NBA Aging: PER Curves

[Latrinsorm]

I've been thinking about Paul George and Kawhi Leonard recently. They each had debutante 7 games series against Miami last year: 19 6 5 on 48% and 15 11 1 on 51% while semi-credibly guarding LeBron. (He still put up 29 7 5 on 51% and 25 11 7 on 45% against them, but that counts.) This led many to proclaim that both were poised to explode into superstardom this year, including a literally incredible article claiming George (http://bleacherreport.com/articles/1880078-which-superstar-would-you-rather-build-around-kevin-durant-or-paul-george) was a better option to build around than Kevin God Damn Durant.

But then... neither really got there.

George improved by 4 points per game but lost a rebound, half an assist, and bumped his poor FG% by a mere .005. Overall, his PER increased 3.3 but his USG increased 4.8.

Leonard's improvement was more overall but less dramatic: 1 point, 0.2 boards, 0.4 assists, .03 FG% to get above 50%. His composite metrics fared better: +3 PER to only +1.9 USG. He also missed 14 games due to a broken finger.

Both PERs are right around 20, and I feel like 25 PER is a pretty reasonable ask for a perimeter superstar. If we look at all players in the past 10 years to put up 2 or more such seasons, we get 12 guys, 7 of whom are listed at 6'10" or under: LeBron, Wade, Paul, Durant, Kobe, Love, Amar'e. Of those, 6 put up their first 25+ PER season by their 4th season - Kobe is a huge outlier in his 7th. 5 of the 7 did so aged 23 or younger - Kobe and Wade were both 24. These cutoffs are interesting because George is going into his 5th season, age 24. It wouldn't be totally unprecedented for him to get there, but 79% of those who became superstars were already there. Kawhi's will be going into season 4 age 23 so he's got more time, only trying to catch up with 64%.

That metric is pretty small, though, so I also wondered what an overall PER aging curve would look like. This is tricky because as I've mentioned in the past there are 3 basic types of NBA careers: 1 year, 3 year, and 11 year. A 1 year player is lousy no matter what his age, that's why he only plays 1 year. If we're talking about superstars, we're talking about guys who stick around. There are the Bill Waltons of the world, but the guys who don't suffer catastrophic injury are far more common. Thus I decided to take any guy who had at least 10 1000+ MP seasons in his career. I'm not sure how many that is yet, but I took the 50 retired players who fit the bill and got this...

http://img.photobucket.com/albums/v456/johnnyoldschool/NBAAgingPERCurve_zps6e6ca4a1.png

Mixed news for George: average player PER doesn't hit an extended peak until 25, but it also only goes up by 2.5 points from the age 23 season. I'm pretty pleased how smooth the curve is already without having to factor in injuries, usage, playing time, etc. I've got a lot more data to get through, then we can do things like see if Paul George starting off at 13.0 PER hurts him, or if Kawhi only improving by 2.8 from 1st year to 3rd hurts him. Too much ground to make up or going too slow to do so, in other words.

Full data:


19 19.6 1
20 39.7 2
21 192.8 11
22 597.9 37
23 765.4 46
24 859.9 49
25 937.7 50
26 936.1 49
27 943.3 50
28 915.2 48
29 934.1 49
30 923.2 50
31 876.2 48
32 884.7 50
33 864.3 50
34 815.1 49
35 719.9 45
36 588.5 38
37 429.2 28
38 266.5 18
39 134.8 8
40 70.7 4
41 23 2
42 13 1
I only graphed the bins with 20+ count pretty much arbitrarily.

[Latrinsorm]

I got out to 150 fellas:

http://img.photobucket.com/albums/v456/johnnyoldschool/NBAAgingPerCurve2_zps5af6f08b.png

Clearly, that first high point tells us we should go by years in league rather than age. The players playing that young are doing so because they're elite, so they skew the average high. When we convert to experience, we get:

http://img.photobucket.com/albums/v456/johnnyoldschool/NBAAgingPerCurve3_zps112028f7.png

Much better! There are a couple little hitches left, but that's probably just noise. The length of peak (defined as within 5% of maximum) is 24-30 by age and 4-9 by year, but year 10 is really close at 94.96% - the metrics should only substantially differ at the fringes rather than the densely populated peak, so that's good news. I think this is a goodly amount of data, next I'll check out the effect of first year performance on the curve.

[Latrinsorm]

First thing is in the experience graph I didn't adjust to ignore partial years. When I do it looks like this:

http://img.photobucket.com/albums/v456/johnnyoldschool/NBAAgingPerCurve4_zpscabd6bf8.png

Which is really nice. You can still see hitches in the second derivative but we're gonna have to live with that.

.

It occurred to me that it would also be interesting to look at the guys who were in the third career distribution but on the short end rather than the long, so I grabbed 100 of them and got this:

http://img.photobucket.com/albums/v456/johnnyoldschool/NBAAgingPerCurve5_zps15d8e30e.png

And relative to their peaks:

http://img.photobucket.com/albums/v456/johnnyoldschool/NBAAgingPerCurve6_zps8382236f.png

Pretty stark. They start off lower and reach their peak sooner, but each peak is about as long (5 years) and the decays are pretty similar too. The difference is that because they don't get as high, 80% of their peak apparently isn't good enough to play in the NBA. And speaking of never getting as high, only one player out of the 100 to play less than 10 seasons of 1000+ MP ever managed a 25+ PER: Zelmo Beaty, and even that comes with an asterisk because he did it in the ABA.

Which brings us back to Paul George and his first year 13.0 PER. If we give him the benefit of the doubt and assume he fits into the long curve better than the short curve, the average such player will see a 1.36x higher peak, or 17.7. He's obviously ahead of that already, but how high might he go? The highest in terms of multiplication in this 150 player sample was 2.10... Karl Malone went from a 13.7 first year PER to a peak of 28.9, and his fourth year PER was 24.4. There was only one other player to have a multiplier as high as 25/13... Cliff Robinson went from a 9.3 to an 18.3.

Another way of thinking about it is to look at the average player who had between a 13 and 14 PER. It turns out there are 22 such players in the sample, which was more than I would have guessed. They reached an average peak of 20.0, almost exactly PG's current peak of 20.1. However, none of those players peaked before their 5th year, and on average it took them 7 years to peak, so he's got that going for him.

[Geijon Khyree]

Sidebar, but Lebron is basically 5 points away from having, potentially, a 4-peat on his hands. Jordan was a Winner, but James is a better baller.

[Atlanteax]

http://espnfivethirtyeight.files.wordpress.com/2014/07/silver-datalab-carmelo-heattable.png?w=610&h=800

Wow @ Karl Malone consistently hovering around 17, and Kobe was consistently solid green as well...

[Latrinsorm]

Yeah, Malone wrecks the curve for everyone.

If I were a Knicks fan I'd be extremely worried. A metric that Kobe performs well in is one that is extremely forgiving to volume scoring, which is Carmelo's only marquee skill. If even that metric predicts Carmelo will do poorly, yikes.

As stated in my analysis I think years of experience make a better model than years of age, but the point that Carmelo is either at or just past his peak stands.