democrats in 2000 and 2016, but it came close to splitting against Republicans in 2004. John Kerry would have won the Electoral College with 273 electoral votes if he had swung 60,000 people in Ohio who voted for Bush, even though Bush would have won the national popular vote by 2.9 million votes.
The Electoral College was not skewed against Democrats during the Obama administration, and Obama had significantly better odds of winning in 2012 than the polls naively suggested because of the Electoral College. It was biased in Trump's favor in both 2016 and 2020. In the 2020 election he managed to lose in spite of his structural advantage, which was the only reason the election was competitive at all. The Electoral College made elections be decided by 60,000 (swing) voters in 2004, 40,000 in 2016, and 22,000 in 2020. The national popular vote margins were 3 million, 2.9 million, and 7 million in those years. Of course, it was once infamously decided by only 269 swing voters, in the year 2000.
The Republican party is ill-posed for (legitimately) winning national elections without the freak luck of a popular vote and Electoral College split. This will be much less likely if and when Texas starts voting "blue" in statewide elections, which may happen by 2024 or 2028. (Beto O'Rourke came within two points of Ted Cruz in 2018. John Cornyn was at risk in 2020. The suburbs of Newt Gingrich now vote Democrat.) Consider the vote margins attained by the two major party nominees since 2000, as compiled by David Frum:
(1) Obama (2008): 52.9%
(2) Biden (2020): 51.3%
(3) Obama (2012): 51.1%
(4) Bush (2004): 50.7%
(5) Gore (2000): 48.4%
(6) Kerry (2004): 48.3%
(7) Clinton (2016): 48.2%
(8) Bush (2000): 47.9%
(9) Romney (2012): 47.2%
(10) Trump (2020): 46.9%
(11) Trump (2016): 46.1%
(12) McCain (2008): 45.7%
The bottom 5 of 12 are all Republicans. Only 1 of the top 7 is Republican. The Republican party is nevertheless unlikely to change course in the immediate future, as the party elites have capitulated so utterly to Trump (who can still run again), and because Republicans performed well down-ballot in spite of Trump's loss as well as record smashing turnout. Though Trump 2020 ranks 10 out of 12 on that list of national vote margins, it was the second highest vote total ever, five million votes higher than Obama in 2008. The 2020 election was very high turnout for both parties. But the high Republican turnout from "fracking for votes" and the pandemic was not enough to win in the Electoral College. While Trump did not come all that close to winning, he did come close to being close enough to steal it.
Trump performed better in urban areas than he did in 2016. Biden's win in the battleground states can be attributed entirely to his improvements on Hillary in the suburbs. However, both Republicans and Democrats had much higher vote totals than 2016, which is a potential source of uncertainty. The extent of turnout uncertainty causing polling error is unclear. But it is straight forward to compare the results if Trump had only received as many votes as he did in 2016.
While this is not plausible for a high turnout election amid a politicized pandemic --- and in fact it is inverting the bogus statistical argument in their court cases, where they claim Biden only had a one in a quadrillion chance of winning, assuming he only got a many votes as Hillary in 2016 but without Trump being similarly constrained --- it is also not unrealistic per se to use Trump's own vote totals from 2016 as a prior for how many votes Trump himself might get in 2020. The population is increasing over time, but it was only four years ago. (In spite of Trump's rhetoric about having won more votes than Reagan did in 1984, that is also true of every major party candidate since 2004.)
In other words, this is similar to what would have been seen in a normal sized polling error in Biden's favor, which has to be regarded as a quasi-realistic possibility (when removed from the context of 2020) given Trump was in both elections. Republican turnout having not been weak, by itself, is sufficient for explaining the lack of Democratic gains in the House and Senate. It is not necessary for the two wings of the Democratic party to blame each other for not winning by enough, because it was not swing voters failing to deliver. The suburbs did swing to Biden. It was the high influx of infrequent or new voters not being especially lopsided.
Trump had 11 million more votes than he did in 2016, but Biden had 15 million more votes than Hillary, who in turn had 3 million more votes than Trump in 2016. It is an important but subtle point that vote margins are not equivalent with the number of voters. While this is not intuitive, the reason for it is straight forward. "Vote margins" are not the same category of quantity as "votes." The margin is not itself a group of ballots. When speaking of "vote totals" there is a one-to-one relationship of "voters" to "votes", each vote in the absence of fraud corresponds to one ballot.
But this is not true, in general, for "vote margins." The margin is the difference between two vote totals. This is a statistical disparity between two aggregates, it is not a simple equivalence to number of voters. The vote margin is a product of multiple factors, and cannot be uniquely back imputed to individual voters.
In the extreme case where the two vote totals would have been identical in the absence of (excess) new voters, then the whole vote margin does equal the number of excess new voters. However, the margin would more generally be impacted by voters who choose to vote for a third party, or write-in, or not even voting for the presidential race at all. In the opposite extreme case where the margin is entirely determined by swing voters, between only the two major party candidates, then the vote margin is exactly twice as large as the number of swing voters. This would be because the one vote total loses a vote, and the other vote total gains a vote simultaneously. These kinds of votes are all mixed together.
Imagine having an election consisting of a room with 100 people in it. The Senate would be such a chamber if it was wholly decided in the same election. In practice several people may switch sides, but these will partly cancel out. If there is one net person switching sides, this causes a 2% shift in the voting margin. The vote totals would become 51-49 instead of 50-50.
Imagine instead a hypothetical state where Biden and Trump each have 1,000,000 votes. Now 50,000 voters decide at the last moment to change their vote from Biden to Trump. Biden has lost 50,000 votes, his vote total is now 950,000. Trump has 1,050,000 votes. There is now a vote margin of 100,000 instead of 0, or equivalently (52.5%-47.5%) a 5% margin of victory. The 100,000 vote margin was only caused by 50,000 ballots, even though the winning side has 100,000 more ballots.
With the incredibly low response rates that pollsters get, and the weighting they have to do to unbias their data, it is really impressive that the polls are able to get elections right to this level of accuracy. The polls have to be read as having unknown systematic biases in each state, but within a few points on average, with similar errors in similar states. Allowing for this offset, polls show the trends, and the level of volatility. This allows us to infer the likelihood of who will win --- not because we assume the polls are right, but because we can be confident about the size of their wrongness.
What will not happen are "Dewey Defeats Truman" severities of polling error. Landslides by the 20th century standard of margins no longer happen. Ideological or cultural partisanship and party tribalism have made the vote very inelastic compared to a few decades ago. Of all the lessons that might be learned from the Trump presidency, throwing away our empirical grounding to factual reality in a fit of hyper-partisan solipsism would be among the worst, as well as the most ironic.
