Who Votes When: Older Areas Vote Early, Wealthier Ones Later, and the 6 p.m. Turnout Does Not Say Who Wins

In 2023, 40.5% of the electoral roll had voted by 2 p.m. and 53.1% by 6 p.m., ahead of a final turnout of 70.4% excluding the overseas vote. The time of day at which people vote is a stable trait of each place: census sections that vote early in one election tend to do so in the next (correlation of 0.44 to 0.66). Turnout at 6 p.m. is a good predictor of final turnout, but not of who wins: the link with the conservative People's Party (PP) that shows up in the raw data in the last seven general elections almost disappears in five of them when comparing similar areas.

25 October 2026 · Miguel · Leer en español

On election day, the Ministry of the Interior publishes two turnout updates: at 2 p.m. and at 6 p.m. (at 1 p.m. and 5 p.m. in the Canary Islands, which is the same moment). They are the only figures known before polling stations close, and someone always reads them as a preview of the result. We took the updates from the last seven general elections, 2008 to 2023, polling station (mesa) by polling station, to see who votes early, who votes late and whether those figures say anything about who is going to win. Everything that follows compares areas, not individuals.

What the national figures say

In 2023, 40.5% of the electoral roll had voted by 2 p.m. and 53.1% by 6 p.m. Final turnout, not counting the overseas vote (CERA, voters registered abroad), was 70.4%. That is, by 6 p.m. three out of four of the day’s voters had already voted.

The curves do not always move together. The July 2023 updates (53.1% at 6 p.m.) are below those of November 2019 (56.8%), and yet final turnout was slightly higher (70.4% versus 69.9%). A good part of the explanation is the postal vote: it counts in final turnout but not in the updates, and in 2023 it was very high because the election fell in mid-summer. That is the first warning for reading the updates: they measure the ballot box on the day, not the whole vote.

Older areas vote early

To separate those who vote early from those who vote a lot, we measured what share of a section’s voters had already voted by 2 p.m. We sorted the census sections (secciones censales, the smallest electoral areas) into ten groups with the same electoral roll, according to their share of population aged 65 or over.

In 2023, in the youngest sections 53.3% of voters had already voted by 2 p.m.; in the oldest, 57.9%. In November 2019 the difference was from 50.3% to 55.4%. When age, income, municipality size, share of foreign-born residents and vote are estimated together, age is the factor that weighs most in all seven elections, with a standardised coefficient of 0.23 to 0.41. Municipality size adds between 0.08 and 0.22 in the general elections from 2008 to November 2019 (people vote earlier in big cities) and disappears in 2023. The model explains between 35% and 53% of the differences between sections.

Wealthier areas vote later

Income runs in the opposite direction: the richer a section is, the smaller the share of its voters who have voted by 2 p.m. The standardised coefficient is negative in all seven elections, from −0.13 to −0.33.

In 2023 the difference is large: 62.2% of the voters in the poorest sections had already voted by 2 p.m., against 52.3% in the richest. In November 2019 it hardly existed (from 53.5% to 53.0%, with almost the whole middle range around 54%). Probably part of that difference is not a timing habit but the postal vote: it is concentrated in wealthier areas, is not in the updates and, once added to final turnout, makes those areas look “late”. That is why the 2023 pattern, in mid-summer, is stronger than in a November election.

The time of voting is a trait of the place

If voting early were a matter of the date or the weather, a section that votes early today would have no reason to do so in the next election. But it does. Between 2011 and 2023, the correlation across elections of the share of voters who had already voted by 2 p.m. is between 0.44 and 0.66 when comparing stable census sections. In consecutive elections it ranges from 0.49 (2019 to 2023) to 0.66 (April to November 2019). After subtracting each province’s average, it drops to a range of 0.32 to 0.57: still clearly positive.

They vote late, but in large numbers

Within that pattern there is a striking group: sections that vote late and yet have high final turnout for their province. We defined them as those in the 20% of sections with the lowest share of early voters and, at the same time, in the 20% with the highest final turnout in their province. In each election there are between 1,863 and 2,254 such sections.

They are sections with higher income than the rest (between 0.5 and 1.3 standard deviations in all seven elections, 1.3 in 2023), somewhat younger (from −0.14 to −0.55) and with fewer foreign-born residents (from −0.35 to −0.49). In their vote, they are alike in giving more weight to the PP (from +0.10 to +0.70 standard deviations, +0.70 in 2023) and less to the PSOE (the Socialists; from −0.24 to −0.94, −0.94 in 2023). In the elections in which Ciudadanos (centre-right liberals, now marginal) ran strongly (2015 to November 2019), they also have more Ciudadanos vote (from +0.63 to +0.84).

And it is not a one-election fluke: if a section is in this group, the probability of being in it again in the next election is 39% to 47%, against 3% or 4% for those that were not. With one important caveat: as the postal vote counts in final turnout but not in the updates, some of these sections do not vote late but by post. The pattern also appears in November 2019, with little postal vote, so it is not explained by that alone.

Why the 6 p.m. snapshot does not say who wins

Every election, someone reads high turnout at 6 p.m. as a sign that a party is winning. At polling-station level, that link exists: among the polling stations won by the PP or the PSOE, the PP wins 26% of those in the decile with the lowest turnout at 6 p.m. in 2023 and 74% of those in the decile with the highest.

But this is a background association, not a signal from the day: some areas have higher turnout and vote more for the PP for reasons that existed beforehand. When age, income, municipality size and region are controlled for, the link vanishes in five of the seven elections.

Without controls, the coefficient is positive in all seven elections, from 0.37 to 0.76: in 2023, one standard deviation more turnout at 6 p.m. multiplies by 1.9 the ratio between the probability that the PP wins the polling station and the probability that the PSOE wins it. With controls, the coefficient is between −0.17 and +0.20 and, in 2023, 0.09 (the ratio is multiplied by 1.1). Only 2008 (+0.20) and 2011 (−0.17) move slightly away from zero. Turnout at 6 p.m. says more about who lives there than about how they will vote.

What the updates do anticipate

There is something the updates are useful for: final turnout. Between consecutive elections, the change in turnout at 6 p.m. explains between 36% and 63% of the change in final turnout across municipalities (median 46%) and between 56% and 88% across provinces (median 82%). By 6 p.m. the day is already largely settled as to how many people will vote.

To know who will vote for whom, no. The change in turnout at 6 p.m. explains a minimal part of the change in the split between parties: the median R² across municipalities is 0.3%, and across provinces 11%, with peaks of 60% in one specific pair of elections (the PSOE between 2011 and 2015) and, apart from the PSOE, signs that change from one election to the next. As an exercise, we trained a model on earlier elections to predict 2023 using only what was known at 6 p.m.: it foresaw a fall of 3.6 points for the PP; the PP rose 11.5 points. The improvement over a simple uniform shift for the PP was 4.5% across municipalities and 8.9% across provinces: little help.

What to keep in mind on 29N

• The postal vote does not count in the updates. Areas with more postal voting can look “late”. In July 2023 the effect was large; in a November election it should be smaller.
• The overseas vote (CERA) is not included. Both the updates and the figures in this article exclude the votes of residents abroad.
• In the Canary Islands, one hour earlier. The 2 p.m. and 6 p.m. updates are published at 1 p.m. and 5 p.m. there, the same moment.

How we calculated it

• Updates. Turnout at 2 p.m. and 6 p.m. by polling station, from the Ministry of the Interior polling-station files for the 2008–2023 general elections (avances_mesas.py). Only polling stations with consistent data (2 p.m. update ≤ 6 p.m. update ≤ voters ≤ electoral roll) and without overseas vote.
• Early voters. Update divided by the polling station’s voters, which include the postal vote. That denominator makes early voting look lower where there is more postal voting.
• Deciles. Of census sections with the same electoral roll in each decile; percentage aged 65 or over and net income per person from the section panel (2021 Census and the INE Atlas of Household Income).
• Models and persistence. Least squares weighted by electoral roll, with region fixed effects and standardised variables; correlations across stable sections, with an electoral roll between 0.8 and 1.25 times the previous one (avances_analisis.py).
• Polling stations and who wins. Deciles of turnout at 6 p.m. and logistic regressions with standard errors clustered by municipality (foto18.py).
• Same-day usefulness. Regressions on pairs of consecutive elections, at municipality and province level (avances_senal.py).
• Code and data. articulo_horas.py, in the altnalysis repository; the results cited are in site/src/data/articulos/horas.json.
• Caveat. These are associations between areas, not behaviour of individuals: from the fact that a section with more older residents votes earlier, it does not follow how any individual older person votes.