民调不是选票,别急着庆祝Polls Aren't Ballots. Don't Celebrate Yet.
这篇稿子讲的是民调为什么不可靠,但它自己就是民调叙事的一部分——它靠拆穿别人的预测来卖自己的谨慎,谨慎在这里也是一种商品。
它列的问题都是真的:答题的人偏老、偏有文化、偏关心政治,躲开问卷的人偏年轻、偏穷、在最近的总统选举里更可能投特朗普。事后加权能补一点,补不干净。今年后期民调还少了,密歇根州那位民主党众议员 Haley Stevens 突然背书一个共和党人,阿拉斯加出了选票堆叠丑闻,这些都可能落在数据之外。
但整篇文章从头到尾没有出现一个具体的人。没有选民,没有候选人,没有谁在为这场选举付代价。有的是「民主党人现在大概心情不错」,有的是「我不会提前开香槟」。这是一篇关于预测准确性的文章,不是关于任何人的处境的文章。
这正是民调作为一种表达被武器化的地方。它把「谁会被算进样本」包装成一个技术问题,交给统计学家去修。可样本里少掉的那些人不是随机少掉的,是结构性的少掉——他们本来就在公共表达里被挤到边上,现在连「民意」这个词里都没有他们的位置。谁被问到,谁就是民意;谁没被问到,谁就不存在。这不是误差,这是定价。
文章末尾还附了一条建议:跟政见不同的亲戚别吵,问一句「这为什么对你重要」,然后用「我」开头讲自己的看法。这条建议本身没错,但它出现在一篇讲民调失真的稿子后面,读起来像是:既然数据说不准,那就回家好好说话。把结构问题降级成家庭沟通技巧,是文化暴力最舒服的形态。
真正没被问的是:为什么决定参议院归属的七个州里,有三个在农村州,而农村州的选民最难被民调够到?为什么难够到的人,恰好是这些年被主流叙事反复写成「不可预测」的那批人?把他们写成不可预测,比去理解他们便宜得多。
八小时后投票开始。民调会错,或者不会错。但无论哪种结果,被样本筛掉的那部分人,明天依然不在任何一张图表里。
This piece is about why polls are unreliable, and it is itself part of the polling narrative: it sells its own caution by dismantling other people's predictions. Caution, here, is also a product.
The problems it lists are real. Respondents skew older, more educated, more invested in politics; the poll-avoiders skew younger, less educated, and in recent presidential contests more likely to back Trump. Post-hoc weighting patches some of it, not all. There were fewer late-cycle surveys this year, and late developments—Democratic Rep. Haley Stevens's surprise endorsement of a Republican in Michigan, a ballot-stacking scandal in Alaska—may sit outside the data.
But not one specific person appears in the whole piece. No voters, no candidates, no one paying a price for this election. What we get is that Democrats are probably feeling good, and that the author won't pop champagne early. This is an article about predictive accuracy, not about anyone's situation.
That is exactly where polling gets weaponized as expression. It packages the question of who gets counted as a technical problem for statisticians to fix. But the people missing from the sample are not missing at random; they are missing structurally. They were already pushed to the edge of public expression, and now they don't even have a seat inside the word "public opinion." Whoever gets asked is the public. Whoever doesn't, doesn't exist. That isn't error. That's pricing.
The piece closes with advice: don't fight your relatives, ask "why does that matter to you," use "I" statements. Fine advice. But placed after an essay on polling distortion, it reads as: since the data is fuzzy, go home and talk nicely. Downgrading a structural problem into a family-communication tip is cultural violence in its most comfortable form.
The question never asked: why are three of the seven Senate-deciding states rural, where reaching a representative sample is hardest? And why are the hardest-to-reach voters precisely the ones mainstream narrative keeps writing as "unpredictable"? Writing them off as unpredictable is cheaper than understanding them.
Voting starts in eight hours. The polls will be wrong, or they won't. Either way, the people the sample filtered out will still be absent from every chart tomorrow.