Somewhere past the middle of last season, I tried some math to see how good all the young Mavericks (OJ Mayo, Darren Collison, Jae Crowder, Bradan Wright) would likely end up being at their peak for the SBNation Mavericks site that I also write for. What I finally ended up doing was vaguely interesting, but limited, and I couldn’t run the numbers for Crowder or Wright due to sample size issues.
Recently, I figured out a way to do a similar thing for both Crowder and Wright, and it seems to have worked about as well as predictive stats crunching like this can, and it seems like a great tool to use to try and predict the potential of the Grizzlies’ young talent as well.
The Way it Works
What I realized, is that just looking at a player’s PER and then saying “well, he’s had PERs of 16, 14, 15, 21, and 22, so if I punch those numbers into a calculator I’ll get a vaguely applicable graph/equation” ends up being difficult because we have no idea when that player will actually peak.
So, for all we know, the calculator will just spit out an equation that increases until Mike Conley or Ed Davis has a PER of 35 in his 13th year in the league. As wonderful as that would be, I think we can all agree that it’s wildly unlikely.
According to some wonderful basketball statisticians on Basketball Reference’s blog, the average basketball player actually peaks at the age of 27 with a standard deviation of 3.6. Meaning that 2/3 of players peak between the ages 24 ½ to 30 ½.
Age is obviously not the only factor in peak performance, though; experience and context play a huge role. Some players just peak when they move to a team that fits them better, or after getting over injury troubles, or after getting more playing time. Others come into the league either late or early and just learn as they play more and more.
Context is impossible to predict mathematically, in this case, so I’m assuming that it’ll be accounted for in the error, and at any rate, it’ll show up again later, I promise.
Regarding experience, though, after combing Basketball Reference for all players who played more than ten seasons, I found that most players actually peaked after 5.8 seasons in the league with a standard deviation of 2.2 (meaning that 2/3 of players peak between the 3.6 years in the league and 7 years).
What I did, then, using the average peak age and experience data, is I plugged the players’ growth1 from one year to another, along with the probable peak season, into my calculator to get a basic equation describing the players’ growth. Using that equation, I could then do just a little bit of calculus to get a projected career performance arc.
I did this for PER, Net Efficiency in Points per 100 Posessions (Points scored per 100 possessions minus Points allowed per 100 possessions), and Win Shares per 48 minutes. I used those three because they’re all “comprehensive” metrics to tell us how good a player is, but they all tell us slightly different things.
PER tries to tell us how good a player is, objectively, but is still remarkably bad at accounting for defense. As well, because PER doesn’t really take minutes into account, it’s bad at telling us the impact a player has on a team. Net Efficiency, on the other hand, does tell us about defense and impact, but since it’s basically just an upgraded +/- system, it does a very poor job of telling us just how good the player is, isolated on his own merits. Win Shares is a nice mix of the two, really.
To test that this would, in fact, work, I tried it on a player who’s already retired: the Mavericks’ Mark Aguirre. I chose Aguirre because he was a particularly good case for this analysis, since he peaked at right around the average point in both age and experience. As well, he spent most of his time on one team and without major injury, which helps reduce context error. I only used Aguirre’s stats from his first 4 seasons (one less than Jerryd Bayless has played, by the way). Then, I compared my projection to his actual career stats.
Here are the projections for all three metrics, compared to his real stats in all of those categories:
This one worked pretty well, I think: notice that it projects his peak PER and how long he maintains that peak to within .6. It doesn’t predict the sharp drop-off in Aguirre’s 8th season, but at least it predicts the downslide starting in that season.
This one seemed to be the least helpful of the group, I think, but it still has some real value. Aguirre’s actual net efficiency only peaked at 5, which is really a mediocre team impact. 20+ would be the typical superstar impact. So, in the spectrum of efficiencies, the projected peak does actually a decent job of letting us know that he was both in the negative in impact for awhile, and that even at his peak, efficiency wasn’t his strong point. Efficiency in general is hardest to project for, because it’s the most context dependent out of these stats.
To be able to say that these all of these pairs of graphs are describing the same relationship, the correlation coefficient for each would need to be at least 40%. To be able to say that these two graphs are definite, absolute projections, the correlations would need to be above 90%.
For PER, the correlation was 70.79%. For Efficiency, it was 71.70%. For Win Shares, it was 83.69%.
It’s flawed, but it has some pretty darn good value as a predictor of peak production.
The Niche Role Player: Jerryd Bayless
Bayless is a prime example of the kind of player whose stats have been affected by context: in his five seasons in the NBA, he’s played for three whole different teams, and his first season’s stats were so bad that I had to adjust the starting point on the EFF and Win Shares stats to make Bayless “peak” at the right time.
Nonetheless, Bayless is a great player to start this with, because he’s such a simple role player for the squad. He’s logged enough years that the stats can be reliable and he doesn’t do too much outside of his role to throw the stats off. He’s the epitome of a role player: he’s a gunner.
Here were Bayless’ regressions, with comparisons to his real stats for the 5 seasons he’s played:
I would say that I might be surprised if Bayless could peak at a PER of 17 or above if he hadn’t done it already. I think this makes sense for Bayless: to get better, he primarily needs to learn how to pass more while turning it over less, he needs to shoot just a little better, and he needs to get to the line more.
Next season, with smarter backup point guards and more floor spacing, he should be primed to make all of those personal leaps. As a result, I don’t think that it would be unreasonable to expect to get a 16 PER out of him for the next few years, but I also think we can look for him to peak soon. Look for Bayless to be a really good role player/sixth man for the next couple years, but not more than that, or not for any longer.
This betrays how bad Bayless’ defense is. Up until his last two seasons, his defense has been so bad that his high-octane offense was not enough to compensate and make his play “valuable.” With the Grizzlies, there’s the obvious benefit of having the league’s best defenses behind him. So, to that point, while he’s playing on this team, expect his offense to be (marginally) more valuable than his bad defense is detrimental.
The nice thing about the win share graph is that it’s a little more encouraging towards Bayless’ overall impact to the team than the efficiency stats are. That said, 13 Win Shares per 48 minutes — while good — is certainly not world shaking.
The takeaway from this, regarding Bayless, is that he is (and will continue to be) a very good and talented individual player and gunner. His team impact, however, is very limited due to his terrible defense and limited court vision. Conveniently, he projects to still have a positive impact for the next few years while he is with the Grizzlies. So, that’s good news.
So there’s your projection for Jerryd Bayless; check back tomorrow to see how Quincy Pondexter projects to be!