開票觀察 Watch PartyElection-night needles · 開票即時預測

Who's going to win, while the votes are still being counted

票還在開,誰會贏?

票猶咧開,啥人會贏?

On 28 November 2026 Taiwan elects its 22 mayors and county magistrates, and every city and county council. This site follows the count as it comes in: for each race, a needle with the chance each contender wins, the projected margin and turnout, and who would hold each council seat if counting stopped now. Until then it replays real past nights — 2018 and 2022 local, 2020 and 2024 presidential — minute by minute, exactly as the model would have seen them.

2026 年 11 月 28 日,台灣將選出 22 位縣市長,以及各縣市議員。本站在開票過程中即時追蹤:每場縣市長選舉有一支指針,顯示各候選人的勝選機率、預估差距與投票率;議員則顯示若此刻停止開票、各席會由誰拿下。在那之前,本站以真實的過去開票夜(2018、2022 地方選舉,2020、2024 總統選舉)逐分鐘重播,完全照模型當時能看到的資料運算。

2026 年 11 月 28 日,台灣欲選出 22 位縣市長,佮各縣市的議員。本站佇開票的時陣即時追蹤:逐場縣市長選舉有一支指針,顯示逐位候選人的當選機率、預估差距佮投票率;議員就顯示若這馬停止開票,逐席會予啥人提去。佇彼進前,本站用真正的過去開票暗(2018、2022 地方選舉,2020、2024 總統選舉)一分鐘一分鐘重播,完全照模型當時看會著的資料來算。

How it works · 怎麼算

The Central Election Commission (CEC) reports results polling station by polling station. Each station covers a few villages (村里), and every village has voted before. The model leans on that.

  1. Two sides. Each mayoral race is scored as one number: the KMT contender's share of the top two (or, where the KMT fields no one, the share of whoever faces the DPP). Other candidates are set aside; races where a third candidate is close get a three-way treatment (below).
  2. A baseline for every village. How the same village split last time: for mayors, the previous mayoral and presidential races averaged; for president, the previous presidential race; for legislative districts, the previous presidential and legislative races averaged. Stations are matched by village, so renumbered, split or merged stations don't matter (98–100% of electors match).
  3. The swing so far. At the stations already counted, how far tonight's share has moved from the baseline, city-wide and district by district. That swing is applied to every station still out.
  4. What's left. Each station still out has a known number of electors; the model estimates how many will have voted from the turnout at counted stations compared with their own past turnout.
  5. Uncertainty. The projected margin has a range that allows for the whole city swinging differently, for districts differing, for station-to-station noise, and for turnout running above or below its projection. The win chance comes from that range. It starts wide and narrows as stations report.
  6. Decided. When a lead is bigger than every elector at the stations still out, the race is decided, whatever the turnout.

Three-way races

When a third candidate is close (in 2024, 柯文哲 in eight counties), the two candidates with a past election to compare with are split by the swing model, and the third candidate's share is projected from the counted stations in each district, adjusted for station size. The page shows a two-way needle between the two likeliest winners and the third candidate's chance beside it, from the 5th station counted.

Turnout, council seats and the count watch

Turnout is projected the same way as the vote: counted stations against their own past turnout. Council seats aren't projected: the page shows who would hold each seat if counting stopped now, with the women's quota, and marks a seat decided only when no split of the votes still out could change it. The count watch (2024) sets the TV channels' on-screen counts next to the CEC count and the model's projected final, and flags any count that goes above what can plausibly be reached.

中央選舉委員會(中選會)以投開票所為單位公布結果。每個投開票所涵蓋幾個村里,而每個村里過去都投過票。模型就是利用這一點。

  1. 兩方。每場縣市長選舉化為一個數字:國民黨候選人在前兩名中的得票比例(國民黨未提名時,則為與民進黨對決者的比例)。其他候選人先不計;第三人票數接近的選舉另以三強方式處理(見下)。
  2. 每個村里的基準。同一村里上次怎麼投:縣市長選舉取上一屆縣市長與總統選舉的平均;總統選舉取上一屆總統選舉;立委選區取上一屆總統與立委選舉的平均。投開票所以村里對應,所以投開票所重新編號、拆分或合併都不影響(98–100% 的選舉人可直接對應)。
  3. 目前的擺盪。在已開出的投開票所,比較今晚的比例與基準差了多少,分全市與各區計算,再把這個擺盪套用到每個尚未開出的投開票所。
  4. 剩下多少。每個尚未開出的投開票所,選舉人數是已知的;投票人數則依已開投開票所與其過去投票率的比較來估計。
  5. 不確定性。預估差距有一個區間,涵蓋全市整體擺盪不同、各區不同、投開票所之間的雜訊,以及投票率高於或低於推估。勝率就從這個區間算出:一開始很寬,隨投開票所回報而收斂。
  6. 確定。當領先票數大於所有尚未開票投開票所的選舉人數,不論投票率如何,勝負已定。

三強鼎立

第三位候選人票數接近時(2024 年柯文哲在八個縣市),有過去選舉可比較的兩人依擺盪模型拆分,第三人的得票率則依各區已開投開票所推估,並依投開票所規模調整。頁面顯示最可能勝選兩人之間的指針,第三人的勝率另外標示,從第 5 個投開票所開出後開始。

投票率、議員席次與開票數字監看

投票率的推估方式與得票相同:已開投開票所與其自身過去投票率比較。議員席次不做預測:頁面顯示若此刻停止開票、各席會由誰拿下(計入婦女保障名額),只有在剩餘票數無論如何分配都無法改變結果時才標示為確定。開票數字監看(2024)把各電視台畫面上的票數與中選會票數、模型預估的最終票數並列,若有票數超出合理可能達到的範圍就會標示出來。

中央選舉委員會(中選會)用投開票所做單位公佈結果。逐个投開票所包含幾若个村里,逐个村里過去攏投過票。模型就是靠這點。

  1. 兩爿。逐場縣市長選舉化做一个數字:國民黨候選人佇頭前兩名內底的得票比例(國民黨無提名的時,就是佮民進黨對決彼个人的比例)。其他候選人先無算;第三人票數相倚的選舉,另外用三強的方式處理(看下跤)。
  2. 逐个村里的基準。仝一个村里頂擺按怎投:縣市長選舉用頂一屆縣市長佮總統選舉的平均;總統選舉用頂一屆總統選舉;立委選區用頂一屆總統佮立委選舉的平均。投開票所是用村里來對,所以投開票所重新編號、拆開抑是合併攏無影響(98–100% 的選舉人會使直接對著)。
  3. 這馬的擺盪。佇已經開出來的投開票所,比今暗的比例佮基準差偌濟,分全市佮各區來算,才閣共這个擺盪套用到逐个猶未開出來的投開票所。
  4. 賰偌濟。逐个猶未開出來的投開票所,選舉人數是知影的;會有偌濟人投票,就照已開的投開票所佮伊過去的投票率比較來估。
  5. 無確定性。預估差距有一个區間,包含全市規个擺盪無仝、各區無仝、投開票所之間的雜訊,佮投票率比推估較懸抑是較低。贏面就是對這个區間算出來的:起頭真闊,投開票所愈回報就愈收縮。
  6. 確定。贏的票數比所有猶未開票的投開票所的選舉人閣較濟的時,毋管投票率按怎,輸贏已經定矣。

三強鼎立

第三位候選人票數相倚的時(2024 年柯文哲佇八个縣市),有過去選舉通比的兩人用擺盪模型分,第三人的得票率就照各區已開的投開票所推算,閣照投開票所的大細調整。頁面顯示上有可能當選的兩人之間的指針,第三人的贏面另外標,對第 5 个投開票所開出來才開始。

投票率、議員席次、立委佮開票數字監看

投票率的推算方式佮得票仝款:已開的投開票所佮伊家己過去的投票率比。議員席次無做預測:頁面顯示若這馬停止開票,逐席會予啥人提去(有算婦女保障名額),干焦佇賰的票按怎分攏改變袂著結果的時,才標做確定。立委(2020、2024):73 个區域選區逐个有指針,34 席不分區用上有可能的席次分配佮機率呈現。開票數字監看(2024)共各電視台畫面頂懸的票數,佮中選會的票數、模型預估上尾的票數排做伙,若有票數超過合理會當到的範圍,就會標出來。

How accurate was it? · 準確度

Each past night was replayed with the station report times the CEC logged, and the model was run at every minute using only what had been counted by then. Below, each race is scored at the moment it reached a share of its stations, against its own final count.

每個過去的開票夜都依中選會登錄的投開票所回報時間重播,模型在每一分鐘只用當時已開出的票運算。以下以每場選舉開出一定比例投開票所的時刻,與該場最終結果比較。

逐个過去的開票暗攏照中選會登錄的投開票所回報時間重播,模型逐分鐘干焦用彼陣已經開出來的票來算。下跤用逐場選舉開出一定比例投開票所的時刻,佮彼場上尾的結果比。

Night選舉選舉

What it can't do · 限制

Races to watch on 28 November · 11/28 焦點選戰

Why these races matter, from news coverage and analysts' previews as of 10 November. This is not a forecast: on the night, the count decides. It is the first nationwide vote since the 2025 recall votes, and the midpoint of 賴清德's presidency.

這些選戰為何重要,整理自截至 11 月 10 日的新聞報導與分析者的選前評估。這不是預測:當晚以開票為準。這是 2025 年大罷免投票後第一次全國性投票,也是賴清德總統任期的期中考。

這寡選戰是按怎重要,整理自到 11 月 10 日為止的新聞報導佮分析者的選前評估。這毋是預測:當暗以開票為準。這是 2025 年大罷免投票了後頭一擺全國性的投票,嘛是賴清德總統任期的期中考。

Sources:

資料來源:

資料來源: