What 2023 Taught Us: The Vox Voter Who Stayed Home, Late Deciders and the Mail Voting Record

Three years after the July 23 election there is plenty of data to review what happened: three CIS surveys with 66,000 interviews, ten post-election surveys since 1996, private polls and some twenty academic studies. Three conclusions. Vox was the party that lost the most potential voters to abstention, out of disaffection more than logistics. Late deciding, which now affects one in four voters, hurt the PP as always but this time not the PSOE; even so, it explains about one point of the polls' miss on the PSOE, not the full three and a half. And mail voting broke every record, with a bias toward Sumar that moved the result by a few tenths of a point.

31 October 2026 · Miguel · Leer en español

Update, October 31, evening: what we changed in the model. We used what we learned here to review the forecast. (1) We checked against the 2015–2023 general elections whether its uncertainty was right, forecasting each election using only the polls available 0, 7, 14, 21, 28, 35 and 42 days out: the 90% interval was right 96% to 100% of the time, too wide. It is now right 88% to 96% of the time, and the forecast intervals are somewhat narrower: the PP goes from 106–156 seats to 109–154 and the PSOE from 83–136 to 85–134. The central seat counts do not change. (2) People who do not say whom they voted for in 2023, and whom the model assigns to a party, now get the same probability of voting as those who declare that party; estimated abstention among the PP’s 2023 voters falls from 14% to 12.4%, level with the PSOE’s. (3) We tested allocating undecided voters away from the PP, as the historical series suggests, and dropped it: in 2023 it makes the forecast worse, because some of those undecided voters are PP voters who do not say so. (4) Nor did we find in the provincial record that turnout favors any major party: when it rises, local candidacies gain. Details are in the model specification.

The sources and their limits

We have combined four types of sources:

• The 2023 CIS microdata. Two pre-election surveys (29,201 and 27,058 interviews) and the post-election survey (10,101). The second pre-election survey was conducted while mail voting was already under way and asked those who had already voted whom they had voted for.
• The ten CIS post-election surveys since 1996, which always ask when the vote was decided.
• Private polls from our database and what the firms have published since.
• The academic literature, especially the CIS book on the 2023 elections and subsequent journal articles.

None of the surveys is a panel: the same person was not interviewed before and after voting. And all share a known bias. In the 2023 post-election survey, 34.3% of those who voted for a party say they voted for the PSOE (the result was 31.7%), 28.6% for the PP (33.1%) and 9.0% for Vox (12.4%). A study published in 2025 in the Revista Española de Investigaciones Sociológicas finds the same in 16 CIS barometers after July 23: recalled vote falls 8.2 points short of the PP’s result and 3.1 points short of Vox’s. That is why we give each figure both raw and corrected, weighting each electorate by its actual size. The correction moves the numbers; it does not change the order of the conclusions.

One: the Vox voter who stayed home

Before the vote, the Vox voter did not look less mobilized than the rest. In the CIS’s two pre-election surveys, 85% of those planning to vote for Vox said they were completely sure to vote, the same as PP and PSOE voters. And their direct vote intention was not inflated: 11.2% and 11.6% of those naming a party, against a result of 12.4%.

The post-election survey tells a different story. It asked non-voters whether they had considered voting and for whom they would have voted. For every 100 votes each party won, these were the non-voters who had considered voting for it:

Vox is the party with the most potential voters lost to abstention in both versions: 25 per 100 votes raw and 34 corrected, against 15–17 for the PP, 17–18 for the PSOE and 11–16 for Sumar. The CIS book on these elections reaches something similar by another route: of those who voted for Vox in November 2019, 7.4% abstained in 2023, compared with 4.6% of PP voters and 6.4% of PSOE voters.

What stands out are the reasons:

Non-voters who leaned toward Vox cite distrust, discontent and not feeling represented more often, and problems with mail voting less often than those of the other parties: 12%, against 29% for Sumar and 21% for the PP. This is abstention out of disaffection, not logistics.

The absolute figures should be read as a ceiling. Applied to the 10.4 million non-voters, they would imply some 750,000 people who considered voting for Vox and did not, but non-voters who answer surveys are more interested in politics than the rest. The comparison between parties is more reliable than the number.

Two: late deciding and the PSOE

People decide later and later. Since 1996, the CIS post-election survey has asked when the vote was decided:

In 2023 it was 26% of voters, as in 2015, compared with one in ten between 1996 and 2004. And in all ten elections a pattern repeats: late deciders vote less for the PP than those who decide earlier.

The PSOE also tends to weigh less among late deciders. There are two exceptions: 2004, the election after the March 11 attacks, when it clearly won among those who decided at the end, and 2023, when it tied (33.0% among late deciders and 33.4% among the rest). The space to the left of the PSOE wins among late deciders almost every time.

Corrected for the result, this is how late deciders and earlier deciders split in 2023:

If late deciders had voted like everyone else, the PSOE would have won about one point less, Sumar a point and a half less, the PP 1.7 more and Vox 0.7 more. It is a real effect, but it does not explain everything the polls got wrong:

The 20 pollsters with fieldwork between July 10 and 17 fell short on the PSOE by 3.5 points on average. The nine that kept asking until the day before still fell 2.9 points short. The six pollsters that measured both before and after the start of the pre-election polling blackout saw the PSOE rise half a point in that last week. There was a late swing, a small one, and underneath it a level bias of about three points shared by the whole industry. That common bias is what our forecast tries to capture with its industry error.

The literature points the same way on the why. A University of Murcia survey during the campaign finds that rejection of Vox multiplied the odds of voting for the PSOE by 2.5, with all controls, while rejection of the PSOE or Sumar did not push people toward Vox. Other work describes left-wing mobilization and strategic voting between the PSOE and Sumar in small provinces. And the most common doubt before voting was precisely between the PSOE and Sumar: 36% of undecided voters who hesitated between parties.

Three: the mail voting record

On July 23, 2.6 million mail voting requests were accepted, and 2.47 million votes reached the ballot box: 94.2%. It was a record on both counts.

The pattern is the usual one, on a new scale. Summer elections send mail voting soaring: 4.2% of the electorate in June 2016, against the usual 2–3% in autumn and winter. In July 2023 it was 7.4%. There were incidents: the Electoral Board extended the delivery deadline twice, and on July 18 some 9% of envelopes were still waiting to be collected. Even so, only 5.8% of those who requested it failed to vote, the lowest figure on record.

Mail votes are mixed in the ballot box with the rest, so there is no official data on whom they went to. The only measurement is the CIS’s:

More Sumar voters and fewer Vox voters voted by mail. The CIS book confirms this with the post-election survey: 15% of center-left voters used mail voting and 6% of right-wing voters, and 17% of university graduates against 4% of those with only primary education. Since mail voting was 10% of the vote, that difference moved the result by a few tenths of a point: about half a point in Sumar’s favor and three tenths against Vox. A claim by GAD3 before the vote, without published data, held that among mail voters the right led the left by twelve points; the CIS data point the opposite way. No study has measured whether mail voting increased turnout or merely changed how people vote.

What it means for November 29

• Abstention out of disaffection is Vox’s risk. It currently stands at about 19% in the forecast, and its potential voter is the one who most often stays home when not feeling represented. Our model already computes turnout by party; we will watch it in the final polls.
• Undecided voters do not split like decided ones. In all ten elections since 1996 they have voted less for the PP than the rest. Our forecast does not impose this, and the 2023 data suggest the effect is about one point. We will study it before the November 23 freeze.
• Mail voting will return to its usual level. In November it should be around 2–3% of the electorate. And on election night it is worth remembering that the 2:00 p.m. and 6:00 p.m. turnout updates do not include mail voting: in 2023 they gave 53.1% at 6:00 p.m., against 56.9% in November 2019, and part of that gap was mail voting.

How we calculated it

• Microdata. CIS 3411 and 3416 (pre-election) and 3420 (post-election) of 2023, weighted. The corrected figures multiply each electorate by the ratio between its official result and its recalled vote in the post-election survey; the same response bias is assumed among voters and among non-voters who leaned toward the same party.
• Historical series. CIS post-election surveys from 1996 to 2023 (cis_poselectorales.py). Deciding late: in the last week, on the day of reflection (the day before the vote) or on election day.
• Private polls. Our poll database (data/encuestas.csv), without the CIS; last poll from each pollster in each window.
• Mail voting. Requests from the Electoral Census Office; votes admitted according to Correos (July 2023).
• Literature. Book Elecciones generales de 2023 (CIS, 2024), chapters 7, 15 and 18; Alaminos and Alaminos-Fernández (REIS, 2025); Rojo-Martínez (Revista Española de Ciencia Política, 2025); Rodon and Rodríguez López (South European Society and Politics, 2024). Full notes with links in research_notes/estudio_2023/.
• Doctoral and master’s theses. We searched Dialnet (which covers Spain’s TESEO thesis database) and the repositories of about a dozen universities. We found none on 2023 mail voting, PSOE–Sumar strategic voting, the polling error or late deciding; most theses on the July 2023 election deal with political communication. Some repositories did not respond, so a few may be missing.
• Code. estudio_2023.py, in the altnalysis repository.