Američki izbori 2020.

Kirko Skaddeng wrote:inače meni se čini da je ovogodišnja greška mnogo veća nego 2016. kada je relativno dobro procenjen popular vote, i po državama (tamo gde su pravljene dobre ankete) i na saveznom nivou. ove godine procenti su waaaay off jer pollsteri ne umeju da ukalkulišu ljude koji ih iz nepoverenja i direktnog otpora vuku za nos.

rekavši to, ne znam zašto se od polstera očekuje potpuna preciznost. polovi pre svega služe akterima procesa da vide opšte trendove (i naravno, one na mikronivou) i ravnaju se u skladu s njima, a ono što dolazi u medije je clickbait igračka koja nudi opšti pogled na proces ali ne može da proriče budućnosti i da čita misli.

to je vrlo slično meteorologiji. čak i najprecizniji modeli imaju granice.

Now, as a potential Biden victory morphs from a landslide to a slim possibility the question is: What the hell went wrong? “I can not believe that four years later they make the same polling errors,” a Democratic adviser to Biden texted me last night. “The American fucking errors. Across the board. All the House races and states races and national races: WRONG, WRONG, WRONG.” Another Democratic government official messaged me to rant about the incompetence of the polling firms, noting that “Pollsters are good at calling easy elections, bad at calling close ones, and worse still if there are changing demographics. This means they are only useful when you don’t need them; that’s the same as being useless.”

Of course, the pollsters, and the polling aggregators, like FiveThirtyEight, see it differently, defending themselves in the face of the morass of hatred being thrown their way. As one Republican market analyst pointed out to me, “When these guys are wrong, they fall back into the ‘Well, we said there was a 10% chance’ when it makes no sense to apply probability to an event like that.” For example, the analyst said, if these aggregators run their simulations on a state like California, which has a 95% chance of going for Biden, there is no real 5% chance of Trump getting the vote, but that slim margin allows them to hedge. “These guys aren’t paid to be right,” the Republican market analyst said, “they are paid for clicks.” And if there’s one thing people have been doing a lot of over the past few months, it’s clicking on sites like FiveThirtyEight and the New York Times’ The Upshot, run by Nate Cohn.

The Nates, as many refer to Cohn and Silver, and their supporters, often plead no contest when their predictions prove to be off, saying that it’s not their math at issue but the raw polling numbers. This is called a cop-out. Imagine for a moment that Nate Silver was an oncologist, and patients went to him to deduce whether they had cancer. Doctor Silver drew blood, performed a slew of tests, and the lab results came back clean. Except—oops—you did have cancer; the lab results were wrong. But Doctor Silver continued to use the same lab for his tests. Again and again, patients were told they were healthy, when they were not. In this scenario, you wouldn’t say that the lab was at fault; Silver would be to blame for repeatedly using faulty labs. The same is true in terms of polling. At some point, the Nates should either find a new lab to analyze their tests, or quit their jobs entirely. (As of press time, Silver had not responded to a request for comment.)