'A graveyard for pollsters': Exit polls, surveys miss the mark again

Audio report: written by reporters, read by AI

Major discrepancies between election forecasts and the final results of the June 3 local elections and National Assembly by-elections have sparked growing doubts about the accuracy of exit polls and pre-election surveys.

Changing voting patterns, declining survey participation and methodological shortcomings all contributed to the misses. The result has been a growing debate over whether Korea's polling industry needs to rethink how it measures public opinion.

In the Seoul mayoral race, incumbent Oh Se-hoon of the People Power Party (PPP) won with 49.15 percent of the votes compared to his contender Chong Won-o of the Democratic Party (DP), who won 48.13 percent of the votes as of 5 p.m. on Thursday. 

However, the exit poll had inaccurately projected that Chong would win with 51.4 percent and Oh at 46 percent.

A similar discrepancy emerged in the South Gyeongsang governor's race. The exit poll showed Kim Kyoung-soo of the DP leading with 54.3 percent, ahead of the incumbent Park Wan-soo of the PPP at 45.7 percent. Park ultimately won the race with 51.28 percent of the votes compared to Park's 48.71 percent.

The discrepancies were striking in races that had been expected to be closely contested.

The Daegu mayoral race had been projected as a razor-thin contest. Instead, Choo Kyung-ho of the PPP defeated Kim Boo-kyum of the DP. Choo received 53.92 percent of the vote to Kim's 45.05 percent.

The North Jeolla governor's race, another contest expected to be a nail-biter, also ended with a wide margin. Lee Won-taeg of the DP secured 51.22 percent of the vote, comfortably ahead of independent candidate Kim Kwan-young, who received 41.78 percent.

Forecasts for National Assembly by-elections showed similar discrepancies. 

In Gyeonggi's Pyeongtaek-B constituency, PPP candidate Yu Eui-dong won with 34.83 percent of the votes, far ahead of DP's Kim Yong-nam at 28.77 percent and Cho Kuk of the Rebuilding Korea Party at 27.24 percent.

Pre-election polls also differed significantly from the final results.

A telephone survey conducted by KBS and Hankook Research between May 24 and 27 showed Cho would win at 24 percent and DP's Kim Yong-nam would earn 22 percent, with Yu at at 20 percent. Despite the survey results, Yu ultimately won.

A separate ARS survey conducted by CBS and the Korea Society Opinion Institute between May 22 and 23 in Gyeonggi's Hanam-A constituency showed DP candidate Lee Kwang-jae at 48.8 percent and PPP candidate Lee Yong at 39.1 percent. The final result was far closer. Lee Kwang-jae won by a margin of just 1.56 percentage points and earned 49.68 percent of the votes and Lee Yong earned 48.12 percent.

The telephone poll results from the JoongAng Ilbo–Kstat survey from May 26 to 27 projected that Chong would win with 44 percent of the votes compared to Oh's 36 percent, with a margin of error of plus or minus 3.5 percentage points. However, Oh ultimately won the race. Details are available on the website of the National Election Survey Deliberation Commission.

Why did exit polls and poll surveys fail to capture voter sentiment accurately?

"Godeok-dong, Godeok-myeon and Paengseong-eup [in Pyeongtaek, Gyeonggi] are grouped into a single polling area, but too few people answer phone surveys, which makes it difficult to accurately reflect local voter sentiment," said Yu, the winner of the National Assembly by-election in Pyeongtaek-B constituency, on SBS Radio on Thursday.  "Many rural conservatives are tied up [with farm work] through the end of May [and can't pick up the phone for telephone surveys]."

"After conducting polls in Pyeongtaek-B and Busan Buk-A constituencies, I came to a troubling realization," said Kim Ou-joon, the pro-government YouTuber who operates polling firm Flower, during a broadcast on Thursday. "We conducted these polls ourselves, but even we can't trust them. It showed me that there are serious flaws in the polling and exit-poll methods that have been used for years."

Analysts point to changes in early-voting patterns as one of the primary reasons for the forecasting failures.

It is illegal to conduct exit polls during early voting under the Public Official Election Act. Consequently, pollsters combine exit-poll results from election day with survey data on early voters when producing their projections. How those figures are adjusted can significantly impact the final forecast.

"There used to be a belief that high early-voting turnout favored the DP and high election-day turnout benefited the PPP," said Park Dong-won, a representative of the political consulting firm Policom. "That pattern has broken down as more conservative voters have begun participating in early voting."

Experts also pointed to the phenomenon of "shy conservatives" as a key source of distortion in polling results.

"Following the Dec. 3 martial law declaration [in 2024] and the impeachment of former President Yoon Suk Yeol, conservative voters have become increasingly reluctant to participate in both telephone surveys and exit polls," said Yoon Wang-hee, a professor at the Academy of Korean Studies. "As more conservatives avoid pollsters altogether, the number of hidden voters has grown."

Voter fatigue caused by an overload of surveys also reportedly undermined polling accuracy.

"Voter fatigue was particularly severe in battleground races," said a DP official. "With an overload of telephone surveys, only those with a strong political interest or a desire to influence the outcome kept answering."

Analysts also pointed to limitations in the way polling samples are constructed.

Another limitation lies in how pollsters build their samples. Under current guidelines, polling firms use resident-registration data from the Ministry of the Interior and Safety to set quotas by sex, age and region and then apply statistical weighting to the results.

The problem is that actual voter turnout by age groups often differs from the population makeup. As a result, relying on demographic weighting alone can introduce structural errors into poll results.

Experts are increasingly warning against treating polls as infallible.

"This election will be remembered as a graveyard for pollsters," said Park of Policom. "Inaccurate polls created a false sense of momentum and distorted the race."

"Polling may be reaching a breaking point," said Prof. Yoon. "Rather than reflecting voter sentiment, polls can sometimes distort it simply by publishing the numbers. We need to fundamentally rethink how public opinion is measured."

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.

BY YEO SUNG-KUK [lee.jiwon10@joongang.co.kr]