If we're wrong: replaying past Polish polling misses on today's polls
Abstract. Nine readings of today's Polish polls: as they stand (also our projection, because the correction is off) and each Sejm election's final-poll miss since 2001 replayed in full, all through the same seat calculation and line-up. The replays move seats mostly through the thresholds: 2001 pushes Lewica under 5%, 2005 and 2019 push Rozwój Plus under it. We explain why every okręg changes under at least one replay.
Summary
A lens is one stated assumption about how wrong today's polls are. For Poland the site shows nine readings of the same polls: the polls as they stand, which are also our projection because the correction fails its test (Reading Polish polls: why they are not corrected), and seven replays that take each Sejm election's final-poll miss since 2001 and apply it in full to today's average. Every reading goes through the same seat calculation under the same line-up (Line-ups and thresholds: 5%, 8% and the Trzecia Droga question), and we show them side by side without averaging them.
Two things follow from the Polish record. The replays move seats mostly through the thresholds: the 2001 miss pushes Lewica under 5%, the 2005 and 2019 misses push Rozwój Plus under it, and each time the lost votes change everyone else's seats. And on today's figures no replay gives the governing coalition 231 under the default line-up, while the parties to its right reach it under all but one. 41 of the 41 okręgi change seats under at least one replay.
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.
- The lenses on the map (November). The replays are also on the map as "The polls miss again", beside "You set the vote".
- When the polls missed (2 November): Ten points clear, then second: how the polls missed Duda's first-round win in 2015.
What a lens is
Let \(M\) be the parties the forecast models, \(v_p\) today's average for party \(p\) (each pollster's latest poll within 21 days of the newest, with don't-knows taken out, equally weighted) and \(S = \sum_{p \in M} v_p\). The record measures the five 2023 lines, not today's parties, so a replay works on the lines. Each line's parties are added together; for a past election \(t\) with error \(e_{L,t}\) on line \(L\), the line's total becomes
the lines are rescaled to their original combined share, and each line's total is shared among its parties in their ratio today. So Rozwój Plus moves with PiS, KKP with Konfederacja, Razem with Lewica, and the PSL and Polska 2050 with Trzecia Droga's line. Konfederacja's line was measured only in 2015, 2019 and 2023, and is not moved by the earlier replays.
Each replay applies a whole past miss to the polls as they stand. The replays are what the Sejm would look like if this election's polls missed exactly as a past one's did.
After the replay, every reading goes into PolandForecast.Run under the default line-up: proportional swing to the okręgi, the thresholds and D'Hondt (From votes to seats: the Kodeks wyborczy).
The readings
The polls as they stand, and our projection. The same numbers, because the correction is off.
Seven past misses. One replay for each of 2001, 2005, 2007, 2011, 2015, 2019 and 2023, built from 50 final polls.
The reader's own. The what-if tool lets a reader set an error for each party and runs the same calculation. Its presets are the readings above. The prediction tool does the same from shares the reader types in.
Every reading through the same calculation
| Reading | KO | PiS | Konfederacja | KKP | Lewica | Governing coalition | PiS + R+ + Konf. + KKP |
|---|---|---|---|---|---|---|---|
| The polls as they stand | 191 | 131 | 73 | 41 | 24 | 215 | 245 |
| Our projection (the polls, uncorrected) | 191 | 131 | 73 | 41 | 24 | 215 | 245 |
| If the polls miss like 2001 | 194 | 139 | 75 | 43 | 0 | 194 | 266 |
| If the polls miss like 2005 | 153 | 122 | 78 | 39 | 49 | 202 | 239 |
| If the polls miss like 2007 | 195 | 129 | 72 | 41 | 23 | 218 | 242 |
| If the polls miss like 2011 | 181 | 135 | 78 | 40 | 26 | 207 | 253 |
| If the polls miss like 2015 | 195 | 127 | 73 | 40 | 25 | 220 | 240 |
| If the polls miss like 2019 | 199 | 120 | 80 | 41 | 20 | 219 | 241 |
| If the polls miss like 2023 | 197 | 128 | 61 | 38 | 19 | 233 | 227 |
| Our rule, learned from the four votes since 2024 | 188 | 140 | 71 | 41 | 20 | 208 | 252 |
| If the polls miss like the 2024 regional assemblies (sejmiki) | 191 | 148 | 58 | 34 | 18 | 209 | 251 |
| If the polls miss like the 2024 European Parliament election | 190 | 137 | 68 | 40 | 16 | 206 | 254 |
| If the polls miss like round one of the 2025 presidential election | 174 | 134 | 81 | 46 | 16 | 190 | 270 |
| If the polls miss like the 2025 presidential run-off | 183 | 139 | 73 | 41 | 24 | 207 | 253 |
The table is recomputed whenever a poll is added. Four features follow from the shape of the record and should survive most changes in the polls.
2001 takes Lewica out. The polls overstated the left by 6.8 points in 2001; taking that off today's Lewica leaves it under 5%, and its seats go mostly to the other large lists. The governing coalition is at its lowest under this reading.
2005 and 2019 take Rozwój Plus out. Both years overstated PiS's line, by 2.8 and 2.6 points; Rozwój Plus, sharing the line's correction, falls under 5%. In 2005 the same replay also takes 6.5 points off KO and adds 5.6 to Lewica.
2007, 2011 and 2015 change little. Those elections were polled closely, and their replays move a few seats.
2023 is the government's best replay. It adds 3.7 points to the line of the PSL and Polska 2050 and takes some off PiS and Konfederacja. The PSL and Polska 2050 stay under 5% each, so their gain is lost, but KO gains and the right loses. It still leaves the governing coalition short of 231.
Why the seats jump
Two things make Polish seat counts jump. The thresholds are cliffs: a list that crosses 5% wins a dozen seats or more at once, and its votes stop being lost. And D'Hondt in okręgi of seven to twenty seats gives the last seat in each okręg to whichever list is closest, so a small national shift moves the last seat in many okręgi at once. 22 okręgi have a last seat within a point of changing hands in our projection (From votes to seats: the Kodeks wyborczy).
Okręgi at risk
For each okręg the site counts how many of the seven replays give it a different set of seats, by list, from the projection:
An okręg is "at risk" if \(F_k \geq 1\); today 41 of the 41 are. That is every okręg, because a replay that moves a list across a threshold changes seats everywhere. It is a count of historical scenarios, not a probability; the simulation's chance for each seat is shown on the okręg pages and the seat map.
Limitations
Seven past misses are not a distribution. The replays show what known misses would do, not how likely each is.
Old misses, new parties. The largest misses come from 2001 and 2005, from lists that no longer exist. Applying Lewica's 2001 miss to today's Lewica assumes the polls would miss a party of 8% as they missed an alliance of 41%.
The line-up is fixed. Every reading uses the default line-up. The difference between line-ups (Line-ups and thresholds: 5%, 8% and the Trzecia Droga question) is as large as the difference between readings.
Splinters move with their parents. A replay cannot tell whether the polls miss Rozwój Plus the way they missed PiS.
References
Jennings, W. and Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.
How to cite
Polling Forecast (2026). If we're wrong: replaying past Polish polling misses on today's polls. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/poland-lenses. Live figures change as new polls arrive; the date above is when the text was last revised.


