This article is sent from the Kiko Llaneras newsletter, a newsletter for EL PAÍS subscribers: .
Good morning! Today I am sending you a dozen curious facts about the elections in the United States, in telegraphic format.
1. Be careful with blaming abstention
It is being written, analyzed and repeated that the Republican has won the popular vote by four points – the total number of votes at the national level – but that figure is hasty: there are millions of votes left to count. In California alone, about six million are missing. According to , Donald Trump will indeed win in votes, but by one or two points over Kamala Harris.
It would also stop explanations of the result based on participation. It is said that Harris has lost millions of votes, and that is why she is defeated. It is a premature analysis because we do not yet have all the votes. The final participation could be equal to or greater than that of 2020. Patience.
2. Trump has not swept the votes
In 2020, Joe Biden won by 4.5 points over Trump, so the turnaround that the country as a whole has taken will be about six points. The change is equivalent to 3% of the electoral roll moving between parties.
Society has not changed: four years ago the Democrats won 51% to 47%, while now the Republicans win 50% to 48%. Why then is there talk of a red wave? Because the map of States turns red, when half a dozen territories turn, but there are not that many votes. The number of Americans who voted Democrat or Republican is really similar to that of 2020, and also to that of 2016. Power changes hands—a lot of power—and that is important, but we exaggerate the swings of voters.
3. The result was not a surprise
I’m surprised to see perplexed people! The polls—not to mention the prediction markets—have been saying for months that there was equality between the candidates or that Trump was one step ahead in his chances of winning the presidency.
We said it clearly in our last analysis, published very large in all our headlines: . Counterintuitively, the red and blue waves were the most likely results, due to how the American country’s electoral system works. These days I used the simile of throwing a coin in the air: you can’t tell if it will land heads or tails, but that doesn’t mean it will come up flat! The tie was rare. As we reported on Tuesday, when the elections were simulated, the most likely results were the Trump wave (23%) or the Harris wave (16%). That’s why I put Trump 306 electoral votes in the newspaper’s baton.
In El País we were not the only ones who issued this warning. They explained similar things in , , or . Many people will have been surprised by the result, because not all of us have to follow current events up to the minute, but the information was there.
4. What happened to Fredi9999?
You will remember that I have followed the forecasts of , the technology-based prediction platform blockchain and cryptocurrencies, where thousands of people bet small and large amounts of money. I also told you about Fredi9999, an anonymous user who bet millions in favor of Trump, and whose activity we could trace, because it was recorded in the blockchain. At first there was speculation about his identity, fearing a conspiracy to favor the Republican. But it seems confirmed that it was a person: a trader French. Fredi9999 moved the market with his injection of millions, in Polymarket and beyond. However, because they did not flood the platform with liquidity to counteract its appearance.
Well, Fredi999 has come out very well with his radical bet on Trump’s victory: according to , the trader has made $50 million in profits.
5. Let’s not exaggerate with prediction markets
During these months I have shared: models based on surveys (like those of and ), market of enthusiastic forecasters without money (), and markets with money (Polymarket or ). I value all three methods and that’s why I included them in our probabilistic prediction average. It is absurd to want to choose between them – because you can combine them – but some of their supporters only want to compete to put their method above the others.
In that fight, the Polymarket market has won this battle. They gave Trump a 60% chance of winning, while models based on surveys and Metaculus gave him a 50% chance. Go ahead and I’m glad that there is a prediction market with sufficient size and liquidity, because it can be useful for predicting lots of things, from football matches to political conflicts, that are beyond those that you can try to predict with polls. But the cheers from Polymarket itself and its followers are being excessive.
Because it is a statistical error to now assume that Polymarket predicts elections better than Metaculus or Nate Silver. In fact, applying Bayes’ theorem, one’s judgment on Monday—before the election—about the relative accuracy of Polymarket, Metaculus, and Nate Silver should hardly change after Trump’s victory. Because? Because 60% and 50% are similar predictions!
An example – for the more mathematical ones -: suppose that a week ago you were 70% sure that Polymarket was incorrectly adjusted and that Metaculus was correctly adjusted (that is, that the probability of Trump was really 50%). So, knowing now that Trump won, you should lower your confidence to 66%. In other words, you must update your judgment, but only little. And the same thing in reverse: if a week ago you thought with 70% confidence that Polymarket was better (and that the probability of Trump was 60% and not 50%), now you should reaffirm yourself, but only until you are 74% confident. Not at all safe! We feel that after a success we should confirm our beliefs, but this is not true, as demonstrated by this simple but counterintuitive law of probability.
Plus, there’s another key: marketplaces like Polymarket are powered by people who watch surveys. They are not really alternatives. If all polls disappeared, prediction markets would perform worse, because they would have lost a source of information. The proof is that when a relevant or surprising survey was published, the numbers on Polymarket moved.
6. The error in the surveys was not large, but it is problematic
The error of the polls was normal. Our national polling average is closed, so your error will be 2.5 or 3 points. Is it little or a lot? It’s basically a typical error. The deviation was 2 points on average. In 2016 it was 2.2 points, and in 2020, despite “getting” Biden’s victory right, they were wrong by 4.5 points. And the polls of the key states? In 2024 it seems that the average error will be 2.5 or 3 points, which means improving its average error of 4.2 points since 1976.
But US polls do have a problem. The worst of all: they have chained three presidential elections—2016, 2020 and 2024—with failures in the same direction, underestimating Trump. Furthermore, this time they have also deviated in the same direction in practically every State, which reinforces the thesis of systemic error. Pollsters can defend themselves by remembering that in the 2022 midterm elections they underestimated the Democrats, but I think that is insufficient. The question is obvious: Why is Trump underestimated?
Then there are more notable failures. Renowned pollster Ann Selzer caused a stir last week with her final poll in Iowa—her specialty—which . In the end, Trump won that state by 13 points. That is not a normal mistake, but a setback for their methods.
—
PS. I leave you two other articles from the team where we analyze the elections. First we count the Republican’s progress among voters by age, sex, race and income: . The next day, Almudena Á. Blacksmiths He put the magnifying glass on one of the great keys: .
Can you help me? Forward this newsletter to whoever you want, and if you are not subscribed, It is an exclusive newsletter for EL PAÍS subscribers, but anyone can receive it for a trial month. You can also follow me on Twitter, at or write to me with clues or comments, at kllaneras@elpais.es.
