Eric Neufeld writes:
A Guardian review of a recent book about Biden’s victory, entitled “Lucky” that’s that “Seven million votes more was almost not enough. Had 45,000 gone the other way in Arizona, Georgia and Wisconsin, Donald Trump would still be president.”
It strikes me that if *those* 45,000 votes had counterfactually gone the other way, many other things would have been different. Would any of your readers have the data and machinery to answer this?
Yeah, this sort of question has come up before. My quick response is that it makes sense to assume a uniform national swing compared to what happened. According to wikipedia, the electoral margins in Arizona, Georgia, and Wisconsin were 0.31%, 0.24%, and 0.63%. Flip these and you get a 269-269 tie. To flip all three states would require a 0.63%*(1/2) swing from the Democrat to the Republican, which at the national level would correspond to (81.3 million + 74.5 million)*0.0063*0.5 = 500,000 votes.
So, to first approximation, to bring Trump to a tie would’ve required a switch of half a million votes. Or a million additional Trump votes, nationwide.
There’s no good reason to consider a scenario in which he’d win 45,000 votes in those three states and not have done better elsewhere.
Umm, targeted voter fraud? Perhaps with the help of some Fancy Bears?
A lot of people argued after the 2016 election that Clinton would’ve won if she’d actively campaigned/invested in WI and MI, which she arguably took for granted and ended up being surprise, narrow victories for Trump. If they were correct, that would be a scenario where those individual states might have moved without a national trend. But WI and MI are relatively similar in terms of geography and demographics, so it makes more sense that they might’ve moved together against the national trend. In the case of 2020, though, AZ, GA, and WI have neither geography nor demographics in common–all they have in common is that they’ve been close recently.
I guess one could argue that there is a stochastic element in every (contested) state race, due to things like weather, local GOTV, general butterfly effects, etc.; that these effects usually cancel but, at some point in the long run, will inevitably line up in the same direction in a given state or states; and that states where the margin is tight are more likely to change outcomes in that case. So maybe Biden “luckily” won because random factors in close states happened to line up in his favor. But who’s to say he didn’t win in spite of those factors lining up for Trump?
It’s impossible to say that a winning candidate’s tight margin is evidence that they should’ve built a larger margin in case the race had been close, because clearly they had the margin they needed for a close race! In a way, it’s actually analogous to post hoc power estimation. In both cases, it assumes you can know the direction and magnitude of errors, which by definition are confounded with the result.
> A lot of people argued after the 2016 election that Clinton would’ve won if she’d actively campaigned/invested in WI and MI, which she arguably took for granted and ended up being surprise, narrow victories for Trump.
Two interesting non-conflicting facts
1. Clinton campaign’s mistake of ignoring Wisconsin looks really really bad in hindsight.
Wisconsin has been the pivotal electoral college state for two straight elections (in 2016 and 2020).
Even in 2012, it wasn’t very far off. Colorado was the pivotal state with margin D+5.4% while Wisconsin was D+6.9%. Other “close” states on the D side were Pennsylvania (D+5.4%), New Hampshire (D+5.6%), Iowa (D+5.8%), and Nevada (D+6.7%)
2. Clinton campaign’s mistake of ignoring Wisconsin did not cost her the election.
Ultimately, even if she won Michigan and Wisconsin, she still needed Pennsylvania, Florida, or both Arizona and North Carolina.
She definitely focused on those states and lost all of them.
Yes to the general point of both of these: Given that campaigns invest at the state or sub-state level, and that a candidate can only be in one place at a time and thus necessarily chooses where to invest his/her campaign time, I don’t think it’s at all true that the first-order model must consider electorate-wide swings, with second-order effects coming off of that. I suppose it depends on what you are holding equal in an ‘all else equal’ sort of analysis. Or perhaps “first order in what parameter(s)?”
Presumably Trump could have invested more time and money in those states and less in others, perhaps attaining an electoral college victory while earning the same (or fewer) total votes as happened in reality. I don’t see why that’s a less valid way to look at the issue than looking at electorate-wide swing. Indeed, it seems more realistic.
Alls I can say it’s a good thing Russia didn’t bring out it’s top cybergame. Hilary gets it: they did just enough to get The Trump over The Hump. It’s really down to the precise application of assets. The Russians got that pretty much.
> A lot of people argued after the 2016 election that Clinton would’ve won if she’d actively campaigned/invested in WI and MI,
it’s a nice theory and all, but it doesn’t hold up. Russ Feingold got fewer votes than Clinton, and presumably he was campaigning and invested in WI!
What differences of consequence would there be?
Anon:
If those 45,000 votes had gone the other way (along with the expected half million swing elsewhere in the country, conditional on that shift), the electoral vote would have been a tie, and it’s likely that the election would’ve gone to Trump, and then I think there would’ve been many differences of consequence.
This is off the point of this thread, but the fact that Biden could have gotten six and a half million more votes than Trump but still lost the election is a reminder how far we are from a real democracy.
+1 There is a simple solution. Congress has the power to set the number of representatives. In fact, it has an obligation to do so under the Constitution. Originally, one Congressman represented about 70,000 people. Today it is around 700,000. Congress could increase the number of representatives and then the bias of the number of electors who represent relatively small states (two for each state) would be nullified.
“Biden could have gotten six and a half million more votes than Trump but still lost the election is a reminder how far we are from a real democracy.”
You can’t be serious. :)
The “democracy” that you seem to think is the ideal reality doesn’t now and never will apply to anyone, anywhere in the universe.
In my state there are over 30 counties, but if you win in one you win in the state. Maybe you call that “Democracy” but I don’t. There’s a reason that there is a geographic component to presidential elections. That reason is that if the part of the country that gets excluded – like the 30+ counties in my state – leave the union, the union would be substantial weakened, to the point that its viability would be in doubt. Imagine a country that runs from the Rio Grande at the Big Bend to the Arctic ocean, and two other countries on either side of it.
So you can keep bloviating about “democracy” but the reality is that without geographic concurrence you lose anyway. I mean I **am** impressed by your heroic effort and chutz, trying to make it look like liberals somehow would have been cheated by a marginal loss when they should have won by a landslide against a dolt like Trump, but sorry I don’t think people are quite buying that angle, which suggests a reason why the left hasn’t been able to sell it.
Anon:
You say that John’s idea (electing the president using the popular vote rather than an electoral collect) “doesn’t now and never will apply to anyone, anywhere in the universe.” But actually many countries elect presidents by popular vote. There are historical reasons why the U.S. uses the system it uses; other countries use other systems.
I get that. The most important things going on now are high inflation along with shortages then Russia invading Ukraine.
Both are things that have been brewing for awhile, so
I don’t see how Trump would have made much of a difference there.
Perhaps China would have attempted to invade Taiwan already if Trump had won?
How can interest rates rise above the inflation rate without requiring nearly the entire tax revenue of the US to pay the interest on the national debt? It is already $500B /yr at ~2% interest rate.
Do either have a plan, or even mention this?
Anon:
You’ve made an argument I’ve heard many times before, that both the Democrats and the Republicans are bad, featuring some combination of incompetence, corruption, and inappropriate goals. That may be so, but that does not mean there are no differences between the two parties in a wide range of policy areas including taxes, spending, immigration, foreign policy, environmental policy, health care policy, abortion rights, gun rights, and lots of other things. The fact that both parties annoy you does not mean there are no important consequential difference between who is elected. Or, to put it another way, the fact that the election outcome has major consequences does not imply that the winning candidate will implement good policies or have a good plan for debt, climate change, risks from terrorism, and other challenges. We may be going to hell in a handbasket, but there are many different paths to get there, as well as many different sorts and degrees of ruin.
> How can interest rates rise above the inflation rate without requiring nearly the entire tax revenue of the US to pay the interest on the national debt?
Easy, just sell the electricity generated by Irving Fisher’s spinning corpse.
If Trump were president, these crises would still be going on, but I suspect he’d make them worse, plus we’d have additional crises of his creation/contribution. Trump loved to create problems and fan flames if he thought it would distract people, defer blame, or rally his base. Think of all the crises we’ve avoided because the feds have responded reasonably and/or weren’t fanning the flames. Were Trump in charge, we’d be talking about how:
*The feds still haven’t given states enough funding and organizational support for vaccines, and their messaging is mixed at best and undermining confidence at worst, so vaccination rates are still low;
*Jan. 6-style, armed occupations of statehouses in blue states are a recurring problem;
*At Trump’s urging and with the support of federal law enforcement, trucking convoys are blockading ports, border crossings, DC, and big liberal cities, further disrupting supply chains;
*The US is tying up efforts to punish Russia using global monetary policy and sanctions, undercutting efforts to send military aid, refusing to take refugees, buying up Russian oil, and trying to play EU countries against one another;
*Trump blames China, Canada, and Mexico for supply chain problems, and his solution is to ratchet up tariffs and trade barriers;
*Confidence in US elections is at an all-time low thanks to the messaging and actions from Trump’s latest commission on voter fraud/corrupt Secretaries of State/hacked voting machines;
*Entire cabinet departments (Ed, HHS, HUD) are running on skeleton crews because Trump won’t appoint permanent officials, has fired most existing bureaucrats for not being “loyal,” and won’t draw down funds approved by Congress;
*The Bidens (father and son) are under federal investigation for trying to overthrow the government in collusion with foreign powers;
*An executive order leads to US soldiers being dishonorably discharged for providing transportation for pregnant women from bases to abortion clinics;
*A recording of Trump making a handshake deal not to oppose Israeli annexations so he can build the next Trump Tower there is the center of a third push for impeachment; And so on, and so on…
Okay, that last one is a stretch, but it’s a placeholder for whatever legit corruption scandal he’d be in that we can’t even predict right now because it would be 100% Trump’s creation (like the Ukraine weapons thing in his first term). You MAY be able to claim business-as-usual Republicans aren’t worse than Dems, but that man *created* disasters.
Can you expand on this? Do you mean you expect inflation to soon drop back near 2%?
> There’s no good reason to consider a scenario in which he’d win 45,000 votes in those three states and not have done better elsewhere.
I’m not sure what constitutes a “good reason” but if you didn’t already know the actual outcome would the alternative scenario be so inconceivable?
If you knew only the results in the rest of the states could you predict with high certainty that Arizona, Georgia and Wisconsin did fall on one side and not on the other?
Phil, Carlos:
Don’t forget that the campaigns were already trying their best to win and were campaigning hard in swing states. One way to get a sense of alternative outcomes is to consider the joint predictive distribution of outcomes in all 50 states, based on the models fit before the election. Any reasonable forecast would have high predictive correlations between states. All things are possible, but the most natural way to get a swing of X% in three particular states would be from something approximating an X% national swing. Yes, there’s one particular scenario where a 45,000-vote swing would’ve changed the outcome of the electoral college. There are also lots of scenarios where a swing of 1 million or more toward Trump would not have changed the outcome. Or, for that matter, scenarios where there was a 45,000-vote swing toward Biden with Biden losing the election, depending on how the swing is distributed. I don’t see any reason to privilege the one particular potential outcome where there’s a fixed swing in those three states and a zero swing everywhere else. A more reasonable starting point would seem to me to be a uniform national swing representing national changes in vote attitudes and turnout.
> One way to get a sense of alternative outcomes is to consider the joint predictive distribution of outcomes in all 50 states, based on the models fit before the election. Any reasonable forecast would have high predictive correlations between states. All things are possible, but the most natural way to get a swing of X% in three particular states would be from something approximating an X% national swing.
Is it the most likely way – according to those reasonable forecasts? It’s not a rhetorical question, I don’t know.
Is the ex-ante probability – at every time prior time – higher for the counterfactual where Arizona, Georgia, and Wisconsin results move by 0.31%, 0.24%, and 0.63% or for the counterfactual where every state swings 0.63%?
And that’s one way to frame it but not the only one. One could also look at the ex-ante probability of the actual outcome and the counterfactual where Arizona, Georgia, and Wisconsin results change by 0.31%, 0.24%, and 0.63%. Maybe the latter wouldn’t have been more surprising than the actual outcome, I don’t know.
There are many other potential outcomes, sure, and other questions more interesting and relevant. But that specific “what if” could still be a legitimate scenario.
Carlos:
It’s a legitimate scenario, one of thousands. After the fact, it’s an easily visualized scenario but it’s still just one of many possibilities. Kinda like if you go the casino and the roulette ball lands on 30, 12, 17, 4, in that order, then you can think, “What if I’d bet on 30, 12, 17, 4, in that order? Then I’d be rich!” But it’s just one of 38^4 things you could’ve done.
Agreed! The scenario where you switch 500’000 votes instead of 45’000 is also one of thousands.
> So, to first approximation, to bring Trump to a tie would’ve required a switch of half a million votes. Or a million additional Trump votes, nationwide.
That’s the part I don’t necessarily agree with. I’m not convinced that it’s the more natural or plausible scenario where Trump wins. (It may be – with some reasonable assumptions – but it doesn’t seem obviously so.)
Carlos:
I’m basing this on the forecast distribution and on my general understanding of U.S. elections. I think the half-million-vote switch is pretty much the center of the distribution of national vote swing, conditional on a local swing that would’ve been just enough to swing those three states.
I agree that the foreknowledge to target those three specific states is a big stretch and is not the right way to look at it…but I still disagree that looking at an electorate-wide swing is the right way to look at it. More votes in North Dakota weren’t going to help Trump any, and more votes in Washington State weren’t going to help Biden any. Both campaigns knew that. To the extent that they could announce policies that would move voters their way (or, perhaps more likely, move some voters their way and some voters into the other column) they were presumably trying to choose policies that were most likely to swing close states to their side. That’s also true of choices of where to campaign, where to have the biggest get-out-the-vote efforts, and so on.
We can imagine all kinds of counterfactuals. Uniform swing is one extreme; swing only in these three states in the other extreme. I think a more reasonable comparison is somewhere in between, perhaps something like “out of the ten states that were considered most competitive prior to the election, how big would the swing have had to be?”
The last forecast published by The Economist predicted (mid-point) that Biden would lead by 8.8 percentage points in the two-party popular vote. The actual outcome was a 4.6 points lead, 4.2 p.p. below the forecast.
That’s approximatively what happened in the “average” state. The mean decrease in Biden’s advantage (or increase in disadvantage) – relative to the forecast for the state – was 4.3 points. The median was 4.4 points.
A uniform change of 4 points (2% of votes switching from Biden to Trump) would have flipped not only Florida and North Carolina – as it happened – but also Georgia and Arizona.
Interestingly, these are the four states assigned to Biden classified in the forecast as “competitive” – with Georgia the most as risk.
From that point of view the question would be how Trump didn’t win a few more thousands votes in Georgia and Arizona – having done better elsewhere.
Biden resisted remarkably well in Georgia. The predicted advantage was less than one percentage point. There are only three states were Biden did better relative to the forecast – as expected in Vermont and a gain of a couple of points in Colorado and Maryland.
The case of Wisconsin is completely different. The predicted lead for Biden was 8 percentage points but he ended less than one point ahead. It was the seventh worst result relative to the forecast. While Biden losing almost all of his predicted advantage was quite surprising, maybe losing a little more would have been only a little more suprising.
It’s interesting to look at the other “big misses”. In West Virginia, North Dakota, South Dakota and Montana a large lead for Trump had been predicted anyway. But in Iowa and Ohio – the most competitive states together with Georgia according to The Economist – there was only half a point between them according to the forecast and Biden ended 8 percentage points behind.