近来，新闻中充斥着基于民调的种种预期，认为民主党将在下个月的中期选举中大获全胜。《今日美国》近日宣称：“预期的‘蓝色浪潮’可能比预想的还要猛烈，不仅可能让民主党掌控众议院，甚至还能拿下参议院。”而《赫芬顿邮报》的预测则更为夸张：“民主党似乎即将取得历史性的胜利。”
The news is awash these days with poll-inspired expectations about Democrats winning big in next month’s midterm elections. USA Today declared the other day that “the expected blue wave could be even bigger than expected, and could not only give Democrats control of the House, but also the Senate.”Even more expansively, Huffington Post proclaimed: “Historic victories are seemingly within grasp” for Democrats.
然而，鉴于民意调查本质上是一种难以衡量瞬息万变选民情绪的脆弱工具，我们有充分理由对这类基于民调的叙事保持谨慎，甚至警惕。数据记者内特·西尔弗曾指出：“对选举民调持适度——甚至可能不止适度——的怀疑态度，并非不合理。”当然，这并不意味着民调驱动的预测必然错误。它们或许准确无误，民主党也或许真能在下个月迎来一场“蓝色海啸”般的选举胜利。
And yet, solid reasons exist for treating such poll-based narratives cautiously, even warily, given that surveys are inherently fragile instruments for measuring an ephemeral electorate. As data journalist Nate Silver has noted, “It’s not unreasonable to approach election polls from a position of mild — maybe even more-than-mild — skepticism.”This is not to say poll-driven expectations are bound for error. They might be accurate; the Democrats might enjoy a “blue tsunami” of electoral success next month.
但确实存在令人信服的理由，让我们应当克制选举前的过度兴奋——这些理由根植于民调领域屡经考验的复杂历史。以下对近期选举民调缺陷的总结，可为我们今年对民调所带来的高期望保持审慎提供背景与依据。
But there are compelling reasons for tempering pre-election headiness — reasons that are rooted in polling’s checkered past. The following summary of recent shortcomings in election polling offers context and justification for tapping the brakes about poll-driven high hopes this year.
选举民调长期低估共和党支持率。在2024年，民意调查机构连续第三次低估了唐纳德·特朗普的民众支持度。他的得票数在三次总统选举中持续上升：从2016年的近6300万张，到2020年的约7400万张，再到2024年的超过7730万张。
Election polls have been underestimating Republicans Pollsters in 2024 underestimated Donald Trump’s popular support for the third successive presidential election. His vote totals increased in each of the three races, from almost 63 million in 2016 to about 74 million in 2020 to more than 77.3 million in 2024.
当然，特朗普本人并未出现在2026年的选票上，但他已象征性地鼓励中期选举的选民“闭上眼睛假装”他参选。
Of course, Trump is not on the ballot in 2026, but he has encouraged midterm voters to “close their eyes and pretend” that he is, symbolically.
其假设是，特朗普出现在选票上会吸引那些原本可能不会投票的共和党人参与选举。然而，低估共和党人投票率的问题并不仅限于特朗普的参选。
The hypothesis is that Trump’s presence on the ballot attracts participation by Republicans who otherwise might not vote. But the problem of underestimating Republican turnout goes beyond the Trump candidacies.
根据《华盛顿邮报》的民意调查主管斯科特·克莱门特发布的数据，在过去六次全国性选举中，有五次的民意调查结果都高估了民主党候选人的支持率。克莱门特指出，如果今年的中期选举民意调查普遍高估民主党候选人的支持率（平均高估3个百分点），那么民主党很可能会赢得众议院的控制权，但无法掌控参议院。
More broadly, polls collectively have overstated support for Democratic candidates in five of the past six national elections, according to data posted by Scott Clement, polling director for The Washington Post. If the midterm polls this year systematically overestimate support for Democratic candidates by an average of 3 percentage points, Clement wrote, Democrats likely will win the House but fail to gain control of the Senate.
即便是那些以高准确性著称的民意调查专家，也难免会出现重大失误。事实上，自1936年总统选举以来，这一现象就一直存在：当时被公认为民意调查权威的《Literary Digest》杂志预测共和党人阿尔夫·兰登会轻松击败富兰克林·D·罗斯福总统，但其预测结果严重偏离了实际情况（该杂志的民意调查实际上低估了罗斯福的支持率近20个百分点）。兰登最终以微弱差距落败，仅赢得了两个州的选票。
Even the best pollsters fail Pollsters with exceptional records for accuracy have still been known to err dramatically. Reputation brings no insulation from polling embarrassment. This reality has been apparent since the presidential election of 1936, when Literary Digest magazine — the acknowledged polling oracle of its day — predicted Republican Alf Landon would easily defeat President Franklin D. Roosevelt. The Digest’s projection was based on its huge mail-in poll, which understated Roosevelt’s popular support by nearly 20 percentage points. Landon lost in a landslide, carrying just two states.
更近期的例子是J·安·塞尔泽——她此前出色的民意调查工作为她赢得了“民意调查女王”和“政治领域最佳调查专家”的称号——在2024年的选举中也出现了重大失误。她在选举前的最后一次爱荷华州民意调查中声称民主党人卡玛拉·哈里斯已“明显取得领先优势”，在爱荷华州的支持率比特朗普高出3个百分点。这一结果既令人震惊又出乎意料。与此同时，埃默森学院进行的另一项民意调查却显示特朗普在爱荷华州的支持率领先10个百分点，但这一结果几乎被所有人忽视了。
More recently, J. Ann Selzer, whose previously stellar survey work won for her acclamation as the “queen of polling” and “best pollster in politics,” erred dramatically in 2024. Her final Iowa poll before the presidential election said Democrat Kamala Harris had “clearly leaped into a leading position” and was 3 points ahead of Trump in the Hawkeye State. The poll result was as stunning as it was unexpected. All but overlooked in the frenzy were results of a concurrent survey by Emerson College that showed Trump leading in Iowa by 10 percentage points.
最终，特朗普以13个百分点的优势赢得该州，这意味着塞尔泽的民调出现了16个百分点的偏差。不久之后，塞尔泽宣布计划退出选举民调领域。她说：“民调是一门估算的科学，而科学总会周期性地让科学家意识到自己的局限。”大型媒体的民调结果向来令人存疑。由主要新闻机构委托或开展的选举民调自然吸引了广泛关注，其结果也迅速被大量观众和读者所知晓。然而，尽管这些民调备受瞩目，但它们也可能存在诸多问题。
In the end, Trump carried the state by 13 points, which meant Selzer’s poll was off by 16 points. Soon afterward, Selzer made known her plans to quit election polling. “Polling is a science of estimation,” she said, “and science has a way of periodically humbling the scientist.”Major media polls have been iffy Election polls commissioned or conducted by major news organizations unsurprisingly attract wide attention. The results are promptly shared with large audiences of viewers and readers. But for all their prominence, media polls can be problematic.
2024年，在竞选活动后期，CNN、《纽约时报》/锡耶纳学院以及为PBS和NPR进行民调的玛里斯学院，在所有摇摆州均显示哈里斯领先，且领先幅度足以让她赢得选举人团。如果这些媒体民调与摇摆州的最终选举结果一致，哈里斯本应当选总统。
In 2024, swing-state polling late in the campaign by CNN, The New York Times/Siena College, and Marist College for PBS and NPR all had Harris leading in enough swing states to win the Electoral College. Had those media polls aligned with election outcomes in swing states, Harris would have been elected president.
然而，特朗普赢得了全部七个摇摆州，尽管其中一些州的优势仅很微弱。
Trump carried all seven swing states, if by narrow margins in some of them.
CNN和《纽约时报》/锡耶纳学院的民调显示，竞选结束时全国范围内的选情处于胶着状态。玛里斯学院的民调则估计哈里斯在全国领先4个百分点。而特朗普最终以1.5个百分点的优势赢得了全国普选。
CNN and the Times/Siena poll estimated the race nationally was tied at campaign’s end. Marist estimated that Harris led nationally by 4 points. Trump won the national popular vote by 1.5 points.
未能成形的“红色浪潮”四年前期中选举时，人们曾预计共和党将在国会掀起一场“红色浪潮”，即大幅扩大其席位优势。2022年秋季，民主党面临的政治局势看似一片黯淡：当时通货膨胀率高达7.7%，非法跨境移民激增，拜登总统的支持率也跌至40%。
The red wave that wasn’t The midterm elections four years ago were expected to produce a “red wave” of Republican gains in Congress. The political landscape looked bleak for Democrats in the fall of 2022. Inflation was raging then at 7.7 percent, unlawful cross-border migration was surging, and President Biden’s approval rating was down to 40 percent.
然而，“红色浪潮”最终并未出现。民主党在参议院净增一个席位，并继续保持对参议院的控制。共和党在众议院仅赢得九个席位，勉强获得微弱多数。
The red wave never materialized. Democrats scored a net gain of one Senate seat and kept control of the upper chamber. Republicans picked up just nine seats in the House, enough to give them a wobbly majority.
民主党超出预期的表现，促使拜登做出了2024年寻求连任的糟糕决定。拜登当年赢得了所有民主党总统初选，但在2024年7月与特朗普的辩论惨败后，退出了竞选。
The better-than-expected outcome for Democrats contributed to Biden’s ill-fated decision to seek reelection in 2024. Biden won all the Democratic presidential primaries that year but quit the campaign in July 2024 following a disastrous debate against Trump.
对于民意调查机构来说，它们在2022年的表现可谓“喜忧参半”——既有一些成功的案例，也有一些令人失望的结果。
For pollsters, their performance in 2022 was more spotty than spectacular.
在亚利桑那州、科罗拉多州、佛罗里达州、新罕布什尔州、宾夕法尼亚州和华盛顿州等关键州的参议院选举中，民意调查的预测结果与实际胜出差距均达到了4个百分点或更多。在新罕布什尔州，民主党候选人玛吉·哈桑的实际得票率比RealClearPolitics的民意调查平均值高出近8个百分点；在华盛顿州，民主党参议员帕蒂·默里的实际得票率比民意调查预测的高出11个百分点。
Polls overall missed the margins of victory by 4 percentage points or more in key Senate races in Arizona, Colorado, Florida, New Hampshire, Pennsylvania and Washington. In New Hampshire, Democrat Maggie Hassan exceeded the RealClearPolitics polling average by nearly 8 points. In Washington state, Democratic Sen. Patty Murray outperformed the polls by 11 points.
当然，这并不意味着民意调查的失败历史就一定预示着未来的选举结果也会同样不准确。不过，我们确实有理由对选举结果保持警惕，因为民意调查可能无法准确预测选举的最终结果。
None of this, of course, is to say that election polling’s flawed past is necessarily prologue. But it is reasonable to be braced for surprises, as polls might not offer an accurate indication about elections outcomes.
正如已故的民意调查领域创新者沃伦·米托夫斯基（Warren Mitofsky）曾经指出的那样：“在民意调查领域，我们完全有理由保持谦逊的态度；每次过于自信时，都可能带来失败。”
As the late polling innovator Warren Mitofsky once said about placing excessive confidence in survey data, “There’s a lot of room for humility in polling. Every time you get cocky, you lose.”
W·约瑟夫·坎贝尔（W. Joseph Campbell）是位于华盛顿特区的美国大学（American University）的荣誉教授。他独立撰写的七本书中，包括《Lost in a Gallup: Polling Failure in U.S. Presidential Elections》（《盖洛普民意调查的失败：美国总统选举中的失误》）。
W. Joseph Campbell is a professor emeritus at American University in Washington, D.C. His seven solo-authored books include “Lost in a Gallup: Polling Failure in U.S. Presidential Elections.”