Skip to content

保障措施

什么是选举审计

中级阅读时间 8 分钟机器翻译

本页为机器翻译,正在等待人工审校。英文版本为正式记录版本。 阅读英文正式记录版本

An election audit checks whether the reported result is right by comparing it against the paper record. The strongest kind — a risk-limiting audit — hand-counts a random sample of paper ballots, often only a few hundred, and expands only if the sample casts doubt on the reported winner, giving high confidence at low cost.

本文内容

审计不是重新计票

这两个词常被交替使用,但它们是用于不同任务的不同工具。重新计票会重新统计每一张选票——即完整重做,通常由极小的差距或正式挑战触发。审计则是较轻量、常规的核查:它在认证之前,查验纸质记录的一部分,以确认已报告结果正确,无论是否有人对此提出争议。

可以把它理解为体检与手术之间的区别。审计是例行每次都会进行的检查;重新计票则是只有在检查发现问题时才升级采取的干预。设计良好的审计,正是决定是否需要重新计票的机制——而不是等待政治争议来要求重新计票。

风险限制审计

最强有力的形式是**风险限制审计(RLA),其核心理念确实巧妙:你必须人工清点多少张选票,取决于差距**,而不是选民规模。

为什么?因为审计检验的是一个结果,而不是每一个数字。如果某位候选人以十个百分点获胜,只有极其严重、系统性的错误才可能改变真实胜者——而如此大的错误,即使在一个很小的随机样本中也几乎肯定会显现出来。如果竞选是以寥寥数票决定胜负,审计就会自动抽取更多选票。其结果是显著的效率:一场拥有数百万张选票的选举,往往只需人工清点几百张纸质选票即可得到确认。

审计能够证明什么

RLA 只有两种可能的结局,而且两者都很有用:

  • 确认。 样本与已报告胜者在事先约定的置信水平上相符,因此审计停止,结果得到确认。
  • 升级。 样本相符,因此审计扩大——并且,如果疑点持续存在,则一直升级到全面人工重新计票。

请注意,审计并不声称什么:它并不证明每一项单独计数都毫无瑕疵。它证明的是,在既定的确定性水平下,结果是正确的。这既诚实,也足够——需要被捍卫的是谁赢了,以及赢了多少。

审计与其回应的主张

审计的现实价值,在它们面对舞弊指控时最为明显。2020 年后,曾有说法称 Michigan 的 Antrim County 的投票机调换了选票。官员随后对纸质选票进行了人工清点。人工审计与机器结果一致,而最初的报告错误——很快被发现并更正——被证明是人为失误,而非舞弊。

这就是应当内化的模式。审计把无法证伪的“机器被黑了”转化为一个具体、可核查的问题:纸质记录是否一致? 这相当于在如何核查一项主张中的纪律要求所对应的保障措施——用测试取代感觉。而且,单纯的统计异常只能提出问题,审计则能够真正回答它。

审计需要纸质记录

没有可供核对的实体记录,这一切都无法成立。审计将电子结果与独立的纸质记录进行比较;如果没有纸质记录,就无从审计。这就是为什么这两项保障实际上是一个理念的两个部分:纸质记录创造证据,审计读取证据。二者结合,才能让选举有底气地说:我们核查过了——而且你也可以这样核查。

How a risk-limiting audit works, step by step

  1. 1

    Fix the contest and the reported outcome

    Start from the machine-reported result — who won, and by what margin. The audit's job is to test that specific outcome against the paper, not to re-litigate every number.

  2. 2

    Set the risk limit

    Choose in advance the maximum chance you are willing to accept of confirming a wrong outcome — say 5%. This risk limit is the audit's guarantee, agreed before any ballots are drawn.

  3. 3

    Draw a random sample

    Using a public, verifiable random process, select ballots to hand-count. A wide margin needs only a small sample; a razor-thin margin needs a large one — the sample scales with how close the race was, not how many voters there are.

  4. 4

    Hand-count the sample and compare

    People — not machines — read the selected paper ballots and compare them against what the machines recorded. The question is whether the paper is consistent with the reported winner.

  5. 5

    Stop or escalate

    If the sample supports the reported outcome to the required confidence, the audit stops: the result is confirmed. If it does not, the audit expands — drawing more ballots and, if needed, escalating to a full hand recount.

常见问题

What is the difference between an audit and a recount?

A recount re-tallies every ballot, usually after a very close margin or a formal request. An audit checks a sample of the paper against the reported totals to confirm the outcome is right. An audit is the routine confidence check; a recount is the full re-count it can trigger when the check fails.

How can counting a few hundred ballots verify millions of votes?

Because a risk-limiting audit tests the outcome, not every number. If a candidate won by a wide margin, only a large, consistent error could have changed the winner — and even a small random sample would very likely reveal it. The closer the race, the more ballots the audit draws.

What does an audit actually prove?

It gives strong statistical confidence that the reported winner is correct — or clear evidence that a full recount is needed. It does not prove every individual number is perfect; it proves the outcome is right to a pre-agreed level of certainty, which is what matters for trust in the result.

Do audits need paper ballots?

Yes. An audit compares the electronic result against an independent physical record, so it is only possible where a voter-verified paper trail exists. Without paper there is nothing to audit against — which is why paper and audits are two halves of the same safeguard.

来源

  1. 1.Securing the Vote: Protecting American Democracy (2018)U.S. National Academies of Sciences, Engineering, and Medicine (2018)
  2. 2.Joint Statement: the 2020 election was the most secure in American historyCISA (U.S. Cybersecurity and Infrastructure Security Agency)
撰写者 VoteRights Editorial审阅者 VoteRights standards desk最后审阅

继续学习

数据库中的相关案例

分享链接会打开该网络自身的撰写器。此处不嵌入任何内容,因此不会有第三方得知您阅读了哪些页面。