Zakaria: Why Trump won

[Igel]

10. It's the "last point" phenomenon.

You're in a basketball game against your school's archrival, and they've won 10 of the last 10 times your school played them. There's 20 seconds until the final buzzer and the score's tied, 120-120. Mr. Q has been on the bench and is brought in for the last 20 seconds.

At 19 seconds Mr. Q is fouled and makes a 1-point shot. The score's 121-120.

Quick: Who's primarily responsible for winning the game? He's responsible, though, for 1/121 of the points.

If any other player on the winning side hadn't made a shot, at best Mr Q would be responsible for a tie. If any of the players who made a 2- or 3-point shot hadn't made that shot, Mr Q would be noted as the guy who brought them up from 118-120 to 119-120. In other words, unimportant. As it is, when Mr Q is picked up and carried in celebration he's going to take the praise. Only later will he say that he owes it all to his other teammates, his coach, his elementary school teacher, friends, his dog, etc.

It took a lot for Newton to say that if he could see farther it was only because he stood on the shoulder of giants.

In Mr Q's case, he's one of the lesser important people. But he's the last one, so all those who went before, those who scored the 120-based that made his point mean anything, are lost in the shuffle.

So if Trump won with 46% of the vote (not gonna bother to check, not important) and 6% is attributed to Comey, there's still the other 40%, every point of which is equally important to each of those 6 points.


So Comey and Nate Silver. A couple of things things. First, on election day the undecideds in the last polls who actually went to vote stopped being undecided. People tend to ignore the undecideds, as well as degrees of certainty We don't know how they shifted at the last minute or how they voted. They're lost in the shuffle. (2) We do know that every poll is based on a small sample scaled up to represent the entire population based on a model of the population and a model of who votes how. You poll 83 blacks for a total of 9% of your sample, you assume more than 9% of the black population will vote so you adjust the figures--for that, and income, education, geography, etc., etc. We also know that the model was wrong, as it often is, so that that particular scaling, that adjustment, was incorrect. That makes it really hard to tell how many (D) voters shifted to (T) voters. Too many moving parts all working all at once.