Just saw this crazy stat

[87&91ALUM]

Win and all this is in the rearview mirror. Tennessee can win if they play well, get off the field on defense, do not put the defense in bad positions, keep the refs out of the game by not making motion and targeting types of mistakes, and hold onto the football. LSU's stadium isn't called Death Valley for no reason. Gonna be a tough game, weather 80-86, NE wind 10 mph, 30-40% humidity, and if we let their fans get in the game early, even tougher. I'll be rooting for my Vols! Just win.

 

[BanditVol]

No. They really don't. And no, the team is not 75% the same. Even like UT when you have a lot of returning players, you are essentially saying that players do not get better or develop. If my election example isn't good where you have controllable factors then your poker example is far worse. Improvement of a roster isn't a random shuffle of the cards or "flop".

BTW, if you believe they're that accurate then you should stop playing poker and bet on teams they say have 60% chance to win when Vegas has them as underdogs.

You can believe in it if you like. But they make too many assumptions in the treatment of the "data".

So what you are saying is you don't believe the numbers ESPN published? I do. I see no reason to doubt them. They started the FPI in 2014 and state that they have tweaked it all along. I think 8 years is plenty of time to get a model like that to line up pretty well with reality.

Their table says that for predictions between 50% and 60% the favorite wins 57% of the time, which is pretty darn good. You can doubt it if you want, that's your business, but you are now saying they are lying, which is highly unlikely IMO.

You misunderstood my poker analogy. In elections, a percentage is not a prediction of who will win but literally the "score" by which politicians win. So if a politician wins 51% to 49%, it's a squeaker. However if politician wins 60% to 40% that's a butt kicking. I think we can agree on that. Now, if I was to PREDICT the outcome of an election, and the polls PREDICT that one politician will win 60% to 40%, then the PROBABILITY that he will win is far higher than 60%, because to overcome a 60 to 40 advantage in the polls is very difficult, that's like being behind by 3-4 TDs. The percent vote expected to go for a politician is not the the prediction, it's an "expected score". If on the other hand one politician is predicted to have a 51-49 advantage, his expectation of winning is still greater than 51% but far less than the 60-40 case.

Poker, coin tosses, and football on the other hand don't use percentages as score keeping devices therefore there is no confusion between the "score" and the PREDICTED OUTCOME. If I wanted to predict the outcome in an election I wouldn't take the percent of people likely to vote for a politician as the thing to bet on, I would go broke.

For the record I bet on poker all the time, and do quite well with it, thank you.

 

[sjt18]

So what you are saying is you don't believe the numbers ESPN published? I do. I see no reason to doubt them. They started the FPI in 2014 and state that they have tweaked it all along. I think 8 years is plenty of time to get a model like that to line up pretty well with reality.

I'm not inclined to believe a media organization who claims they've verified their own system... and that may especially include ESPN.

That being said. They introduce bias into their own system as I mentioned above.

Their table says that for predictions between 50% and 60% the favorite wins 57% of the time, which is pretty darn good. You can doubt it if you want, that's your business, but you are now saying they are lying, which is highly unlikely IMO.

Did I say they were lying? They're "marketing".... because they aren't doing this for "free". If it weren't making them money then they wouldn't be doing it.

You misunderstood my poker analogy. In elections, a percentage is not a prediction of who will win but literally the "score" by which politicians win. So if a politician wins 51% to 49%, it's a squeaker. However if politician wins 60% to 40% that's a butt kicking. I think we can agree on that. Now, if I was to PREDICT the outcome of an election, and the polls PREDICT that one politician will win 60% to 40%, then the PROBABILITY that he will win is far higher than 60%, because to overcome a 60 to 40 advantage in the polls is very difficult, that's like being behind by 3-4 TDs. The percent vote expected to go for a politician is not the the prediction, it's an "expected score". If on the other hand one politician is predicted to have a 51-49 advantage, his expectation of winning is still greater than 51% but far less than the 60-40 case.

The limit of my example was simply that percentage wise 14% is significant. That is NOT basically a "toss up".

Poker, coin tosses, and football on the other hand don't use percentages as score keeping devices therefore there is no confusion between the "score" and the PREDICTED OUTCOME. If I wanted to predict the outcome in an election I wouldn't take the percent of people likely to vote for a politician as the thing to bet on, I would go broke.

For the record I bet on poker all the time, and do quite well with it, thank you.

Now you are just evading. Poker is a game of probabilities based exclusively on the cards and "chance" events. That isn't football. Football is a product of human decisions and actions.

 

[BanditVol]

Now you are just evading. Poker is a game of probabilities based exclusively on the cards and "chance" events. That isn't football. Football is a product of human decisions and actions.

I'm not saying there isn't a profound difference in calculating the probabilities, there is. But in each case, it IS a probability and not a predicted score like it is in an election.

Yes, there is random chance in cards, but it's very predictable, moreso than football. The odds in cards are exact because there are only 52 cards in the deck and everyone knows what they are. Football is actually far more random because it involves actions of human beings, not to mention the ball can bounce funny.

It's only qualitative similarity in that the percent quoted by the FPI is a prediction as is the predicted outcome of a card game. The card game is an exact percentage and mathematically accurate however. My point in bringing in the poker analogy is that anyone who has played a lot of hold em knows that on any given hand, the favorite can and often does lose (and it can be very painful when it does!). The averages only apply to very large sample sizes, on a given hand literally anything can happen.

Having said that, from their blog ESPN is saying they base the prediction on statistics from previous games and that they use regression, and that they have repeatedly tweaked the system to make it more accurate. It's a reasonable way to do probability for sports, since it's based on a very large sample size.

It's not perfect, nothing is, but their reported performance is pretty good.

 

[chuckb1448]

I don't know, it feels like the right spot for a letdown. I think Tennessee wins, but this is about to be a boring, slow-moving, close game that will stir up some bad feelings. I believe CBK when he said he plans to slow Tennessee down. The clock will be inside the ten every time LSU runs a play.

I hope he doesn't pull Flops on us

 

[sjt18]

I'm not saying there isn't a profound difference in calculating the probabilities, there is. But in each case, it IS a probability and not a predicted score like it is in an election.

Yes, there is random chance in cards, but it's very predictable, moreso than football. The odds in cards are exact because there are only 52 cards in the deck and everyone knows what they are. Football is actually far more random because it involves actions of human beings, not to mention the ball can bounce funny.

It's only qualitative similarity in that the percent quoted by the FPI is a prediction as is the predicted outcome of a card game. The card game is an exact percentage and mathematically accurate however. My point in bringing in the poker analogy is that anyone who has played a lot of hold em knows that on any given hand, the favorite can and often does lose (and it can be very painful when it does!). The averages only apply to very large sample sizes, on a given hand literally anything can happen.

Having said that, from their blog ESPN is saying they base the prediction on statistics from previous games and that they use regression, and that they have repeatedly tweaked the system to make it more accurate. It's a reasonable way to do probability for sports, since it's based on a very large sample size.

It's not perfect, nothing is, but their reported performance is pretty good.

The limits of trusting a media organization's fact checking of its own accuracy.

This is what I was talking about concerning their built in bias. Someone said they consider 247 rankings, history prior to this year, and previous match ups this year. Missing: development, coaching, schemes, changes in culture, etc. Those are HUGE gaps.

 

[photovol]

Y'all, UT has beat a likely overrated Pitt on the road in OT and a decent UF at home and the other two wins were over cupcakes. Not only that, UT has struggled for years and has not yet shown it can deal with success. Even more, it is on the road in Baton Rouge against an LSU program that has had ups and downs but has been very good overall in the 2000s and still has talent. I think the game is a toss up and 58% chance for LSU at home is a pretty good indication of that being the case rather than some kind of smear against UT. I mean, it isn't 90% that LSU will win. I also think the algorithm that feeds the FPI is a stronger indicator later in the season after more data is available about current teams and it is still relatively early. And so, I don't think this is some kind of ESPN bias against UT as it is an algorithm and UT is going to be challenged in Baton Rouge and the game could go either way. The Vols better bring their A game if they want to win.

Somebody thought wrong!

 

[BanditVol]

The limits of trusting a media organization's fact checking of its own accuracy.

This is what I was talking about concerning their built in bias. Someone said they consider 247 rankings, history prior to this year, and previous match ups this year. Missing: development, coaching, schemes, changes in culture, etc. Those are HUGE gaps.

Those other things you mention aren't reduce-able to numbers but are very subjective. They really almost entirely on statistics. It's all in the blog above, you can read it yourself.

The one that I do think is week is the 247 ratings. Those are subjective to begin with and often don't pan out, but they use them anyway. I guess it must work though or they manage to manipulate the numbers enough to make it work.

As for trusting them...in an earlier post you stated that they are making millions off the FPI, unless I misunderstood you. I don't see how...they make millions off of broadcasting games, and FPI is just one of many tools they use to generate interest in the games.

As such, I don't see how misrepresenting how accurate it is helps them. To be clear, the numbers they present don't reveal much and could be easily manipulated to look better than they are, but at the same time they match my experience with FPI over the years, which is that it's a pretty good predictor as far as those things go. About the best free one you will find IMO.

 

[sjt18]

Those other things you mention aren't reduce-able to numbers but are very subjective. They really almost entirely on statistics. It's all in the blog above, you can read it yourself.

Maybe difficult to quantify but indispensable to accurately doing something like FPI.

The one that I do think is week is the 247 ratings. Those are subjective to begin with and often don't pan out, but they use them anyway. I guess it must work though or they manage to manipulate the numbers enough to make it work.

You don't have to read many of my posts to know I think the recruiting ratings are overplayed. That said, they have "general accuracy". IOW's, I doubt their accuracy for any given player but do tend to project the talent of 85 guys. That is far more legit than using previous years.

As for trusting them...in an earlier post you stated that they are making millions off the FPI, unless I misunderstood you. I don't see how...they make millions off of broadcasting games, and FPI is just one of many tools they use to generate interest in the games.

Don't recall saying millions but yeah they wouldn't do it if it didn't help them make more money. That makes any assessment they do of their own accuracy... biased as well.

As such, I don't see how misrepresenting how accurate it is helps them.

Because much like the votes of preseason poll voters or preseason projections by talking heads... they're never truly held accountable for being wrong. Their purpose really isn't accuracy as you allude to above. Their purpose is to generate interest in their "product".

To be clear, the numbers they present don't reveal much and could be easily manipulated to look better than they are, but at the same time they match my experience with FPI over the years, which is that it's a pretty good predictor as far as those things go. About the best free one you will find IMO.

Not sure what to tell you except that 57% LSU just turned into 40-13 UT. I didn't look but what was the FPI for UK/SC or Bama/TAM?