Poland Parliamentary electionby 11 Nov 2027in 355 days Government majority 10% Right-wing majority 86% Calendar Track record Research
Research · Poland
EN PL

Reading Polish polls: why they are not corrected

Abstract. The core methods paper for Poland. Taking off half the average past miss, with each election predicted from the others, was further out on average at the seven Sejm elections since 2001, so Polish polls are taken as published. Only Trzecia Droga's line was missed the same way each time; the others' large misses changed sign. We set out the test election by election, how the don't-knows are removed, what CBOS's declared vote offers, and the seat calculation that reproduces every election since 2005.

Polling Forecast · updated 9 November 2026 · 5 min read · Poland, method, leave-one-out, correction, don't-knows, CBOS, Kodeks wyborczy

Summary

This paper sets out how Polling Forecast reads Polish opinion polls for the next Sejm election, and why, as in Germany and Spain, it does not correct them. Final polls at the seven Sejm elections since 2001 missed by about 1.7 points per list, but the direction of the large misses changed from one election to the next. Taking off half the average past miss, with each election predicted from the others, was further out on average, so our projection takes the polls as they are, with the don't-knows taken out. The seats then depend on the Kodeks wyborczy and on who runs with whom; our seat calculation reproduces every Sejm election since 2005 exactly, and the line-up is an explicit assumption (Line-ups and thresholds: 5%, 8% and the Trzecia Droga question).

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 Poland the 1993 to 1997 seat check is still shown but no longer counts.
  • Our figures beside the polls (26 October). The correction is off in Poland, so our projected shares are the poll average, and the pages say so rather than repeating the figures.

The electoral system

Each voter marks one candidate on one committee's list in their okręg; the vote counts first for the list. The rules are in the Kodeks wyborczy (Dz.U. 2025 poz. 365).

  1. Okręgi. 460 seats in 41 okręgi of 7 to 20 seats, fixed in annex 1 of the Kodeks and unchanged since 2011. Votes cast abroad and on ships are counted in okręg 19, Warsaw.
  2. Thresholds. Only lists with 5% of the valid national vote take part in the share-out, or 8% for a coalition committee (art. 196). Committees of registered national minorities are exempt (art. 197).
  3. D'Hondt. In each okręg the qualifying lists share the seats by D'Hondt (art. 232).

A majority is 231. Three fifths of the votes, 276 when all 460 vote, override a presidential veto (The President's veto and the 276 line).

The data

Results by okręg for every election since 2001, and national results since 1991, come from the Państwowa Komisja Wyborcza's data repository. Polls come from the pollsters' own releases (grade A) and the commissioning outlets' articles (grade B) where their figures could be matched, and otherwise from Wikipedia's poll tables, graded C and marked on the site. The Polish data: results, maps, statistics, MPs and polls gives every source.

The seat calculation

Our implementation of the law, given each election's official okręg votes, reproduces the seats the PKW declared for every list at every election from 2005 to 2023:

ElectionLists with seatsLists under the thresholdSeats that differEvery seat matches
2023500yes
2019600yes
2015630yes
2011600yes
2007500yes
2005710yes
Live The seat law run on each election's official okręg votes, against the seats the PKW declared Computed from the database, polls to 19 November 2026.

2001 used a modified Sainte-Laguë count and is not part of the check. From votes to seats: the Kodeks wyborczy gives the details.

How far Polish polls have missed

Final-poll average minus result, in points, by 2023 lineage:

Party2001200520072011201520192023Average
KO+1.2+6.5−0.6+1.7−1.2−1.8−0.9+0.7
Lewica+6.8−5.6+0.10.0−0.2+1.2+1.2+0.5
PiS−1.3+2.8−0.1+0.1+0.3+2.6+1.5+0.8
TD (PSL + Polska 2050; earlier PSL and Kukiz'15)+0.2−2.1−1.3−1.4+0.4−1.6−3.7−1.4
Konfederacja––––−0.1−1.0+2.0+0.3
Live Sejm: final-poll average minus result, by 2023 lineage (points) Computed from the database, polls to 19 November 2026.

The final polls are taken three to five days before the vote, before the election silence (Measuring Polish polling error, 2001–2023). The lists that became Trzecia Droga were understated at five of seven elections, and by every pollster in 2023. The other lines changed direction: the left was overstated by 6.8 points in 2001 and understated by 5.6 in 2005; KO's line was overstated by 6.5 in 2005 and understated by 0.9 to 1.8 at the last three elections.

The test

Each election was predicted with the correction learned from the other six: half the mean error, for lines measured at two or more of them, followed by a rescale so the shares keep the same total. The lines scored are PiS, KO, Trzecia Droga, Lewica and Konfederacja.

ElectionPolls' average missCorrected, leave one outVerdict
20012.372.69further out
20054.274.31further out
20070.530.47closer
20110.810.70closer
20150.430.73further out
20191.631.65further out
20231.861.81closer
Average1.701.77
Live Correcting Polish polls, each election left out in turn Computed from the database, polls to 19 November 2026.

The correction was closer at 3 elections and further out at 4, and further out on average: 1.77 points per list against 1.70 for the polls as published. That fails the rule, so our Polish projection does not use the correction.

Election by election:

  • 2001 and 2015: further out, clearly. In 2001 Trzecia Droga's line was slightly overstated and PiS's understated, the opposite of their usual pattern, so the correction moved both the wrong way. 2015 was polled closely, 0.43 points per list, so the correction moved lists that were already about right.
  • 2005 and 2019: further out, by a few hundredths. In 2005 the left, understated by 5.6 points, is moved further down by a correction learned mostly from 2001, when it was overstated. In 2019 KO, understated, is moved further down by a correction learned mostly from 2005, when its line was overstated.
  • 2007, 2011 and 2023: closer, narrowly. Trzecia Droga's line was understated in all three, and the correction adds to it, which helps most in 2023.

So the only steady signal is the smallest line's, and it is not enough to offset the damage done by correcting lines whose misses change sign. The code checks the rule each time the forecast is built: after the next election, if the correction beat the polls on average, it would switch on without any change to the method.

Don't-knows and CBOS's declared vote

Most Polish pollsters publish a don't-know share inside the headline figures; we take it out and rescale the parties to 100 before averaging:

PollsterFieldwork endDon't know, as publishedPolls since 2023 with a don't-know figure
Fundacja Centrum Badania Opinii Społecznej (CBOS)4 Nov 2026–39/46
Instytut Badań Pollster19 Nov 2026–1/53
Instytut Badań Rynkowych i Społecznych (IBRiS)7 Nov 20268.556/58
OGB (Ogólnopolska Grupa Badawcza)5 Nov 2026–0/24
Opinia24 (Opinia24 Sp. z o.o.)4 Nov 20265.557/59
Research Partner (Ariadna panel)16 Nov 20269.518/20
Social Changes10 Nov 2026–6/20
United Surveys (by IBRiS)9 Nov 202610.173/73
Live Don't-knows in the headline: each pollster's latest poll in today's average Computed from the database, polls to 19 November 2026.

In today's average the pollsters that publish one put it at 8.4 points on average. Removing it assumes that the undecided will split like the decided; that assumption is made in every final poll of the record too, so any error it causes is part of the misses measured above.

In Britain the forecast starts from what respondents say before the pollster's adjustments (Why we start from British pollsters' raw answers rather than the published headline). CBOS publishes the nearest Polish equivalent, the declared vote of everyone asked, and since late 2025 some of its reports add an imputed distribution without the undecided. We store both. There are no CBOS pairs of the two at past elections, so there is no raw-figure test for Poland yet.

Today's average

PartyPolls2023 listRuns on
KO32.6KOKO
PiS22.0PiSPiS
Konfederacja14.0KonfederacjaKonfederacja
KKP8.8KonfederacjaKKP
Lewica7.4LewicaLewica
Rozwój Plus4.7PiSRozwój Plus
Razem4.3LewicaRazem
PSL3.4TDPSL
Polska 20501.8TDPolska 2050
Live Parties in today's average, the 2023 list they come from and the list they run on in our line-up Computed from the database, polls to 19 November 2026.

The average is of parties, not lists. Each party's share goes to the list it runs on in the line-up assumed (Line-ups and thresholds: 5%, 8% and the Trzecia Droga question); a joint list gets the sum of its parties' shares.

From the average to the seats

Limitations

  • Seven elections, two of them (2001 and 2005) from a party system that no longer exists.
  • The lines joining past lists to today's parties are rough, and today's party system has split again since 2023.
  • The line-up is an assumption, and it changes the seats more than the polls do.
  • About a third of this term's polls reach us through Wikipedia's tables.
  • What we don't know yet: limitations of the Polish forecast collects the rest.

References

Kodeks wyborczy (Dz.U. 2025 poz. 365), articles 115, 196, 197 and 232, and annex 1.

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

Konstytucja Rzeczypospolitej Polskiej, article 122.

How to cite

Polling Forecast (2026). Reading Polish polls: why they are not corrected. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/poland-method. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like