Best Method to Predict Future Strength of a Team

[In the Deed the Glory]

Beware, this is a pretty nerdy post.

 

 

If you want to see the data, here it is: https://spreadsheets.google.com/ccc?key=0Avzywsc6VKg9dFRaYS1VR0kzakZVLU1Gb0RWREpFTnc&hl=en&authkey=CPea6uAF

 

On a separate note, why can't we have dynamic redistricting, with rivalries protected, every few years, based on a formula?

 

 

Math CAN'T perdict the football future folks.

 

 

 

Well, that is true...

 

If you are going to be an asshat to somebody, maybe check your spelling first.

[deedsker15]

 

Math CAN'T perdict the football future folks.

 

 

you have never heard of edward lorenz have you?

 

man tried to set up a 12 formulas to explain weather patterns...ends up being able to predict chaotic behavior of large complex systems(kinda like college football teams in the NCAA)

 

oh yeah...math does predict the weather, eclipses, tides, chemical reactions...the list just keeps going...so why can't it predict basic trends in college football over large periods of time

[irafreak]

I do believe vegas does fairly well with their formulas...including sports gambling

[Huskerzoo]

It's interesting but what's the error you have going? Would a more complete analysis include recruiting ranks in the formula? I wonder how much total variance could be accounted for by various models.

[Igetbored216]

 

That's the goofiest stuff I think I've ever seen.

 

tell ya what...

 

EVEN WITH THE FULL VISION OF 20/20 HINDSIGHT and THE SPREADSHEET RIGHT IN FRONT OF YOU..

 

show me on those funny charts, where it predicts nebraska falling from grace...

 

where it predicts NU won't beat a top ten team on the road for a decade...

 

or where it predicts Boise State rising to the top ten...

 

or where it predicts USC cheating and getting caught.

 

or where it predicts the future qbs and rbs at LSU and their success.

 

how about where it predicts Bradford's injury and OU falling down?

 

Math CAN'T perdict the football future folks.

 

 

Ain't happening!

 

Seriously....

 

Wow. Really? You are freakin brilliant. Please look up the meaning of "statistics" because you obviously have no clue. I was going to respond further, taking advanced classes in statistics myself, but the post below sums everything up nicely.

 

It can if you are intelligent enough to know what you're talking about.

 

Events happen for a reason, and if you can prove that the same reason (other than luck) consistently is linked to an event, it's fairly easy to understand that the two are related (ie, correlated). It doesn't matter one bit WHY two things correlate, as most logical hypotheses (in this case) don't attempt to discover how something happens, but what the implications of its occurance are.

 

Like in our case, we have three variables: Success over the pasat 25 years, Success over the past 10 years, and Success over the next 10 years. 25 year success happens to correlate more with future success than 10 year success does; who cares why, it does. There's plenty of reasons one could guess are to attribute to this (good leadership within the coaches and athletic department, facilities, tradition), but that isn't what the math is even intending to prove. That's a lot more subjective and complicated.

 

And no, statistics could not have proved that Bradford would get injured, but again, you prove you do not understand enough about the subject to have a reasonable opinion. Statistics can only answer what happens to a set of data on average, not on an individual basis. The average life expectancy for females is about 7% longer than male life expectancy. Does that mean that when a husband outlives his wife that the world is full of crap? Of course not. Look at the big picture instead of pointing out every insignificant detail and maybe you'll understand more about how things work.

[BeltwayHusker]

It's interesting but what's the error you have going? Would a more complete analysis include recruiting ranks in the formula? I wonder how much total variance could be accounted for by various models.

 

 

Since I didn't actually try to build a predictive model, there's not really an error to test. All I did was determine whether 10-year or 25-year past history correlated better with success in the next 10 years, using the year 2000 as a reference point. As it turned out, the 25-year data had a better correlation.

 

Certainly, the deviations were still pretty big; on average, winning percentage was plus or minus 0.108 between the 25-year result and the result in the future 10 years (e.g., from 0.600 to 0.708 or to 0.492). The average deviation in winning percentage rank among 100 teams was about 22 (e.g., from 40th best up to 18th best or down to 62nd best). These are big swings, proving that this should not be used to predict the success of a given team.

 

I'm sure recruiting rankings would help predict success over the next few years to some extent, but I'm not sure how well they'd work for a 10-year future time frame. I also imagine that some sort of weighting system, whereby more remote years are weighted less heavily than recent years, could work better. But that seems like a lot of work for a model with pretty limited use... all I wanted to test was whether Iowa and Wisconsin's recent successes mean they should be treated as superior to Michigan and Penn State when trying to create competitively balanced divisions in the Big 10.

[Ponderosa]

It's interesting but what's the error you have going? Would a more complete analysis include recruiting ranks in the formula? I wonder how much total variance could be accounted for by various models.

 

 

Since I didn't actually try to build a predictive model, there's not really an error to test. All I did was determine whether 10-year or 25-year past history correlated better with success in the next 10 years, using the year 2000 as a reference point. As it turned out, the 25-year data had a better correlation.

 

Certainly, the deviations were still pretty big; on average, winning percentage was plus or minus 0.108 between the 25-year result and the result in the future 10 years (e.g., from 0.600 to 0.708 or to 0.492). The average deviation in winning percentage rank among 100 teams was about 22 (e.g., from 40th best up to 18th best or down to 62nd best). These are big swings, proving that this should not be used to predict the success of a given team.

 

I'm sure recruiting rankings would help predict success over the next few years to some extent, but I'm not sure how well they'd work for a 10-year future time frame. I also imagine that some sort of weighting system, whereby more remote years are weighted less heavily than recent years, could work better. But that seems like a lot of work for a model with pretty limited use... all I wanted to test was whether Iowa and Wisconsin's recent successes mean they should be treated as superior to Michigan and Penn State when trying to create competitively balanced divisions in the Big 10.

 

Would you be able to factor in coaches? Or at least a coaching change?

[General Blackshirt]

Beltway, what was the R-squared value comparing 10 years and 25 years?

[BeltwayHusker]

Beltway, what was the R-squared value comparing 10 years and 25 years?

 

For 10 years (1990-1999 vs. 2000-2009), it's 0.236.

For 25 years (1975-1999 vs. 2000-2009), its 0.272.

 

Weak effects, and arguably a wash between the two. But enough to say that a) there is some correlation between past and future success in general and b)25-year success is at least as good, if not better, than "recent" success.

 

But now we're getting super nerdy.

[clone]

 

That's the goofiest stuff I think I've ever seen.

 

tell ya what...

 

EVEN WITH THE FULL VISION OF 20/20 HINDSIGHT and THE SPREADSHEET RIGHT IN FRONT OF YOU..

 

show me on those funny charts, where it predicts nebraska falling from grace...

 

where it predicts NU won't beat a top ten team on the road for a decade...

 

or where it predicts Boise State rising to the top ten...

 

or where it predicts USC cheating and getting caught.

 

or where it predicts the future qbs and rbs at LSU and their success.

 

how about where it predicts Bradford's injury and OU falling down?

 

Math CAN'T perdict the football future folks.

 

 

Ain't happening!

 

Seriously....

 

Wow. Really? You are freakin brilliant. Please look up the meaning of "statistics" because you obviously have no clue. I was going to respond further, taking advanced classes in statistics myself, but the post below sums everything up nicely.

 

It can if you are intelligent enough to know what you're talking about.

 

Events happen for a reason, and if you can prove that the same reason (other than luck) consistently is linked to an event, it's fairly easy to understand that the two are related (ie, correlated). It doesn't matter one bit WHY two things correlate, as most logical hypotheses (in this case) don't attempt to discover how something happens, but what the implications of its occurance are.

 

Like in our case, we have three variables: Success over the pasat 25 years, Success over the past 10 years, and Success over the next 10 years. 25 year success happens to correlate more with future success than 10 year success does; who cares why, it does. There's plenty of reasons one could guess are to attribute to this (good leadership within the coaches and athletic department, facilities, tradition), but that isn't what the math is even intending to prove. That's a lot more subjective and complicated.

 

And no, statistics could not have proved that Bradford would get injured, but again, you prove you do not understand enough about the subject to have a reasonable opinion. Statistics can only answer what happens to a set of data on average, not on an individual basis. The average life expectancy for females is about 7% longer than male life expectancy. Does that mean that when a husband outlives his wife that the world is full of crap? Of course not. Look at the big picture instead of pointing out every insignificant detail and maybe you'll understand more about how things work.

 

 

statistics for the geniuses out there are numbers from the past. how many yards rushing did player X get, team x get, etc.

 

even those are flawed, because they don't account for home field, weather, turnovers, injuries or strength of opponents.

 

using them to predict future events, especially years out, when players, coaches, schedules, injuries and countless other unknowns can't possibly be factored in is....

 

<sarcasm> BRILLIANT!! CARRY ON! </sarcasm>

 

 

tell ya what...

 

please provide the AP top ten as they will finish in 2010.

 

when ya crunch those numbers, please provide the AP top ten as they will finish in 2011 and 2012.

 

let's see how accurately the OPs spreadsheets are.

 

predict (perdict) away.

 

if laughing at something as preposterous as trying to predict how teams will fare in the future using stats ...

 

is trolling

 

you folks need some help.

[zoogs]

I have a funny feeling that Texas will be more successful for the next twenty years than Central Michigan.

 

Why? Based on recent and semi-recent performance.

 

C'mon, you can't really say you have NO idea what will 'probably' happen in the future? There's such thing as 'reasonable expectations', which are all these and any 'predictions' are, within a certain confidence interval.

 

Nobody's claiming to know exactly what will happen, so demanding that is sort of silly.

 

So, this would suggest that it's reasonable (though of course no guarantee) to expect Michigan and Penn State to win more than Iowa and Wisconsin over the next 10 years, despite their better records in the last 10 years.

 

And while models can only be so good at taking in all the variables, they pretty often use 'past data' to predict 'future trends', so...really not sure where you are coming from, or if you are serious.

 

Also I'm no statistician but those look like REALLY weak R^2 values, which you'd probably expect from the small sample size...which on the whole, means the predictions probably don't amount to that much outside of gut feeling, which is there's no reason to suspect some of the traditional winningest programs in college football will continue to be outshone by these 'other' schools. But maybe it really is the end of an era; who knows?

 

I will say that clone's post is just an opinion, albeit strongly worded, and there's no need to call him a troll for voicing it.

[huKSer]

statistics for the geniuses out there are numbers from the past. how many yards rushing did player X get, team x get, etc.

 

even those are flawed, because they don't account for home field, weather, turnovers, injuries or strength of opponents.

 

using them to predict future events, especially years out, when players, coaches, schedules, injuries and countless other unknowns can't possibly be factored in is....

 

<sarcasm> BRILLIANT!! CARRY ON! </sarcasm>

 

 

tell ya what...

 

please provide the AP top ten as they will finish in 2010.

 

when ya crunch those numbers, please provide the AP top ten as they will finish in 2011 and 2012.

 

let's see how accurately the OPs spreadsheets are.

 

predict (perdict) away.

 

if laughing at something as preposterous as trying to predict how teams will fare in the future using stats ...

 

is trolling

 

you folks need some help.

**sigh**

 

The failure of the American education system

[Lil' Red]

 

That's the goofiest stuff I think I've ever seen.

 

tell ya what...

 

EVEN WITH THE FULL VISION OF 20/20 HINDSIGHT and THE SPREADSHEET RIGHT IN FRONT OF YOU..

 

show me on those funny charts, where it predicts nebraska falling from grace...

 

where it predicts NU won't beat a top ten team on the road for a decade...

 

or where it predicts Boise State rising to the top ten...

 

or where it predicts USC cheating and getting caught.

 

or where it predicts the future qbs and rbs at LSU and their success.

 

how about where it predicts Bradford's injury and OU falling down?

 

Math CAN'T perdict the football future folks.

 

 

Ain't happening!

 

Seriously....

 

Wow. Really? You are freakin brilliant. Please look up the meaning of "statistics" because you obviously have no clue. I was going to respond further, taking advanced classes in statistics myself, but the post below sums everything up nicely.

 

It can if you are intelligent enough to know what you're talking about.

 

Events happen for a reason, and if you can prove that the same reason (other than luck) consistently is linked to an event, it's fairly easy to understand that the two are related (ie, correlated). It doesn't matter one bit WHY two things correlate, as most logical hypotheses (in this case) don't attempt to discover how something happens, but what the implications of its occurance are.

 

Like in our case, we have three variables: Success over the pasat 25 years, Success over the past 10 years, and Success over the next 10 years. 25 year success happens to correlate more with future success than 10 year success does; who cares why, it does. There's plenty of reasons one could guess are to attribute to this (good leadership within the coaches and athletic department, facilities, tradition), but that isn't what the math is even intending to prove. That's a lot more subjective and complicated.

 

And no, statistics could not have proved that Bradford would get injured, but again, you prove you do not understand enough about the subject to have a reasonable opinion. Statistics can only answer what happens to a set of data on average, not on an individual basis. The average life expectancy for females is about 7% longer than male life expectancy. Does that mean that when a husband outlives his wife that the world is full of crap? Of course not. Look at the big picture instead of pointing out every insignificant detail and maybe you'll understand more about how things work.

 

 

statistics for the geniuses out there are numbers from the past. how many yards rushing did player X get, team x get, etc.

 

even those are flawed, because they don't account for home field, weather, turnovers, injuries or strength of opponents.

 

using them to predict future events, especially years out, when players, coaches, schedules, injuries and countless other unknowns can't possibly be factored in is....

 

<sarcasm> BRILLIANT!! CARRY ON! </sarcasm>

 

 

tell ya what...

 

please provide the AP top ten as they will finish in 2010.

 

when ya crunch those numbers, please provide the AP top ten as they will finish in 2011 and 2012.

 

let's see how accurately the OPs spreadsheets are.

 

predict (perdict) away.

 

if laughing at something as preposterous as trying to predict how teams will fare in the future using stats ...

 

is trolling

 

you folks need some help.

 

I agree that numbers can't predict games and I wish the BCS could comprehend that. But numbers can give us predictions and a general idea on what may happen. College football is filled with numbers we analyze all the time such as rankings, yards, strength of schedule, and etc. So although numbers can't predict actual results they can give us general ideas and all this thread was saying is that a team that has been good since 2000 is less likely to be as good the next ten years as a team that has been good from 1975-1999.

[clone]

I have a funny feeling that Texas will be more successful for the next twenty years than Central Michigan.

 

Why? Based on recent and semi-recent performance.

 

C'mon, you can't really say you have NO idea what will 'probably' happen in the future? There's such thing as 'reasonable expectations', which are all these and any 'predictions' are, within a certain confidence interval.

 

Nobody's claiming to know exactly what will happen, so demanding that is sort of silly.

 

So, this would suggest that it's reasonable (though of course no guarantee) to expect Michigan and Penn State to win more than Iowa and Wisconsin over the next 10 years, despite their better records in the last 10 years.

 

And while models can only be so good at taking in all the variables, they pretty often use 'past data' to predict 'future trends', so...really not sure where you are coming from, or if you are serious.

 

Also I'm no statistician but those look like REALLY weak R^2 values, which you'd probably expect from the small sample size...which on the whole, means the predictions probably don't amount to that much outside of gut feeling, which is there's no reason to suspect some of the traditional winningest programs in college football will continue to be outshone by these 'other' schools. But maybe it really is the end of an era; who knows?

 

I will say that clone's post is just an opinion, albeit strongly worded, and there's no need to call him a troll for voicing it.

 

have a funny feeling that Texas will be more successful for the next twenty years than Central Michigan.

 

Why? Based on recent and semi-recent performance.

 

yup. and I bet ya didn't need a spreadsheet. you picked a pretty obvious comparison!!

 

but if you say that texas will be more successful for the next twenty years than Boise st or Ohio St or Nebraska....

 

ya just might be wrong.

 

even with some sort of mathmatical formula. the more closely matched the compared teams are the more important all the variables are that come into play.

 

you can probably apply some common sense and football savvy and do better short term.

 

 

long term.... it becomes a crapshoot. who wants a predictable college football future anyway. what fun would that be??

[General Blackshirt]

even those are flawed, because they don't account for home field, weather, turnovers, injuries or strength of opponents.

 

using them to predict future events, especially years out, when players, coaches, schedules, injuries and countless other unknowns can't possibly be factored in is....

 

Actually, you're completely wrong about this. It's actually quite easy to "adjust" any type of statistic based on any particular variable. See Park Factor as a general example of how baseball does the exact same thing that you are claiming is a load of crap. If you can adjust a statistic for park in baseball, it's really not that difficult to adjust for it in football; same goes with weather, injuries, strength of schedule, etc.

 

 

Also I'm no statistician but those look like REALLY weak R^2 values, which you'd probably expect from the small sample size...

 

Correct and correct. Beltway said it best: yeah, the correlation isn't strong, and maybe the difference between the two isn't enough to be significant...but a correlation still exists, which is pretty interesting. Maybe Athletic Directors should consider this a bit more before firing a coach who isn't winning MNCs at schools with no tradition (Hello ISU, talking to you here).