Math to pick a starter. (1 Viewer)

[th3f00l]

Okay, first of all I am not asking for advice on who to start.

I couldn't decide between who I thought should start with Tony Romo on a BYE. My options were Mark Sanchez or Matt Hasselbeck. I chose to use math to predict their fantasy performance. What are your opinions on this method for predicting points? (please limit any responses about who will have the bigger day to variables that can be defined mathematically, I know what Rex said)

Hasselbeck: Averaging 17 points against defenses that cumulatively allow an average of 14.1875 points to quarterbacks.

Pittsburgh: Allowing an average of 9.75 points to quarterbacks that cumulatively score an average of 12.5 points.

so if Hasselbeck is averaging 120% of the average points against a defense, that puts his numbers at 11.68 against PIT's 9.75 avg.

and if PIT is allowing 78% of a QB's average points scored that estimates them to allow 13.26 points to Hass's 17 point avg.

The mean of these two numbers is 12.47. That is what I would predict for Hass using this method.

Sanchez: Averaging 13 points against defenses that cumulatively allow an average of 13.625 points to quarterbacks.

Patriots: Allowing an average of 21 points to quarterbacks that cumulatively score an average of 15.3125 points.

Sanchez 95.4% * 21 = 20.03

NE: 137% * 13 = 17.83

(20.03+17.83) / 2 = 18.93. The predicted number for Sanchez.

Using this equation I could clearly see that Sanchez has the mathematical advantage.

 

[shadyridr]

this has gotta be shtick. were u the same guy who wanted to set ur lineups based on refs?

 

[th3f00l]

this has gotta be shtick. were u the same guy who wanted to set ur lineups based on refs?

 

[ready5]

you realize those teams have played different defenses, right?

 

[th3f00l]

you realize those teams have played different defenses, right?

I averaged each of the defenses points allowed to QB's that the players respectively played. The defenses that Hass has played against (Jac, Bal, Den, Cle) allowed an average of 14.1875 points to quarterbacks. The defenses that Sanchez has played against (Dal. Jac, Oak, Bal) allowed an average of 13.625 points to quarterbacks.

 

[Warrior]

...I like it

Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.

That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

 

[warpedone]

Seems as good a way as any.

What do the numbers look like in games they were favored and/or won vs. games where they were dogs and/or lost? Might be a relationship between those and how often the (had to) pass.

 

[th3f00l]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

If you are talking about the Patriots and Pittburgh defenses that is precisely what I did. The Patriots have allowed 37% more points than a QB's average, and the Steelers have allowed 22% less.

 

[SoBeDad]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

If you are talking about the Patriots and Pittburgh defenses that is precisely what I did. The Patriots have allowed 37% more points than a QB's average, and the Steelers have allowed 22% less.

This is precisely why most football fans would start Sanchez over Hasselback, even though he's scored fewer points than Hasselback.

 

[th3f00l]

Seems as good a way as any.What do the numbers look like in games they were favored and/or won vs. games where they were dogs and/or lost? Might be a relationship between those and how often the (had to) pass.

I'll have to figure out where I can plug that into the equation. I would have to analyze trends form previous seasons, and find an average of point difference for players by each position in games that they are favored and/or winning, as well as games they are predicted to lose and/or losing. I assume that there is a definite point difference in favor of QB/WR stats playing from behind, and RB playing with a lead.

 

[ready5]

you realize those teams have played different defenses, right?

I averaged each of the defenses points allowed to QB's that the players respectively played. The defenses that Hass has played against (Jac, Bal, Den, Cle) allowed an average of 14.1875 points to quarterbacks. The defenses that Sanchez has played against (Dal. Jac, Oak, Bal) allowed an average of 13.625 points to quarterbacks.

and you realize those different defenses played against different quarterbacks, right?univariate analysis is a minefield and playing fast and loose with arbitrary averages is a great way to slam into a simpson's paradox.

 

[th3f00l]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

If you are talking about the Patriots and Pittburgh defenses that is precisely what I did. The Patriots have allowed 37% more points than a QB's average, and the Steelers have allowed 22% less.

This is precisely why most football fans would start Sanchez over Hasselback, even though he's scored fewer points than Hasselback.

I also wanted to test the opposite. How much more/less a QB was scoring against a defense's avg vs QB. I may try modifying and testing this equation on select players throughout the season and see how close it comes to predicting actual stats.

 

[CapnJB]

Not sure if anyone noticed, but Hasselbeck has the same number of letters as Pittsburgh and if you mix up the letters in 'Matt Hasselbeck' it spells 'The clam baskets'. And we all know what that means.

 

[ready5]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

and then take those previous defenses and compare them to other quarterbacks, etc, etc, ad infinitum.and don't forget about injuries, gameplans, and the weather.

 

[Warrior]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

If you are talking about the Patriots and Pittburgh defenses that is precisely what I did. The Patriots have allowed 37% more points than a QB's average, and the Steelers have allowed 22% less.

Maybe the defenses that Hass or Sanchez have faced went up against creampuff QB's the other 3 weeks? Maybe they went up against three stud QB's?That might skew things a little IF the other QB's they faced were Brady, Rodgers and Brees...if it's pretty similar, I wouldn't worry about it.In other words, you say "Team A, B, C, D all gave up 15.75 points to other QB's on average". Well, maybe those QB's were Grossman, Tarvaris, Campbell and Painter. In which case, the results are skewed.

 

[Warrior]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

and then take those previous defenses and compare them to other quarterbacks, etc, etc, ad infinitum.and don't forget about injuries, gameplans, and the weather.

exactly!

 

[th3f00l]

you realize those teams have played different defenses, right?

I averaged each of the defenses points allowed to QB's that the players respectively played. The defenses that Hass has played against (Jac, Bal, Den, Cle) allowed an average of 14.1875 points to quarterbacks. The defenses that Sanchez has played against (Dal. Jac, Oak, Bal) allowed an average of 13.625 points to quarterbacks.

and you realize those different defenses played against different quarterbacks, right?univariate analysis is a minefield and playing fast and loose with arbitrary averages is a great way to slam into a simpson's paradox.

I thought about analyzing per each QB's average and how they fared against each defense, then I would also have to analyze each of the defenses that played against each of those QB's and so on and so on. Then I would eventually have a number for every player/defense which would change every time I got back to them. Do you have any suggestions for analyzing the data in a more pragmatic manner.

 

[ready5]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

and then take those previous defenses and compare them to other quarterbacks, etc, etc, ad infinitum.and don't forget about injuries, gameplans, and the weather.

exactly!

also referees, sarcasm, and the assumption that the past is the future.

 

[Sarnoff]

this has gotta be shtick. were u the same guy who wanted to set ur lineups based on refs?

What? He's trying to come up with a mathematical reasoning to project points. When Dodds predicts Tom Brady to score 23.7 points, you think he's pulling them out of one of those rotating drums from a bingo hall?

 

[th3f00l]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

If you are talking about the Patriots and Pittburgh defenses that is precisely what I did. The Patriots have allowed 37% more points than a QB's average, and the Steelers have allowed 22% less.

Maybe the defenses that Hass or Sanchez have faced went up against creampuff QB's the other 3 weeks? Maybe they went up against three stud QB's?That might skew things a little IF the other QB's they faced were Brady, Rodgers and Brees...if it's pretty similar, I wouldn't worry about it.In other words, you say "Team A, B, C, D all gave up 15.75 points to other QB's on average". Well, maybe those QB's were Grossman, Tarvaris, Campbell and Painter. In which case, the results are skewed.

Noted. So to make this analysis more accurate I would need to assign a definitive average for each QB and defense in the NFL. To achieve this average would I need a method that goes more in depth? Would it be entirely false to pull data only form this season? I ask, because I feel like the differences in the final number, of predicted points, may be slight. But since you were willing to humor me I may be willing to test this.

 

[ready5]

you realize those teams have played different defenses, right?

I averaged each of the defenses points allowed to QB's that the players respectively played. The defenses that Hass has played against (Jac, Bal, Den, Cle) allowed an average of 14.1875 points to quarterbacks. The defenses that Sanchez has played against (Dal. Jac, Oak, Bal) allowed an average of 13.625 points to quarterbacks.

and you realize those different defenses played against different quarterbacks, right?univariate analysis is a minefield and playing fast and loose with arbitrary averages is a great way to slam into a simpson's paradox.

I thought about analyzing per each QB's average and how they fared against each defense, then I would also have to analyze each of the defenses that played against each of those QB's and so on and so on. Then I would eventually have a number for every player/defense which would change every time I got back to them. Do you have any suggestions for analyzing the data in a more pragmatic manner.

that recursion is half the problem, and the other half is that you're assuming that any given defense is identical week-to-week.there are simply way too many moving parts to do this sort of analysis, especially if you want to switch from description of the past to prediction of the future. who says that struggling defense can't improve? who says coaches always employ the exact same philosophy/personnel week after week? just because something happened in the past doesn't necessarily mean it therefore will happen again in the future, especially if we're only considering a single variable.

 

[th3f00l]

you realize those teams have played different defenses, right?

I averaged each of the defenses points allowed to QB's that the players respectively played. The defenses that Hass has played against (Jac, Bal, Den, Cle) allowed an average of 14.1875 points to quarterbacks. The defenses that Sanchez has played against (Dal. Jac, Oak, Bal) allowed an average of 13.625 points to quarterbacks.

and you realize those different defenses played against different quarterbacks, right?univariate analysis is a minefield and playing fast and loose with arbitrary averages is a great way to slam into a simpson's paradox.

I thought about analyzing per each QB's average and how they fared against each defense, then I would also have to analyze each of the defenses that played against each of those QB's and so on and so on. Then I would eventually have a number for every player/defense which would change every time I got back to them. Do you have any suggestions for analyzing the data in a more pragmatic manner.

that recursion is half the problem, and the other half is that you're assuming that any given defense is identical week-to-week.there are simply way too many moving parts to do this sort of analysis, especially if you want to switch from description of the past to prediction of the future. who says that struggling defense can't improve? who says coaches always employ the exact same philosophy/personnel week after week? just because something happened in the past doesn't necessarily mean it therefore will happen again in the future, especially if we're only considering a single variable.

This same data is used to create a Strength of Schedule chart, and determine what match ups are "juicy". My goal here was to create a way to view the data and make a more accurate analysis of it, and use that to assist in predicting a players performance. More speculative things such as a coaches philosophy, and a variable's potential improvement, would be taken in to account were I to make a final prediction. However, For the sake of trying to figure out an equation, what is the best route?

 

[Warrior]

...I like it Now you just have to take the QB's that played against those defenses previously, and figure out how many points they scored against that defense, compared to other defenses...to see if these defenses held those QBs to more or less points.That will give you an idea of how well these defenses did against solid QBs, against poor QBs, etc. etc.

If you are talking about the Patriots and Pittburgh defenses that is precisely what I did. The Patriots have allowed 37% more points than a QB's average, and the Steelers have allowed 22% less.

Maybe the defenses that Hass or Sanchez have faced went up against creampuff QB's the other 3 weeks? Maybe they went up against three stud QB's?That might skew things a little IF the other QB's they faced were Brady, Rodgers and Brees...if it's pretty similar, I wouldn't worry about it.In other words, you say "Team A, B, C, D all gave up 15.75 points to other QB's on average". Well, maybe those QB's were Grossman, Tarvaris, Campbell and Painter. In which case, the results are skewed.

Noted. So to make this analysis more accurate I would need to assign a definitive average for each QB and defense in the NFL. To achieve this average would I need a method that goes more in depth? Would it be entirely false to pull data only form this season? I ask, because I feel like the differences in the final number, of predicted points, may be slight. But since you were willing to humor me I may be willing to test this.

Unless you want to go reeeeeaaaally deep down the rabbit hole, I suggest only going this one more step. Pull these QB names (3 for each side) and see if they stack up reasonably well against each other. If they do, stop there. Go with Sanchez. If one set of three is much better than the other set, you may either be more sure of your current hypothesis or you may be back to square one.

 

[GregR]

Personally I would want my decision to also take into account things like the effect on the game plan of the loss of Kenny Britt and the likely return to football shape of Chris Johnson who has now had the equivalent of a training camp's worth of time back with the team. Hasselbeck threw the ball 34, 42 and 36 times in the first 3 games. He threw it 20 times with Britt out of the lineup and with CJ rushing for over 100 yards (compared to 24, 53, and 21 the first three games for CJ).

 

[Dinsy Ejotuz]

Pretty sure I can save you some time and tell you to use FBGs weekly projections.

In addition to the stuff that's been mentioned in this thread, I think they actually simulate the games many 1000s of times, using actual injury and suspension info, in order to account for the different possible game flows before making the projection.

Going to be pretty hard to beat that process, especially since they've been doing it for a long time now.

 

[th3f00l]

Personally I would want my decision to also take into account things like the effect on the game plan of the loss of Kenny Britt and the likely return to football shape of Chris Johnson who has now had the equivalent of a training camp's worth of time back with the team. Hasselbeck threw the ball 34, 42 and 36 times in the first 3 games. He threw it 20 times with Britt out of the lineup and with CJ rushing for over 100 yards (compared to 24, 53, and 21 the first three games for CJ).

Of course taking all of these things into account is important.

 

[pittstownkiller]

Not sure if anyone noticed, but Hasselbeck has the same number of letters as Pittsburgh and if you mix up the letters in 'Matt Hasselbeck' it spells 'The clam baskets'. And we all know what that means.

 

[th3f00l]

Pretty sure I can save you some time and tell you to use FBGs weekly projections.In addition to the stuff that's been mentioned in this thread, I think they actually simulate the games many 1000s of times, using actual injury and suspension info, in order to account for the different possible game flows before making the projection.Going to be pretty hard to beat that process, especially since they've been doing it for a long time now.

I can also buy a birdhouse. But isn't it fun to make something sometimes.ETA: I hope that didn't sound rude. It was my way of saying that I am doing this for fun.

 

[ready5]

you realize those teams have played different defenses, right?

I averaged each of the defenses points allowed to QB's that the players respectively played. The defenses that Hass has played against (Jac, Bal, Den, Cle) allowed an average of 14.1875 points to quarterbacks.

The defenses that Sanchez has played against (Dal. Jac, Oak, Bal) allowed an average of 13.625 points to quarterbacks.

and you realize those different defenses played against different quarterbacks, right?univariate analysis is a minefield and playing fast and loose with arbitrary averages is a great way to slam into a simpson's paradox.

I thought about analyzing per each QB's average and how they fared against each defense, then I would also have to analyze each of the defenses that played against each of those QB's and so on and so on. Then I would eventually have a number for every player/defense which would change every time I got back to them. Do you have any suggestions for analyzing the data in a more pragmatic manner.

that recursion is half the problem, and the other half is that you're assuming that any given defense is identical week-to-week.there are simply way too many moving parts to do this sort of analysis, especially if you want to switch from description of the past to prediction of the future. who says that struggling defense can't improve? who says coaches always employ the exact same philosophy/personnel week after week? just because something happened in the past doesn't necessarily mean it therefore will happen again in the future, especially if we're only considering a single variable.

This same data is used to create a Strength of Schedule chart, and determine what match ups are "juicy". My goal here was to create a way to view the data and make a more accurate analysis of it, and use that to assist in predicting a players performance. More speculative things such as a coaches philosophy, and a variable's potential improvement, would be taken in to account were I to make a final prediction. However, For the sake of trying to figure out an equation, what is the best route?

accurate analysis? analyzing data isn't the problem, it's the presumption that past performance guarantees future results. so if by accuracy you mean predictive power, then we've still got a problem... and any post-hoc tweaks you make is proof that the analysis was meaningless or you picked the wrong variable (or should've gone multivariate, etc).and why do you think improvement is speculative? you're using averages because performance varies week-to-week, right? if there's an upward trend, that's called improvement and it means that a mere average is probably understating things going forward.

and i think you misunderstood what i meant by philosophy/personnel if you think it's also speculative. key injuries matter, so coaches adjust. different opponents are different, so coaches adjust. sometimes touchdowns are thanks to dumb luck, so coaches adjust. pulling an average of that chaos and cramming it through an equation just isn't going to work.

also, watch out for black swans. you never know what new wrinkle some innovative coach is going to debut -- and it's a 'new' wrinkle because it's not in your data so your statistical analysis can't even see it.

 

[Ignoratio Elenchi]

Okay, first of all I am not asking for advice on who to start.

I couldn't decide between who I thought should start with Tony Romo on a BYE. My options were Mark Sanchez or Matt Hasselbeck. I chose to use math to predict their fantasy performance. What are your opinions on this method for predicting points? (please limit any responses about who will have the bigger day to variables that can be defined mathematically, I know what Rex said)

Hasselbeck: Averaging 17 points against defenses that cumulatively allow an average of 14.1875 points to quarterbacks.

Pittsburgh: Allowing an average of 9.75 points to quarterbacks that cumulatively score an average of 12.5 points.

so if Hasselbeck is averaging 120% of the average points against a defense, that puts his numbers at 11.68 against PIT's 9.75 avg.

and if PIT is allowing 78% of a QB's average points scored that estimates them to allow 13.26 points to Hass's 17 point avg.

The mean of these two numbers is 12.47. That is what I would predict for Hass using this method.

Sanchez: Averaging 13 points against defenses that cumulatively allow an average of 13.625 points to quarterbacks.

Patriots: Allowing an average of 21 points to quarterbacks that cumulatively score an average of 15.3125 points.

Sanchez 95.4% * 21 = 20.03

NE: 137% * 13 = 17.83

(20.03+17.83) / 2 = 18.93. The predicted number for Sanchez.

Using this equation I could clearly see that Sanchez has the mathematical advantage.

Hasselbeck scored 16.4 points.Sanchez scored 18.3 points.

Conclusion