Research · When the polls missed

The sorpasso that never came: Spain's June 2016 polling miss

Abstract. All 22 final polls had Unidos Podemos second, ahead of the PSOE, and the PP on under 30%; the coalition finished third with 21.2% and the PP won 33.0%. The polls counted the December voters of Podemos and Izquierda Unida as if they would all back the joint list, and more than a million did not.

Polling Forecast · 2 November 2026 · 6 min read · Spain, June 2016 general election, polling error, final polls, inquiry, pollster methods

Every election the polls got badly wrong →

Summary

Every one of the 22 final polls before the Spanish general election of 26 June 2016 had Unidos Podemos in second place, ahead of the PSOE. Their average put the new coalition of Podemos and Izquierda Unida on 24.9% and the PSOE on 21.3%. On the night the coalition took 21.2% and finished third, 1.5 points behind the PSOE, while the PP won 33.0%, 3.3 points more than the polls had given it. The error on the PP's lead over the PSOE was 2.0 points; the error on the race for second place was 5.1. The polls treated the December 2015 voters of Podemos and Izquierda Unida as if they would all turn out for the joint list, and more than a million of them did not.

What the polls said

Spanish law bans publishing polls in the five days before the vote, so the last legal day was Monday 20 June. The final polls, a selection of the 22:

Pollster (client)	Fieldwork end	PP	Unidos Podemos	PSOE	Cs
GAD3 (ABC)	20 Jun	30.7	24.9	21.2	14.1
GESOP (El Periódico)	19 Jun	28.3	24.0	21.9	14.9
Invymark (laSexta)	19 Jun	30.1	26.1	19.9	13.8
TNS Demoscopia	19 Jun	29.6	24.7	21.0	14.8
NC Report (La Razón)	17 Jun	31.1	25.4	21.2	13.2
Celeste-Tel (eldiario.es)	17 Jun	30.5	25.3	21.8	13.3
Sigma Dos (El Mundo)	16 Jun	30.5	24.8	20.0	14.1
MyWord (Cadena SER)	16 Jun	28.9	24.8	20.4	14.4
Metroscopia (El País)	15 Jun	29.0	26.0	20.5	14.5
Simple Lógica	15 Jun	29.2	23.5	22.5	15.4
CIS	22 May	29.2	25.6	21.2	14.6
Average of all 22		29.7	24.9	21.3	14.2

The Unidos Podemos figure includes its regional allies: En Comú Podem, Compromís-Podemos-EUPV and En Marea. The CIS, with 17,488 interviews between 4 and 22 May, gave the coalition and its allies 88 to 92 seats against 78 to 80 for the PSOE.

What happened

The Junta Electoral Central's official count, published in the Boletín Oficial del Estado on 22 July 2016:

Party	Votes	Share	Seats	Polls minus result
PP (with PP-PAR, PP-Foro, UPN-PP)	7,941,236	33.0	137	−3.3
PSOE (with PSC, PSOE-NCa)	5,443,846	22.6	85	−1.4
Unidos Podemos and allies	5,087,538	21.2	71	+3.7
Ciudadanos	3,141,570	13.1	32	+1.2

Shares are of the valid vote, 24,053,755 including blank votes. Turnout fell from 69.7% in December 2015 to 66.5%. The PP gained 704,271 votes on December and 14 seats. The PSOE lost five seats but kept second place.

Why it happened: what the pollsters found

Spain has no industry body that investigates a polling miss, and no formal inquiry was held. The nearest thing was a public debate, "Aprendiendo del 26J", at the Fundación Alternativas in September 2016, reported by Nueva Tribuna on 22 September. Four pollsters spoke.

Belén Barreiro (MyWord) said "el error gordo del 26J fue Podemos". Podemos voters, she said, exaggerate how likely they are to vote, and pollsters misread them at the estimation stage, the "cocina", with no electoral history for a new party to lean on.
José Miguel de Elías (Sigma Dos) put the fault in the estimation too: the PP, PSOE and Ciudadanos figures were right, but for Unidos Podemos "la tendencia no estaba lo suficientemente ajustada". He added that the PP is usually under-represented in samples.
Narciso Michavila (GAD3) blamed two novelties: the first repeat election in Spanish history and a coalition that had never stood before.
José Juan Toharia (Metroscopia) said polls describe a mood and are not meant to predict. He and Michavila said the votes the coalition lost went to abstention.

Two pieces on Politikon went further. Pablo Cabrera Álvarez (29 June 2016) rejected a late swing driven by the Brexit referendum of 23 June as the main cause, because the same bias showed in the latest polls and in the election-day surveys of GAD3 and Sigma Dos. He pointed to an overstated turnout among Unidos Podemos supporters, to their greater willingness to answer surveys, to quota samples and to herding. Antoni-Ítalo De Moragas and Berta Barbet (7 July 2016) modelled the CIS pre-election survey and found the coalition's losses concentrated among centrist voters with little interest in politics, former IU voters and older voters. The CIS post-election study later showed, as Delgado-Fernández and Cazorla report, that only 54% of those who recalled voting for IU-Unidad Popular in 2015 had voted for Unidos Podemos.

What our data shows

It was the whole industry. All 22 final polls overstated Unidos Podemos, by between 2.3 and 5.0 points, and all 22 understated the PP, by between 1.9 and 4.7. All 22 had the coalition ahead of the PSOE. Our Spanish error paper finds that more than 90% of each party's error was common to every pollster. Averaging more polls would not have helped.

It was not a late trend the polls missed. The CIS finished fieldwork five weeks before the vote and had the coalition on 25.6; GAD3 finished six days before and had it on 24.9. Polls taken across the last month tell the same story. Any movement came after the ban, and the election-day surveys suggest that most of the error was in the measurement, not the timing.

The polls added up two parties. In December 2015 Podemos and its allies won 5,212,711 votes and IU-Unidad Popular 926,783: together 24.4% of the valid vote. The polls' average for the merged list in June was 24.9%, almost exactly the sum. The real total was 5,087,538, more than a million fewer than December's combined vote. The pollsters' turnout and recall models assumed the merger kept both electorates. It did not.

The sign flipped in six months. At the December 2015 election the same industry had understated Podemos and its allies by 3.1 points. A pollster that leaned against repeating that error in June would have made the June error larger.

The blocs were missed too. PP plus Ciudadanos averaged 43.9% in the polls and won 46.1%; Unidos Podemos plus the PSOE averaged 46.1% and won 43.8%.

Our paper scores the PP without the UPN-PP list in Navarra, which puts the PP-PSOE gap error at 1.5 points rather than the 2.0 measured here against the PP's full total. Either way, 2016 is one of the two industry-wide misses in Spain since 2004, along with 2004 itself.

What the pollsters changed

No pollster published a revised method after 2016 that we could find. Most firms defended their work in public, and the debate stayed with turnout filters and the estimation stage. The CIS is the one case with a documented change. In 2016 its estimate already weighted the data by imputed recalled vote, as its own technical note for study 3141 says. After José Félix Tezanos became its president in 2018, the CIS announced that its barometers would drop the "cocina"; Maldita reported that its pre-election surveys for Andalusia in 2018 and the general election of 2019 instead used a "projection model" of more than 100 variables. Our records also show the CIS moving its final poll from face-to-face interviews, as in 2016, to telephone interviews by 2023.

What it means for our forecast

Our forecast corrects polls only when a correction passes a test: take off half the average past miss, and keep it only if doing so would have improved past forecasts, each election predicted without knowing its own result. In Spain the correction fails, and 2016 is a large part of the reason. The left lineage was understated by 3.1 points in December 2015 and overstated by 3.7 six months later; a rule built on the average of past errors would have pushed the wrong way in June.

So we do not correct Spanish polls. The record, 2016 included, sets the size of the simulated polling error instead, with an RMS error of about two points for the left lineage and two and a half for the PP. Because 2016 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 each pollster's 2016 final poll counts in our ratings.

Sources

How to cite

Polling Forecast (2026). The sorpasso that never came: Spain's June 2016 polling miss. Retrieved 2 November 2026 from https://www.pollingforecast.com/research/missed-es-2016. Live figures change as new polls arrive; the date above is when the text was last revised.