So I posted Who Is He? Nate Silver on Thursday, June 30. In it I included this paragraph:
FiveThirtyEight.com posted its first general election analysis, Who Will Win the Presidency?, within the last couple of days. It says Hillary Clinton, has a 79.2% chance of winning. Unpacking this analysis matters. First, a 79.2% chance of winning does not mean Ms. Clinton will win 79.2% of the votes. Rather, it means an election, held 100 times, would see Ms. Clinton winning about 80 times.
On July 1, in the evening, regular reader SC wrote me privately—for sure, because he didn’t want to embarrass me—about my statement. Paraphrasing his several words, he wrote “Wrong!” And, after checking in with my math major sister, my daughter the chemist, and her boyfriend the engineer, I agree with SC … I think!
Apologies for the error. I try to get things right but, from time to time, I mess up. And right now, I’m wishing I’d paid lots more attention in Statistics 101, taken at 40 years ago.
Before I proceed, I offer you a discussion at FiveThirtyEight.com, An 80 Percent Shot Doesn’t Mean Clinton Is A Sure Thing, about just exactly what the 79.2% means. In it Nate Silver says:
It can change, sure. But let’s be clear — 80 percent is the forecast Clinton has to win on Nov. 8. That’s our best estimate of her chances, accounting for the uncertainty between now and then, based on the historical accuracy of presidential polling. If the election were held today instead, she’d be a safer bet still.
The polls can change a lot between now and Nov. 8. And they probably will. But there’s a chance those changes benefit Clinton, and not Donald J. Trump. And since she’s up by about 7 points now, there’s the chance they help Trump … but not enough to allow him to win.
And that’s the thing. Of the 80 percent of the time Clinton wins — PLENTY of those times are going to involve her sweating. Either because Trump makes it very close at the end or because there are some periods in which things look very tight along the way, as they did for Obama against McCain and against Romney.
But Clinton will win a lot of those close calls, along with her share of landslides.
If you really want to get behind all of this, read A User’s Guide To FiveThirtyEight’s 2016 General Election Forecast by Nate Silver. I wish I had read and fully understood it before I posted Who Is He? Nate Silver. (And, again, I’d like another crack at Stats 101.)
I have a few additional comments. First, the User’s Guide reflects lots of rigor. As I noted a few days ago, the modeling might be bad—again, I don’t think so—but a system exists.
Second, the system applies to Clinton and Trump in the same way. (Former New Mexico Governor Gary Johnson, the Libertarian Party candidate, gets treated differently because, historically, lesser party candidates underperform by Election Day. Again, there’s rigor.)
Third, the model assumes an election on November 8. So, it takes into account expected changes, but it gets reviewed with the issuance of each new poll. (As of today, the new number is 78 – 22.)
Finally, and this is where I messed up, I think, the prediction reflects a roughly 80% likelihood that Hillary Clinton will be elected on November 8. That does not mean she wins the election 80 times out of 100, although Nate Silver does say “Of the 80 percent of the time Clinton wins … .” It means the model says there is an 78% likelihood that Ms. Clinton will win the 2016 Presidential Election, held on November 8, 2016, as of today.
PS Thanks as always to SC for reading carefully and commenting.