How close the simulation is to real people.

Compared with real surveys

We asked the simulated people, with the same panel as Valyron's studies, the same questions as two published surveys, with the same answer options, and compared what they answer with what real people answered. The questions, the sample sizes and what would count as a hit were fixed before asking anything. We publish what came out.

We measured it with the current version, the one that answers in Spain's studies today after the change told in «Against real sales», and with the previous one. In both, in option questions, each person spreads their answer across the options according to how likely each one is, and the report gives the average. We show both, side by side.

In this comparison the current version answered in the six countries of the European survey; in studies outside Spain the simulated people still answer with the previous one. Both are measured with the same questions, the same surveys and the same yardstick for a hit, and we publish both grades as they come out, including where the current one does worse.

What to bear in mind

• The electric car survey asks people who hold a driving licence; Valyron asks all adults. Somewhat more “No” and “Don't know” are to be expected in Valyron.
• That survey covers the 27 EU countries; Valyron is compared in six: Germany, France, Italy, Spain, Poland and the Netherlands. Romania, the seventh country the survey highlights, is not in the panel.
• The electric car survey does not publish the literal wording of its questions: we asked them with its same answer options.
• Part of the difference is chance in the samples: even if Valyron were exactly right, the average distance would come out at about 3.2 points for electric cars and about 1.6 for purchase intention. The purchase intention survey itself moves 1.0 points on average from one month to the next.
• Neither survey publishes how it weights its answers: their intervals are approximate.
• The simulated people do not know the news of the month in which they are asked.

Sources

Consumer Monitor 2025: EU drivers' view on electric cars. European Alternative Fuels Observatory (EAFO), © European Union, 2026. Authors: Nikolaas Van den Steen, Lieselot Vanhaverbeke, Dennis Verbist. Licence CC BY 4.0. Changes: answer categories grouped as in figure 21 of the report, and comparison with Valyron.

Source: El Observatorio Cetelem, Barómetro de consumo, septiembre de 2026 (consumer barometer).

Measured on 10 October 2026. It will be repeated with every change to the method.

Against real consumption data

We asked the simulated people, with the same panel as Valyron's studies and its previous version (the one before the change told in «Against real sales»), about consumption facts in Spain whose real answer is published: which phone they use, which car gets bought, where the household's money goes… Each question carries the same options as the real figure. The questions, and what would count as an improvement, were set before asking anything. Here are the 9 facts that were set aside for checking and never used to try out any change. We publish what came out.

On average, each Valyron figure is 8.9 points from the real one; spreading the answers evenly across the options would land 15.5 points away. The order of the options matches with an average correlation of 0.75 (1 would be the same order), and the most chosen option matches the real one in 4 of 9 facts.

We tried four changes to bring it closer. None improved enough on these facts. Later, another change did pass its check («Against real sales», below) and since then the simulated people in Spain answer with it: these figures are from before that change. With the current version only a small test was run, with very few people per fact, which is not enough to publish as a result.

Valyron Real figureBars in %: for Valyron, the mean of what each simulated person spreads across the options, with the thin lines of the 95 % confidence interval; the real figure, as published.

What to bear in mind

• Some real figures do not count people: the share of web use (heavier users weigh more), car registrations (which include company and rental fleets) or households. Valyron asks people.
• Several sources ask only part of the public (licensed drivers, people about to buy a car, users of paid services); the simulated people are asked to put themselves in that situation.
• Part of the gap is chance: Valyron asks 300 people for each fact.
• The simulated people do not know the news of the month they are asked in.
• The brands shown belong to their owners; none takes part in Valyron or endorses these results.

The sources of these data, with the attribution their licences require, are under «Sources and licences», at the foot of the page.

Measured on 10 October 2026.

Against real sales

We asked the simulated people to choose between two similar products sold in Spain, of the same kind and at a similar price (33 pairs of cars and 7 of motorbikes), each with its spec sheet and its price and without telling them which one sells more. Then we compared that with which one really sold more to private buyers between January and August 2026. 60 simulated people answered per pair. The pairs, what would count as a hit and what would count as an improvement were set before asking anything. We publish what came out.

Valyron picks the best seller in 78% of the 40 pairs (31). In the 30 where one clearly sells more (at least 1.5 times the other) it gets 87% right; in the 10 where they sell about the same, 50%: there it is at chance. What the simulated people prefer and how sales really split resemble each other with a correlation of 0.61 (1 would be the same split).

This check decided a change. With the previous version of Valyron, measured on the same pairs and the same people, the hit rate was 58% (63% in the clear pairs) and the correlation 0.42: in nine pairs the new version is right and the previous one is not, and in one, the reverse. Since 12 October 2026, the simulated people in Spain answer with the new version.

Previous version NowBars in %: pairs in which it picks the best seller, with the thin lines of the 95 % confidence interval.

What to keep in mind

• These are well-known products, and what answers for the simulated people may know which ones sell more: this measures whether they choose like the market, not whether they work it out from the spec sheet alone.
• When two products sell about the same, a hit or a miss is close to chance; that is why those pairs are shown apart.
• Part of the improvement over the previous version comes from pairs where the preference ended up close to half and half: with other simulated people, some of them could flip.
• Sales also depend on the dealer network, delivery times, stock and discounts, which the spec sheet does not carry, and the sheet is the same for every version of each model.
• There are 40 pairs and some products appear in two: each percentage carries its 95 % confidence interval, which is wide.
• The brands belong to their owners; none takes part in Valyron or endorses these results.

The source of the sales, with the attribution its licence asks for, is in «Sources and licences», at the foot of the page.

Measured on 12 October 2026.