In these times of polarization and noise, where information and junk messages mix, debates arise where there should be none. One case is the precision of the CIS. The matter hit the headlines again this week when the government confirmed José Félix Tezanos as head of the organization. Executive Spokesperson and Minister of Education, Pilar Alegría, explained this to the press: “You will understand that the sociologist who is right in the polls will be confirmed.”
Some experts have said (correctly) that talking about failures and successes is not the best way to judge a gauge under uncertainty. But it's even worse: the statement is wrong. I dare say that we are not dealing with a problem of opinion, but rather with a fact: the CIS estimates with Tezanos were inaccurate and biased.
1. The CIS did not “receive” the result of 23-J.
Although more precise than in the past, its estimate of seats assumed a different scenario than later: the CIS gave the PSOE and Sumar about 171 or 172 seats, which would have allowed them to govern without the support of Junts and ERC , but in reality they remained at 152 and it was necessary to negotiate with these forces for months. Our poll average deviates in the opposite direction, but less: it gave about 177 seats for the sum of PP and Vox, which remained at 170 and had no government options. Which poll came closest? The 40 dB query. that this newspaper published that PP and Vox had 173 seats and the sum of PSOE and Sumar 148 seats.
2. The CIS estimates would still be questionable even if they had been correct for 23-J
The reason is simple: since Tezanos came to the center in 2018, the center has accumulated failures in several elections, so a single success was not enough to offset this history. To measure this, I created the following table comparing the deviation of half a dozen pollsters at a dozen election events since 2018.
The data is eloquent. The CIS with Tezanos was, on average, more wrong than any other pollster. On four occasions he was more wrong than ever before – three times between 2018 and 2021, which was his worst time – and he was always wrong by more than our poll average.
Calculation in three steps: (1) For each election and each pollster, I calculate the average error per party (for those with 2% of the vote or more); (2) I calculate the pollsters' average error for each election. and (3) for each pollster and each election, I now calculate the relative error by subtracting the average error in that election from the pollster's error. So a positive error means that the pollster deviated more from the pollster average, and a negative error means that she made fewer errors.
3. Most importantly: The biggest problem with CIS estimates is not the size of their errors, but their direction
Since Tezanos took office, the CIS has overestimated the left's vote in 36 of 37 elections. Also on the 23rd. In July the Left surprised with a better result than all private polls had expected, but even under these circumstances the CIS again overestimated them. On the contrary, he was surprised: his estimate was that PSOE and Sumar would prevail by 5 points over PP and Vox, but in reality they were 1.4 points lower.
Yesterday, an editorial in EL PAÍS criticized the ratification of José Félix Tezanos, arguing that his speech tarnished the image of the Center's impartiality, “regardless of the success or failure of its forecasts.”
To go further, I think these errors are an indication of bias. Because while it is true that there are technical reasons that explain why the CIS polls have deviated to the left on one and several occasions, when this deviation becomes systematic and repeats itself after four years and 30 elections, the question is unavoidable. : Why is this bias not corrected?
In other news
⚽ Rodrygo returns to average. I was interested in this data from David Álvarez about the Real Madrid striker. The graphic shows the evolution of Rodrygo's successes this season and compares his accumulated goals (dark blue) with his expected goals, according to the number and characteristics of his shots (light blue).
The graph says that Rodrygo's random production was more or less constant. The player got into the target position with regular frequency, resulting in four expected goals in his first ten games and another four expected in the next ten. What changed wasn't that, but its accuracy. Rodrygo managed to “owe” more than three goals in the 11th game, when he scored only one goal from four shots. But in the last five games his success has been extreme – seven goals – to offset this debt and improve the statistics in his favor.
There are many ways to explain this back and forth. Part of it will be simple coincidence: an exceptionally bad series must be followed by a normal, i.e. better, series. But it's also useful to think about a player's mindset. In the way a failure or a goal affects your game and in the advantage of finding calm, that automatic mode that José Adán, in charge of a neurotraining session, calls: “a state in which everything flows.” The brain does what it needs to do at all times, without obstacles, without ruminations, worries or expectations.”
🔥 2023 is the hottest year in millennia. Why is this so and how do we know this? Manuel Planelles, Clemente Álvarez and Laura Navarro tell the story in this visual analysis of climate evolution, following the human footprint and the traces that temperatures leave in the trees and glacial ice.
✏️ Has AI already taken away some jobs? A recent study examined advertisements on a website for freelance positions for editors and writers. A month after ChatGPT was released, a year ago, a decline in offers was observed: fewer authors were wanted and they were paid less. He told it John Burn Murdoch in the Financial Times.
You help me? Forward this newsletter to anyone you want. If you are not yet subscribed, sign up here. 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 @kikollanor write to me with tips and comments at [email protected].
Subscribe to continue reading
Read without limits
_