The difference this year is that a normal-sized polling error in Democrats’ direction would merely make the race for the Senate close. (Likewise, a normal-sized polling error in the GOP direction would make the House close, but Republicans would still have to fight it out on a district-by-district basis). A sports analogy, for those so inclined: In 2016, Trump was doing the equivalent of driving for the game-winning touchdown with the odds somewhat but not overwhelmingly against him. If enough undecided voters in the Midwest broke toward him, he was going to win the Electoral College. In the Senate this year, by contrast, it’s more like Democrats are driving for the game-tying touchdown; they still have to win in overtime even if they score.
By a systematic polling error, I mean one that occurs in a correlated way across every race, or in certain groups or races — not merely errors that happen on a one-off basis. Our models account for the possibility of several different types of systematic errors, but in this article, I’m going to focus on the simplest type of systematic error, which is a uniform swing that applies to every race. In certain simulations, for example, our model will randomly simulate a 4-percentage-point uniform swing toward Republicans, in which it adds 4 points to the Republican margin in every state and district. From there, it proceeds to consider the other types of error and uncertainty.
Join the conversation as a VIP Member