Statistician with near-perfect election formula: prepare yourself for President Trump

[mjr]

He's only "missed" on one election since 1912...the Kennedy election in 1960...

http://www.seattlepi.com/elections/a...la-6856643.php

He predicts that if Trump is the Republican nominee, there's a 97% chance he'll win vs Hillary, and a 99% chance vs Sanders.

[wolfie]

The article doesn't mention one important fact - WHEN did he develop this "almost perfect" system? If it was after the 2012 election, then it hasn't successfully predicted even one election.

He's fallen into one of the biggest traps for statisticians developing a method of predicting ANYTHING based on past results - he has the luxury of being able to pick and choose from various statistics in order to make his model fit the historical results, so he can come up with an amazingly accurate algorithm, but you NEVER test an "expert system" against the same data set that was used to train it. You ALWAYS keep some data out of the training set so that you can use that excluded data for testing.

If he had "trained" his system against results from 1912 to 1988, and it had then predicted the 1992 through 2012 elections (3 switches of party, 3 presidents re-elected), I'd give his system more weight.

[mjr]

The article doesn't mention one important fact - WHEN did he develop this "almost perfect" system? If it was after the 2012 election, then it hasn't successfully predicted even one election.

If he had "trained" his system against results from 1912 to 1988, and it had then predicted the 1992 through 2012 elections (3 switches of party, 3 presidents re-elected), I'd give his system more weight.

All valid, wolfie. It would be interesting to know the actual methodology.

Although, I was thinking about this the other day. The last time an incumbent party held the White House for more than two consecutive terms was the two terms of Reagan, and the 1 term of George H. W. Bush.

Before that, you have to go back to FDR/Truman. FDR had 3+ terms (died in office during his fourth term). Of course, this was before the 22nd Amendment. Truman completed FDR's fourth term, then had one of his own, then Eisenhower got elected.

I'm guessing that's part of the "pattern" the guy used. I have no idea.

[Jetfire]

Basically that. Having a party hold the Presidency for more than 2 terms in a row is already a very uncommon event. By all accounts, this election should be the GOP's election to lose.

And all other signs are pointing to the GOP doing their damndest to lose it at all costs.

So I'm fairly confident that this guy's method (whatever it may be) is about to hit its second miss in a Century and change.

[Gravekeeper]

Although, I was thinking about this the other day. The last time an incumbent party held the White House for more than two consecutive terms was the two terms of Reagan, and the 1 term of George H. W. Bush.

So I read his paper. This literally IS the guy's methodology. There's no complex statistical analysis going on. This is quite literally what he is basing it on. That historically neither party holds the White House for more than 2 consecutive terms most of the time. Thats it. That is all his model is. The only thing he did was average out the vote margins for one term incumbent parties vs two term incumbent parties to add numbers to the idea to make it sound fancy.

His model is so simplistic you actually guessed the entire thing just pondering about it to yourself.