民意调查再次成为人们关注的焦点，一些人甚至急于宣称这些调查已经“失效”或“不再可靠”。但实际上，并非民意调查本身的质量下降了，而是多种因素共同作用，给民意调查本身以及试图理解调查结果的公众带来了前所未有的挑战。
Polls are again under scrutiny, with some rushing in to proclaim their demise. But it isn’t that polls have gotten worse. Rather, a mix of factors is creating unprecedented challenges for polling and for the public trying to understand it. Why?
首先，那些长期存在的问题一直被人们误解。民意调查确实非常适合用来衡量人们对各种问题的看法和态度，也能较为准确地反映受访者对候选人的偏好。然而，预测实际投票结果始终是一项具有挑战性的任务——因为人们往往难以准确预测自己的行为，而且他们通常会表达自己的愿望而非真实想法。
First, there are the perennial challenges that are consistently misunderstood. Surveys are well suited to measure beliefs and attitudes about issues, and they can capture respondents’ candidate preferences pretty accurately. But predicting who will actually turn out to vote will always be an imperfect effort, largely because people are bad at predicting their own behavior and tend to report aspirations rather than realities.
这种挑战在中期选举和初选中尤为明显：这些选举的投票率通常较低，地区间的情况差异较大，且可获取的信息也较为有限。然而，许多公众却期望民意调查能够提供高度精确的预测结果，而这显然是不现实的。因此，人们一次次感到失望。
This challenge is magnified in midterms and primaries, contests that often have lower turnout, localized dynamics and low information. And yet, many consumers expect a level of precision that polls can’t deliver. And the cycle of disappointment repeats.
其次，过去十年间一些新的问题进一步加剧了这一困境。随着低成本、自愿参与的民意调查方式的出现，民意调查领域出现了大量新的参与者。与此同时，高质量民意调查的成本急剧上升，导致这类调查的数量相对减少。因此，人们越来越难以判断哪些民意调查机构是可靠的；一些地区和群体因此无法获得可靠的数据。
Second, there are slow-growing issues that have compounded the problem over the past decade. The polling field has exploded with new entrants as low-cost opt-in polling has developed. At the same time, the rapid increase in the cost of high-quality polling has led to fewer of these polls, relative to cheaper opt-in polls. It’s harder to know which pollsters to trust, and some states and localities are left without reliable data for consumers.
虽然有些提供自愿参与式民意调查的服务机构有着不错的记录（可能是由于他们的建模技术有所改进，或者他们的预测基于合理的分析），但我们并不清楚这些机构的调查方法究竟如何运作（因为他们的调查方法往往不够透明）。不过，人们对这类民意调查的信任度正在逐渐下降。多项最新研究表明，有大量虚假受访者混入了这些调查中——这些虚假受访者可能是由人工智能驱动的机器人，也可能是试图牟利的人。
Some opt-in pollsters have decent track records — their modeling techniques may have improved, or they make educated guesses. We don’t know because their methodology is often opaque. But trust in opt-in polling is increasingly in question. Numerous recent studies have documented “bogus respondents” slipping into these surveys. Fake respondents — whether artificial intelligence-enabled bot farms or people trying to make money — are more sophisticated and harder to eliminate.
这种情况尤其影响了对西班牙裔成年人和年轻人的投票行为的预测，而这两个群体的投票意愿和投票率在11月的选举中可能起到决定性作用。
It particularly affects estimates of Hispanic adults and young people — groups whose preferences and turnout may be decisive in November.
与此同时，随着投票规则（何时、何地以及如何投票）随时间和司法管辖区的变化而变化，以及竞选活动的微定位能够以民意调查往往无法捕捉的方式影响选民群体，预测谁会出现投票变得越来越困难。
Meanwhile, predicting who will show up to vote keeps getting harder as voting rules (when, where and how to vote) change over time and across jurisdictions, and as microtargeting by campaigns can move segments of the electorate in ways broad polls often can’t.
如果这些情况让你有所犹豫，我们还没有谈到今年出现的三个新不确定因素。
If any of this gives you pause, we still haven’t touched on three new wildcards this year.
首先，我们正在看到越来越多的完全“合成调查”——即声称测量公众意见但实际上并未调查真实人类的民调。相反，它们使用人工智能来模拟人们可能的回答。我们对合成调查的研究发现，它们不能作为公众意见的可靠替代品。与此同时，我们还看到一些完全虚构且未披露来源的民调正在被传播。
First, we’re seeing more completely “synthetic surveys,” — polls that purport to measure public opinion but don’t actually survey humans. Instead, they use AI to model what people would say. Our research on synthetic surveys finds they are not reliable substitutes for public opinion. Separately, we have also seen polls being shared that are entirely fabricated without disclosure.
其次，对投票机制的不确定性——例如邮寄选票、某些州的选举法变动，甚至对选民恐吓的担忧——可能以难以建模的方式影响投票率。广泛进行的中期重新划分选区打乱了选民对自己选票上谁在参选的认知。而且，2026年的选举选票上的在职现任者人数自1992年以来的任何一次大选中都是最少的。
Second, uncertainty about voting mechanics, such as mail-in ballots, changing election laws in some states and even concerns about voter intimidation could influence turnout in hard-to-model ways. Widespread mid-cycle redistricting has scrambled voters’ awareness of who’s on their ballots. And the 2026 race has fewer incumbents on the ballot than any general election since 1992.
第三，人工智能和数据中心的问题已经进入公众讨论，但尚未完全按党派两极分化。这使得人工智能对选举的影响更依赖于当地条件，例如附近是否有规划中的或现有的数据中心，以及候选人对此议题的表态。
Third, the issues of AI and data centers have swelled into the public dialogue but haven’t yet become completely polarized by party. This makes the impact of AI on elections more dependent on local conditions, such as planned or existing data centers nearby, and what candidates are saying about the issue.
简而言之，在投票率较低的选举中，预测谁会投票一直很困难。随着时间的推移，这种挑战变得更大，而2026年充斥着各种未知因素，这些因素相互叠加，加大了预测难度——就像一场完美风暴，易于引发混乱。其中任何一个因素都可能导致即使是最高质量的民调也无法预测的惊喜。
In short, predicting who will vote has always been hard in lower turnout elections. It’s become more challenging over time, and 2026 has a swirl of unknowns compounding the challenge — a perfect storm ripe for confusion. Any one of these factors could yield surprises that are impossible to predict, even for the highest quality polls.
那么，在这一轮选举中，面对民调的洪流，我们该如何理解它们呢？首先，在考虑选举结果时，不要过度依赖单一民调，并警惕异常值。
So, what can you do to make sense of the deluge of polls in this cycle? To start, don’t put too much stock in a single poll when thinking about election outcomes, and be wary of outliers.
对您所考虑的民意调查提出明智的问题：谁进行了这项调查？该机构是否有良好的记录和声誉？他们是否有既得利益？您能否确定该调查是在对真实的人进行抽样？如果是，抽样对象是谁？该机构是否愿意公开其问题和方法？透明度不等同于质量，但缺乏透明度绝对是一个危险信号。
Ask smart questions of the polls you do consider: Who conducted the poll? Does the organization have a good track record and reputation? Do they have a dog in the fight? Can you be sure the poll is surveying real humans? If so, which ones? Does the organization readily share information about its questions and methodology? Transparency is not the same as quality, but a lack of transparency is certainly a red flag.
将民意调查用于其最擅长的方面：帮助我们了解国家的情绪、公众关注的议题以及什么在分裂和团结人们 — — 简而言之，选举可能意味着什么。不要指望民意调查做它们无法做到的事情，尤其是今年。
Use polls for what they are best at: helping us understand the mood of the country, the issues on the public’s mind and what’s dividing and uniting people — in short, what the election is likely going to mean. Don’t expect polls to do what they cannot, especially this year.
最后，不要因小失大：不要让一次糟糕的民意调查彻底动摇您对民意调查的信任。严谨的民意调查仍然是了解公众所思所想的最佳方式，即使它们并不总是能准确预测初选和中期选举。
Lastly, don’t throw the baby out with the bathwater: Don’t let a bad poll destroy your trust in polling altogether. Rigorous polls are still the best way to understand what the public thinks and believes even if they don’t always accurately predict primaries and midterms.
如果民意调查中没有公众的声音，我们只能剩下政客们自称代表我们说话。迈克尔·迪莫克是皮尤研究中心总裁。
Without the public voice in polls, we’re left only with politicians claiming to speak for us. Michael Dimock is president of Pew Research Center.