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

Measuring Polish polling error, 2001–2023

Abstract. We measure the error of the final polls at the Sejm elections from 2001 to 2023, taken a few days before the vote because of the election silence: which polls count, how the don't-knows are removed, how lists that merged and split are measured, the error list by list and on the gap between the two largest lists, each pollster's record and the long view. The lists that became Trzecia Droga were understated at five of seven elections; the other lines' large misses changed direction, from the left's in 2001 and 2005 to KO's.

Polling Forecast · updated 9 November 2026 · 7 min read · Poland, polling error, final polls, cisza wyborcza, don't-knows, CBOS, Trzecia Droga, PiS, KO, pollsters

Summary

This paper sets out how we measure polling error in Poland: which polls count as final polls, how the don't-knows are handled, how lists that merged and split are measured, how the sources are graded, how the error is defined and what the record shows. At the seven Sejm elections from 2001 to 2023 the average of the final polls missed each list by between 0.4 and 4.3 points, 1.7 on average. The misses were large in 2001 and 2005, small in 2007, 2011 and 2015, and larger again in 2019 and 2023. One line was understated again and again: the PSL and the lists it later joined, measured as Trzecia Droga, at five of seven elections, by 3.7 points in 2023, when every final poll had it too low. The others changed direction too often for a correction to learn from them, and the correction fails its test (Reading Polish polls: why they are not corrected).

The record has three uses on the site: its direction was tested for a correction and rejected (Reading Polish polls: why they are not corrected); its size sets the width of the ranges (Simulating Polish polling error: where the chances come from); and its pollster-by-pollster detail feeds the ratings (Rating Polish pollsters: accuracy, the lead and openness).

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.
  • When the polls missed (2 November): Ten points clear, then second: how the polls missed Duda's first-round win in 2015.
  • The track record across countries (28 September). Final polls at 133 elections in nine countries, measured the same way, are at /accuracy/elections.

What is measured, and against what

The quantity that decides the seats is each list's share of the valid national vote: it decides which lists clear their thresholds, and, through the okręgi, how many seats they win. We measure against that share, from the results of the Państwowa Komisja Wyborcza (The Polish data: results, maps, statistics, MPs and polls).

Lists are measured by their line to the five lists of 2023:

  • PiS: PiS, with the AWS list of 2001 (AWSP) and the smaller right-wing lists that later joined it.
  • KO: the Platforma (PO), with the Unia Wolności in 2001 and Nowoczesna in 2015; the Koalicja Obywatelska from 2019.
  • Trzecia Droga (TD): the PSL, with Kukiz'15 in 2015 and 2019 (which ran on the PSL's list in 2019), and Polska 2050 in 2023.
  • Lewica: the SLD and its allies in each year (SLD–UP in 2001, the SLD and SDPL in 2005, LiD in 2007, the SLD and Ruch Palikota in 2011, the Zjednoczona Lewica and Razem in 2015), the Lewica list from 2019.
  • Konfederacja: the Korwin-Mikke lists in 2015, Konfederacja from 2019.

These lines are our coding, and they are rough: a line's result at an election is the sum of the lists we put in it, and a poll's figure is the sum of the same lists. They measure whether the polls had each political family right, which is what matters for today's parties, most of which descend from one of them.

Definitions

Let \(v_{p,t}\) be list \(p\)'s share at election \(t\) and \(y_{j,p,t}\) its share in final poll \(j\). The error of a poll is \(e_{j,p,t} = y_{j,p,t} - v_{p,t}\): poll minus result, so a positive error means the polls overstated the list. The industry error is the error of the simple average of the final polls,

\[ e_{p,t} = \bar{y}_{p,t} - v_{p,t}, \qquad \bar{y}_{p,t} = \frac{1}{n_{p,t}}\sum_{j} y_{j,p,t}, \tag{1} \]

with each pollster counting once. The average miss per list at an election is

\[ \text{MAE}_t = \frac{1}{P_t}\sum_{p} \left| e_{p,t} \right|, \tag{2} \]

and each list's record is summarised by its mean error (direction) and root mean square (size):

\[ \bar{e}_p = \frac{1}{T_p}\sum_{t} e_{p,t}, \qquad \text{RMS}_p = \sqrt{\frac{1}{T_p}\sum_{t} e_{p,t}^{2}}, \qquad \text{RMS}_p^2 = \bar{e}_p^{\,2} + s_p^2. \tag{3} \]

The gap error on the two lists that finished first and second, \(a\) and \(b\), is \(e^{\text{gap}}_t = e_{a,t} - e_{b,t}\); a positive value means the polls put the winner too far ahead.

The evidence: final polls before the election silence

What counts as a final poll

A pollster's final poll is its last poll whose fieldwork ended within 21 days of polling day. Polish law forbids publishing polls from midnight at the start of the day before the vote until voting ends (Kodeks wyborczy, art. 115), so the last polls are published two days before at the latest:

ElectionFinal pollsMedian days before the voteClosest to the voteFurthest
20015517
20054437
20078327
201155120
20157326
20199329
202312427
Live When the final polls were taken: days between the end of fieldwork and polling day Computed from the database, polls to 19 November 2026.

The typical final poll finished fieldwork three to five days before the vote: later than a Spanish or Italian final poll, earlier than a Dutch or British one.

Don't-knows

Most Polish pollsters publish a share of respondents who don't know how they will vote inside the headline figures. We take it out and rescale the named parties to 100, in final polls as in today's polls. Older poll tables sometimes merged "others" and "don't know" in one column; there, at most three points of it are counted as other parties and the rest dropped. The Centrum Badania Opinii Społecznej (CBOS) reports the declared vote of everyone asked, including those who don't know; we use its figures the same way, after taking the don't-knows out.

The elections

The database holds 50 final polls at the seven elections from 2001 to 2023: 5 in 2001, 4 in 2005, 8 in 2007, 5 in 2011, 7 in 2015, 9 in 2019 and 12 in 2023. We found no final polls for 1991, 1993 or 1997. Only CBOS and Kantar (earlier OBOP and TNS) have a final poll at all seven.

Grading the source

Every figure carries a source grade. Grade A is the pollster's own release; grade B the commissioning outlet's article; grade C a secondary transcription, here Wikipedia's poll tables. Each final poll's figures were compared with the release Wikipedia cites, where it could be read: 4 are grade A, 19 grade B and 27 grade C.

A worked example: 2023

The 2023 election, held on 15 October with a record turnout of 74.4%, had twelve final polls:

Pollster Fieldwork end PiS KO TD Lewica Konf. Poll's miss per list
Opinia24 13 Oct 34.8 29.9 12.2 8.6 9.6 1.19
IPC 8 Oct 34.3 31.4 10.3 9.5 8.6 1.64
Pollster 12 Oct 35.8 31.4 10.7 10.5 8.7 1.65
IBRiS 12 Oct 36.6 29.2 11.9 10.5 8.9 1.77
PGB Opinium 12 Oct 33.2 30.6 11.9 11.6 8.5 1.82
Estymator 11 Oct 36.7 29.9 10.1 9.7 9.7 2.01
United Surveys 10 Oct 36.7 30.5 10.2 11.1 9.1 2.02
Kantar 12 Oct 33.7 28.3 13.0 12.0 8.7 2.07
Social Changes 10 Oct 36.6 28.0 11.8 9.7 10.8 2.23
Ipsos 10 Oct 39.1 30.4 8.7 8.7 9.8 2.48
Research Partner 9 Oct 38.6 29.2 9.1 9.6 9.7 2.71
CBOS 11 Oct 46.1 28.9 7.9 6.6 7.9 4.34
Average 36.85 29.81 10.65 9.84 9.17
Result 35.38 30.70 14.40 8.61 7.16
Error +1.5 −0.9 −3.7 +1.2 +2.0

Table: Final polls and result, Sejm 2023, national list shares with don't-knows taken out, points. Rows ordered by each poll's own miss.

The average missed by 1.86 points per list. Every final poll had Trzecia Droga too low, by between 1.4 and 6.5 points, and every one had Konfederacja too high. CBOS, whose survey is partly face to face and whose headline includes the undecided, had PiS at 46.1% once the don't-knows were taken out, more than ten points too high. Without it the average would have missed PiS by less than a point.

The record, 2001–2023

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.

Four points stand out.

2001 and 2005 were the worst. In 2001 the polls had the SLD–UP near 47%; it won 41%. In 2005 the polls had the Platforma ahead or level; PiS won by almost three points, and KO's line was overstated by 6.5.

Trzecia Droga's line is understated. At five of seven elections, and by every pollster in 2023.

The others change sign. PiS was overstated by 2.8 points in 2005 and 2.6 in 2019 and understated by 1.3 in 2001. The left was overstated by 6.8 in 2001 and understated by 5.6 in 2005. KO's line was overstated in 2001, 2005 and 2011 and understated in 2007, 2015, 2019 and 2023.

2007 to 2015 were close. The average missed by 0.4 to 0.8 points per list.

Using equation (3):

List Elections Mean error RMS error Share of RMS² that is bias
Lewica 7 +0.50 3.39 2%
KO 7 +0.70 2.74 7%
TD 7 −1.36 1.87 53%
PiS 7 +0.84 1.63 27%
Konfederacja 3 +0.30 1.29 5%

Table: Summary of industry errors, 2001–2023, points. The last column is \(\bar{e}_p^2 / \text{RMS}_p^2\).

For every line but Trzecia Droga, almost all of the error is noise around a mean close to zero. A correction that takes off half the mean can help only where the steady part is large, and here it is large only for the smallest line. That is why the test fails (Reading Polish polls: why they are not corrected).

The gap between the two largest lists

ElectionTop twoReal gapPolls' gap missWorst poll
2001Lewica / KO25.3+5.5+9.7
2005PiS / KO0.4−3.7−8.5
2007KO / PiS9.4−0.5−5.4
2011KO / PiS9.3+1.5+8.3
2015PiS / KO5.9+1.5+3.7
2019PiS / KO16.2+4.4+13.6
2023PiS / KO4.7+2.4+12.4
Live The gap between the two largest lists: final polls' average error and the worst poll (points) Computed from the database, polls to 19 November 2026.

In Poland the largest list does not necessarily govern: in 2023 PiS came first and lost power. But the gap between the two largest lists shows the same pattern as the list errors. In 2019 and 2023 the polls put PiS too far ahead of KO, by 4.4 and 2.4 points on average, and the worst polls (CBOS both times) by more than twelve.

Pollster by pollster

Pollster Elections Average miss per list
Millward Brown 3 1.43
IBRiS 3 1.76
Kantar 7 1.82
Estymator 3 1.88
PBS 3 2.63
PGB Opinium 3 2.77
CBOS 7 3.02

Table: Average miss per list of each pollster's final polls, 2001–2023, pollsters with three or more, points. The track record page lists all of them.

CBOS stands apart: its final polls missed by three points per list on average, and it overstated PiS at every election from 2015. Kantar, the only other pollster with seven final polls, missed by 1.82. The ratings (Rating Polish pollsters: accuracy, the lead and openness) use these records, pulled towards a poor Polish final poll.

How the record is used

Our rule takes off half the average miss and keeps the correction only if it would have improved past forecasts, each election predicted without knowing its result. In Poland it does not: 1.77 points per list with the correction against 1.70 without (Reading Polish polls: why they are not corrected). The record still sets the size of the simulation's error for each party (Simulating Polish polling error: where the chances come from).

The long view

ElectionFinal pollsAverage miss per partyBiggest miss
200152.37Lewica +6.8
200544.27KO +6.5
200780.53TD (PSL + Polska 2050; earlier PSL and Kukiz'15) −1.3
201150.81KO +1.7
201570.43KO −1.2
201991.63PiS +2.6
2023121.86TD (PSL + Polska 2050; earlier PSL and Kukiz'15) −3.7
Live Final polls against the result, every Sejm election we could source Computed from the database, polls to 19 November 2026.

We found no final polls for the elections of 1991, 1993 and 1997, so the long view starts in 2001.

Limits

Seven elections, two of them very different. The record is dominated by 2001 and 2005, when the party system was changing fast.

Lines are rough. Lewica's line joins lists as different as the SLD–UP of 2001 and Ruch Palikota; TD's joins the PSL and Kukiz'15.

Don't-knows. Removing them assumes the undecided split like the decided. Older tables that merged others and don't-knows are handled by a rule of thumb.

Secondary sources. More than half the final polls are Wikipedia transcriptions.

Final polls only. These errors say nothing directly about a poll taken a year out; that is From today to polling day in Poland.

References

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

Kodeks wyborczy (Dz.U. 2025 poz. 365), article 115.

Shirani-Mehr, H., Rothschild, D., Goel, S. and Gelman, A. (2018). Disentangling bias and variance in election polls. Journal of the American Statistical Association, 113(522), 607–614.

How to cite

Polling Forecast (2026). Measuring Polish polling error, 2001–2023. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/poland-polling-error. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like