中期选举即将到来，民调机构正在磨亮他们的数字铅笔，以预测谁将胜出。众议院和参议院的控制权对两党都处于争夺之中。总统的频繁言论进一步使局势扑朔迷离，因为他所说的往往与人们的实际体验不一致。这为已经混乱的进程增添了额外的不确定性。
The midterms are fast approaching, and pollsters are sharpening their digital pencils to forecast who will come out on top. Control of both the House and the Senate is in play for both parties. Frequent comments by the president further cloud the situation, as what he says is often not consistent with what people are experiencing. This adds additional uncertainty to an already chaotic process.
鉴于所有这些模糊因素，民调机构几乎肯定会再次出错，但原因不同。
Given all this ambiguity, pollsters will almost certainly get it wrong once again, but for a different reason.
民调的核心是通过抽取少量人群来推断大多数人可能的行为。有四个问题增加了其不可预测性，每个问题都更难以克服。
Polling is about sampling a small group of people to determine how a large group of people is likely to act. There are four issues that add to its unpredictability, each more challenging to overcome.
首先，民调机构会询问被调查者是否有投票资格。接下来，他们会询问该人是否可能投票。第三，他们会询问该人可能支持哪位候选人，或他们认为谁会赢得选举。最后一个问题是民调机构几乎无法控制的：投票率差异和欠投票。这是指人们实际投入选票时的行为，包括他们可能有意选择不参加投票或在特定选举中不投票。
First, pollsters ask if a person polled ised to vote. Next, they ask if the person is likely to vote. Third, they ask which candidates a person is likely to vote for or who they think will win the elections. And the final issue is one over which pollsters have little control: turnout discrepancy and undervoting. This is what a person does when they actually cast their vote and place it in the ballot box, including how they may intentionally choose to not show up or not vote in particular races.
民调机构也发现，直接提问可能无法提供最有用的信息。这就是为什么他们通常会提出一系列问题，这些问题共同指示一个人可能的行为。
Pollsters have also learned that asking questions directly may not provide the most useful information. That is why they typically ask a series of questions that collectively indicate what a person is likely to do.
鉴于所有这些因素，民调机构只能通过扩大样本规模来减轻抽样误差，尽管这无法纠正隐藏的偏差或样本目标不准确的问题。关键挑战在于将被调查的人群样本与实际前来投票的人群匹配起来。
Given all such factors, pollsters are constrained to mitigate sampling error by taking larger polling sample sizes, though this cannot correct hidden biases or inaccurate sample targets. The key challenge is matching the sample of people polled with who shows up to vote.
目前，民主党似乎已稳操胜券，将赢得众议院，并且重新夺回参议院的可能性也在增加。鉴于共和党在众议院仅以微弱多数优势占据上风，即便在2025年进行了大部分旨在保护共和党现任议员的重新划区，这一结果也不足为奇。而几个月前，重夺参议院的想法还被认为是不可想象的。
At present, Democrats appear set to win the House and have a growing chance to also take back the Senate. The former is not surprising, given the slim majority held by Republicans in the House, even with all the additional redistricting that occurred in 2025 to mostly protect Republican incumbents. The latter was unthinkable just a few months ago.
一些共和党评论员仍抱希望，认为共和党能够保住两院。然而，面对一位历来不受欢迎的总统敦促选民像他就在选票上那样行事，共和党大获全胜的希望正在逐渐消退。
Some Republican pundits continue to hold hope for Republicans to hold onto both chambers. Yet with a historically unpopular president asking voters to act like he is on the ballot, hopes for a Republican sweep are fading.
现任总统所属的政党通常在中期选举中会失去众议院席位。在过去半个世纪里，只有两位现任总统所在的政党曾赢得众议院席位（比尔·克林顿在1998年和乔治·W·布什在2002年）。2018年，在特朗普第一个任期期间，民主党赢得了41个众议院席位。由此导致的共和党不利局面很可能今年仍将持续。
The party of the sitting president typically loses House seats in the midterm elections. Over the past half century, only two sitting presidents have seen their party gain House seats (Bill Clinton in 1998 and George W. Bush in 2002). In 2018, during Trump’s first term, Democrats gained 41 House seats. The resulting Republican headwinds are likely to continue this year.
尽管民调机构倾向于预测民主党胜利，但影响民调进行方式的因素表明，民主党今年具有一种固有优势，而民调可能无法完全捕捉到这一优势。
As much as pollsters are leaning toward Democratic victories, the factors that play into how polls are conducted suggest that the Democrats have an inherent advantage this year that polls may not fully capture.
这是因为人们在手中握有选票时的行为，可能不是任何人（包括他们自己）在选举日之前就能确定无疑的。将决定民主党得失的群体不仅不是民主党人（他们肯定会投票给民主党候选人），也不仅仅是独立人士（他们可能比以前更倾向于支持民主党而非共和党）。对共和党候选人而言，最大的风险在于共和党选民——尤其是得克萨斯州等地的拉美裔等非MAGA选民——究竟会如何投票，或者是否会投票。
This is because what people do when they have their ballot in hand may not be something that anyone (including themselves) can ascertain with certainty prior to Election Day. The one group that will determine Democratic gains will not beed Democrats (they will certainly vote for Democratic candidates) and independents (they may be inclined to vote for more Democrats than Republicans compared to previous elections). The greatest risk for Republican candidates is how unhappy Republican voters — particularly non-MAGA voters like Latinos in states like Texas — will vote or not vote.
相当一部分这类选民仍会继续支持共和党候选人。尽管其中许多人深受汽油价格飞涨、生活成本难以负担（即所谓的“可负担性问题”）以及美国移民与海关执法局不当的移民管控措施的影响。而另一部分非“让美国再次伟大”阵营的共和党支持者则可能选择不参加投票，或者在众议院及参议院选举的选票上留白。这种弃票行为实际上等同于把胜算拱手让给民主党，使得不投票也成了对民主党候选人的助力，尤其是在势均力敌的选区。
A large faction of such voters will continue to support Republican candidates, even though many of them have been personally affected by soaring gasoline prices and their inability to pay their bills (the so-called affordability issue), as well as misguided immigration control by ICE. The group that may stay away are non-MAGA Republicans, who may opt to just not show up to vote, or leave House and Senate races blank. By not voting, they are effectively giving the election to Democrats, making a non-vote a de facto boost for Democratic candidates, particularly in close races.
通常情况下，中期选举的投票率远低于总统选举。今年很可能也是如此；不过现任总统支持率低迷，这或许能激发民众的参与热情，促使他们走上投票站以表达自身诉求。
Midterms typically attract significantly fewer voters than presidential elections. This year will likely be the same, though an unpopular president may provide ample motivation for people to show who up and express their concern at the ballot box.
综合上述种种因素来看，即便存在选区划分不公等结构性不利条件，民主党仍有望在各选区全面获胜。那些依据市场投注情况来预测选举结果的预测平台也得出了相同的结论。
All these factors point to a clear Democratic victory across the board, even with the structural headwinds of gerrymandering. Prediction markets that report odds based on how people bet in a marketplace are coming to the same conclusion.
民意调查机构拥有众多分析手段与工具，但他们无法准确预判选民的言行差异——即人们口头上怎么说，选举当天又会如何行动。正是这种不确定性让今年的中期选举预测变得极为棘手；最终，那些尚未投出的选票很可能决定2027年和2028年国会的权力归属。
Pollsters have a large number of levers and tools available to them. What they cannot foresee is the discrepancy between what people say and how they will act on Election Day. It is such uncertainty that will make their job exceedingly difficult in this year’s midterms, and why uncast ballots may ultimately decide who wields power in Congress in 2027 and 2028.
谢尔登·H·雅各布森博士是伊利诺伊大学厄巴纳-香槟分校的计算机科学教授。他运用自己在数据驱动型风险决策方面的专业知识来评估并推动公共政策的制定。
Sheldon H. Jacobson, Ph.D., is a professor of Computer Science at the University of Illinois Urbana-Champaign. He applies his expertise in data-driven risk-based decision-making to evaluate and inform public policy.