Nate Silver's projections

[Grant Green]

There's nothing else to test that model against.

Exactly. So there is no way to quantify how far off he was. Unless there is a concrete way to measure the actual probability, there is no way to determine if he was off or not.

People are looking at his prediction and saying, well trump won so he was off. But he gave trump a 28% chance to win.

 

[fairgambit]

Another pet peeve of mine. I love when people complain about weathermen being wrong that one time out of ten and forget the other 9 times that they nailed it. They are predicting the freaking future and people think they are idiots when the miss one prediction.
If I told you there was a 16% chance of rolling dice once and hitting a six, would you tell me I was way off if it hit?

Ah, but I did not complain. You persuaded me with your argument and I said he should have been a weatherman because I admire their accuracy. (Nate Silver says the chances I actually believe this are 14.73%).

 

[Erial_Lion]

Exactly. So there is no way to quantify how far off he was. Unless there is a concrete way to measure the actual probability, there is no way to determine if he was off or not.

People are looking at his prediction and saying, well trump won so he was off. But he gave trump a 28% chance to win.

Actually, one good way to look at the model is to look at the betting line and test it against that. If you used Nate's numbers, you would have been loading up on Trump. It would be like someone creating a college football model that set PSU as an 11 point dog at home against Ohio State this year. If Penn State wins, I'm sure that some trash the model saying that it doesn't know what's it doing because it picked the wrong team? I'd be looking at how it did against it's peers (ie other prognosticators and the betting line).

I also laugh at people that complain about weathermen. I'll hear "how can it be raining, they said it only had a 20% chance of rain today", thinking that somehow 20% means that it shouldn't rain today.

 

[Grant Green]

Actually, one good way to look at the model is to look at the betting line and test it against that. If you used Nate's numbers, you would have been loading up on Trump. It would be like someone creating a college football model that set PSU as an 11 point dog at home against Ohio State this year. If Penn State wins, I'm sure that some trash the model saying that it doesn't know what's it doing because it picked the wrong team? I'd be looking at how it did against it's peers (ie other prognosticators and the betting line).

I also laugh at people that complain about weathermen. I'll hear "how can it be raining, they said it only had a 20% chance of rain today", thinking that somehow 20% means that it shouldn't rain today.

Ps. I noticed nate gave psu a 52% chance to beat wisky. I ended up getting psu at+3 but now I'm thinking of sprinkling on the money line

 

[Mixolydian]

Exactly. So there is no way to quantify how far off he was. Unless there is a concrete way to measure the actual probability, there is no way to determine if he was off or not.

People are looking at his prediction and saying, well trump won so he was off. But he gave trump a 28% chance to win.

His model predicted Clinton winning by 67 electoral votes so there is a way to tell how far it was off...a lot.

 

[PSU Mike]

His model predicted Clinton winning by 67 electoral votes so there is a way to tell how far it was off...a lot.

Here's a question that I hope you can answer. If you can't maybe you should soften your responses on this thread.

Given Trump won, how much more likely is it that the "true" probability of Trump winning was, say 51%, than the 28% that you guys are attributing to Silver?

How high would Silver's probability have needed to be for you to call him "right?"







Cheater hint: Google "likelihood ratio test" and see where it leads you.

 

[Mixolydian]

Here's a question that I hope you can answer. If you can't maybe you should soften your responses on this thread.

Given Trump won, how much more likely is it that the "true" probability of Trump winning was, say 51%, than the 28% that you guys are attributing to Silver?

How high would Silver's probability have needed to be for you to call him "right?"







Cheater hint: Google "likelihood ratio test" and see where it leads you.

I'm not "attributing" anything, nor am I softening my response. I have a question for you. How low did Silver's probability need to be for you to call him wrong? 0%? Good work if you can get it...

 

[TenerHallTerror]

Silver is a fraud.

and no one knows more about sample sizes than I do!

I surveyed one person and he said so.

 

[royboy]

http://projects.fivethirtyeight.com/2016-college-football-predictions/

Looking at Silver's projections this morning, Washington is almost certainly in, 99%+
PSU needs to win, and would be helped greatly by a Clemson loss