2018 Pre-Season Probability Poll Discussion Thread

[ee8384]

Nerds. Everywhere, Nerds.

 

[coit]

Nerds. Everywhere, Nerds.

 

[18in32]

I have updated the spreadsheet to have the @GoGATech modification as follows. I take the @ThisIsAtlanta scoring and I assign the result of each game as follows:

0.4 = >20 point victory
0.3 = >15 point victory
0.2 = >10 point victory
0.1 = >5 point victory
0.0 = toss up
-0.1 = >5 point loss
-0.2 = >10 point loss
-0.3 = >15 point loss
-0.4 = >20 point loss

I then take each voter's probability, subtract 0.5 to normalize it with the values above, and take 1 - abs(vote-result). That produces a weighting factor between 0.2 and 1 depending on how close your vote was to the result. Then I multiply that weighting factor by the TIA factor.

As for expectations going in, I could use the ESPN FPI numbers to determine a predicted point differential, factoring in 3 point home field advantage. But I wonder if using the results for the weighting factor wouldn't be better? Otherwise a voter would be penalized for predicting a high probability in a game we are heavily favored to lose. Right?

I'm not sure what I think of this yet. My only stats intuitions are in the vein of "lies, damned lies, and statistics". But it does seem like the margin-of-victoy breakpoints should be more connected to football scoring (ie, FG's and TD's) and less to base ten. On your calculation, winning a game by two possessions (9 pts) is the same magnitude of victory (0.1) as winning by one possession (7 pts). In football reality, trying to close out a game with a two possession lead feels way different than trying to close out a game with a one possession lead. So I'd just say make it more connected to typical football scoring.

0.4 = three possession (>16) pt win
0.3 = two TD (>11) pt win
02. = two possession (>8 pt) win
0.1 = TD (>3 pt) win

 

[GoGATech]

I'm not sure what I think of this yet. My only stats intuitions are in the vein of "lies, damned lies, and statistics". But it does seem like the margin-of-victoy breakpoints should be more connected to football scoring (ie, FG's and TD's) and less to base ten. On your calculation, winning a game by two possessions (9 pts) is the same magnitude of victory (0.1) as winning by one possession (7 pts). In football reality, trying to close out a game with a two possession lead feels way different than trying to close out a game with a one possession lead. So I'd just say make it more connected to typical football scoring.

0.4 = three possession (>16) pt win
0.3 = two TD (>11) pt win
02. = two possession (>8 pt) win
0.1 = TD (>3 pt) win

I'm on board with this.

 

[coit]

I'm not sure what I think of this yet. My only stats intuitions are in the vein of "lies, damned lies, and statistics". But it does seem like the margin-of-victoy breakpoints should be more connected to football scoring (ie, FG's and TD's) and less to base ten. On your calculation, winning a game by two possessions (9 pts) is the same magnitude of victory (0.1) as winning by one possession (7 pts). In football reality, trying to close out a game with a two possession lead feels way different than trying to close out a game with a one possession lead. So I'd just say make it more connected to typical football scoring.

0.4 = three possession (>16) pt win
0.3 = two TD (>11) pt win
02. = two possession (>8 pt) win
0.1 = TD (>3 pt) win

Yeah I think that's a good point. I'll go with that!

 

[coit]

OK I've updated the OP with the scoring rules.

To be successful in the contest you will need to have voted on each game. If you have missed a vote and want to add votes in, just put them in this thread and I'll included them.

 

[GT76]

OK I've updated the OP with the scoring rules.

To be successful in the contest you will need to have voted on each game. If you have missed a vote and want to add votes in, just put them in this thread and I'll included them.

Since you’ve created new rules unrelated to the initial premise of probability of winning, now including margin of victory, can votes already cast on closed threads (Alcorn, etc.) be changed?

 

[eg1]

I will just nod sagely and pretend that I understand any of this ...

 

[coit]

Since you’ve created new rules unrelated to the initial premise of probability of winning, now including margin of victory, can votes already cast on closed threads (Alcorn, etc.) be changed?

Sure, if anyone wants to change votes just post them here!

 

[18in32]

Since you’ve created new rules unrelated to the initial premise of probability of winning, now including margin of victory, can votes already cast on closed threads (Alcorn, etc.) be changed?

You think margin-of-victory predictions and probability-of-winning predictions are "unrelated"? Seems to me like they track the same thing. What would support a belief that Team A is extremely likely to beat Team B but by only a few points?

 

[coit]

You think margin-of-victory predictions and probability-of-winning predictions are "unrelated"? Seems to me like they track the same thing. What would support a belief that Team A is extremely likely to beat Team B but by only a few points?

Past history, perhaps?

That being said, for folks that are voting realistically I think the new weighting will only serve to penalize the folks that vote 0.9 or 0.1 every time. And it will also help sort out a vote for 0.5 that in the old system was counted as voting for a loss. So rather than get no credit for a win you can now get full weighting for a close game.

As long as folks are voting in the range of 0.3 to 0.7 or so on most games the weighting likely won't have an effect. The results from 2017 with the weighted values still had the same folks at the top of the list for the most part.

And it still makes sense to vote higher for a win if you think we are going to win the game. Since that is the initial weight you get, and it is modified by the ratio.

If you vote 0.9 for a game that we win by less than a field goal your score would be:

0.9 * (1 - (0.9 - 0.5 - 0)) = 0.54

If you vote 0.8 for that game your score would be:

0.8 * (1 - (0.8 - 0.5 - 0)) = 0.56

If you vote 0.7 for that game your score would be:

0.7 * (1 - (0.7 - 0.5 - 0)) = 0.56

If you vote 0.6 for that game your score would be:

0.6 * (1 - (0.6 - 0.5 - 0)) = 0.54

If you vote 0.5 for that game your score would be:

0.5 * (1 - (0.5 - 0.5 - 0)) = 0.5


So it seems that confidence in winning or losing is more important than voting low for a game you might expect to be close. And you can also see that you probably shouldn't vote 0.9 unless you really think it will be a multiple score win.

 

[ThisIsAtlanta]

You think margin-of-victory predictions and probability-of-winning predictions are "unrelated"? Seems to me like they track the same thing. What would support a belief that Team A is extremely likely to beat Team B but by only a few points?

If team A and B are Alabama and LSU.

 

[GT76]

You think margin-of-victory predictions and probability-of-winning predictions are "unrelated"? Seems to me like they track the same thing. What would support a belief that Team A is extremely likely to beat Team B but by only a few points?

Maybe a bad choice of words but I don’t micro analyze these things. I pretty much go with we should win big, close, maybe, doesn’t look good and pick a probability. Margin of victory of 17 or 21 or whatever is a new factor. The probability of me changing any of my votes is approximately 0.

 

[18in32]

Past history, perhaps?

That's gonna require some explanation.

If you vote 0.9 for a game that we win by less than a field goal your score would be:
0.9 * (1 - (0.9 - 0.5 - 0)) = 0.54

If you vote 0.8 for that game your score would be:
0.8 * (1 - (0.8 - 0.5 - 0)) = 0.56

If you vote 0.7 for that game your score would be:
0.7 * (1 - (0.7 - 0.5 - 0)) = 0.56

If you vote 0.6 for that game your score would be:
0.6 * (1 - (0.6 - 0.5 - 0)) = 0.54

If you vote 0.5 for that game your score would be:
0.5 * (1 - (0.5 - 0.5 - 0)) = 0.5

Well, you've gone and confused me again. Why does the 0.8 voter get more points than the 0.9 or 0.5 voters for a game won by less than a FG? Shouldn't the 'toss up' voter emerge with the most points? Or (on the dubious theory that you can be absolutely certain of a close win) at least the 0.9 voter?

 

[ThisIsAtlanta]

We should calculate the portion of the graph under the line on those "probability of winning" graphs for GT, then your points awarded for the game can be based on how close you got to that. Boom.

 

[gtnap]

It isn't rocket science. Just an attempt to predict the season record. I am only tracking individual performance on a W/L basis where anything .6 or higher means you think we will win and .5 or less means you think we will lose.

I think the most interesting thing about this is how the probabilities versus different opponents change year by year.

It is now...

 

[coit]

It is now...

 

[coit]

That's gonna require some explanation.

Well, you've gone and confused me again. Why does the 0.8 voter get more points than the 0.9 or 0.5 voters for a game won by less than a FG? Shouldn't the 'toss up' voter emerge with the most points? Or (on the dubious theory that you can be absolutely certain of a close win) at least the 0.9 voter?

Here's a set of curves that hopefully illustrate what I mean. You shouldn't be rewarded for picking 0.5 for a game. And the plots for toss-up win and toss-up loss bear that out I think. You get more points for more strongly picking a win or a loss if you are right. Your high confidence overrides the fact that you are off on the margin up to the 0.9 or 0.1 extreme values. The only times you are rewarded for picking 0.9 or 0.1 is if the games are indeed blowouts. As for the high probability but low margin thing I think TIA answered it pretty well with the Bama/LSU example.

 

[18in32]

You shouldn't be rewarded for picking 0.5 for a game. And the plots for toss-up win and toss-up loss bear that out I think. You get more points for more strongly picking a win or a loss if you are right. Your high confidence overrides the fact that you are off on the margin up to the 0.9 or 0.1 extreme values. The only times you are rewarded for picking 0.9 or 0.1 is if the games are indeed blowouts.

You're deliberately penalizing someone for being accurate in his predictions (if the prediction is "too-close-to-call")? And you're deliberately encouraging people to be more confident than they are? Why?

As for the high probability but low margin thing I think TIA answered it pretty well with the Bama/LSU example.

I don't know how TIA meant that, but I consider it to be a joke, something to do with the mystical Curse of Saban and the supposedly impenetrable defenses of the SEC West. I don't see how else one evaluates a certainty of victory prediction other than by using margin of victory? (Isn't that what you're doing, except for your desire to suppress too-close-to-call fence-sitters?)

And – not that it matters — in Alabama's current win streak it has beaten LSU by an average margin of victory of almost two TDs.

 

[coit]

You're deliberately penalizing someone for being accurate in his predictions (if the prediction is "too-close-to-call")? And you're deliberately encouraging people to be more confident than they are? Why?

Yeah in looking at that graph I think the system isn't doing what we want it to do although I'm not sure there's a way to do it since the point of the probability votes is to represent how confident you are that we will win a game.

In reality, as some have pointed out, a voter's confidence in our likelihood of winning or losing a game is somewhat independent of the scoring outcome. One could be equally highly confident in us beating Bowling Green and UVA, yet you might have different expectations regarding the score differential.

I'm certainly open to other ideas if they can be implemented with too much additional complexity.