Drinen's Double-Edged Sword (1 Viewer)

[gheemony]

On the PFR Blog, Drinen wrote a great article about aging curves/patterns for each position.

I can't get a date on the article, but based on the use of Brad Meester as the example, I assume it's relatively old. Does anyone know of an update? Has anyone done their own update of the data?

Thanks.

 

[JamesTheScot]

An interesting article.

I'm not being a smart ### by asking this, please understand that clearly...

...Is this knowledge helpful when drafting a team?

As Drinnen acknowledges, "Betting on a player to buck a strong trend is fine, as long as you have a specific reason for doing so". But at the level where most of us play, you always have several reasons for expecting a player to improve or worsen. Injuries and opportunity changes being HUGE factors.

Ward might get more carries this season than he ever has before. Westy and LT, if healthy, could rebound from their 2008 numbers and surpass someone heading into their peak years.

It seems to me that the caveat to the rule, having a "reason" to buck the trend, always boils down to the multiple variables in play every off season. Which means the caveat essentially swallows up the rule.

On draft day, you are not choosing between LT circa '05 and LT circa '07 with your pick, you're choosing between different RB's who are in completely different team situations and those situations are themselves in flux from the previous year. Like LT circa '07 and Frank Gore circa '07.

It reminds me of the old saw that the stock market has never lost money over a 10+ year ride. Except that individual stocks can easily do so. And if you win or lose based upon how you did that particular year in isolation, the trick is very much so determining which ones will and won't during that particular year. So the old saw doesn't really provide any guidance.

Is it possible to use the findings in the article to reconstruct a draft strategy for past seasons and then test to see is the strategy held any advantage when predicting the next season? By that I mean that a % of increase or drop-off should be determinable for each year of aging. So if we apply that % to every RB in a fantasy draft based upon the previous season's results the rule should predict situations where an older RB who scored more points in the previous season would be surpassed by a younger RB on the upswing who would then be a better pick at the same draft position. We could them look at what happened each successive year to determine if the % of increase or drop-off as predicted had any correlation to the actual results. If so, then there may be an appreciable predictive value to the rule. If not, then we are back to square one.

 

[gheemony]

It' is useful to MY drafting strategy, especially in a dynasty league. For example, I'll project a PPG for each player. Using Drinen's research, I'll extrapolate that two PPG for future years. You can then do a net present value of those future years (e.g., you may expect zero fantasy relevance from a QB when he turns 40). Up to each drafter how they weight these factors.

Will there be exceptions? Sure, someone like Favre would have blown up my expected value of him a few years back. But in the long run, it helps me make decisions about long-term value of players.

Another example is that I'll use Drinen's numbers to determine whether a player will see a decline in PPG in the coming year. For example, if 31-year old RB scored 11 ppg last year, and Drinen's formula suggests a 20% drop off in production this year, I'll think long and hard before I project 11 ppg or higher for that player this year. This is my first year using it, but I'll probably take 08 PPG stats and apply Drinen's research to it to come up with a baseline projection for this year. I may have reasons to increase or decrease that projection, but at least I have a place to start.

A good caveat on all of this is fantasy football is game that involves a lot of luck. Let's assume you have a 10% chance of winning a 10-team league. Optimizing your draft strategy may get you to 12 or 13%. If so, the improvement is really in the margins. Is it worth doing? Depends on each person. Why read this thread or this forum? Presumably you enjoy it and/or think it gives you an advantage. Whether it's worthwhile to factor in age ... I think it is.

 

[CalBear]

I think these kinds of studies are fundamentally flawed--or at least, the way people use them is fundamentally flawed. They start with valid (but generally obvious and expected) populational behaviors, and try to use them to predict future performance of individuals. That's not how it works. The fact that the population of 31-year-old RBs had an average 20% decline in PPG wouldn't say anything at all about whether Curtis Martin or Tiki Barber would have a 20% decline in PPG at age 31. They're individuals, not populations.

Look at it this way; if you measured the income of the population of 65-year-olds, you'd find a populational decline in year N+1. But could you use that to predict that Warren Buffett would make less money at age 66? No; Warren Buffett was still completely functional as the CEO of Berkshire Hathaway, and projections for his income that year should have been based on his own merits, not the behavior of the general population.

 

[gheemony]

I think these kinds of studies are fundamentally flawed--or at least, the way people use them is fundamentally flawed. They start with valid (but generally obvious and expected) populational behaviors, and try to use them to predict future performance of individuals. That's not how it works. The fact that the population of 31-year-old RBs had an average 20% decline in PPG wouldn't say anything at all about whether Curtis Martin or Tiki Barber would have a 20% decline in PPG at age 31. They're individuals, not populations.Look at it this way; if you measured the income of the population of 65-year-olds, you'd find a populational decline in year N+1. But could you use that to predict that Warren Buffett would make less money at age 66? No; Warren Buffett was still completely functional as the CEO of Berkshire Hathaway, and projections for his income that year should have been based on his own merits, not the behavior of the general population.

I think your example of Warren Buffet is good because it does highlight how this type of information can be misapplied. I'm not proposing that anyone apply this information blindly to every player. But how is it flawed to include this as one factor in analyzing a player's future performance? Age seems like an obvious factor to include. I think baseball sabremetricians have done a good job of doing so and I think football could benefit by doing the same. Just don't over do it.

 

[CalBear]

I think your example of Warren Buffet is good because it does highlight how this type of information can be misapplied. I'm not proposing that anyone apply this information blindly to every player. But how is it flawed to include this as one factor in analyzing a player's future performance? Age seems like an obvious factor to include. I think baseball sabremetricians have done a good job of doing so and I think football could benefit by doing the same. Just don't over do it.

Statistics apply to baseball much differently than they apply to football. In baseball there are many events which involve only a single individual, and in fact, the sabremetricians tend to discount events which involve multiple individuals. (For example, they dismiss RBIs as a meaningless stat because they depend on game state and other offensive players). A player's batting average (or rather, on-base and slugging percentages, if you want to get all sabremetric on it) is dependent on almost nothing other than his own performance; therefore you can measure his performance, and make predictions about it.The only situation in football that looks at all like a player at the plate is field goal kicking. You can probably do sabremetrics on field goal kickers relative to age (although you'll run into the fact that the data sets are way, way smaller). But RBs and QBs? There are a million conflating factors, which means the predictions are harder and that the source data has a ton of noise on top of that. Among the many problems with the source data is that it includes all RBs in all situations. Many RBs are no longer the full-time starter at 30, so if you're looking at Tomlinson, who is guaranteed to be the full-time starter, you pretty much have to discount all the data for anyone who wasn't the full-time starter. The fact that Lamont Jordan didn't have a good year at age 30 really has no relevance for whether Tomlinson will; Jordan wasn't a starting RB in the league at that point.

 

[JamesTheScot]

I think these kinds of studies are fundamentally flawed--or at least, the way people use them is fundamentally flawed. They start with valid (but generally obvious and expected) populational behaviors, and try to use them to predict future performance of individuals. That's not how it works. The fact that the population of 31-year-old RBs had an average 20% decline in PPG wouldn't say anything at all about whether Curtis Martin or Tiki Barber would have a 20% decline in PPG at age 31. They're individuals, not populations.

Look at it this way; if you measured the income of the population of 65-year-olds, you'd find a populational decline in year N+1. But could you use that to predict that Warren Buffett would make less money at age 66? No; Warren Buffett was still completely functional as the CEO of Berkshire Hathaway, and projections for his income that year should have been based on his own merits, not the behavior of the general population.

I think your example of Warren Buffet is good because it does highlight how this type of information can be misapplied. I'm not proposing that anyone apply this information blindly to every player. But how is it flawed to include this as one factor in analyzing a player's future performance? Age seems like an obvious factor to include. I think baseball sabremetricians have done a good job of doing so and I think football could benefit by doing the same. Just don't over do it.

With respect, Gheemoney, how do you apply it if not blindly?By that I mean, how do you determine when age will cause a decline and when it won't?

I guess that my concern is, as CalBear stated, the results from one year to the next are so dependant upon a multitude of variables, trying to look a population trend where so many variables are in play and then correlate that age is the cause of the trend seems too much of a reach to quantify in any meaningful way that can be applied to an individual.

If the numbers mean anything, shouldn't they have some predictive value that could be tested against seasons that have already occured? If not, all we are left with is the general proposition that a player's potential declines as he ages with no real "math" application to help us predict how this will impact any particular player comparative to any other player.

If some of the FBG's use this in their rankings, I'd love to know how they incorporate it in.