FanPost

Enes Kanter Projections

With apologies to vjl110 who has written several superior articles to this one in his "Projecting:" series...

All Wolves fans know that with David Kahn, where there's draft rumor smoke, there's probably fire.  Since multiple sources have reported the Wolves are interested in Enes Kanter, I decided the other day to try to talk myself into the pick.  Since I'm not a professional scout, like many posters on Hoopus I really primarily on draft sites and quantitative data to get a feeling for prospects, and the oft-repeated problem with evaluating Kanter this way is that these sites have no data about his performance.

It turns out, however, that there ARE some data about Kanter's performance in competitive basketball games.  In particular, at fibaeurope.com, we can find 17 games and 524 minutes worth of data about Kanter's performance in the 2008 and 2009 European Men's under 18 championships.  His performance there was very good: Per 36 minutes, Kanter averaged about 22 points on 0.627 TS%, 18 rebounds, 1 assist, 1.5 blocks, 1 steal, 2 fouls and 3 turnovers.  Considering the level of competition, what do these numbers mean, and do they have any relevance to the NBA?

First: Kanter wasn't just good in these 17 games, he was dominant.  His overall raw production is better than any other player in the online history of the tournament - fibaeurope has box score stats for the top division going back to 1996 and no player (out of 1,874) has posted a higher Win Scores / minute than Kanter did in 2008.  The next closest player to Kanter's 25.4 WS/40 that year was Valanciunas in 2009 and 2010, with 23.8 and and 24.2, respectively; and besides Kanter's 20.6 in 2009, only two other players have posted a WS/40 above 20 (E. Lorbek and M. Raduljica, with 21.9 and 20.5, respectively).  That's pretty dominant.

Fine, you say, but any player who can play in the NBA should dominate against this level of competition.  Well, it turns out that 15 players 6'9" or taller did play in the tournament (during the period that stats are available) and later log minutes in the NBA: Alexis Ajinca, Andris Biedrins, Omri Casspi, Semih Erden, Pau Gasol, Marc Gasol, Victor Khryapa, Andrei Kirilenko, Kosta Koufos, Maciej Lampe, Ian Mahinmi, Dirk Nowitzki, Johan Petro, Darius Songaila, and Ronnie Turiaf.   None of them dominated the way Kanter did.  But more importantly, this gives us the opportunity to see if we can learn something about Kanter's likely NBA performance.

Forecasting Kanter's NBA Stats

Using the data from FIBA and BB-ref career per 36 averages for each of the 15 players above, I produced the following prediction for Kanter's TS% and Per 36 numbers:

TS% PTS/36 REB/36 STL/36 BLK/36 AST/36 TOV/36 PF/36
0.576 16.8 10.1 1.2 1.6 1.4 2.4 3.8

 

Scoring - True Shooting Percentage and Points

Here's a plot of each player's TS% in the under-18 championships versus the players TS% in the NBA.  Two "outliers" were removed: Erden, whose TS% increased dramatically (from 0.44 to 0.59) due to a drastic reduction in usage; and Lampe, who had the most extreme drop in TS% in the data set (from 0.54 to 0.44) for no reason that I could see.

The best-fit line (shown in the image) has an r^2 of about 0.4, but a p-value of 0.04, so U18 TS% is statistically significant but only explains about 40% of the variation in NBA TS% among this sample.  In any case, if you plug in Kanter's 0.627 TS% to the linear model found by regression, you get a predicted NBA TS% of 0.576. 

The regression for points has an even worse r^2:

I think the relatively low r^2 in both of these models can be explained primarily by variation in usage: most, but not all, of these players had high usage in the U18 tournament but they had a wider range of usage in the NBA. Most of the points above the best-fit line represent players with consistent usage between the two leagues, while it is more of a mixed bag below the best-fit line. Since I would expect Kanter to have high usage in the NBA (due to being drafted early) I think the model gives a reasonable prediction for his TS% and scoring.

Rebounds 

Kanter's 18 Rebounds per 36 minutes is off the charts both among the players in the NBA sample and all players with at least 40 minutes in the U18 tournament since 1996 - the next highest is Valanciunas with 15.  So the model may be underpredicting his rebounds/36 at 10.1.  (This model excludes one outlier, Biedrins, who somehow saw his rebounds/36 increase by 50% - from 8 to 12 - between the U18 tournament and the NBA, despite playing his first season in the NBA only a few months after playing in the tournament)

 

Assists, Steals,  and Blocks

Several of the older players in the sample appeared to have no blocks data (or just had really bad luck collecting no blocks in 100-200 minutes across 7-9 games) but otherwise these are pretty straightforward translations with pretty good r^2 values:

To show better detail, I truncated one point from the Steals plot: AK47 is way off to the right at 5.4 steals/36 in U18 play. (His NBA STL/36 of 1.6 almost perfectly fits the model)

Kanter's numbers are right  around the median for these stats: they don't look great or terrible.  (Although given his high usage the AST/36 numbers are a little low.  He might be kind of a black hole)


Turnovers and Fouls

Linear models for TOV/36 and PF/36 had a low r^2, mostly because there does not seem to be much variation between the players in the sample at the NBA level: they all had between 1.6 and 2.4 TO/36 (the r^2 for the "best fit" line here was 0.02) so to be safe I went with the maximum value among the NBA players.  The model for PF/36 looks like this:

Like TS%, turnovers and, to a lesser extend fouls, will depend on usage.  Since Kanter already had a high usage rate in the U18 tournaments and we would expect him to have a high NBA usage rate based on being picked early in the lottery, his PF/36 may slightly exceed the 3.8 predicted by the model.

Exploring Sample Size

While 500+ minutes and 17 games is not exactly a small sample, it's not huge either; and 15 players are a pretty small set to build a translation model from.  In order to estimate the potential for error in the model due to sample size, I used resampling ("the bootstrap") to compute confidence intervals.  The 10th percentile of predictions for each stat came out like this:

TS% PTS/36 REB/36 STL/36 BLK/36 AST/36 TOV/36 PF/36
0.530 13.5 9.2 0.8 1.3 1.4 2.4 3.9

 

What does it mean?

In resampling, the 10th percentile of predicted WP48 for Kanter was 0.14 - so roughly there is a 90% chance of producing a higher prediction, depending on randomness in the sample. That is a very good center, similar by Berri's metrics to players like Al Jefferson (WP48 of 0.127) and Elton Brand (0.146).  Especially given Kanter's dominance on the boards, I think he is likely to produce along the lines of these two players, and that is reasonable, if not exciting, production for a non-superstar pick.  I'm not sure if Kanter is a good fit for the wolves, especially because of rumors of his defensive inadequacy, but he seems like a good value generally. 

Of course, these data are old.  More recent observations of Kanter seem to have been mixed, which is why it is more important than usual to combine scouting data with stats in this case.  Finally, there is one cautionary tale in the U18 data set:

Player TS% PTS REB AST STL BLK PF TOV
Kanter 0.627 22.0 18.2 1.1 1.2 1.7 2.2 3.1
Koufos 0.541 27.4 13.4 0.4 1.4 3.6 1.7 1.9
X
Log In Sign Up

forgot?
Log In Sign Up

Forgot password?

We'll email you a reset link.

If you signed up using a 3rd party account like Facebook or Twitter, please login with it instead.

Forgot password?

Try another email?

Almost done,

Join Canis Hoopus

You must be a member of Canis Hoopus to participate.

We have our own Community Guidelines at Canis Hoopus. You should read them.

Join Canis Hoopus

You must be a member of Canis Hoopus to participate.

We have our own Community Guidelines at Canis Hoopus. You should read them.

Spinner

Authenticating

Great!

Choose an available username to complete sign up.

In order to provide our users with a better overall experience, we ask for more information from Facebook when using it to login so that we can learn more about our audience and provide you with the best possible experience. We do not store specific user data and the sharing of it is not required to login with Facebook.

tracking_pixel_9347_tracker