Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Sept 3, 2012 9:52:01 GMT -8
He has an it is an experiment. he can go back to the old way any time. If hsi special gteams shape up he will. Thye wer getting way too many blocked kicks in practice low tragectory He alos is abamboozler and will do what he can ot screw up the oppiosition using the press,particularily the naive kids at Rivals and Scout. HIs quick kick was downed on the seven. I do not know what you guys are thinking. Rocky is a sly fox and uses all sorts of tricks to get an advantage. His chart tells hmi when to kick when not to,so tha tmeaqns sometimes he will kick extra points. If the Kickers are amking them in practice. he also like the go for itmentality on offense to help with fourth down conversions.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Sept 3, 2012 12:10:29 GMT -8
If you're a good offensive team with a good OL you should be able to make a 2-point conversion 50% of the time or better. Oregon seems to think so. Boise goes for it a lot more than other teams also. And everything over 50% becomes "gravy" points which you would never get unless you go for it. For example, assume you score 4 TD's with 4 PAT = 4 points while 3 of 4 two-point conversions would be 6 points. This. Even if you make it 50% of the time you'll get the same amount of points. As therealeman said, empirical data is only worthwhile if you have a sufficient sample size. Also, you have to control for opponent quality. Converting a two-point PAT against New Mexico is a helluva lot easier than doing so against TCU. How good is Army at defending the two-pointer? I have no idea but I have a feeling we're going to find out. BTW, has it occurred to anyone else that in college ball the defense can earn two points by taking a turnover on a two-point conversion all the way back the other way? I don't recall ever seeing that happen on a botched kick but I've seen it a dozen times on an intercepted pass. Just sayin.
|
|
|
Post by aardvark on Sept 3, 2012 19:22:47 GMT -8
If Rocky is truly saying he can't recruit and coach a kicker to perform PAT's and chip shot FG's at a D1 level, than that is Rocky not doing his job as a head coach. I don't want to knock on Long after a the first game against an opponent that was obviously better, but I have to agree with this statement. He's not even really putting it on the players, but is instead relying on a chart for in-game calls. The thing about the type of empirical data is that Rocky is using, is that you need large samples to get to that "50% of the time" he is talking about when going for the 2 point conversion. With that in mind, we would need large samples in the game as well, and two shots at the endzone is not enough "data". Not considering the variables he is totally ignoring like QB performance, O-Line production, environment, etc, which no mathematical formula will ever account for. In his defense, though he did say that it was a week-to-week strategy. Something tells me he will kick a field goal or two next week; at least give the effing kicker a shot. I would rather the Aztec placekicker kick about 8 extra points.
|
|
|
Post by Strawberry Puppy Kisses on Sept 3, 2012 19:49:41 GMT -8
This. Even if you make it 50% of the time you'll get the same amount of points. As therealeman said, empirical data is only worthwhile if you have a sufficient sample size. Also, you have to control for opponent quality. Converting a two-point PAT against New Mexico is a helluva lot easier than doing so against TCU. How good is Army at defending the two-pointer? I have no idea but I have a feeling we're going to find out. BTW, has it occurred to anyone else that in college ball the defense can earn two points by taking a turnover on a two-point conversion all the way back the other way? I don't recall ever seeing that happen on a botched kick but I've seen it a dozen times on an intercepted pass. Just sayin. There are a lot of different ways to look at the problem. Hair et. al. (2006) argues that the ratio between independent variables in the variate should be, at a minimum, 5:1. So, five observations for each predictor variable (i.e, the variables you don't care about re: the outcome). In my experience, between 15 and 20 observations for each independent variable yields a more narrow model in terms of the alpha used. So, if you had fifteen variables (e.g., weather, time, home/away, mean OL weight, etc.) plus an extra Boolean variable called "converted 2pt?", you'd need a sample size of at least 225 games to produce a sufficient sample of training data for the model. Controlling for opponent quality isn't the right way to look at it. Training your model against logical variables to measure quality is the correct approach. Thinking about this, I would argue that there is probably enough data and variables to develop a simple multivariate regression model that would be statistically significant to an alpha of .10. Another way to interpret that - my first guess is that Rocky has a statistical simulator that produces a type II error (i.e., "wrong" answer) 10% of the time. Beyond that, alphas down to .01 or even .05 would probably be overfit to the training data... and unreliable.
|
|
|
Post by freshtodeath on Sept 3, 2012 20:02:49 GMT -8
So which mathematical formula did this dude use? Can someone post the exact formula so we can see if this formula is for real or is it just bullsheet?
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Sept 3, 2012 20:03:08 GMT -8
Puppy, you're talking way over my head. I haven't studied that stuff since like 1980 when in grad school and have never used it since. However, you obviously do and I'll certainly take your word for it that Rocky probably has sufficient reliable data that we can conclude our kickers are so bad they can't be counted on to convert more than 81% of their PATs - or am I overstating it to say that?
|
|
|
Post by k5james on Sept 3, 2012 20:08:30 GMT -8
As therealeman said, empirical data is only worthwhile if you have a sufficient sample size. Also, you have to control for opponent quality. Converting a two-point PAT against New Mexico is a helluva lot easier than doing so against TCU. How good is Army at defending the two-pointer? I have no idea but I have a feeling we're going to find out. BTW, has it occurred to anyone else that in college ball the defense can earn two points by taking a turnover on a two-point conversion all the way back the other way? I don't recall ever seeing that happen on a botched kick but I've seen it a dozen times on an intercepted pass. Just sayin. There are a lot of different ways to look at the problem. Hair et. al. (2006) argues that the ratio between independent variables in the variate should be, at a minimum, 5:1. So, five observations for each predictor variable (i.e, the variables you don't care about re: the outcome). In my experience, between 15 and 20 observations for each independent variable yields a more narrow model in terms of the alpha used. So, if you had fifteen variables (e.g., weather, time, home/away, mean OL weight, etc.) plus an extra Boolean variable called "converted 2pt?", you'd need a sample size of at least 225 games to produce a sufficient sample of training data for the model. Controlling for opponent quality isn't the right way to look at it. Training your model against logical variables to measure quality is the correct approach. Thinking about this, I would argue that there is probably enough data and variables to develop a simple multivariate regression model that would be statistically significant to an alpha of .10. Another way to interpret that - my first guess is that Rocky has a statistical simulator that produces a type II error (i.e., "wrong" answer) 10% of the time. Beyond that, alphas down to .01 or even .05 would probably be overfit to the training data... and unreliable. Dude, all I see is this...
|
|
|
Post by Strawberry Puppy Kisses on Sept 4, 2012 6:07:11 GMT -8
Puppy, you're talking way over my head. I haven't studied that stuff since like 1980 when in grad school and have never used it since. However, you obviously do and I'll certainly take your word for it that Rocky probably has sufficient reliable data that we can conclude our kickers are so bad they can't be counted on to convert more than 81% of their PATs - or am I overstating it to say that? Sorry - hope my response didn't come off condescending. The topic fits an area of expertise and I ran with it. There is no way we can know exactly what Rocky has concluded without looking at the data and/or tools used by the coaching staff. If we see 2pt conversions against Army again, it's safe to say that he's drawn a conclusion about the season-long yield of going for two. Should that happen I'll definitely be more interested in the data... I'd actually like to work with it and see if my conclusions differ
|
|
|
Post by hbaztec on Sept 4, 2012 6:29:24 GMT -8
Puppy, you're talking way over my head. I haven't studied that stuff since like 1980 when in grad school and have never used it since. However, you obviously do and I'll certainly take your word for it that Rocky probably has sufficient reliable data that we can conclude our kickers are so bad they can't be counted on to convert more than 81% of their PATs - or am I overstating it to say that? Sorry - hope my response didn't come off condescending. The topic fits an area of expertise and I ran with it. There is no way we can know exactly what Rocky has concluded without looking at the data and/or tools used by the coaching staff. If we see 2pt conversions against Army again, it's safe to say that he's drawn a conclusion about the season-long yield of going for two. Should that happen I'll definitely be more interested in the data... I'd actually like to work with it and see if my conclusions differ Well, it was not condescending. It is actually the right response to the thread title.
|
|
|
Post by Section T(urn Up) on Sept 4, 2012 8:00:24 GMT -8
Puppy, you're talking way over my head. I haven't studied that stuff since like 1980 when in grad school and have never used it since. However, you obviously do and I'll certainly take your word for it that Rocky probably has sufficient reliable data that we can conclude our kickers are so bad they can't be counted on to convert more than 81% of their PATs - or am I overstating it to say that? Sorry - hope my response didn't come off condescending. The topic fits an area of expertise and I ran with it. There is no way we can know exactly what Rocky has concluded without looking at the data and/or tools used by the coaching staff. If we see 2pt conversions against Army again, it's safe to say that he's drawn a conclusion about the season-long yield of going for two. Should that happen I'll definitely be more interested in the data... I'd actually like to work with it and see if my conclusions differ What scares me is it's hard to imagine Rocky working his way through this in an objective manner. It sure sounds like he just thinks "hey, my kickers suck, go for 2 and if we make it 50% of the time it's the same!" Which, of course, is accurate (assuming our kickers suck) so I don't necessarily fault him. However, if the kickers suck, that's on him too. The buck stops with Rocky and what a lot of people call "honesty" just seems like that thing kids start doing when they realize there's less punishment for straight up admitting to breaking things--or outing themselves for wrong behavior.
|
|
|
Post by AztecBill on Sept 4, 2012 9:05:01 GMT -8
If we score a touchdown with 2 seconds left to tie the game and Rocky goes for 2, then I will agree with everything being said. There are 4 main variables needed to decide whether to go for the first down or opt for one of the foot related choices.
Rocky's Chart contains the first two variables. 1. Yards to go 2. Field Position
But there are two others 3. Time to go 4. Score differential
No "chart" can be used to determine a line of action when 4 variables are important to the decision.
Rocky needs an app - not a chart.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Sept 4, 2012 9:09:53 GMT -8
Sorry - hope my response didn't come off condescending. The topic fits an area of expertise and I ran with it. There is no way we can know exactly what Rocky has concluded without looking at the data and/or tools used by the coaching staff. If we see 2pt conversions against Army again, it's safe to say that he's drawn a conclusion about the season-long yield of going for two. Should that happen I'll definitely be more interested in the data... I'd actually like to work with it and see if my conclusions differ Well, it was not condescending. It is actually the right response to the thread title. I didn't take it that way, either. In fact, reading about the alpha and that stuff opened up some door in the innermost portion of my pea brain that hadn't been opened in years. If I hadn't sold my statistics book back I would have used it to refresh my memory about that stuff. I mean, I'm sure it's just like riding a bike. Once you learn it, it all comes back immediately. (Yeah, right.)
|
|
|
Post by k5james on Sept 4, 2012 9:12:27 GMT -8
Well, it was not condescending. It is actually the right response to the thread title. I didn't take it that way, either. In fact, reading about the alpha and that stuff opened up some door in the innermost portion of my pea brain that hadn't been opened in years. If I hadn't sold my statistics book back I would have used it to refresh my memory about that stuff. I mean, I'm sure it's just like riding a bike. Once you learn it, it all comes back immediately. (Yeah, right.) Pffft, I burried that door under useless football and beer stuff many moons ago.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Sept 4, 2012 10:10:37 GMT -8
Pffft, I burried that door under useless football and beer stuff many moons ago. I took that class when you were just out of diapers so I've got an excuse for not being able to remember something that old. No excuse for you. Well, too much beer might qualify . . .
|
|
|
Post by TheSanDiegan on Sept 4, 2012 11:53:36 GMT -8
Can you base your decisions this year from statistical information from last several years considering you have a different quarterback and a different kicker? Don't you have to analyze the current roster statistical data? Billy beane thinks Rocky is crazy. The short answer is yes, and yes. You can use old data as input in your decision making process, but it will not result in good decisions if your data does not accurately reflect your current performance capability. However, as SPK has alluded to, we do not have enough data to accurately assess the 'goodness' of Rocky's statistical model. Furthermore, an effective model should be adaptable to a given set of circumstances. Even in the absence of quantifiable variability, there should still exist Boolean tests to determine when to use the model, and when used, if and when to tailor it. IMO Rocky's quotes follwing the game demonstrated the model lacks maturity in this regard.
|
|
|
Post by Strawberry Puppy Kisses on Sept 4, 2012 12:01:36 GMT -8
Can you base your decisions this year from statistical information from last several years considering you have a different quarterback and a different kicker? Don't you have to analyze the current roster statistical data? Billy beane thinks Rocky is crazy. The short answer is yes, and yes. You can use old data as input in your decision making process, but it will not result in good decisions if your data does not accurately reflect your current performance capability. However, as SPK has alluded to, we do not have enough data to accurately assess the 'goodness' of Rocky's statistical model. Furthermore, an effective model should be adaptable to a given set of circumstances. Even in the absence of quantifiable variability, there should still exist Boolean tests to determine when to use the model, and when used, if and when to tailor it. IMO Rocky's quotes follwing the game demonstrated the model lacks maturity in this regard. That's exactly right. A model built to work with a specific data set is overfit. The simple method is to obtain data and partitition it three ways - training, validation, and test data sets. If your model shows no statistical significance against the test data set, you have no measure to express confidence the model will work on actual data. There are a lot of possiblities here and the discussion makes me interested in seeing what conclusions I can derive on my own. But the essence of what we care about is what Rocky is looking at... I wish I had that data and understood the analysis performed. Think he'll share it with me if I ask nicely?
|
|
|
Post by untitled on Sept 4, 2012 12:09:06 GMT -8
well if we're going for two every time we'd better come up with some better damn plays
|
|
|
Post by AztecBill on Sept 4, 2012 12:11:27 GMT -8
The short answer is yes, and yes. You can use old data as input in your decision making process, but it will not result in good decisions if your data does not accurately reflect your current performance capability. However, as SPK has alluded to, we do not have enough data to accurately assess the 'goodness' of Rocky's statistical model. Furthermore, an effective model should be adaptable to a given set of circumstances. Even in the absence of quantifiable variability, there should still exist Boolean tests to determine when to use the model, and when used, if and when to tailor it. IMO Rocky's quotes follwing the game demonstrated the model lacks maturity in this regard. That's exactly right. A model built to work with a specific data set is overfit. The simple method is to obtain data and partitition it three ways - training, validation, and test data sets. If your model shows no statistical significance against the test data set, you have no measure to express confidence the model will work on actual data. There are a lot of possiblities here and the discussion makes me interested in seeing what conclusions I can derive on my own. But the essence of what we care about is what Rocky is looking at... I wish I had that data and understood the analysis performed. Think he'll share it with me if I ask nicely? That is not the correct method to use in situations like this. Coaches have to make assumptions. Those assumptions must be used as input to whatever decisions they come to. Coach "A" Assumptions: Everyone kicks extra points so it must be correct. Rocky's assumptions: We will make a 2 point conversions against Washington at a 43% rate. So if we are down 2 TDs, I will go for 2 instead of 1.
Things change too fast and there are too many variables that are not constant to use the method you describe. This is called real life. You can't use some soft mamsy pamsy social science model of life. This isn't science, it is game theory.
|
|
|
Post by AztecBill on Sept 4, 2012 12:24:14 GMT -8
well if we're going for two every time we'd better come up with some better damn plays We will go for 2 when it is the correct thing to do. So far we haven't run across any situation where kicking was correct.
|
|
|
Post by Strawberry Puppy Kisses on Sept 4, 2012 12:49:00 GMT -8
That's exactly right. A model built to work with a specific data set is overfit. The simple method is to obtain data and partitition it three ways - training, validation, and test data sets. If your model shows no statistical significance against the test data set, you have no measure to express confidence the model will work on actual data. There are a lot of possiblities here and the discussion makes me interested in seeing what conclusions I can derive on my own. But the essence of what we care about is what Rocky is looking at... I wish I had that data and understood the analysis performed. Think he'll share it with me if I ask nicely? That is not the correct method to use in situations like this. Coaches have to make assumptions. Those assumptions must be used as input to whatever decisions they come to. Coach "A" Assumptions: Everyone kicks extra points so it must be correct. Rocky's assumptions: We will make a 2 point conversions against Washington at a 43% rate. So if we are down 2 TDs, I will go for 2 instead of 1.
Things change too fast and there are too many variables that are not constant to use the method you describe. This is called real life. You can't use some soft mamsy pamsy social science model of life. This isn't science, it is game theory. You're a good guy and enjoy RPI calculations. You seem to have a genuine interest in application of some mathematical derivations relative to sports. I respect you, in that regard. But, Bill, you're stepping all over your logic. Further, you're making inane statements relative to my comments. You stated above that using the foundation of data mining, i.e. establishment of supervised learning for multivariate models, is not the correct method to use in situations "like this." Further, you can't use "mamsy pamsy social science" here. I can't tell if you're being ignorant on purpose... or if you truly think that analysis of data using multivariate statistical methods qualifies as social science. Game theory, as you alluded to, is rooted in the correlation of data equillibrium and spatial data analysis. Such a technique is extremely valuable in practical data mining application. It is also realized as part of the inception done through unsupervised learning, i.e., the association rules that derive the correlations for your multivariate model! There is linear progression here in the logic - are you sure you understand what I'm talking about? Your post above is so incredibly inane that, given you seem like a reasonably intelligent person, you must be simply misunderstanding. Right?
|
|