W niedzielę wyborczą, tuż po zamknięciu lokali wyborczych, miliony Polek i Polaków z zapartym tchem czekają na wyniki exit poll. To badanie wzbudza niespotykane emocje i zainteresowanie, ponieważ niemal natychmiast pokazuje, jak mogą wyglądać wyniki wyborów, a wkrótce potem konfrontowane jest z oficjalnymi danymi PKW. Ta wyjątkowa sytuacja sprawia, że exit poll staje się jednym z najbardziej wyczekiwanych i jednocześnie najbardziej wymagających badań społecznych.
W takim badaniu wynik musi być trafiony. Poza błędem oszacowania – choć nawet tego niektórzy nam nie wybaczają – nie ma tu miejsca na potknięcia. I ludzie, i technologia nie mogą zawieść. Każda pomyłka, czy to w doborze próby, zbieraniu danych, czy w ich analizie, może znacząco wpłynąć na precyzję wyników i podważenie wiarygodności nie tylko wykonawcy badania, ale całej badawczej branży.
Exit poll – ze swoimi wyśrubowanymi standardami – nie tylko prognozuje wyniki wyborów, ale także dostarcza cennych lekcji na temat jakości i rzetelności badań w ogóle. To też lekcja tego, jak czytać wyniki i gdzie są granice wyciągania z nich wniosków. Bo choć niemal wszystkim wydaje się, że znają się na sondażach, to zdecydowana większość z nas nie jest w tym tak dobra, jak zakłada.
Watch this WaysConf session, then continue with related talks or explore the current programme.
In this WaysConf presentation, Joanna Skrzyńska, Public Affairs Leader at Ipsos, explores the methodology, challenges, and misinterpretations of political polling, focusing on the high-stakes environment of exit polls. Using the second round of the June 1, 2025 presidential election as a case study—where the exit poll initially favored Rafał Trzaskowski before the late poll correctly shifted to Karol Nawrocki—Skrzyńska explains how statistical errors, non-response bias, and media sensationalism distort public perception. She details the rigorous logistics of exit polling, which involves over a thousand field interviewers and massive sample sizes, and contrasts it with the "Late Poll" method that uses actual committee results. Ultimately, she emphasizes that while data is invaluable, researchers and the public must maintain humility, understand the inherent margins of error, and prioritize robust processes over the illusion of absolute precision.
Political polls are not predictive forecasts but rather snapshots of public sentiment at the exact moment of data collection. They often include a significant margin of undecided voters and are heavily influenced by the methodology used, such as telephone versus online surveys.
Watch from 9:41Reporting survey results with decimal places creates a false sense of accuracy. Given standard statistical error margins of several percentage points, minor fluctuations of one or two percent are statistically insignificant and should not be interpreted as meaningful trends.
Watch from 17:51Unlike pre-election polls that rely on declarations of intent, exit polls survey actual voters immediately after they leave the polling booth. This methodology eliminates the uncertainty of voter turnout, though it still carries a declared error margin of about two percent.
Watch from 22:16The refusal rate for exit polls has risen to twenty percent, and because these refusals are systematic rather than random—often concentrated among specific demographics like older, rural men—researchers must use advanced analytical modeling to correct the data.
Watch from 32:04Voting patterns are highly time-dependent, with older, conservative voters typically casting ballots early in the morning, often after church. This creates a temporary morning lead for right-wing candidates, which progressive candidates gradually close as younger voters turn out later.
Watch from 41:27These concise notes were edited from automatic captions and checked against the talk structure. They are not a verbatim transcript.
During the second round of the presidential election on June 1, 2025, Ipsos warned television networks that the race was too close to call, with results falling well within the two-percent margin of error. Despite this warning, candidate Rafał Trzaskowski publicly declared victory at 21:00 based on a slim 50.3% lead in the initial exit poll. Within two hours, as actual data began to stabilize, the projected winner flipped. This dramatic shift highlighted the danger of treating early, statistically non-definitive exit poll numbers as absolute certainties.
Watch from 3:48While telephone surveys remain the most reliable method for tracking political preferences, online polling is increasingly common. However, online panels fail to accurately represent voters over fifty, meaning raw online data would project a completely different government than reality. Furthermore, while pre-election polls boast a sample size of one thousand, this number only reflects the initial question about voting intent. Once filtered for respondents who actually plan to vote, the active sample size for party preferences drops to around seven hundred, widening the margin of error.
Watch from 10:54Media outlets frequently publish political poll results with two decimal places, creating a false illusion of mathematical precision. In reality, with a standard margin of error of plus or minus three percentage points, decimal differences are completely meaningless. Journalists also tend to report minor fluctuations of one or two percent as dramatic surges or devastating drops. This hyper-detailed reporting ignores basic statistical principles, leading the public to believe that minor, statistically insignificant shifts represent real changes in voter behavior.
Watch from 17:30Ipsos utilizes a multi-stage process on election night, starting with the exit poll at 21:00, which relies on paper questionnaires filled out by voters. By 23:00, this is updated to a "Late Poll 50%," where half of the survey data is replaced with actual, official results posted on the doors of sampled polling stations. By early morning, the "Late Poll 90%" is released, relying almost entirely on real committee counts. This progression steadily reduces the margin of error from two percent down to just half a percent, providing a highly accurate final projection.
Watch from 23:30Modern exit polling faces a significant hurdle as the refusal rate has climbed to twenty percent. Because these refusals are systematic rather than random, they threaten the validity of the sample. For instance, older men, individuals with basic vocational education, and rural residents are statistically more likely to decline participation. To counter this bias, field interviewers record the estimated age and gender of everyone who refuses, allowing central analysts to apply advanced demographic modeling and correct the final calculations.
Watch from 31:43Election day data reveals that voter turnout is not demographically uniform throughout the day. Early morning voters are predominantly older citizens who often cast their ballots immediately after attending church services. Because this demographic heavily favors conservative candidates, right-wing politicians typically enjoy a massive lead in the early hours. Progressive candidates and parties generally catch up later in the afternoon and evening as younger, urban voters finally make their way to the polling stations.
Watch from 40:43
