altnalysisEspañol
← Analysis

What History Says About Tragedies and the Vote: Tenths of a Point, Not Points, Unless a Story of Blame Takes Hold

Before any calculation come the 69 people who died on October 28 in Urda. No figure in this article is worth one of those lives. But the campaign goes on, and it is inevitable that voters will react to a tragedy like this one. What we know about how they do so is limited, and we say so plainly: in 23 cases from Spain and other countries, accidents with an uncertain technical cause moved the polls by tenths of a point; those that became a story of attributed political responsibility moved them by 3 to 14 points. Today there is no poll with fieldwork after the accident. Until there is, this is a framework for reading them, not a forecast.

31 October 2026 · Miguel · Leer en español

Update, October 31, afternoon. GAD3 president Narciso Michavila wrote on X that his firm detects “an abrupt shift in voting since Wednesday,” the day of the accident, and that “it is now impossible for Sánchez to win the most votes.” This is not a published poll: there is no methodology statement, no fieldwork dates and no percentages, except that Vox “does not reach 19%.” It is the first signal from a pollster with daily tracking, and it points in the direction of this article’s central scenario, a drop for the PSOE. Its size is unknown, and past cases suggest that the first measurements after a tragedy tend to overstate the change. Our forecast will not include it until GAD3 publishes figures; then it will enter as one more poll, corrected for its house effect. Much of the rest of what Michavila points to matches today’s forecast: Vox doubles its seats without reaching 19% (we have 66 seats and 18.7%), the PP is the largest party, the Sumar space falls and Bildu wins in the Basque Country.

First, the victims

The collision of two high-speed trains in Urda on October 28 has caused 69 confirmed deaths and more than a hundred injuries, and dozens of people from 23 countries are still to be identified. The official mourning period ends today. This article is not about them or about the causes of the accident, which are being investigated by the railway authority and by a court, under judicial seal. It assigns no responsibility, political or criminal: we do not know it. What it measures is something else: the responsibility that voters perceive and attribute, deserved or not, which is what moves votes.

It is about something else, which is also real. There are 29 days until the election, the campaign has resumed, and many people are asking whether a tragedy like this one changes how people vote. The honest answer is that it can, that it almost never does so as much as it seems to in the first days, and that what determines the size of the effect is not the tragedy itself but what happens afterward.

Second: voters will inevitably react, and what we know is limited

A tragedy weighing on the public mood is neither an anomaly nor irrational. Research on retrospective voting has described this for decades: voters judge those in government by what happens to them, even by what is beyond the government’s control. Achen and Bartels estimated that U.S. presidents lose about one point in states hit by droughts or floods. Other authors, such as Fowler and Hall, find no such effect for purely random disasters. And there is a more interesting reading than irrationality: Ashworth, Bueno de Mesquita and Friedenberg argue that an event like this can reasonably move votes, because it reveals something about the government’s preparedness that nobody would have seen without it.

What we know has clear limits. Elections are few, the tragedies that coincide with them are fewer, and each case mixes the event with everything else that was going on. We have gathered 23 cases from Spain and other countries with polls before and after, or with an election nearby. It is the best evidence available, not a law.

How much the governing party fell in the polls after each eventPoints of vote lost in the following weeks, with the range across the different measures. Gray: accidents with an uncertain technical cause. Orange: events that became a political story. Red: catastrophe with political responsibility attributed from day one, with no court ruling (regional level).
−15−10−50+5+10+15
Adamuz (Jan 2026)
Angrois (2013)
Tempi, Greece (2023)
Lisbon funicular (2025)
Iraq (2003)
Prestige (2002–03)
11-M (2004)
Valencia floods (2024)

Bar: range across the different measures. Dot: estimated change, in points of vote.

Third: we do not know the direction, and there are three families of cases

The 23 cases fall into three groups, and the difference between them is not the number of victims.

Transport accidents with an uncertain technical cause. This is the family most like Urda today. In Spain, Angrois (2013, 80 dead) left no trace: the Galician PP was rising in the polls four months later. Spanair (2008, 154 dead) did not either. The most direct precedent is Adamuz, on January 18 of this year: 46 dead in another rail collision, with the same ministry and the same operator. The two polls that compared the same firm before and after measured tenths of a point: the PSOE fell 0.3 or 0.4 points, and Vox rose 0.4 or 0.6, within the margin of error. Outside Spain the pattern is the same, although the initial gap is larger. In Greece, after the Tempi crash (February 2023, 57 dead), the governing party lost 3 or 4 points in the polls, recovered part of it within five weeks and won the May election with 40.8%, above what it had before the accident. In Lisbon, five weeks before the 2025 municipal elections, the Glória funicular accident left the mayor in a statistical tie in the polls; he won with 3 points more than four years earlier.

Events that became a political story. These are the ones that the opposition, or the facts, managed to link to a prior decision by the government. The Prestige oil spill (November 2002) was not just a spill: it was the response of the Galician regional government (Xunta) and the national government, and an enormous mobilization. The Galician PP fell 2.6 points in the polls within four months and 3.9 in the municipal elections of May 2003, while across Spain as a whole it barely moved. The Iraq war, in those same months, gave the PSOE a lead of 6 or 7 points in April 2003 that had vanished by the fall. And the March 11, 2004 Madrid bombings (11-M) are the limiting case: the attack and the government’s handling of information, three days before the vote, cost the PP about 5 points by the most-cited estimate (Montalvo), with a range of 1 to 7.5 depending on the method. There was no time for the effect to fade. By contrast, the ETA attacks of the 1990s raised turnout but did not move support for the government.

Catastrophes with political responsibility attributed from day one. This is the Valencia floods (DANA) of October 2024 (more than 220 dead). A distinction is needed: we are talking about political responsibility as perceived by voters, not criminal responsibility; the court case over the Valencia floods remains open and there is no ruling. The Valencian PP went from 38–42% to 25–32% in regional polls within two weeks, and two years later it remains 4 to 10 points below 2023. Three features explain this: political responsibility attributed personally and immediately to the regional president, a demonstration of 130,000 people eleven days later, and a story that was renewed with each anniversary and each judicial step. And two nuances matter for today: the punishment stayed in the regional arena, because in general elections the Valencian PP is barely one seat below 2023; and the lost vote went mainly to Vox, not to the PSOE.

The conclusion of the literature is that what matters is not the damage but the response and the responsibility that voters attribute, which is political and does not necessarily coincide with what the courts eventually establish. Healy and Malhotra found that voters reward spending on relief afterward and do not reward prevention. Gasper and Reeves, that a good response is worth more than the damage costs. And Malhotra and Kuo, studying Hurricane Katrina, that blame is filtered through party: each voter tends to blame the opposing side. In Spain the data fit. After Adamuz, 79% of PP voters, 89% of Vox voters and 51% of PSOE voters considered that the accident “could have been avoided.” That bounds the effect: most of the electorate has already decided whom to blame before the cause is known.

Fourth: what is visible today, three days after the accident

If the size of the effect depends on what happens afterward, it is worth looking at what is happening. As of October 31:

Given that picture, Urda today looks more like Adamuz and Tempi than like the Valencia floods or 11-M. What could change that is the story shifting from “accident” to “pattern,” because of the two crashes in nine months, and something keeping it alive through November: an appearance before Parliament, an indictment, a revelation from the investigation or a mobilization by the families.

Fifth: the effect fades within weeks, unless something sustains it

This is the most useful finding for an election 29 days away. In every case with an election one to two months later, the polling gap closed partly or entirely before the vote. In Greece, the governing party lost 2.9 points after nine days, recovered part of it after five weeks and won 82 days later above its previous level. In Lisbon, the statistical tie in the polls turned into a clear victory. In Spain in 2003, the Socialist lead of April had disappeared by May. And in the three cases with a nearby election (Greece, Lisbon and South Korea after the 2014 Sewol sinking) the polls overestimated the punishment: they measured a gap that the ballot box did not confirm.

The exceptions share a mechanism: something kept the issue alive. In Greece, two years later, the widespread belief in a cover-up and the anniversary protests again took 3 or 4 points from the government, and the beneficiary was a new party, not the opposition. In Serbia, the collapse of the Novi Sad station took 23 months of student mobilization to translate into a defeat for the governing party. In Valencia, the anniversaries and the trial reactivated the punishment. And the Prestige offers the most nuanced lesson: two and a half years later, in the regional elections of June 2005, the Galician PP lost its absolute majority and the government, going from 51.6% and 41 seats to 45.2% and 37. But the Prestige was only one part: in the 2004 general elections the PP fell in Galicia by the same amount as in the rest of Spain, and in the 14 municipalities of the Costa da Morte, ground zero, it lost less in 2005 than the Galician average (4.4 points versus 6.2). What decided 2005 was, above all, the rise of the Galician Socialists (PSdeG) by 11 points amid a favorable national cycle and the wear of an 82-year-old leadership. The tragedy opened a crack; other things widened it.

What it would mean for November 29, if it happened

Given all of the above, our reading is that the central scenario is a loss of about one point of vote for the PSOE on election day, with a plausible interval from half a point in its favor to three against, and a tail of up to five points if a story of negligence or cover-up crystallizes and is sustained until the polling blackout. This is a judgment based on the cases, not a statistical estimate, and it does not move our forecast: the model changes only with polls.

To translate this into seats, we ran the calculation on the 50,000 simulations of the October 31 forecast, removing that fraction of vote from the PSOE and redistributing it according to where it goes:

Seats the PSOE would lose depending on the vote loss and where it wentAverage seat difference across the 50,000 simulations of the October 31 forecast. Assumptions: uniform loss across Spain, on election day, with decay already accounted for.

PSOE −0.5 pt

To abstention−1.7
To Vox−2.5
Half to Vox, half to abstention−2.1

PSOE −1 pt

To abstention−3.5
To Vox−5.0
Half to Vox, half to abstention−4.2

PSOE −2 pt

To abstention−7.0
To Vox−9.9
Half to Vox, half to abstention−8.5

PSOE −3 pt

To abstention−10.7
To Vox−14.8
Half to Vox, half to abstention−12.8

PSOE −5 pt

To abstention−18.2
To Vox−24.5
Half to Vox, half to abstention−21.5

What we will watch

When polls with fieldwork after October 28 arrive, we will look at five things and report on them here:

  1. The same firm before and after, as with Adamuz. Differences between firms are usually larger than the effect, and the aggregator already corrects each firm’s bias.
  2. Where the lost vote goes: to Vox, to the PP or to abstention. After Adamuz and the Valencia floods it went to Vox.
  3. The share of 2023 PSOE voters who attribute responsibility to the government. After Adamuz it was half; if it rises clearly, the effect will be larger than in January.
  4. The curve: if the fourth week is smaller than the second, the effect is fading as in Greece and Lisbon; if not, something is sustaining it.
  5. The milestones: the prime minister’s appearance before Parliament, the state funeral, the court’s decisions and the investigation’s reports.

And a rule for reading it: in precedents with a nearby election, the jump in the first polls was larger than the one on voting day. We will treat that jump, if it appears, as a short-term movement and not a change of level, until several polls confirm it.

How we calculated it