Does Spain Have Its Ohio? Zaragoza Is Always Right, but the Best Thermometer Is Five Provinces Together

In the United States, Ohio voted for the winner of the White House in every election from 1964 to 2016. We looked for its equivalent in the seven general elections from 2008 to 2023. Single units hold up: municipalities of more than 5,000 voters that picked the winner six times in a row did so again in 2023 in 78% of cases, against 54% for the rest. But a set of 100 "Ohio" census sections predicts 2023 worse than 100 chosen at random. The exception is five provinces: they were off by 0.6 points in 2023.

25 October 2026 · Miguel · Leer en español

In US politics, Ohio was for decades the state that “always gets it right”: from 1964 to 2016 it voted for whoever ended up winning the presidency, until it failed in 2020. The question is obvious in a country with 8,106 municipalities, tens of thousands of census sections (secciones censales) and 52 provinces: is there anywhere in Spain whose result is a small-scale mirror of the national one? We looked at the seven general elections from 2008 to 2023 at four scales (polling station, census section, municipality and province), and the answer has two halves.

What we measured

For each unit we calculated the vote share of the PP (the conservative People’s Party) and the PSOE (the Socialists) over valid votes in each election and compared it with the national one. Two tests: whether the unit’s most-voted party matches the national one, and how far off it is on average, in points. In those seven elections the PSOE won three times (2008 and both 2019 elections) and the PP four (2011, 2015, 2016 and 2023).

We left out the CERA vote (voters registered abroad) at every scale, provinces included, because it is counted separately from the polling stations (mesas). The national reference shares do include it; the sum of all polling stations, without CERA, stays within 0.3 points of them in all seven elections. And this is an ecological analysis: it describes places, not people.

Single units: the winner persists

Zaragoza is the best example among municipalities of more than 5,000 voters. It picked the winner all seven times and was off by an average of 1.5 points on PP and PSOE, the lowest of the 1,053 municipalities that pass the filter.

The pattern repeats at the other scales. We selected the units that picked the winner in the first six elections (from 2008 to November 2019) and checked whether they got it right again in 2023, when the PP regained first place. Those that had done so six out of six did better than the rest.

For polling stations, 64% of those that were right six times were right again in 2023, against 47% of the total; for census sections, 62% against 46%; and for municipalities of more than 5,000 voters, 78% against 54%. For provinces the difference is small (9 out of 10, against 79%) and with so few units it cannot be told apart from chance.

A set of “Ohios” does not work

A bellwether, however, is not a single unit but a set. If the sections that best imitated the country between 2008 and 2019 keep doing so, added together they should give a result almost equal to the national one. We tested it: we selected with 2008–2019N, measured in 2023 and compared with 1,000 sets of the same size chosen at random.

The error is the mean of four absolute deviations, in PP, PSOE, Vox (the far right) and the parties to the left of the PSOE, in points. With 100 sections chosen for their similarity, the error is 1.9 points; the median of the random sets, 0.8. Of the random sets, 98% do better. With 100 polling stations the same happens (1.7 against 0.8). The reason lies in what the criterion did not watch: we selected on PP and PSOE, and those sections share a bias in the others. In 2023 the 100 best Ohios were 5.2 points above the average for the parties to the left of the PSOE and 1.8 above it for the PP.

The exception: five provinces

Provinces are a different matter. With 2008–2019N, the five with the lowest error on PP and PSOE were Las Palmas, Zaragoza, Illes Balears, Málaga and Santa Cruz de Tenerife. In 2023, added together, they were off by 0.6 points on average. The median of random sets of five provinces was 2.4; only 0.1% of them did better.

None of the five does well on its own: Las Palmas was off by 4.5 points in 2023 and Málaga by 3.2. They work as a set because their errors offset each other. The result is only partly robust. If we select with the other six elections and measure in April 2019, the set again beats chance comfortably (0.5 points against 2.1; only 0.8% of random sets do better). If we measure in November 2019, its error (1.8) is still below the median of chance (2.5), but one in five random sets does better.

What these results do not say

There are four caveats. The first is that there are few clean exams: the test against chance says the 2023 hit is not luck, but one hit does not make a rule, and in November 2019 the advantage was much smaller. The second is that if, instead of selecting with 2008–2019N, we use all seven elections, Castellón replaces Málaga: the set changes with the data used. The third is that Vox did not stand for the Congress in Santa Cruz de Tenerife in 2023, which alters the split in that province that year. And the fourth is that the vote of these provinces is a thermometer, not a cause: we do not know why they resemble Spain.

Nor do they resemble it in early turnout. At 18:00 in 2023, with the Ministry of the Interior’s turnout update, turnout in the five provinces was 49.2% of the electorate, against 53.1% for the total; the 100 Ohio sections, by contrast, were at 53.0%. The provinces that best reflect the vote do not reflect the hour at which people vote.

Nor does their turnout anticipate the result. We checked whether the change in turnout at 14:00 or 18:00 in these five provinces, relative to the previous election, leads the change in the PP’s lead over the PSOE, in them and in Spain as a whole, controlling for age, income, foreign-born population and municipality size. With only six pairs of elections, their 18:00 turnout tracks the national change just like that of almost any group of provinces (correlation of −0.85, against −0.82 for Spain as a whole), because turnout everywhere rises and falls together with the climate of each election; at 14:00 they do not stand out at all. And in the panel of the 52 provinces, the link between their turnout and their own result is even weaker in them than in the rest.

On the night of November 29, when the count begins, we will look at those five provinces. Not as a forecast, but as one more reading.

How we calculated it

• Data. Polling-station results from the Ministry of the Interior, 2008–2023, validated against the BOE (the official gazette). Shares over valid votes (voters minus null votes). Families: PP (with UPN), PSOE (with PSC), Vox, Cs, left (IU, Unidas Podemos, Sumar, Más País, Compromís, Equo), nationalists and rest. National reference: official totals, with CERA (ohio.py).
• Units. Polling stations and sections present in all seven elections, with an electorate between 0.8 and 1.25 times that of 2023 and at least 300 valid votes on average: 26,038 polling stations and 21,906 sections. Municipalities of more than 5,000 voters in 2023: 1,053. Provinces: 52. The CERA vote is excluded at every scale, provinces included.
• Error. Mean of the absolute deviations of the PP and PSOE shares from the national one in the elections considered.
• Validation. Selection with 2008–2019N, measured in 2023, against 1,000 random sets of the same size (fixed seed). Percentile: share of the random sets with an error less than or equal. Measuring in April or November 2019 repeats the test selecting with the other six elections.
• 18:00 update. Second turnout update from the Ministry of the Interior, excluding postal votes.
• Code. ohio.py and articulo_ohio.py, in the altnalysis repository.