WHEN AN UNSTOPPABLE force meets an immovable object, one or the other has to give. For most of Donald Trump’s second term, pundits(评论员，权威人士) have assumed that the object, Republicans’ seemingly impregnable(坚不可摧的，难以攻克的) Senate map, had the edge over(比……略占优势) the force, Democrats’ lead in national polls. Last year Sabato’s Crystal Ball, a leading forecasting outfit(机构，团队), called Republicans “strong favourites” to hold their Senate majority, and classified(把……列为，归类) Alaska, Iowa and Kansas as “safe” red seats. As late as this June, the group still found enough races at least “leaning” towards Republicans for the party to control the chamber.
当不可阻挡的力量遇上不可移动的物体，总有一方要让步。在唐纳德·特朗普第二任期的大部分时间里，评论员们一直认为，作为“物体”的共和党参议院版图看似坚不可摧，比作为“力量”的民主党全国民调领先更具优势。去年，领先的预测机构萨巴托的水晶球将共和党称为保住参议院多数席位的“强力热门”，并将阿拉斯加、艾奥瓦和堪萨斯列为“安全的”红州席位。直到今年六月，该机构仍认为有足够多的选情至少“倾向”共和党，足以让该党控制参议院。
Now, with midterm elections less than a week away, conventional wisdom(普遍看法，传统智慧) holds that the force has claimed the upper hand(占据上风，取得优势). Crystal Ball’s latest update says Democrats “have a slight advantage in the battle for the Senate”, while betting markets make the party a 67% favourite to flip(翻转，使翻转) the chamber. Nate Silver, America’s most prominent quantitative(定量的，数量上的) election forecaster, gives Democrats a 76% chance. But The Economist’s own prediction model is more cautious, putting the probability of the Senate changing hands(易手，转手) at just 54%.
如今，距离中期选举不到一周，普遍观点认为“力量”已占据上风。水晶球的最新更新称民主党“在参议院争夺战中略有优势”，而博彩市场认为该党有67%的可能性翻转参议院。美国最著名的定量选举预测师内特·西尔弗给民主党76%的胜算。但《经济学人》自己的预测模型更为谨慎，认为参议院易手的概率仅为54%。
The main source of this disagreement is our model’s intentional breach(违反，破坏) of a common forecasting maxim(格言，准则): thou shalt not attempt to predict the direction of polling errors. Although “generic-ballot” surveys, which ask respondents which party they plan to support for Congress, have an impressive record of accuracy, they also have a marked tendency to overshoot: wave elections usually prove less towering than advertised.(尽管“通用选票”调查有着令人印象深刻的准确记录，但它们也明显倾向于高估：浪潮选举通常没有宣传的那么声势浩大。) On the eve of the 2006 midterms, for example, Democrats led these surveys by 13 percentage points on average, while their final margin in the national House vote was a mere seven. And when national polls misfire(失准，失败), that error usually appears in state-specific surveys as well.
这一分歧的主要来源，是我们的模型有意打破了一条预测戒律：不可试图预测民调误差的方向。尽管“通用选票”调查——即询问受访者计划在国会选举中支持哪个政党——有着令人印象深刻的准确记录，但它们也明显倾向于高估：浪潮选举通常没有宣传的那么声势浩大。例如，在2006年中期选举前夕，民主党在这些调查中平均领先13个百分点，而他们在全国众议院投票中的最终优势仅为7个百分点。而当全国民调失准时，这种误差通常也会出现在各州的调查中。