What we don't know yet: limitations of the Spanish forecast
Abstract. Every known weakness of the Spanish forecast in one place: eight elections with two exceptional misses, the last seat in small provinces, independent errors in the simulation, untested proportional swing with no regional layer, the split left, new and local parties, the election date, second-hand polls and the gap between seats and investiture. For each, how much it matters and what evidence would change our mind.
Summary
The Spanish forecast makes few claims. It does not correct the polls, because a correction failed its test; it takes the poll average, runs it through a seat calculation that reproduces 2023 exactly, and puts its uncertainty where Spanish seats are decided, in the last seat of each province. That leaves little to go wrong in the method, but the evidence under it is thin in places. It rests on eight general elections, two of which (2004 and 2016) carry most of the error. It moves every province by proportional swing from 2023, with no regional layer. It splits the left between Sumar and Podemos the same way everywhere. It cannot give a seat to a local list. And many of its polls reach us second-hand.
This paper sets out each weakness, how much we think it matters, and what evidence would change our mind. The British counterpart is What we don't know yet: limitations and open questions in our UK forecast; the method itself 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.
- An early election (5 October). Pedro Sánchez dissolved the Cortes; the general election is on Sunday 29 November 2026. The forecast, the election-day model and the days-to-go figures use that date, and polls may not be published from 00:00 on 24 November (Madrid time) until the polls close.
- 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.
- Pollster grades rescaled (13 October) (Rating Spanish pollsters: accuracy, the lead and openness).
Eight elections, two of them exceptional
Everything the forecast knows about polling error comes from 140 final polls at eight general elections.
| Party | 2004 | 2008 | 2011 | 2015 | 2016 | April 2019 | November 2019 | 2023 | Average |
|---|---|---|---|---|---|---|---|---|---|
| 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 | +0.8 | 0.0 | −0.3 | – | −0.7 | +0.2 | +0.2 | 0.0 |
| Junts | +0.1 | −0.2 | −0.9 | −0.2 | – | −0.5 | −0.6 | +0.3 | −0.3 |
| PNV | +0.1 | +0.3 | −0.1 | 0.0 | – | −0.2 | −0.1 | +0.1 | 0.0 |
| 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 |
Two of the eight carry most of the error. In 2004 every final poll had the PP ahead; the PSOE won by 5.4 points, three days after the Madrid train bombings, which no final poll could have measured because the ban on publishing polls had already begun. In 2016 every final poll overstated Unidos Podemos and understated the PP, by 3.7 and 2.9 points on average. Without 2004 the PP's and the PSOE's typical misses (the root mean square) fall from about 2.5 points to about 1.9.
The leave-one-out test (Reading Spanish polls: the correction test and the CIS's raw figures) is the guard against learning a pattern from these, and it worked: the correction was further out at 5 of eight elections. But the two elections still shape the simulation. The PP's and the PSOE's national shares are drawn with standard deviations of about two and a half points, set by their records, and the seat ranges on the home page are correspondingly wide. Whether that width is right depends on whether a miss like 2004's can happen again, and eight elections cannot say.
The misses are also not independent draws. All the polls at an election share the same electorate and the same last week, so 140 polls carry roughly eight elections' worth of information, not 140.
What would change our mind. Each new election is added to the record and the backtest rerun automatically; if a correction began to beat the plain average, it would switch on without a change to the method. The long view back to 1977 (Measuring Spanish polling error, 2004–2023) could be brought into the test once its results are official rather than Wikipedia's.
The last seat in small provinces
The seat forecast is most fragile at the last seat of each province. 29 of the 52 provinces elect five deputies or fewer, and in those the last seat can turn on a fraction of a point. At any time a dozen or so provinces have a party less than a point short of it:
| Province | Seats | Last seat | Next in line | Points short |
|---|---|---|---|---|
| Almería | 6 | PP | PSOE | 0.15 |
| Bizkaia | 8 | PSOE | EH Bildu | 0.20 |
| Barcelona | 32 | PP | Sumar | 0.24 |
| Palencia | 3 | Vox | PP | 0.42 |
| Ceuta | 1 | PP | Vox | 0.42 |
| Madrid | 37 | Vox | PP | 0.46 |
| Pontevedra | 7 | PP | PSOE | 0.47 |
| Alicante/Alacant | 12 | PP | Sumar | 0.53 |
| Zamora | 3 | PP | Vox | 0.56 |
| La Rioja | 4 | PSOE | Vox | 0.56 |
| Araba/Álava | 4 | PNV | PSOE | 0.58 |
| Navarra | 5 | EH Bildu | Vox | 0.67 |
| A Coruña | 8 | PSOE | PP | 0.68 |
| Girona | 6 | ERC | PP | 0.69 |
| Asturias | 7 | PSOE | Vox | 0.79 |
| Lleida | 4 | PSOE | ERC | 0.94 |
We handle this in the simulation, not in the central forecast: each run moves every party's national share and every party's share in every province, and counts the seats again (Simulating Spanish polling error: where the chances come from). Three simplifications matter here.
National errors are drawn independently. Spanish misses have often come in pairs, one large party up and another down. The simulation's parties move independently, so the uncertainty of the gap between the PP and the PSOE is a little smaller in the simulation (a standard deviation of about 3.6 points) than in the record (a root mean square of about 4.1).
The local factor was chosen. Each party's provincial share is multiplied by a random factor with a standard deviation of 8%. That figure was not estimated from past elections. A backtest of provincial swings, possible with the results we hold, would give a better one.
The 3% line and ties. Both are implemented as the law says. Exact ties in votes, settled by lot, are not modelled; none has arisen.
What would change our mind. A study of how provincial shares have moved relative to national ones between 2004 and 2023 would give the local factor an estimated size, and show whether some regions (Catalonia and the Basque Country above all) should move apart from the rest.
Proportional swing and no regional layer
Each party's 2023 share in each province is scaled by its national change. That keeps every share above zero and keeps the regional parties in their own provinces, but it assumes every province moves in proportion. Spanish elections often have regional swings of their own; Catalonia, where the PSC, ERC and Junts compete for much of the vote, is the clearest case. Proportional swing has not been tested against a uniform or mixed swing for Spain, and the test, predicting each election from the previous one with the true national shares, is possible with the provincial results we hold (From votes to seats: D'Hondt in 52 provinces).
Regional polls exist for the Congreso, in Catalonia above all, but not often enough, or from enough pollsters, to build a regional layer.
What would change our mind. The backtest above. A regional layer would have to beat proportional swing on it before replacing it.
The split left
Sumar and Podemos stood together in 2023 and are polled apart now. The forecast scales their combined 2023 vote in each province by their combined national change and splits the result in their national ratio (From votes to seats: D'Hondt in 52 provinces). Two things could make that wrong.
The ratio may differ by region. If Podemos is relatively stronger in some provinces and Sumar in others, each would win seats in different places from those shown. No poll gives provincial figures often enough to measure it.
The line-up may change. The two could stand together again, or the space to the left of the PSOE could be reorganised under another name. On today's polls, standing apart costs the two 4 seats between them compared with one list. The prediction tool lets a reader put the two together by adding Podemos's share to Sumar's and setting Podemos to zero.
The history treats them as one lineage throughout, so the polling-error record says nothing about how well each is measured separately.
What would change our mind. A decision by the parties, which we would follow at once; and regional polls showing the two apart, if enough appear.
New and local parties
SALF, which did not stand in 2023, has no provincial pattern to scale and is spread evenly; with a national share below 3% it wins no seats in the model. If its vote were concentrated in a few provinces it could win seats there that the model does not show.
Local lists are in "others", which never wins seats. Teruel Existe won a seat in November 2019; in 2023 Soria ¡Ya! took 19.1% in Soria and Teruel Existe 14.9% in Teruel without winning one. A local list that won a seat at the next election would be missed. The fix is to model the largest local lists as parties in their own provinces, which needs polls that report them, and few do.
UPN, which ran with the PP until 2019, has no polling record of its own.
The date and the drift
The next election must be held by 29 November 2026, 8 days after the latest poll, and can be called earlier. The polling-day forecast widens the ranges by a drift fitted to five elections, at distances up to a year (From today to polling day in Spain):
| Days before polling day | Elections | Polls' miss | Previous result's miss | Pull-back's miss | RMSE |
|---|---|---|---|---|---|
| 30 | 5 | 0.88 | 3.13 | 0.88 | 1.44 |
| 61 | 5 | 0.99 | 3.13 | 0.99 | 1.66 |
| 91 | 5 | 1.21 | 3.13 | 1.21 | 2.00 |
| 182 | 5 | 1.40 | 3.15 | 1.42 | 2.36 |
| 273 | 5 | 1.31 | 3.15 | 1.34 | 2.38 |
| 365 | 5 | 1.49 | 3.15 | 1.52 | 2.97 |
The five elections include April 2019, when a year out the polls missed by 3.7 points per party because the party system changed in that year; the other four missed by one to one and a half. A single drift curve averages over both kinds of year. Pulling parties back towards their last result did not help, so the forecast keeps today's shares and only widens the ranges.
What would change our mind. Each election adds a set of distances. If a pull-back began to beat the polls in the leave-one-out test it would be used, as in Britain.
Publication practice
Many polls are second-hand. Most Spanish pollsters publish through the paper that commissions them, and many of those pages are behind paywalls or rendered by script. Where we could read the page the figures are graded A or B; the rest come from Wikipedia's tables and are graded C (The Spanish data: results, maps, statistics, deputies and polls). The checks we could make agree, but a transcription error in a grade C poll would pass into the average.
Pollsters publish different things. Some give shares including blank votes and some excluding them; some publish an undecided share and most do not; only the CIS publishes its raw figures for every study. Each of these is small, and each makes the average a little less like-for-like.
The CIS leans. The CIS's recent estimates have put the PSOE ahead of the PP while most private pollsters have the PP ahead. In the average it counts once, like every other pollster. Its final polls have been both the closest (2023) and among the furthest (November 2019), so its record gives no reason to weight it up or down.
The pollster ratings use British end points. Spanish grades are not comparable with other countries' (Rating Spanish pollsters: accuracy, the lead and openness).
Investiture is not a seat count
The site reports seats and the chance that combinations of parties reach 176. A government needs an investiture, which can succeed with fewer than 176 if enough deputies abstain, and which depends on bargains between parties that seat counts cannot show. The combinations on the site are arithmetic; they are not predictions of who will govern.
The electoral law
The forecast applies the LOREG as it stands, with the 2023 apportionment of seats to provinces. The next decree will re-apportion the seats by population, and a reform of the law, which is discussed from time to time, would change the calculation. Neither is anticipated.
Summary
| Limitation | How much it matters | What would address it |
|---|---|---|
| Eight elections, 2004 and 2016 dominant | The spread for the PP and the PSOE rests heavily on 2004 | More elections; the backtest reruns itself |
| Last seats in small provinces | The largest single source of seat uncertainty | An estimated size for provincial noise |
| Independent errors | The PP–PSOE gap is a little too certain | Correlated draws from past error patterns |
| Proportional swing untested | Provincial projections may be biased, above all in Catalonia and the Basque Country | Backtest against uniform and mixed swing |
| The split left | Sumar's and Podemos's seats depend on an assumed ratio and line-up | Party decisions; regional polls |
| New and local parties | SALF spread evenly; local lists never win | Model the largest local lists |
| Election date | An early election would narrow the ranges suddenly | None possible; ranges shown for the latest date |
| Second-hand polls | Transcription errors would pass into the average | Pollsters' own series; more readable pages |
| Investiture | Seat counts are not governments | None; stated on every page |
Table: Known limitations of the Spanish forecast.
The forecast's claim is narrow: the Spanish polls, read as they are, carried through a seat calculation that reproduces the last election exactly, with their uncertainty where it matters. The next general election will be the first real test, and we will publish how it went.
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).
How to cite
Polling Forecast (2026). What we don't know yet: limitations of the Spanish forecast. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/spain-limitations. Live figures change as new polls arrive; the date above is when the text was last revised.



