An onslaught of midterm polling is coming. Here’s how to sort through it

Kyle Grillot/Bloomberg via Getty Images

These days, it’s easier than ever to conduct some form of polling and as difficult as it’s ever been to get it right.

Completely faked numbers like a phony “poll” of the Los Angeles mayoral race that made waves last month are thankfully rare. That said, even real polls can have serious flaws in the ways they’re conducted or an agenda behind their release. And the most rigorous polling doesn’t come with any guarantees.

But if read carefully, high-quality polls can still tell you a lot about how this year’s midterms are shaping up – and keeping a few simple questions in mind can help you sort through the overwhelming and sometimes overblown deluge of data.

At CNN, we ask pollsters to answer more than a dozen questions about their work before we cover it, from revealing who paid for the poll to sharing the technical details of their sampling and weighting. Even if you’re not an expert on methodology, there are a few things you can look for to get a sense of whether a poll is worth your time – and to get the most information out of its results.

Here are some of the most important questions to ask:

As a starting point, look for pollsters who are transparent about what they’re doing. If there aren’t many details available about who conducted and sponsored the poll, how they carried it out or what questions they asked, that’s your first sign to be skeptical.

For an idea of what that methodological information might look like, you can read more about how CNN conducts its own polling here or find detailed explanations of how outlets like the Pew Research Center or The New York Times conduct their surveys. The American Association for Public Opinion Research also has a helpful guide to the information you should be looking for .

Fundamentally, polls work by talking to relatively small groups of people who are selected to be demographically representative of a much larger universe (say, “adults in the US,” or “people who are likely to vote in the next election”).

How do pollsters choose which people to talk to? In past decades, many did so by calling a randomly selected set of phone numbers. These days, very few polls are conducted by interrupting Americans’ dinners with calls to their landlines. Instead, pollsters are trying to reach people however they can, whether that’s through cell phone calls, text messages, postcards in the mail, panels of people who’ve agreed to answer surveys or a combination of all these approaches.

What’s critically important across these methods, though, is that the people taking any individual survey are selected by the pollster, not the other way around. Feel free to ignore “polls” that let anyone answer, which can’t reflect the opinions of anybody beyond those who happened to answer.

When looking at the sample size of a poll, keep in mind that a relatively small but representative sample is better than a very large, unrepresentative one. But a small sample does increase the margin for error, particularly among subgroups of people – like young voters, or those from minority groups – that make up only a smaller share of the population.

Polls shouldn’t come out of nowhere. Look for polls that are transparent about the people who conducted them, and the precise survey companies or tools they used to carry them out. If the pollster has a track record of previous work – or if they’ve never released polling before – that’s helpful context, too.

It’s also important for polls to disclose who, if anyone, funded their surveys, whether that’s a media outlet, a think tank, or a campaign. Anyone putting in the time and effort to conduct a survey has a reason for doing so. Sometimes, that’s simply to inform the public. But campaigns are often selective in releasing only internal data that supports the narrative they want to share, or that might drive fundraising and engagement, and advocacy groups are often trying to shape a conversation around the policies they support.

The way a poll question is framed can have major impacts on the results of the poll, particularly around unfamiliar candidates or complicated topics.

So can the order that questions were asked in – for instance, a neutrally worded question about a choice between two candidates can still be thrown out of balance if voters were first asked a series of questions about controversies facing one candidate.

Look for polls that share the exact wording of their questions, so that you can evaluate whether those questions seem balanced and get the clearest sense possible of what the people taking the survey actually meant by their answers.

Pretty much any data point is more useful in context. When looking at new data, it’s helpful to know whether it’s broadly in agreement with other polling, or whether it’s saying something new – or if there isn’t much else out there to compare it to. Just because a poll is an outlier, of course, doesn’t mean it’s necessarily wrong. But sometimes, the variation between polls can tell you something. For instance, if minor changes in wording produce very different reads on an issue, that’s probably a topic where public opinion isn’t very settled.

While surveys can capture how voters are feeling at a given point in time, they can’t predict the future. This can pose a particular challenge in races like primary elections that often see multiple candidates, varied turnout and a lot of movement - see the Democratic primary in Michigan, where polls may have misread the composition of the electorate , and in Wisconsin, where polls also overstated the eventual support for the progressive candidate but accurately reflected that the race remained highly in flux.

Polls also aren’t meant to be precision instruments, which means that in a tight race, they may not be able to reliably measure which candidate is ahead – merely that the race is close. But even in an election year, most surveys also ask questions beyond whom voters are planning to support.

These questions can provide a sense of the dynamics behind the race: For instance, measuring anti-incumbent sentiments, highlighting voters’ ongoing concerns with cost of living, or tracking big partisan shifts . That’s often more illuminating than the horserace itself.

Kyle Grillot/Bloomberg via Getty Images

如今，进行某种形式的民意调查比以往任何时候都更容易，但要做好民意调查却比以往任何时候都更难。

像上个月引发轩然大波的洛杉矶市长竞选虚假“民调”那样完全捏造的数据实属罕见，这着实令人庆幸。即便如此，真正的民调也可能存在严重的缺陷，例如调查方法上的瑕疵或发布背后隐藏的动机。而且，最严谨的民调也无法保证结果的准确性。

但是，如果仔细阅读，高质量的民意调查仍然可以告诉你很多关于今年中期选举形势的信息——记住几个简单的问题可以帮助你从铺天盖地、有时甚至夸大其词的数据中理清头绪。

在CNN，我们会要求民调机构回答十几个问题，然后再进行报道，这些问题涵盖了民调的资金来源、抽样和加权等技术细节。即使您不是方法论专家，也可以通过一些信息来判断一项民调是否值得关注，并从中获取尽可能多的信息。

以下是一些最重要的问题：

首先，要寻找那些公开透明的民调机构。如果民调的执行者和赞助方、调查方式以及调查问题等细节信息很少，这就是你应该产生怀疑的第一个信号。

想了解这些方法论信息可能包含哪些内容，您可以点击此处阅读更多关于CNN如何进行民意调查的信息，或者查找皮尤研究中心或《纽约时报》等机构如何进行调查的详细说明。美国公共舆论研究协会也提供了一份有用的指南，其中包含您应该查找的信息。

从根本上讲，民意调查的运作方式是与相对较小的群体交谈，这些群体经过选择，在人口统计学上能够代表一个更大的群体（例如，“美国成年人”或“可能在下次选举中投票的人”）。

民调机构如何选择访谈对象？过去几十年，许多民调机构会随机拨打一组电话号码进行调查。如今，很少有民调会通过拨打固定电话打扰美国人的晚餐。相反，民调机构会想方设法联系受访者，例如通过手机、短信、邮寄明信片、邀请同意参与调查的人员组成小组，或结合以上所有方式。

然而，这些方法的关键在于，参与任何一项调查的人都是由民调机构挑选的，而不是相反。那些允许任何人作答的“民调”完全可以忽略不计，因为它们根本无法反映那些碰巧参与调查的人之外任何人的意见。

在考察民意调查的样本量时，请记住，相对较小但具有代表性的样本比非常大但缺乏代表性的样本要好。但小样本确实会增加误差范围，尤其是在人口比例较小的群体中，例如年轻选民或少数族裔群体。

民调结果不应该凭空出现。要寻找那些公开透明的民调，包括调查人员、具体调查公司或使用的工具等信息。如果民调机构过往有良好的记录，或者从未发布过民调结果，这些信息也有助于判断其真实性。

民调机构披露其调查的资金来源（如有）也至关重要，无论是媒体、智库还是竞选团队。任何投入时间和精力进行调查的人都有其理由。有时，其目的仅仅是为了向公众传递信息。但竞选团队往往会选择性地发布内部数据，只发布那些支持其想要传达的信息或有助于筹款和提高公众参与度的数据；而倡导团体则常常试图围绕其支持的政策来引导舆论。

民意调查问题的措辞方式会对民意调查结果产生重大影响，尤其是在候选人不熟悉或涉及复杂话题的情况下。

提问的顺序也会影响结果——例如，如果先向选民提出一系列关于其中一位候选人所面临的争议的问题，那么即使是措辞中立的关于在两位候选人之间进行选择的问题，也可能失去平衡。

寻找那些问题措辞完全相同的民意调查，这样你就可以评估这些问题是否平衡，并尽可能清楚地了解参与调查的人们通过他们的回答所表达的真实含义。

几乎任何数据点在特定背景下都更有价值。在查看新数据时，了解它是否与其他民调结果基本一致，或者是否揭示了新的信息，又或者是否缺乏可供比较的数据，都至关重要。当然，民调结果异常并不一定意味着它就是错误的。但有时，民调之间的差异确实能提供一些信息。例如，如果措辞上的细微变化导致对某个问题的解读截然不同，那么这很可能表明公众舆论在该议题上尚未达成共识。

虽然民调可以反映选民在特定时间点的感受，但却无法预测未来。这在初选等通常有多位候选人、投票率波动较大且选情变化频繁的选举中尤为突出——例如密歇根州的民主党初选，民调可能误判了选民构成；又如威斯康星州的初选，民调虽然高估了进步派候选人的最终支持率，但却准确地反映了选情始终处于高度不确定状态。

民调并非精确工具，这意味着在选情胶着的情况下，民调可能无法可靠地衡量哪位候选人领先，只能表明选情接近。但即便在选举年，大多数民调也会询问选民计划支持哪位候选人以外的问题。

这些问题可以帮助我们了解选情背后的动态：例如，衡量选民对现任者的不满情绪，凸显选民对生活成本的持续关注，或者追踪党派支持的重大转变。这些往往比选情本身更有启发意义。