Spain General election29 Nov 2026in 8 days PP + Vox majority 99% 2023 investiture bloc 1% Calendar Track record Research
Research · Spain
EN ES

If we're wrong: replaying past Spanish polling misses on today's polls

Abstract. Ten readings of today's Spanish polls, side by side: the polls as they stand, our projection (the same, since correcting failed its test) and each general election's final-poll miss since 2004 replayed on today's average, all through the same provincial seat engine. The 2004 replay moves the Congreso most; D'Hondt makes the split left's seats jump. We also explain the count of provinces at risk.

Polling Forecast · updated 9 November 2026 · 8 min read · Spain, lenses, scenarios, past polling misses, what-if, provinces at risk, D'Hondt

Summary

A lens is one stated assumption about how wrong today's polls are. For Spain the site shows ten readings of the same polls: the polls as they stand, our projection (which, because correcting Spanish polls failed its test, is the same numbers), and eight replays that take each general election's final-poll miss since 2004 and apply it to today's average. Every reading goes through the same seat engine: proportional swing to the 52 provinces, then D'Hondt in each. We show the readings side by side and do not average them.

The Spanish replays behave in a particular way. Most of the misses are modest, about a point per party, but a few are large and were shared by almost every pollster: 2004, which moves about five points from the PP to the PSOE; 2016, which moves nearly four points from the left towards the PP; and April 2019, which takes three and a half points off the PP. The 2004 replay changes the Congreso more than any other. The other thing that makes the replays jump is D'Hondt: a point on a party near the threshold of exclusion in several small provinces moves several seats at once. The same replays change at least one seat in 37 of the 52 provinces, which the map marks as "at risk".

The British counterpart of this paper is Lenses: reading the same British polls several ways. How the misses were measured is in Measuring Spanish polling error, 2004–2023, and why we do not correct Spanish polls is in Reading Spanish polls: the correction test and the CIS's raw figures.

What has changed since this paper was written

Updated 9 November 2026. The rest of the paper is as first published, except where a sentence below says it was corrected.

Why Spain needs lenses when correcting fails

The average of the final polls has missed by between 0.7 and 2.7 points per party at each general election since 2004:

ElectionPolls' average missCorrected, leave one outVerdict
20041.551.50closer
20080.740.75level
20110.740.78further out
20150.870.90further out
20162.652.72further out
April 20190.931.02further out
November 20190.690.76further out
20230.850.77closer
Average1.131.15
Live Correcting Spanish polls, each election left out in turn Computed from the database, polls to 19 November 2026.

The same table shows why we do not correct them: taking off half the average past miss, worked out each time from the other seven elections, was closer at 2 and further out at 5. The misses do not keep their direction:

Party20042008201120152016April 2019November 20192023Average
PP+4.8−0.3+1.7−0.8−2.9+3.6−0.1+1.0+0.9
PSOE−5.2−1.7+2.5−1.5−1.4+0.6−0.4−3.3−1.3
Sumar (and predecessors)+1.2+1.2−1.0−3.1+3.7−1.8+0.9+1.0+0.3
ERC−0.5+0.80.0−0.3–−0.7+0.2+0.20.0
Junts+0.1−0.2−0.9−0.2–−0.5−0.6+0.3−0.3
PNV+0.1+0.3−0.10.0–−0.2−0.1+0.10.0
EH Bildu––−0.3+0.3–−0.1–−0.1−0.1
Cs–––+5.5+1.2−1.1+2.5–+2.0
CC+0.1–0.0––−0.2––0.0
Vox–––––+0.6−2.6+0.7−0.4
BNG+0.4–+0.1–––––+0.2
Live Congreso: final-poll average minus result, by party (points) Computed from the database, polls to 19 November 2026.

So the headline is the poll average. That does not mean the polls will be right. It means that on the evidence of eight elections nobody can say in advance which way they will be wrong. A lens answers a narrower question: what happens to the Congreso if the polls miss this time exactly as they did at a known past election.

What a lens is

The replay

Let \(M\) be the thirteen parties the Spanish forecast models, \(v_p\) today's average for party \(p\) (each pollster's latest poll within 30 days of the newest, equally weighted), and \(S = \sum_{p \in M} v_p\) their combined share. Everything else is "others", \(100 - S\).

For a past election \(t\), the error of the final polls on party \(p\) is \(e_{p,t} = \bar{y}_{p,t} - r_{p,t}\), poll minus result. The replay of election \(t\) subtracts that error from every party it measured, with a floor of 0.1 points, and rescales the modelled parties to their original combined share:

\[ \tilde{v}^{t}_{p} = \begin{cases} \max\left(0.1,\; v_p - e_{p,t}\right) & p \in P_t,\\[2pt] v_p & \text{otherwise,} \end{cases} \qquad v^{t}_{p} = \tilde{v}^{t}_{p} \times \frac{S}{\sum_{q \in M} \tilde{v}^{t}_{q}}. \tag{1} \]

This is what LensMath.Replay does, wrapped in one Spanish step. The record measures the left of the PSOE as one lineage, so before the replay Sumar and Podemos are added together, the lineage's error is taken off the sum, and the result is split between them in their ratio today. A replay of 2016, when the polls had Unidos Podemos 3.7 points too high, takes 3.7 points off Sumar and Podemos together, not 3.7 off each.

The set \(P_t\) of parties measured at election \(t\) changes. The PP, the PSOE and the left lineage are measured at all eight; Vox from April 2019; ERC, Junts and the PNV at most; EH Bildu, the BNG and Coalición Canaria at some. SALF and UPN are measured at none, so no replay touches them except through the rescale.

The rescale keeps "others" where they are. If a replay's errors do not sum to zero, the difference is spread over all thirteen parties in proportion to their adjusted shares.

Everything after the replay is the same for every reading

The adjusted shares go into SpainForecast.Run, described in From votes to seats: D'Hondt in 52 provinces: each party's 2023 share in each province is scaled by its national change, Sumar's 2023 vote is split between Sumar and Podemos, SALF is spread evenly, and D'Hondt with the 3% threshold is run in every province. No reading changes the baseline, the swing or the law.

The readings on the Spanish pages

The polls as they stand. The poll average with no adjustment.

Our projection. Our rule in every country is the same: take off half the past miss only if that would have made past forecasts better. In Spain it fails, so this lens has no adjustment and is labelled "Our projection (the polls, uncorrected)". Its row is identical to the polls row; we keep both so the table has the same shape as in other countries, and so it would show a difference at once if a future test passed.

Eight past misses. One replay for each of 2004, 2008, 2011, 2015, 2016, April 2019, November 2019 and 2023, built from 140 final polls.

The reader's own. The what-if tool lets a reader set an error for each party, from −6 to +6 points in half-point steps, applies equation (1), runs the same seat engine and shows the seats, whether the combinations reach 176, and every seat that changes hands. Its presets are the replays, the projection and the 2023 result. The prediction tool does the same from shares the reader types in, and can simulate a range around them.

Every reading through the same seat engine

ReadingPPPSOEVoxSumarPodemosPP + Vox
The polls as they stand1301156770197
Our projection (the polls, uncorrected)1301156770197
If the polls miss like 20041071397050177
If the polls miss like 20081291236750196
If the polls miss like 201112610473110199
If the polls miss like 201512411665150189
If the polls miss like 20161401176320203
If the polls miss like April 201911711268140185
If the polls miss like November 20191201148150201
If the polls miss like 20231241306650190
Live Seats under each reading of today's polls Computed from the database, polls to 19 November 2026.

The table is recomputed whenever a poll is added, so the numbers move. Three features follow from the shape of the misses and should survive most changes in the polls.

2004 is the big one. The 2004 replay takes about five points off the PP and adds about five to the PSOE. Nothing else in the record moves the two largest parties that far, and it is the replay under which the PP and Vox have the fewest seats between them. It is also the least likely to repeat in kind: the 2004 miss followed an attack three days before the vote.

2016 moves the left. The 2016 replay takes 3.7 points off the left lineage and adds 2.9 to the PP. Split between Sumar and Podemos, the lineage's loss pushes both below the threshold of exclusion in most provinces, so the replay leaves them with very few seats. The 2015 replay does the opposite, adding 3.1 points to the lineage.

Vox barely moves, except under November 2019. Vox has been measured at three elections, and two of its errors are under a point. The November 2019 replay adds 2.6 points to it, and is the replay under which Vox wins most seats.

Why some replays jump

A seat total under D'Hondt is a step function of the vote share, province by province. In a province with \(M\) seats, a party with more than \(1/(M+1)\) of the vote is sure of a seat, and below that its seat depends on how the rest of the vote splits (From votes to seats: D'Hondt in 52 provinces). Nationally, the steps are smoothed a little by the 52 provinces having different sizes, but not much, because 29 of them elect five deputies or fewer.

The effect is sharpest for Sumar and Podemos. Each is polling at a level where it sits near the threshold of exclusion in many mid-sized provinces at once. A replay that adds a point to the lineage lifts both over the line in several provinces together; one that takes a point off pushes them under together. So their columns in the table move in jumps, and the seats they win or lose come mostly from the PP and the PSOE, which take the last seats they would otherwise have won.

The combinations jump with them. A replay that gives the left a few more seats can take the PP and Vox below 176 without changing their vote by much.

Provinces at risk

For each province \(k\) and each seat in it, let \(w^{L}\) be the party holding the seat under reading \(L\) and \(w^{B}\) the party under the projection. The site counts how many of the eight replays give a different holder for at least one of the province's seats:

\[ F_k = \sum_{t} \mathbf{1}\!\left[\, \exists\ \text{seat } i \in k:\ w^{t}_{i} \neq w^{B}_{i} \,\right]. \tag{2} \]

A province is "at risk" if \(F_k \geq 1\). Today 37 of the 52 are. As in Britain, this is a count of historical scenarios, not a probability. A province that changes under two of eight replays is not "a quarter likely" to change. The simulation gives each province a chance for each party, and the province pages show it next to the count. The replays are also not independent: 2004 and 2023 both understated the PSOE, and 2015 and 2016 moved the left in opposite directions, so a province that changes under one of a pair often changes under the other.

The larger the province, the more replays change it. When we wrote this, every province with eight seats or more changed under at least three of the eight replays: with more seats there are more last seats to fight over, and in Madrid and Barcelona almost any replay moves a seat or two between parties. Among the small provinces the picture is mixed. In some the PP and the PSOE hold their seats under every replay; in others, three-seat provinces such as Huesca, Palencia and Segovia among them, the third seat is fought between the PP, the PSOE and Vox and changes under several replays. The map shows the current count for every province.

Limitations

Eight past misses are not a distribution. The replays show what known misses would do; they do not say how likely each is. The simulation (Simulating Spanish polling error: where the chances come from) is the tool for chances.

One replay dominates. The 2004 replay moves more seats than any other, and that miss followed exceptional circumstances.

The lineage. Replaying a lineage error on a split left assumes that a miss on "the left of the PSOE" would fall on Sumar and Podemos in proportion to their shares.

Parties without a record. SALF and UPN are moved only by the rescale, and Vox has three elections of record. The replays cannot express the most important uncertainty about a new party, which is whether its support is measured well at all.

References

Jennings, W. and Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.

How to cite

Polling Forecast (2026). If we're wrong: replaying past Spanish polling misses on today's polls. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/spain-lenses. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like