Euroleague Now Included in the Box Score Prediction System
Recently I explored the correlation between Euroleague and NBA stats. My next step has been to incorporate Euroleague stats into my Box Score Prediction System.
I used the same process with Euroleague as I did with the NCAA. I developed a formula that projects each NBA stat using Euroleague data and other variables such as height, weight, and experience. These formulas were developed using multiple linear regressions. The adjusted R^2 (http://en.wikipedia.org/wiki/Coefficient_of_determination) values for the different NBA stats are as follows:
FGA: 0.2971
FG%: 0.3004
3PA: 0.6834
3P%: 0.7729
FTA: 0.4111
FT%: 0.6887
REB: 0.8609
AST: 0.7202
STL: 0.5606
BLK: 0.7964
TO: 0.4538
PF: 0.5432
In the next few weeks, I will include projections of European prospects to go along with my projections of college players. I have slightly less confidence about the Euro projections because the data is more unreliable, but the projections will still be useful.
All Euroleague stats were obtained from www.keyhoops.com. NBA stats were obtained from www.basketballreference.com.
How Do Euroleague Statistics Translate to the NBA?
A few weeks ago I explored how well NCAA statistics correlate with NBA numbers. In other words, I wanted to see how much we could predict about a player’s professional career solely using his college performance. Later, using a more complex form of these ideas, I developed my Box Score Prediction System. However, not every NBA player comes from America. Some of the game’s greats are international players that have already gained a lot of experience playing in tough leagues around the world.
At the MIT Sloan Sports Conference in early March, Mike Zarren of the Boston Celtics talked about how one of the things teams haven’t figured out is translating European statistics to the NBA. If a foreign player dominates other foreign players, does that really matter?
Today is the first step in my process of answering that question. I have gathered data about current and former NBA players that previously played in the Euroleague, arguably the second toughest league in the world behind the NBA (although you can argue for NCAA Division I as well).
The Euroleague features teams from all over Europe and some from the Middle East. A lot of times these teams are champions of their respective countries, although this is not always the case. There is a regular season, and then a few rounds of playoffs. Many of the current international players in the NBA have experience in the Euroleague, making it a great league to examine. Ricky Rubio, one of the top prospects in this year’s draft (or next year’s if he doesn’t declare), has Euroleague experience.
For each player, I calculated their per minute box score stats in both NBA and Euroleague play. I ran simple regressions to find out the correlations between the NBA and Euroleague stats. I have expressed the R^2 values below:
(For an explanation of what R^2 is, go to: http://en.wikipedia.org/wiki/Coefficient_of_determination.)
FGA: 0.2207
FG%: 0.2659
3PA: 0.6913
3P%: 0.7173
FTA: 0.354
FT%: 0.633
REB: 0.7448
AST: 0.6949
STL: 0.4654
BLK: 0.5949
TO: 0.4228
PF: 0.2654
As you can see, NBA field goal attempts, field goal percentage, free throw attempts, and fouls are the least predictable based on a player’s Euroleague stats. Three point attempts, three point percentage, rebounds, assists, and blocks are the most predictable. Compared to college correlations, the Euroleague correlations are all slightly lower. However, they are comparable.
The next step I will take is to implement Euroleague stats into my Box Score Prediction System. This involves more complex multiple regressions, so the R^2 values will be higher. In other words, BSPS will do a better job of predicting a foreign player’s performance in the NBA than the simple correlations above. Look for that in the next few days.
Explanation of the Box Score Prediction System (BSPS)
One of the hardest things to project is a college player’s performance in the NBA. So many factors come into play, many of which can’t be measured. However, I have attempted to solve this problem. Using regression analysis, I have developed the Box Score Prediction System (BSPS).
The system works on the basis of a player’s statistical performance at the college level. It also takes into account a player’s height, weight, and NCAA experience.
With BSPS, I can input the various data about a player and it will shoot out the projected NBA numbers. Each NBA box score statistic is calculated based on some combination of the above variables and different coefficients. Certain NBA stats, such as rebounds, can be predicted using many different variables.
The adjusted R^2 values (go to http://en.wikipedia.org/wiki/Coefficient_of_determination for an explanation) for the NBA stats I project are as follows:
Points: 0.4557
Field Goal Attempts: 0.4666
Field Goal Percentage: 0.5879
Three Point Attempts: 0.6765
Three Point Percentage: 0.7972
Free Throw Attempts: 0.3874
Free Throw Percentage: 0.8052
Rebounds: 0.8927
Assists: 0.887
Steals: 0.5904
Blocks: 0.9314
Turnovers: 0.5639
Personal Fouls: 0.5629
As you can see, we can project with the most certainty blocks, rebounds, assists, free throw percentage, and threepoint percentage. Predictions regarding free throw attempts, points, field goal attempts, turnovers, and fouls are the most questionable.
At this time I’m not willing to give out the exact formulas I use, but in the future I will be predicting various college players’ future performances in the NBA using their college stats.
BSPS has the following limitations:
 Player statistics are not adjusted for strength of schedule. Just like with any other statistic, you have to keep context in mind when you look at the results. A guy who has lit up poor competition may project to be better than he actually will do.
 Most of the R^2 values are not extremely high, which just confirms common sense that a lot more goes into NBA success besides college success. Athleticism, game IQ, work ethic, etc. all have an effect.
 The study only includes NBA players that have “made it.” Players that fizzled out or were never good enough to get drafted in the first place aren’t included. Therefore, these numbers are more useful when projecting the guys who are likely to have an NBA future.
How Do NCAA Statistics Translate to the NBA?
As March Madness begins and the NBA Draft approaches, I often wonder how close the college game is to the professional one. It’s clear who the stars in the college game are. But are they just “built” for that style of play, or are they true stars who excel at any level (including the NBA)?
I have attempted to solve this problem by seeing how college stats correlate to NBA stats. To do this, I first took a large sample size of current NBA players’ career statistics and compared it with those sample players’ college stats. Everything was calculated on a perminute basis. Once I had the stats, I ran a series of simple regressions to see how well the NBA numbers correlated with the college ones.
Below I have posted the R^2 values of the different correlations. R^2 basically says how well future outcomes are likely to be predicted by the model and can be thought of as a percentage. For example, if the R^2 of the correlation between college points per game and NBA points per game is 0.3405, then we can say that about 34.05% of NBA players’ PPG can be explained by their college PPG. The higher the R^2, the better.
Below are the R^2’s for the different correlations:
Points per minute: 0.3405
Field goal attempts per minute: 0.3522
Field goal percentage: 0.3436
Threepoint attempts per minute: 0.6391
Threepoint percentage: 0.7941
Free throw attempts per minute: 0.286
Free throw percentage: 0.7615
Rebounds per minute: 0.8312
Assists per minute: 0.8823
Steals per minute: 0.5981
Blocks per minute: 0.9327
Turnovers per minute: 0.4535
Personal fouls per minute: 0.4447
Those numbers are all higher than I expected before I began the study. Specifically, we can predict with pretty good certainty an NBA player’s blocks, assists, rebounds, threepoint percentage, and free throw percentage based on their equivalent college statistics. Free throw attempts, points per game, field goal percentage, and field goal attempts are the weakest.
This all comes with one big caveat. The sample only includes guys that have made it in the NBA. The college stars that fizzled out at the pro level or the guys who NBA teams knew had no chance at the highest level before the draft were not included in this study. In other words, just because a guy is great in college doesn’t mean he will be great in the NBA. However, if he does make the NBA, we can somewhat predict how he’ll end up doing based on his college stats.
This is just the beginning of my study, though. I have developed a model for predicting a player’s NBA stats using multiple variables at a time. As it turns out, even things like NBA assists can be predicted using more than just college assist numbers. In the next few days I will be revealing my system and using it to project some of the stars you’ll be watching in March.

Recent
 How the Shooting Abilities of Shooting Guards, Power Forwards, and Centers Affect Offenses
 At Which Positions Are Bad Players the Most Detrimental?
 At Which Positions Are Great Players the Most Important?
 How the Shooting Abilities of Small Forwards Affect Offenses
 How the Shooting Abilities of Point Guards Affect Offenses
 Developing a Measure for Evaluating Coaches, Part I: Impact on Effort
 Player Projection: B.J. Mullens
 Player Projection: Austin Daye
 Finding the Achilles Heel of Each Remaining Playoff Team
 Euroleague Now Included in the Box Score Prediction System
 How Do Euroleague Statistics Translate to the NBA?
 Explanation of the Box Score Prediction System (BSPS)

Links

Archives
 July 2009 (5)
 June 2009 (2)
 May 2009 (2)
 April 2009 (2)
 March 2009 (3)
 February 2009 (1)
 October 2008 (3)

Categories

RSS
Entries RSS
Comments RSS