Improving aging curves

[Crow]

http://arxiv.org/pdf/1403.7548v2.pdf

Who had seen previously?
Who sees benefits and will apply?

[Crow]

First part on baseball, power hitters age different from others. Second half in basketball. Three aging clusters presented. Some difference on peaking and major inflection points. Looked at position representation in clusters but not much distinction. I think it might help to look at more clusters, more stats. Looking at offensive and defensive bpm and rpm might be interesting.

[willguo]

Age curves in general don't account for dropping out of sample, which is a huge problem that there's no real solution to.

Using any 1# metric to value performance is unideal due to context, but even worse would be to use a cumulative metric, as that's hugely dependent on playing time. Weighted OBA somewhat has this problem too - a platooned player with a higher OBA is not necessarily a better hitter.

Clustering to test for cumulative Win Shares is also curious - ideally, you want age progression to be a % increase/decrease of a rate stat, and not have the clusters be grouped such that one is a much better sample of another. They normalized OBA to avoid this problem, and something similar would be smart.

I'm not sure if I'm expressing myself correctly here - the power hitters vs non-power hitters selection in baseball is unlike anything that they tried in basketball. If we're studying aging in basketball, the output needs to address two things:

1) What role is this player playing now, and what are the probabilities that he's ready for a larger one, or will be forced into a smaller one?

2) What are the expected production of this player in each of those roles?

You obviously need to strip away injuries (not a big deal) and team context (uh oh...) to get a meaningful answer.

[Mike G]

Since 1978, there are 300 players with >20,500 RS minutes.
1/3 of these have OWS > 1.7 X DWS -- Offensively inclined players
Another 1/3 have OWS <.95 x DWS -- Mostly (relative to all players) defensive players.
http://bkref.com/tiny/njIbS

The middle 1/3 would have .95 < OWS/DWS < 1.7 -- Relatively balanced O/D players

It may be that offense-oriented players tend to get fewer (or more) career minutes than defensive players; so these O/D ratios wouldn't define perfect thirds among all player minutes. But it might reveal a distinction in 'type' of player whose skills decline faster or slower than another type.

----

Quickie analysis, part 2:
After age 30 (since 1978), 151 players have totaled 10,000 minutes.
56 of these -- 37% -- are in the 'defensive' group.
52, or 34.4% are from the 'offensive' 1/3.
This leaves just 28.6% from the most balanced 1/3.
http://bkref.com/tiny/sHgoP

Since these are not full careers, but just over-30 years, some players will have jumped a threshold into another classification.
It's not very plausible that more-versatile players are less valuable after a given age; but it may be that they last longer when they decide to concentrate on one end of the floor or the other.

[Crow]

Thanks for the slicing. This dataset is worth thinking about, examining further.

[Mike G]

part 3 --
Since 1978, 50 players have, after age 35, amassed 5000 regular-season minutes.
- Offensiver players -- OWS/DWS > 1.7 -- number just 14 of them: Kareem, KMalone, Stockton, Reggie, Dale Ellis, Nash, Gilmore, English, Schrempf, HoGrant, JNewman, AMiller, Moses, AMason.
- Defensive-minded players, with O/D < .95, are almost half of the sample, with 23. By most minutes, Clifford Robinson, Kidd, Willis, BigE, Deke, Oak, Fish, Dream, Ewing ...
- Leaving 13 middle managers. Mostly Grant Hill, Payton, Ray Allen, Perkins,

[Crow]

I recall reading that defensive performance peaked several (3?) years later than offense but am not confident of my memory for the exact figures.

[Crow]

If you felt inclined, it might be helpful to know the exit rates for offensive and defensive metric performance at x sds from average or below replacement level. My guess is that negative performance on offense causes more and faster exits than comparable metric performance on defense.

[Crow]

I can do a version of this. From 2006-7 thru 2012-13, I looked at last seasons / exits for players with -2 or worse on OBPM, DBPM and both. There were 363 exits with -2 or worse on offense, 147 on defense. But 103 were worse on both making it, 260 below on offense only, 103 both and just 44 on defense only. Of course this says something about DRPM probably pulling many, perhaps many too many, toward the middle but the almost 6-1 ratio of offensive weakness to defense is almost certainly telling a true story, with only the degree uncertain.