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

Simulating Spanish polling error: where the chances come from

Abstract. The seat ranges, each party's chance of winning a seat and the chance of each combination reaching 176 come from 2,000 simulated elections. Each party's national share is moved by a random error the size of its typical final-poll miss, each party's share in each province by a further local factor, and D'Hondt is run in every province. We explain the spreads used, including for Sumar, Podemos and parties with no record, and what independent draws leave out.

Polling Forecast · updated 9 November 2026 · 8 min read · Spain, simulation, uncertainty, seat ranges, D'Hondt, provinces, majority, investiture

Summary

The seat ranges, the chance of each party winning a seat and the chance of each combination of parties reaching 176 all come from 2,000 simulated elections. In each one, every party's national share is moved by a random error whose size is that party's typical miss at the final polls of 2004 to 2023; every party's share in every province is moved again by a small random local factor; and the D'Hondt count is run in all 52 provinces. The outputs are counts across the 2,000 runs.

As in Germany, and unlike in Britain, each party's national error is drawn independently. That is simpler and easier to defend with eight elections whose misses have no stable shape, but it ignores the fact that one party's gain is usually another's loss, and we say below what that does. The simulation's most useful outputs are not the national seat ranges but the province-level ones: which party is likely to take the last seat, and how likely a small or regional party is to win any seat at all.

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.

  • Only elections from 2000 shape the forecast (28 September). Earlier elections stay in the long view and on the history pages but no longer feed any test, correction or error size. In Spain nothing before 2000 was used, so nothing changed.
  • Every outcome (3 October). The seats in every simulated election are shown as histograms at /es/outcomes.

From a best estimate to a chance

Our projection is a single set of national shares, the poll average (Forecasting the Congreso from imperfect polls: an overview), and a single Congreso worked out from it by the seat law (From votes to seats: D'Hondt in 52 provinces). A reader looking at that Congreso will want to know how sure we are. Do the PP and Vox reach 176 if the polls are two points out? Does Podemos keep a seat? Which provinces could go either way?

To answer, we need a model of how wrong the polls might be, pushed through the seat law many times. The model is SpainForecast.Simulate. It uses the same seat engine as the projection, so every run is a complete Congreso: shares in every province, the D'Hondt count, and party totals.

One run, step by step

  1. Move each party's national share. For each of the thirteen parties modelled, add a random error drawn from a normal distribution centred on zero, with the standard deviation given below. A share cannot fall below 0.05%. Others take whatever the named parties leave, with a floor of 0.5%.
  2. Spread the shares to the provinces. Scale each party's 2023 provincial shares by its new national share, exactly as for the projection (proportional swing), splitting Sumar's 2023 vote between Sumar and Podemos in their new national ratio and giving SALF its new national share everywhere.
  3. Add local noise. Multiply each party's share in each province by its own random factor, and rescale the province to 100.
  4. Run the seat law. D'Hondt with the 3% threshold in every province. Record every party's seats nationally and by province, the largest party in each province, and the total of each combination of parties.

The random number generator has a fixed seed, so the same polls always produce the same chances. A change in a published chance therefore reflects a change in the polls, never a different draw.

National error: one draw per party

The draw

Let \(S_p\) be today's national share for party \(p\) and \(e_{p,t}\) the error of the average final poll at past election \(t\) (poll minus result; see Measuring Spanish polling error, 2004–2023). The typical miss is the root mean square over the \(T_p\) elections at which the party was measured,

\[ \text{RMS}_p = \sqrt{\frac{1}{T_p}\sum_{t} e_{p,t}^{2}}. \tag{1} \]

The standard deviation used depends on what the record can say about the party:

\[ \sigma_p = \begin{cases} \max(0.2,\ \text{RMS}_p) & \text{a party measured at past elections,}\\ \max(0.5,\ 0.7\,\text{RMS}_{\text{left}}) & \text{Sumar,}\\ 0.3 + 0.12\,S_p & \text{Podemos and SALF,}\\ \max(0.2,\ 0.3 + 0.1\,S_p) & \text{any other party with no record.} \end{cases} \tag{2} \]

In run \(r\) the simulated share is

\[ S^{(r)}_p = \max\!\left(0.05,\ S_p + \sigma_p\, z_{r,p}\right), \qquad z_{r,p} \sim \mathcal{N}(0, 1) \text{ independently for each } p. \tag{3} \]

The Sumar and Podemos rows need explaining. The record measures the left of the PSOE as one lineage, and its RMS of about two points is the typical miss on the whole of it. Today that vote is split in two, and it would overstate the uncertainty to give each half the whole lineage's error. Sumar, the larger part, takes seven tenths of it, with a floor of half a point. Podemos and SALF have no record as separate parties, so they get a spread that grows with their share: 0.3 points plus 0.12 points per point of share, about 0.7 points for a party on 3%.

Today's values are:

PartyShare todayElections measuredTypical final-poll miss (RMS)Standard deviation used
PP30.882.472.47
PSOE27.682.562.56
Vox19.031.581.58
Sumar6.182.011.41
ERC2.570.470.47
SALF2.00–0.54
Podemos1.50–0.48
EH Bildu1.540.210.21
Junts1.370.480.48
BNG1.020.250.25
PNV0.870.150.20
CC0.430.130.20
Live The simulation's national error: standard deviation drawn for each party (points) Computed from the database, polls to 19 November 2026.

The PP and the PSOE have spreads of about two and a half points, set by misses such as 2004's. The regional parties have spreads of a fifth to half a point: the polls have never missed them by much, and their national shares are small. UPN, which stood with the PP in Navarra until 2019, has no record of its own.

What the independent draw leaves out

In a real election the misses are linked. Shares add up to 100, so if the polls overstate one party they understate others, and Spanish misses have often come in pairs: in 2004 the PP was overstated and the PSOE understated by about five points each; in 2016 the left was overstated and the PP understated; in 2023 the PSOE was understated and the PP, Vox and Sumar all overstated. Our draws are unrelated and the thirteen shares are not rescaled; others absorb the difference.

The effect on each party's own range is probably small. The effect on anything that depends on two parties at once is larger. The gap between the PP and the PSOE is the clearest case. Its error at the eight elections since 2004 had a root mean square of about 4.1 points, much of it from 2004. Two independent draws with the PP's and the PSOE's spreads give a gap with a standard deviation of about 3.6. The simulation therefore understates, a little, how far the order of the two largest parties could move. The British simulation avoids this by replaying whole past misses (Simulating UK polling error: where the chances come from); with eight Spanish elections and a lineage that changes shape, we have not done so here.

Local error: one factor per party per province

A national error alone would move every province in lockstep. Real provinces do not: campaigns, candidates and regional issues move some more than others. In each run, each party's projected share in each province is multiplied by its own factor,

\[ z^{(r)}_{k,p} \leftarrow z^{(r)}_{k,p} \times \max\!\left(0.2,\ 1 + 0.08\, u_{r,k,p}\right), \qquad u_{r,k,p} \sim \mathcal{N}(0,1), \tag{4} \]

and the province's shares are rescaled to 100. A party with 30% in a province moves by about 2.4 points either way from local noise alone; a party with 10%, by about 0.8. Blank votes and others get their own factors. Podemos's share of the split left gets a second factor, so that the two halves of the lineage can move apart in a province.

The size of the factor, 8%, was chosen, not estimated. It is small next to the national error for the large parties and large next to it for the regional ones. For UPN, which wins its one seat in Navarra with a share in the teens, local noise is almost all of the uncertainty.

What comes out

Seat ranges

The 80% range is the band within which a party's seats fell in the middle 80% of runs: the 10th and 90th percentiles of its 2,000 seat totals.

PartyPollsSeats80% range2023Wins a seat
PP30.8130116–142137over 99%
PSOE27.6115101–127121over 99%
Vox19.06759–7833over 99%
ERC2.5118–137over 99%
EH Bildu1.586–106over 99%
Sumar6.175–1431over 99%
Junts1.351–8795%
PNV0.842–65over 99%
BNG1.022–5197%
CC0.410–2173%
SALF2.000–0–6%
Podemos1.500–0–8%
Live Our projection: votes and seats in the Congreso (Sumar's 2023 seats include Podemos) Computed from the database, polls to 19 November 2026.

For the large parties the range is roughly symmetrical and a dozen or more seats wide. For small national parties it is lumpy, because D'Hondt hands out seats in steps: a party near the threshold of exclusion in several provinces either clears it in some of them or in none, so its range can run from one or two seats to a dozen with little in between.

Winning a seat

The "Wins a seat" column is the share of runs in which the party wins at least one seat. For the regional parties it is the chance of holding their own province's seat or seats. For Podemos and SALF it measures something different: whether their national share, spread across provinces, clears the threshold of exclusion anywhere. For SALF, spread evenly at under 3%, the answer is almost always no.

Provinces and seats

For every province the simulation records each party's 80% range and the share of runs in which each party is the largest there. For every seat on the hex map, where each province's seats are ordered by party, it records which party held it in each run. The provinces page and the map's "probability" view show these chances. The most informative are the last seats, where the party next in line is close (From votes to seats: D'Hondt in 52 provinces).

Combinations

The chance of a combination is the share of runs in which its parties' seats add up to 176 or more:

CombinationSeatsMajority
PP + Vox19799%
PP + Vox + UPN + CC19899%
PSOE + Sumar + Podemos1220%
Sánchez's 2023 investiture bloc1531%
PP + PSOE245over 99%
Live Combinations: seats and chance of 176 or more Computed from the database, polls to 19 November 2026.

These are arithmetic, not forecasts of what the parties will do. Two things limit what they mean. An investiture needs 176 votes only at the first ballot; at the second, more yes than no is enough, so a combination short of 176 can govern if others abstain, as Mariano Rajoy did in 2016 with the PSOE's abstention. And the investiture blocs of 2023 were built after the election, by negotiation; the same parties need not line up again.

What we know about the simulation

We can say how it was built, not yet how well it performs. The British simulation was checked by asking how often past results fell inside its ranges; with eight Spanish elections and a seat engine whose projection half is untested (From votes to seats: D'Hondt in 52 provinces), that check would say little. Four points are worth stating.

The spreads are set by the record. A miss larger than any since 2004, or of a new shape, would fall outside the ranges more often than the stated odds suggest.

2004 weighs heavily. Without 2004 the PP's RMS falls from 2.5 to about 1.9 and the PSOE's from 2.5 to 1.9. Whether a 2004-sized miss can happen again is exactly what eight elections cannot say.

Independent draws. The gap between the two largest parties is somewhat less uncertain in the simulation than in the record, for the reason given above.

The local factor is a guess. A larger factor would widen the provincial ranges and the ranges of the regional parties; a smaller one would narrow them.

Limitations

The simulation turns a single projection into chances with the least machinery the evidence supports: one normal draw per party, one local factor per party and province, and the seat law. It does not model linked errors between parties, differences in how pollsters miss in different regions, turnout, or the chance that a party not now in the model wins seats. The polling-day version (From today to polling day in Spain) adds a further error for the time left before the election. What we don't know yet: limitations of the Spanish forecast collects the rest.

References

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

Ley Orgánica 5/1985, de 19 de junio, del Régimen Electoral General (LOREG), articles 162 and 163.

How to cite

Polling Forecast (2026). Simulating Spanish polling error: where the chances come from. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/spain-simulation. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like