The Exit Poll Error Exposed a Deeper Split in Korean Politics

Seoul’s exit-poll controversy and Busan’s cross-voting revealed an electorate dividing its choices by voting time, local office and the politics of property.

Seoul first looked like a polling miss. Busan first looked like a partisan breakthrough. Neither reading held for long.

In South Korea’s June 3 local elections, the joint KBS-MBC-SBS exit poll projected Democratic candidate Jung Won-oh ahead in the Seoul mayoral race, only for People Power Party incumbent Oh Se-hoon to survive the official count. In Busan, Democratic candidate Jun Jae-soo defeated conservative incumbent Park Heong-joon, a result that seemed at first to mark a sharp leftward turn in one of the country’s most important conservative cities.

The fuller picture was less convenient. Seoul’s post-election story was shaped by a gender-and-age exit-poll table that was later found to have omitted early-voter estimates in four regions. Busan elected a Democratic mayor while leaving district halls and the metropolitan council largely in conservative hands. One city showed how an incomplete table could flatten a complicated result into a quick demographic explanation. The other showed, in the actual returns, how voters could separate a citywide choice from local administrative power.

The election was not a clean national swing. It was a split, visible across time, issue, geography and office. Early voters could not be assumed to resemble Election Day voters. Housing politics in Seoul did not speak with one voice. Young voters did not move as a single bloc. Busan did not simply turn Democratic. The exit-poll error mattered because it simplified an election that was never simple.

A failed projection needed a story

Seoul produced the sharpest version of the problem because the election-night expectation and the official result moved in opposite directions.

At 6 p.m. on June 3, the joint exit poll projected Jung ahead of Oh in the Seoul mayoral race, 51.4 percent to 46.0 percent. The official count later delivered a different result: Oh won with 49.22 percent against Jung’s 48.07 percent. The margin was narrow enough to demand explanation, but clear enough to overturn the night’s first reading of the race.

That reversal mattered before any demographic table was examined. Once the candidate shown ahead on television lost the official count, Seoul became more than a mayoral contest. It became a question. Why had the Democratic candidate fallen short in the capital despite an exit-poll lead? Why had Oh survived in a race that had appeared to be moving against him? Why had the city resisted a simple reading of national momentum?

Election analysis moves quickly toward available categories, and Seoul offers many of them. Property, gender, generation, incumbency, redevelopment expectations, renter anxiety, candidate familiarity and partisan fatigue can all be made to explain the same result. The first gender-and-age table narrowed that field of interpretation. It appeared to show that women in their 30s in Seoul had favored Oh over Jung, a finding that fit neatly into a post-election account of younger women moving away from the Democratic camp under the pressure of housing costs and dissatisfaction with the Democratic candidate.

The story was attractive because it reduced a complicated city to a legible bloc. Seoul’s property politics could be read through women in their 20s and 30s. The Democratic defeat could be understood as urban progressive erosion. A close race could be explained as a generational and gendered correction against the left. The explanation was plausible enough to travel, but it carried more certainty than the underlying data could bear.

Days later, the Korea Election Pool, or KEP, acknowledged that the gender-and-age voter analysis tables in Seoul, Daegu, Ulsan and North Chungcheong Province had been produced without early-voter estimate data. The final winner projections, KEP said, had combined Election Day exit-poll data with early-voter estimates. The demographic tables in those four regions had not. They reflected only Election Day exit-poll responses because of an output error in the regions handled by Hankook Research.

The correction changed the political weight of the Seoul table. Women in their 30s no longer appeared in the revised estimate as an Oh-leaning group; the corrected figures put Jung ahead. Women in their 20s also moved more strongly toward Jung in the revised table. The point is not that the corrected crosstab revealed how those voters actually cast their ballots. It did not. Official results do not disclose demographic voting behavior, and the revised figures remained survey estimates. The point is narrower and more damaging to the first interpretation: one of the main tables used to explain Jung’s defeat had been built without a voting layer KEP’s own projection system was supposed to include.

That layer mattered because early voting had become part of the political structure of the election. Voters who cast ballots before Election Day may differ from those who arrive at polling stations on the final day, not only by convenience but by age, organization, partisan intensity, work schedule and campaign mobilization. Treating Election Day respondents as if they could stand in for the whole electorate risked making one part of the city look like the city itself.

The incomplete table did not invent Seoul’s housing politics, nor did it erase Oh’s victory. Property was real in the race, and younger voters may well have divided in ways that deserve close analysis. But the first crosstab made the explanation too clean. It encouraged a reading in which Jung’s loss could be attached to one visible demographic movement, when the result was likely shaped by several overlapping pressures: asset expectations, renter insecurity, redevelopment politics, male youth conservatism, older voters, turnout patterns and the different composition of early and Election Day voters.

The failed projection created a demand for explanation. The incomplete crosstab supplied one. The corrected table did not settle the politics of Seoul, but it made the first story harder to defend.

The early vote was not a side note

The exit-poll correction became consequential because early voting had already changed the structure of election night.

For years, South Korean election broadcasts were built around a familiar image: voters leaving polling stations on the final day, interviewers collecting their answers, and broadcasters turning those answers into projections. That image no longer captures the whole electorate. By the time Election Day arrives, a large share of the vote has already been cast, and the voters who made that choice may not be politically interchangeable with those who wait until the final day.

Early voting is often described in administrative language, as if it were only a matter of convenience. It can also reflect organization, enthusiasm, age, work patterns, partisan discipline, access to polling places and the campaign’s ability to mobilize supporters before the final weekend. A voter who casts a ballot early is not automatically more ideological, more partisan or more loyal to one camp, but the timing of the vote is part of the political act. It tells campaigns who is already locked in, who still needs persuasion and which side has converted intention into turnout.

The 2026 local elections made that point harder to ignore. The final early-voting rate reached 23.51 percent, the highest ever for a local election in South Korea. A layer that large cannot be treated as a small correction attached to the final-day electorate. It becomes part of the electorate’s architecture.

KEP’s own design recognized that reality. Its winner projections combined Election Day exit-poll responses with early-voter estimates. The error came later, in the tables that translated the projection into a demographic account. In Seoul, Daegu, Ulsan and North Chungcheong Province, the gender-and-age tables omitted the early-voter layer even though KEP said the headline projections had included it.

That difference is easy to underestimate. An early-voter estimate is not a decorative supplement to an Election Day exit poll. Once early voting becomes large enough, the survey is no longer one stream of interviews with a small adjustment attached. It becomes a composite picture assembled from different voting moments, each carrying its own demographic and political composition. The question is not whether early voters permanently belong to one party or another. The question is whether a table that leaves them out can still describe the electorate it claims to describe.

In Seoul, the omission mattered because the first table seemed to turn the mayoral result into a story about a particular demographic break. The same risk applied more broadly. In Daegu, Ulsan and North Chungcheong Province, the corrected tables also changed how the electorate could be read. The pattern did not produce a new final truth about every voter group. It showed that Election Day-only tables had presented one version of the electorate with a confidence they did not deserve.

A normal polling miss can come from late movement, turnout imbalance, weighting failure or a sample that fails to capture the final electorate. The KEP case involved something more specific. A data layer that had been collected for the projection system did not appear in the demographic outputs that helped shape the public interpretation of the vote.

The failure sat between measurement and explanation.

That distinction should matter for election coverage. A winner projection asks who is likely to win. A demographic crosstab asks what kind of electorate produced the result. The first question can be answered with a model that combines early and Election Day voters. The second cannot be answered responsibly if the table used for explanation silently drops one of those groups.

Early voting does not make exit polling impossible. It makes election-night interpretation more fragile. When voting is spread across time, the final-day voter is no longer a natural stand-in for the whole. A modern election-night poll has to show not only the projected result but also the layers from which that result was built.

Without that transparency, one voting moment can be mistaken for the electorate.

Housing did not speak with one voice

Housing mattered in Seoul. The error was in making it speak too clearly.

The mayoral race unfolded in a city where property is never only a policy area. It is household balance sheet, generational anxiety, neighborhood expectation and political memory at once. For homeowners, housing can mean asset protection and redevelopment value. For renters, it can mean monthly pressure and a narrowing path into ownership. For younger voters who do not yet own property, it can carry a more conflicted meaning: resentment toward high prices, fear of permanent exclusion and, at the same time, a desire for the market to open a door rather than close it.

That complexity made Seoul vulnerable to an overly clean interpretation after the vote. Once Oh defeated Jung despite the exit poll’s initial projection, housing became the most available explanation. It had been central to the campaign, it was already embedded in Seoul’s political identity, and it could connect individual frustration to a citywide result. If young women in their 20s and 30s had moved sharply toward Oh, the story seemed ready-made: the Democratic camp had lost urban renters and aspiring homeowners to property anxiety.

The corrected crosstab did not make housing irrelevant. It made that version of the housing story less secure.

Both candidates campaigned as housing mayors in different ways. Jung and Oh each pledged large-scale supply and faster redevelopment, while presenting themselves as the candidate better able to move Seoul’s stalled housing machinery. The contest was not between a housing candidate and a non-housing candidate. It was between competing versions of what housing politics should mean in a city where nearly every voter has a stake, even if not the same stake.

That is why the early interpretation around younger women was too thin. A renter in her 30s in western Seoul, a homeowner’s daughter in a reconstruction district, a single professional facing rising jeonse costs, and a young couple calculating whether they will ever enter the market are all touched by housing, but not necessarily in the same political direction. Their choices may respond to price, supply, rent, redevelopment, interest rates, local infrastructure, trust in the mayor’s office or distrust of national parties. “Housing backlash” is broad enough to describe almost everything and too blunt to explain much unless it is broken down.

The first Seoul crosstab gave that broad category a false precision. It appeared to locate the housing revolt inside a visible demographic group: younger women. That made the race easier to narrate. It also made the electorate easier to flatten.

The revised table forced a different reading. If women in their 30s were no longer shown as an Oh-leaning group in the corrected estimate, then the story of Seoul could not rest on a simple claim that young women had abandoned the Democratic camp. The city still had to be explained, but the explanation had to carry more weight. Oh’s victory could involve conservative strength among young men, older voters, district-level margins, turnout composition, redevelopment expectations, incumbency, neighborhood inequality and the uneven relationship between national mood and municipal choice. Housing remained part of the story, but not as a single voice speaking through one demographic bloc.

This is where Seoul differs from a normal partisan swing. A swing suggests a broad movement from one camp to another. Seoul looked more fractured. Jung was competitive across the city and led the election-night projection, while Oh’s final victory depended on margins that cannot be reduced to one social group. That kind of result is rarely explained by a single bloc. It is more often produced by the interaction of margins: where one candidate wins narrowly, where the other wins heavily, which voters turn out early, which voters wait, and which local issues become decisive in particular neighborhoods.

The property question sits inside that interaction. In high-price districts, housing politics may tilt toward asset defense and redevelopment speed. In areas with younger renters or first-time buyer anxiety, it may produce frustration with both parties. In neighborhoods facing stalled renewal, voters may judge candidates less by ideology than by perceived capacity to move permits, infrastructure and local administration. The same word — housing — can pull voters in different directions at the same time.

The flawed crosstab mattered because it reduced those cross-pressures to a cleaner political morality tale. It suggested that a recognizable bloc had defected, and once that suggestion entered the post-election conversation, more complex explanations had to compete against a simpler one.

A stronger analysis of Seoul begins by refusing the easiest sentence. Young women did not become the whole election. Housing did not become the whole city. The corrected exit-poll table did not answer why Oh won, but it closed off one shortcut: the idea that Jung’s defeat could be safely explained by a single demographic break.

Busan split the ballot

Busan offered the clearest counterpoint to Seoul.

If Seoul showed how an incomplete table could make a complicated electorate appear simpler than it was, Busan showed the complexity in the actual returns. The city elected a Democratic mayor, but it did not hand the Democratic Party control of local power in the same way. Voters separated the office from the party, the citywide choice from neighborhood administration, and the desire for change from the habit of local trust.

Jun Jae-soo’s victory over Park Heong-joon was historic enough to invite an easy headline. A Democrat had taken back the Busan mayoralty from a conservative incumbent, giving the party its second elected mayor in the city after Oh Keo-don’s 2018 win. In a city long treated as part of the conservative foundation of southeastern Korea, the result initially looked like a clean break.

The lower-ticket results made that reading too simple.

Across Busan’s 16 districts and counties, Democratic candidates won seven local government chief races, while People Power Party candidates won nine. Democrats took Yeongdo, Nam, Buk, Saha, Gangseo and Sasang districts, along with Gijang County. The People Power Party held or won Jung, Seo, Dong, Busanjin, Dongnae, Geumjeong, Yeonje, Suyeong and Haeundae. The mayoralty had changed hands, but the city’s administrative map had not turned in one direction.

The split became clearer at the district level. Jun led Park in 11 of Busan’s 16 districts and counties, yet in several places where Jun beat Park in the mayoral vote, People Power Party candidates still won the district chief race. In other words, some voters who were willing to choose a Democratic mayor were not willing to transfer that choice to every local office beneath him.

That is the difference between a swing and a split.

A swing implies that voters moved as a bloc from one party to another. Busan’s result suggests something more selective. At the top of the ticket, Jun could stand for change, fatigue with the incumbent, regional development expectations, the appeal of a national governing-party connection, or a belief that Busan needed a new negotiating position with Seoul. At the district level, voters may have judged candidates through different criteria: local organization, administrative familiarity, personal networks, incumbent performance, neighborhood projects and the conservative machinery that remains deeply embedded in many parts of the city.

This does not make the mayoral result smaller. It makes it more interesting. Busan did not abandon its conservative identity, nor did it remain locked inside it. The city divided its judgment by office. Voters could want a new mayor and still prefer conservative district administration. They could support a Democratic figure for the city’s strategic direction while keeping a People Power Party candidate for the everyday machinery of local government. That is not contradiction. It is layered voting.

The same pattern appeared in the legislative layer. Even as the Democratic Party won the mayoralty and made gains in district chief contests, the People Power Party remained dominant in the Busan Metropolitan Council’s district seats. A voter who chose Jun for mayor could still choose a conservative council member or district chief. The old habit of straight-ticket voting had weakened, but it had not been replaced by a new straight-ticket vote in the other direction.

“Busan turned Democratic” is too blunt a sentence. A better reading is that Busan voters unbundled the ballot. They treated the mayoralty as one decision, district government as another and legislative representation as another still. Each office carried a different question. Who should speak for the city? Who should manage the district? Who can deliver neighborhood projects? Who has the organization to keep local administration working?

The geography of the result also matters. Democratic strength in parts of the Nakdong River belt and the old city did not automatically translate into a collapse of conservative local networks. The People Power Party’s continued hold in places such as Haeundae, Suyeong, Geumjeong and other established conservative districts showed that class, development patterns, property interests and neighborhood political memory still mattered. Busan was changing, but unevenly.

That unevenness is the point. The city’s vote was not an ideological conversion. It was a negotiation between change and continuity. Jun’s win showed that the Democratic Party could compete for Busan’s future-facing citywide agenda. The district and council results showed that local conservative infrastructure remained strong enough to survive a mayoral defeat.

Read beside Seoul, Busan helps define the larger election. Seoul warned against using one flawed demographic table to explain a complex defeat. Busan warned against using one mayoral victory to describe an entire city’s political movement. In both places, the easiest story was too clean.

The electorate was dividing its choices by timing, issue, geography and office.

The new Korean voter is harder to map

Seoul and Busan appeared to tell different stories. One kept a conservative mayor after an exit poll had suggested a Democratic win. The other elected a Democratic mayor in a city still anchored by conservative local power. Put together, they pointed to the same problem for political analysis: the electorate was harder to map than the first explanations allowed.

The old language of election swings is useful when voters move broadly from one camp to another. It becomes less useful when voters divide their choices across time, office and issue. In Seoul, the divide ran through early and Election Day voting, homeowners and renters, redevelopment expectations and affordability pressure, young men and young women, citywide margins and district-level concentrations. In Busan, the divide ran through the ballot itself: mayor, district chief and council member did not all ask the same political question.

Party identity has not disappeared. South Korean elections remain deeply partisan, and both major parties still depend on regional memory, national mood and organizational strength. The 2026 local elections showed how those forces now interact with more specific calculations. A voter may support one party’s mayoral candidate because the city needs change, another party’s district chief because local administration feels more reliable, and a different council candidate because neighborhood networks still matter. A voter may dislike housing prices, want redevelopment, fear rent, distrust national parties and still make a municipal choice that cannot be reduced to any one of those pressures.

The exit-poll error mattered beyond the four regions named in the correction because it exposed a weakness in the way elections are explained when a complex result must be turned quickly into a single story. The initial Seoul crosstab seemed to offer a clean demographic key to a confusing outcome. Busan’s returns showed the opposite problem: even when the result was official, the headline could still flatten the city’s split judgment into a simple partisan breakthrough.

The same temptation runs through both cases. A table becomes a coalition. A mayoral win becomes a regional conversion. A housing issue becomes a demographic revolt. A timing difference between early voting and Election Day becomes a technical detail. Each move makes the election easier to explain and less accurate to understand.

A better reading begins with division rather than movement. The June elections did not show one electorate marching left or right. They showed voters sorting decisions into layers. Time mattered because early voters and Election Day voters could not be assumed to have the same composition. Property mattered because housing interests in Seoul pulled in more than one direction. Generation mattered because young voters were not a single political bloc. Office mattered because Busan voters could choose a Democratic mayor while keeping conservative power in district halls and the metropolitan council.

For parties, the lesson is uncomfortable. Democrats cannot read Busan as a full regional realignment when much of the city’s local machinery stayed conservative. Conservatives cannot read Seoul as a simple youth or women’s backlash when the corrected estimates narrowed the basis for that claim. Both parties face voters who may be loyal in one race, conditional in another and strategic in a third.

For media, the lesson is sharper. Election-night data should be treated as provisional not only because projections may miss the winner, but because the categories used to explain the winner may be incomplete. Crosstabs are powerful because they fill the silence left by the secret ballot. That power requires visible limits. When the official count does not disclose how age, gender or early-voting groups voted, estimates should not be allowed to harden into social fact without careful qualification.

The June elections were not unreadable. They were simply less convenient than the first stories allowed.