Research · When the polls missed

The majority that wasn't: Spain's 2023 polling miss and the CIS outlier

Abstract. The final polls had the PP 5.7 points ahead of the PSOE and the PP and Vox close to a majority; the PP won by 1.4 points and the right finished six seats short. Every private pollster understated the PSOE, and only the CIS, which named the wrong winner, did not.

Polling Forecast · 2 November 2026 · 7 min read · Spain, 2023 general election, polling error, final polls, inquiry, pollster methods

Every election the polls got badly wrong →

Summary

The 22 final polls before the Spanish general election of 23 July 2023 put the PP on 34.1% and the PSOE on 28.3%, a lead of 5.7 points, and most of them showed the PP and Vox together at or above the 176 seats needed for a majority. The PP won with 33.1%, but the PSOE took 31.7%, the lead was 1.4 points and the two parties of the right won 170 seats. The error on the lead was 4.3 points, the largest in Spain since 2004. Every private pollster had the PSOE too low, by between 2.3 and 5.4 points; the one poll that did not was the state-run CIS, which had the PSOE ahead and named the wrong winner.

What the polls said

Spain bans publishing polls in the last five days of the campaign, so the last legal day was Monday 17 July. A selection of the 22 final polls:

Pollster (client)	Fieldwork end	PP	PSOE	Vox	Sumar
GAD3 (ABC)	17 Jul	36.9	28.6	11.7	11.5
Sigma Dos (El Mundo)	17 Jul	35.4	28.0	11.2	13.6
YouGov	17 Jul	32.0	28.0	14.0	14.0
AtlasIntel	17 Jul	30.8	29.4	17.0	16.1
NC Report (La Razón)	16 Jul	37.2	28.3	11.1	11.3
40dB (Prisa)	16 Jul	32.9	28.7	13.5	13.7
SocioMétrica (El Español)	16 Jul	33.5	28.3	13.7	13.2
GESOP (Prensa Ibérica)	15 Jul	30.6	27.4	15.0	14.4
Invymark (laSexta)	14 Jul	33.5	26.3	13.8	13.7
Celeste-Tel (Onda Cero)	12 Jul	36.5	28.4	11.4	11.5
CIS	12 Jul	30.8	32.2	11.8	14.9
Average of all 22		34.1	28.3	13.1	13.3

GAD3's final projection gave the PP and Vox 180 seats. The CIS published its "flash" survey, 8,798 telephone interviews taken between 10 and 12 July, on the last legal day; it put the PSOE and Sumar together on 47.1% against 42.6% for the PP and Vox. Its larger pre-election survey, 29,201 interviews between 8 and 27 June, had shown the PP on 31.4% and the PSOE on 31.2%.

What happened

The Junta Electoral Central's official count, published in the Boletín Oficial del Estado on 1 September 2023:

Party	Votes	Share	Seats	Polls minus result
PP	8,160,837	33.1	137	+1.0
PSOE (with PSC)	7,821,718	31.7	121	−3.3
Vox	3,057,000	12.4	33	+0.7
Sumar	3,044,996	12.3	31	+1.0

Shares are of the valid vote, 24,688,087 including blank votes; turnout was 66.6%. The PP and Vox won 170 seats; with UPN and Coalición Canaria they reached 172, four short of a majority. Alberto Núñez Feijóo's investiture failed with 177 votes against him. In Catalonia the PSC took 1,221,335 votes and 19 seats.

Why it happened: what the pollsters found

As in 2016, no inquiry was held. The explanations came from the pollsters themselves, in interviews after the vote.

Narciso Michavila (GAD3), on COPE on 24 July 2023: "Ayer no nos fue bien". The miss, he said, was in the gap between the PP and the PSOE, and "el votante catalán ha premiado mucho a Sánchez". In an interview with ABC reported by eldiario.es on 3 October, he said there had been a change in the last three days, a reaction from voters "al que no le gusta Sánchez pero le gusta menos Vox", and that about 200,000 ERC voters had moved to the PSC. GAD3 had seen the signal and discounted it: "Pensábamos que era ruido, y no era ruido."
Miguel Díaz (Electomanía), in Ara on 25 July: the industry "no ha sabido leer el efecto movilizador de la izquierda", and some firms had adjusted their numbers towards reference pollsters.
Berta Barbet, political scientist, in the same piece: samples miss less likely voters, and people who mean to abstain rarely admit it.
Josep Ribó (Opinòmetre): accurate polling of 52 constituencies needs more than 30,000 interviews, which only the CIS can afford.
Francisco Martín (Research & Insights), in El Independiente on 25 July: a 2,000-interview national poll leaves about 38 interviews per province, and undecided voters drifting to the PSOE and tactical voting in Catalonia were hard to catch.
José Félix Tezanos (CIS), on RNE on 25 July: "Si dicen todas las encuestas fallaron, pongan ustedes también excepto el CIS de Tezanos". He said the flash survey had missed by "unas décimas".

The most detailed later work came from inside the CIS. Antonio Alaminos and Antonio Alaminos-Fernández (REIS 192, 2025) compared recalled vote in CIS barometers after the election with the result. Recalled PSOE and Sumar votes were within half a point of the result; recalled PP vote was 8.2 points short and Vox 3.1 points short. They argue that PP voters hide their vote rather than refuse the survey, and that weighting samples by recalled vote, which most private pollsters do, distorts the estimate.

What our data shows

One party carried the error. The average missed by only 0.85 points per party across the eight parties we measure. The PP, Vox and Sumar were all within a point of the result on average, with polls scattered on both sides. The PSOE was understated by 3.3, and 91% of that error was shared by every poll. Choosing a different private pollster would not have avoided it.

The timing does not explain it. Private polls that ended on 4–6 July had the PSOE on 28.1 to 29.0; those that ended on 17 July had it on 27.5 to 29.4. There is no trend inside the window. More telling, the CIS found the PSOE at 31.2 in June and 32.2 in mid-July, while private pollsters interviewing in the same days found it at about 28. A difference that large between methods, at the same time, points to measurement: who was in the samples and how they were weighted. A late move after the ban, as Michavila describes, may have added to it, but the gap between the CIS and the rest was there a week before the vote.

The CIS was closest on average and still wrong on the winner. Its poll missed by 0.83 points per party, the best of the 22. But it had the PP 2.3 points too low and Sumar 2.6 too high, and its PSOE lead of 1.4 points was a 2.8-point error on the gap, in the other direction. Its record warns against reading 2023 as vindication: at the November 2019 election the CIS had the PSOE on 32.2% against a result of 28.0%. A pollster that leans towards the PSOE will look good in a year when everyone else leans against it.

The gap mattered more than the shares. In a fragmented Congreso a four-point error on one lead turned a projected PP and Vox majority into a result six seats short. That is the largest gap error since 2004, when the polls had the wrong party ahead by ten points.

The PP's error has no direction. The industry understated the PP by 3.3 points in 2016 and overstated it by 1.0 in 2023.

What the pollsters changed

We found no private pollster that published a revised method after 2023. The public lessons were about events (late movement, Catalan tactical voting) and about sample sizes per province, not about weighting. The methodological debate that did follow is the one the CIS researchers opened: whether recalled vote should be a weighting variable at all, given that PP voters in their data under-report it. Their answer is to stop weighting by recall and to treat a hidden vote as a category of non-response in its own right. That is the opposite of the usual private practice. We found no published reanalysis, from either side, showing that a given method would have caught the PSOE's 31.7% without also missing the PP or Sumar.

What it means for our forecast

Our correction rule takes off half the average past miss and keeps the correction only if it would have improved past forecasts, each election predicted without knowing its own result. In Spain it fails. The PSOE was understated by 3.3 points in 2023 but overstated in 2011 and April 2019; the PP was understated in 2016 and overstated in 2023. An error that changes sign cannot be predicted from its own average, so we do not correct Spanish polls.

The record sets the size of the simulated error instead, and our error paper scores the lead between the top two as well as each party's share, because in Spain the lead decides who is asked to form a government and how close the blocs come to 176. Because 2023 is in that record, a miss of its size is part of what the ranges are built from, not an exception to them. The full record is in our Spanish polling error paper, and the 2023 final polls count in our pollster ratings.

Sources

How to cite

Polling Forecast (2026). The majority that wasn't: Spain's 2023 polling miss and the CIS outlier. Retrieved 2 November 2026 from https://www.pollingforecast.com/research/missed-es-2023. Live figures change as new polls arrive; the date above is when the text was last revised.