Drafting a Career in Sports: Determining Underclassmen College Players

You don't have the latest version of Adobe Flash Player.

Please update your flash player.

Please login or register to make a comment!

From the SelectedWorks of Byron J Gajewski June 2004 Drafting a Career in Sports: Determining Underclassmen College Players 9 Stock in the NBA Draft Contact Author Start Your Own SelectedWorks Notify Me of New Work Available at: http://works.bepress.com/byron_gajewski/10 12 VOL. 17, NO. 2, 2004 T he National Basketball Association (NBA) has been slowly shifting how it drafts amateur talent.

Many stu- dents are leaving college early in hopes of making the pros. Teams, once reluctant to choose these underclassmen, are now selecting them regularly due in part to the success of players like Chris Weber and Allen Iverson who were both underclassmen when drafted. Only four players selected in the Krst round of the 2001 NBA draft were seniors in col- lege, turning the draft into a cwho 9s who d of freshmen, sophomore, and junior college athletes.

The lure of professional basketball is not surprising. College athletics are full of pressures, yet offers little Knan- cial reward. Many underclassmen are willing to risk their amateur status for a chance at a NBA contract and poten- tial endorsement deals 4a quick route to cbling-bling, d oth- erwise known as fame and fortune.

Unfortunately, those not selected in the draft lose both the security ... more. less.

of a college pro- gram and the Knancial potential of a NBA deal. One such player, JaRon Rush, entered the 2000 NBA draft after play- ing only two seasons at UCLA. After going undrafted, he spent two unsuccessful years bouncing from semi pro leagues to the NBA developmental league. Had he stayed in college, JaRon would have been a senior in 2002 and a potential star on the nationally ranked UCLA team. On the other hand, staying in school may not always be the best choice. Terence Morris completed four years at the University of Maryland, even though many predicted him a top three draft pick after his sophomore season. Despite a great college career, Morris 9 lack of signiKcant improve- ments from sophomore to senior year resulted in his being chosen as a second round pick in the 2001 NBA draft. Compared to the potential of a guaranteed multi-year, mul- timillion dollar contract that might have come with entering the draft after his sophomore year, Morris 9ultimate pick did not even guarantee a roster spot. Or take the case of Jason Williams, former point guard for Duke. Williams decided to stay and play his senior year of college basketball, despite predictions that he would be the number one pick in the 2001 NBA draft. By staying in school, he opened himself up to potential risks before receiving an NBA paycheck. Many serious debates have ensued over which under- classmen should leave college and which should stay, but thus far, such debates have been based on opinion and con- jecture, resulting in a false perception of a player 9s cstock. d In an attempt to provide college players with a more reliable method on which to base the decision of when to enter the What's the probablity of an underclassman being drafted into the cbling-bling d of basketball? Drafting a Career in Sports: Determining Underclassmen College Players 9 Stock in the NBA Draft Todd Bishop and Byron J. Gajewski Kirk Hatson holds up his new Hornets jersey at a news confer- ence. The Hornets selected Hatson with the 16th pick in the 2001 NBA draft. (AP Photo/Chuck Burton) CHANCE 13 draft, we used principal components, logistic regression, and cross validation to predict a player 9s draft potential. Our model gives each player a numerical statistic, which we call a cbling d number, or draft likelihood. We also obtain a rating of a player 9s physical size, referred to as cbody, d and game skills, referred to as cgame. d Using these ratings we estimate players 9numerical stock for playing at the next level. The Data The dataset is limited to division one players, including all college seniors and any underclassmen that declared them- selves eligible for the draft. We excluded players from lower divisions or junior colleges, high school players or interna- tional players, as these groups are traditionally less preva- lent in drafts (although this trend is slowly changing). We used data from the 2000-2001 season leading into the 2001 draft. Our working data was extracted from the web site www.espn.com (October 2001), which included 1,044 players from 312 programs. Draft information came from www.nba.com (October 2001). Since a Big Ten player averaging certain statistics is more likely to get drafted than a Patriot League player with the same statistics, we identify and take into account each play- ers 9undergraduate college or university. Using Jeff Sagarin 9s NCAA basketball ratings system, we divided all division one schools into three conference tiers based on the strength of the conference. Although we chose not to include them, other possible criteria to consider are T.V. exposure, career statistics, post-season awards, and tournament appearances. Using season-long statistics, we paint an approximate picture of a player 9s ability to succeed at the next level. Our Krst criterion was a player 9s physical body, speciKcally the height and weight of a player. Size is of considerable impor- tance when being considered for the NBA. In fact, from 1990-2000 only 20% of the players drafted in the top ten were under 6 95 9 9( www.nbadraft.net , October 2001), and only one player under this height, the 6 91 d Allen Iverson, was the Krst overall pick. Some of the bigger players were drafted solely on their height potential in the hopes they would develop. But height doesn 9t necessarily guarantee game skills. The Houston Rockets selected Joel Przybilla (7 91 d) as the ninth overall pick, but later traded him to the Milwaukee Bucks. He averaged only 13 minutes a game during the Krst two-years of his career. Our primary criteria were game statistics for each player, speciKcally games played, minutes, points, rebounds, assists, steals, blocks per game, turnovers, assist-to-turnover ratio, Keld goal percentage, free throw percentage, and three-point percentage. We separated the undrafted players from the drafted players. For the statistics between the drafted and undrafted players, we ran Bonferoni adjust- ments and found signiKcant differences for all statistics except assist-to-turnover, free throw percentage, and three- point percentage. Principal Components and Logistic Regression While summary statistics and graphs are a useful way to understand the relationship between variables, they become complicated when there are many variables. Therefore, we transform the multivariate data into lower dimensions using principal components. Variables were standardized to place them on the same scale. We found that two sets of linear combinations explain 67% of the variance in the data, while seven linear combinations explain 90% of the variation in the data. Factor 1 loadings were high for variables associated with playing in a game (minutes, points, rebounds, etc). Factor 2 loadings were high for the variables height, weight, blocks and rebounds. Hereafter we refer to principal component 1 as cgame d and principal component 2 as cbody. d Each play- er received a score for each component. For instance, the scores for Terry Black from the University of Baylor show he has high cgame d (3.863) but a low body score (-0.493). Using our scores, we Knd that no 2001 drafted player had a game score below 0, although some drafted players had a negative body score. However, a high body score did not always result in being drafted. Logistic regression was used to predict the chance that a player was drafted. To obtain the Knal model we used step- wise regression, intuition, and Akaike Information Criterion (AIC). Using results of the logistic regression, we created our bling numbers. To avoid using a player 9s own statistics to predict his cdraftability, d we used a technique called cross validation. This procedure deletes each player from the data set and calculates the player 9s bling number based on the rest of the players in tier 1. We used the bling number to decide the predicted draft cutoff. Our goal was to achieve a fairly high success rate of predicting actual drafted players, while maintaining a low error rate for those who were not drafted yet made our list. We also did not want to far exceed the actual 58 NBA selections. We chose a cutoff bling of .20, which correctly categorized undrafted players at 90% and drafted players at 78%. Other cutoffs would lead to dif- ferent balances between these error rates. Results Table 1 shows the status of 64 players predicted to be draft- ed in the 2001 NBA draft. Players are listed in bling order along with players 9ratings of game and body, and the status of careers. Of these 64 players, 26 were playing in the NBA, 14 play overseas, 14 play pro ball in the United States in leagues such as the NBDL, CBA, USBL and ABA, 8 could not be located, and 2 returned to college (as of April 2002). There are several interesting players in Table 1. First is Eric Chenowith from the University of Kansas. His bling number was a fairly high .47, and he was drafted. However, follow-up revealed that he failed to make his pro team and was playing in the developmental league. As discussed ear- lier, many times teams will draft a player based on potential with hopes that they will develop. Chenowith, at 7 91 9 9 and 270 pounds, has one of the best NBA bodies among the 64 players 4.01) but has a mediocre game (1.47). Our model suggests that Eric Chenowith may have been a draftee based on body rather than proven NBA game. Another player of interest is Terry Black from Baylor University. Despite a high bling number (.56), ranking him one of the top 25 players, he did not get drafted. As our Kndings indicate, Black had the game (3.86), but at 6 95 9 9 14 VOL. 17, NO. 2, 2004 Bling # Body Game Name Status Bling # Body Game Name Status 0.99 2.49 4.35 Michael Brandley 1* 0.49 0.31 2.95 Greg Stevenson Europe 0.96 2.39 4.08 Troy Murphy 1* 0.48 -2.32 2.55 Charlie Bell NBDL/NBA 0.88 2.97 3.70 Eddie GrifDn 1* 0.47 4.01 1.47 Eric Chenowith 1/NBDL 0.87 -1.20 3.17 Jeryl Sasser 1* 0.47 1.36 2.59 Terrence Morris 1* 0.86 0.35 3.47 Sean Lampley 1/CBA 0.44 -0.21 2.96 Ricardo Greer USBL 0.83 4.42 3.36 Alvin Jones 1* 0.40 1.62 2.15 Andre Hutson 1* 0.82 1.87 3.01 Kirk Haston 1* 0.37 3.54 1.51 Ruben Boumtje-Bountje 1* 0.82 -4.10 4.93 Omar Cook 1/NBDL 0.36 -1.52 3.46 Demarcus Minor CBA 0.80 1.89 2.92 Brandon Wolfram Europe 0.36 -2.30 2.58 Jamison Brewer 1* 0.76 2.09 2.40 Jason Collins 1* 0.36 4.36 2.42 Brendon Haywood 1* 0.74 -2.01 3.58 Joseph Forte 1* 0.36 3.01 1.21 Peter Van Paassen Europe 0.74 2.05 2.94 Calvin Bowman USBL 0.35 1.70 1.87 Jarron Collins 1* 0.73 3.13 2.53 Kaspars Kambala Europe 0.35 2.70 0.94 Zach Randolph 1* 0.71 3.53 1.97 Karim Shabazz Europe 0.35 -0.53 1.85 Kenny Gregory NBDL 0.70 -4.22 4.70 Edwin cGreedy d Daniels CBA 0.34 0.58 2.33 Byran Bracey 1/Overseas 0.67 1.76 2.04 Troy Ostler Europe 0.33 0.67 1.96 Brent Wright Europe 0.66 2.33 2.48 Loren Woods 1* 0.32 1.46 2.36 Mekeli Wesley ? 0.65 0.02 2.92 Bobby Simmons 1* 0.32 -0.24 2.07 Jason Richardson 1* 0.60 1.16 2.85 Rodney White 1* 0.31 -0.04 2.70 Victor Thomas ABA 0.59 0.78 3.86 Shane Battier 1* 0.31 1.58 1.78 Greg LaPointe ? 0.57 -2.77 4.19 Jamaal Tinsley 1* 0.29 2.92 2.41 Rahim Lockhart NBDL 0.57 3.64 2.98 Melvin Ely In College 0.28 -2.31 2.93 Leandrew Bass ? 0.56 -0.49 3.86 Terry Black Europe 0.27 1.22 1.98 Will Perkins ? 0.56 -0.18 2.36 Joe Johnson 1* 0.23 -0.90 3.47 Courtney Wallace ? 0.55 0.83 3.55 Damone Brown 1* 0.22 4.43 2.53 Ken Johnson 1/CBA 0.55 -2.08 3.96 Titus Ivory Europe 0.22 3.21 -0.28 Adam Allenspach 0 0.52 2.58 2.52 Derrick Davenport Europe 0.21 1.78 1.71 Isiah Victor Europe 0.51 -1.71 4.22 Cookie Belcher Europe 0.21 1.62 -0.27 Lee Scruggs NBDL 0.51 1.26 2.95 Brian Scalabrine 1* 0.21 -2.69 3.29 Kenny SatterDeld 1* 0.50 2.62 2.67 Ryan Humphrey In College 0.21 0.98 1.67 Anthony Williams ? 0.49 3.59 2.44 Kimani Ffriend NBDL 0.20 0.42 2.03 Richard Jefferson 1* 0.49 1.34 3.15 Gyasi Cline-Heard Europe 0.20 2.28 1.61 Michael Wright 1/Europe Table 1. Status of the 64 players predicted to be drafted in 2001. A c1 d is if the player is drafted and a c* d indicates he is in the NBA as of 4/1/02. orable mentions for the 2001-2002 season, through our model (Table 3). Results show that the player on the All- American Team most likely to be drafted was Drew Gooden from the University of Kansas. In the opinion of many peo- ple close to basketball, Gooden made the best pro player prospect. He declared for the NBA in his junior year (April 2002), a decision our model supports. He had an extremely high bling number (0.99) and may not have beneKted from the associated risks of staying in school, such as a decline in stock or possible injury. Surprisingly, results from our model showed a very low and 200 lbs., he did not have the body (-0.49) for the NBA. As the NBA continues to grow bigger and stronger in all positions, many players like Black will be relegated to play in overseas pro leagues with hopes of one day making it back to the NBA. Table 2 displays players predicted to be undrafted, but who were eventually drafted (April 2002). Many of these players had low body scores, low game scores, or both, caus- ing them to have low bling numbers; however, they were still drafted. For example, at 6 91 d and 180 pounds, Earl Watson from UCLA does not have a large NBA frame, so he had a very low body score. What was not quantiKed for Watson is his demonstrated heart and leadership, which probably led a team to take a chance on him. Similarly, Jeff Trapagnier had a very low game score, a negative body score, and a bling number of .00, yet he was drafted in part due to his 40-inch vertical leap and great athleticism. These factors proved important in Trapagnier 9s case, but they were not included in our model, because they are measured in invit- ed NBA workouts after player draft declaration. Prediction Using the model from 2000-2001, we ran the Associated Press All-Americans, Krst through third teams and the hon- Bling # Body Game Name Status 0.15 -0.84 2.57 Maurice Jeffers 1/NBDL 0.12 -2.72 3.50 Earl Watson 1* 0.11 3.44 0.98 Steven Hunter 1* 0.08 0.89 1.29 Gerand Wallace 1* 0.08 3.66 1.03 Saumual Dalembert 1* 0.07 -2.19 3.08 Will Solomon 1* 0.06 2.69 1.25 Alton Ford 1* 0.05 -1.50 2.27 Gilbert Arenas 1* 0.00 -0.04 0.51 Jeff Trepagnier 1* Table 2. Listed are drafted players that have low bling numbers. Rank Bling # Name School Rank Bling # Name School 1 0.99 Drew Gooden* Kansas 24 0.48 Fredrick Jones Oregon 2 0.98 Luke Walton Arizona 25 0.47 Kareem Rush Missouri 3 0.98 Curtis Borchardt Standford 26 0.45 Predrag Savovic Hawaii 4 0.96 Melvin Ely Fresno St. 27 0.44 Erwin Dudley* Alabama 5 0.92 Darius Songaila Wake Forest 28 0.43 Anthony Grundy NC St. 6 0.80 Carlos Boozer* Duke 29 0.42 Lonny Baxter Maryland 7 0.78 Jason Williams* Duke 30 0.37 Tayshan Prince* Kentucky 8 0.77 Nick Collison Kansas 31 0.36 Preston Shumpert Syracuse 9 0.76 David West* Xavier 32 0.34 Dan Dickau* Gonzaga 10 0.76 Caron Butler Uconn 33 0.34 Lynn Greer Temple 11 0.75 Casey Jacobson* Standford 34 0.33 Frank Williams Illinios 12 0.73 Brandin Knight* Pitt 35 0.30 Dajuan Wagner Memphis 13 0.73 Udonis Haslem Florida 36 0.29 Josh Davis Wyoming 14 0.73 Steve Blake Maryland 37 0.23 Juan Dixon* Maryland 15 0.73 Dwayne Wade Marquette 38 0.22 Troy Bell B.C 16 0.71 Matt Bonner Florida 39 0.21 Rod Grizzard Alabama 17 0.68 T.J. Ford Texas 40 0.20 Luke Ridenour Oregon 18 0.68 Mike Dunleavy* Duke 41 0.16 Jason Kapono UCLA 19 0.65 Steve Logan* Cinn. 42 0.13 Luke Recker Iowa 20 0.64 Reggie Evans Iowa 43 0.09 Jarvis Hayes Georgia 21 0.62 Jared Jefferies* Indiana 44 0.07 Jason Garner* Arizona 22 0.62 Kirk Hinrich Kansas 45 0.02 Hollis Price Oklahoma 23 0.52 Sam Clancy* USC 46 0.01 John Linehan Providence Table 3. Bling numbers of the All-Americans for 2001-2002 for the 2002 draft. A c* d denotes a 3rst or second team selection. bling number (0.07) for Jason Garner from Arizona, despite his All-American status. It 9s important to remember that our model does not reLect the level of college play, but rather how a player will fare in the NBA, although players with lower bling numbers can get drafted and remain in the NBA. Interestingly, we look at the comparison of players on Maryland 9s 2002 National Championship team. Juan Dixon (bling .23), the MVP of the NCAA tournament, and Lonny Baxter (bling .42), a big strong inside force, were the 8super- stars 9on the team. However, in our model, Steve Blake, the point guard, Knished with the highest bling number (.73) because of his excellent passing skills. Now we turn to two Duke players, Jason Williams and Mike Dunleavy. Jason Williams 9 2002 bling number (.78) dropped from 2001 (.93). According to our model, his stock declined during an opportunity to mature and improve his overall skills before entering the challenges of NBA life. Many people would choose Mike Dunleavy as the best small forward in the nation for 2002. Dunleavy (bling .68) had a majority of his games televised and received publicity for being a great player at a highly successful basketball pro- gram. Sometimes this exposure can overshadow other great players, like Dwayne Wade, a small forward at Marquette University. Marquette received less national T.V. exposure than Duke players, and without this national media expo- sure, many basketball enthusiasts were simply unaware of his talents. Wade actually ended up with a higher 2002 bling number (.73), and is an example of how clow-expo- sure d players can be identiKed with our model. Finally, our model prediction for pre-season College Player of the Year for 2002-2003, Luke Walton of Arizona, was .98. He had unusual passing skills and is a tall player. Despite this, Walton returned for amateur play and delayed the NBA riches one more year. Conclusion No statistic can measure the mind, heart, or work ethic that are often important factors in the quality of a player. The numbers generated by our model do not necessarily reLect a player 9s college talent, only NBA potential. With each college season and NBA draft, we can collect a larger database and use it to update the model. As time goes on, more informa- tion will become available to allow better predictions. Our model and the techniques described here provide an under- classmen player one more tool for making a decision that can affect the rest of his life. At the very least, our mention of his cbling d number at the ofKce water cooler or college gym may inspire sports fanatics to study statistical techniques such as logistic regression and principal component analysis! Additional Reading Berry, S. 2001. cDo You Feel a Draft in Here, d Chance Magazine , Volume 14, (2), 53-57. Johnson, R.A., and Wichern D.W. (1998), Applied Multivariate Statistical Analysis , Prentice Hall, Upper Saddle River, N.J. Mendenhall, W. and Sinich, T. 1993. A Second Course in Statistics, Regression Analysis , Fifth Edition, Prentice Hall, Upper Saddle River, N.J. White, C., and Berry, S. 2002. cTiered Polychotomous Regression: Ranking NFL Quarterbacks, d The American Statistician , 56(1), 10-21. CHANCE 15

less